Powertrain Modeling and Range Estimation of The Electric Bus

Donkyu Baek* [donkyu at cbnu.ac.kr] School of Electronics Engineering Chungbuk National University Presenter

Dr. Donkyu Baek

Assistant Professor in Chungbuk National University, Korea Got Ph.D. in KAIST (Korea Advanced Institute of Science and Technology), Korea Post doc experiences Mar. 2018 ~ Dec. 2018: KAIST, Korea Jan. 2019 ~ Jan. 2020: Politecnico di Torino, Italy Research interest EV/drone powertrain modeling EV/drone runtime optimization EV/drone design-time optimization

2 EVs Are Invading the ICEV Market

Higher efficiency Average 141 kWh for ICEVs and 20 kWh for EVs when drive 100 km Environmental friendliness + government promotion Quiet and comfort ride

Electric vehicles Gasoline vehicles 400400 Veyron

300300 www.fueleconomy.gov Escalade GL63 AMG Mustang M5 200200 Corvette

(kWh/100 mi) (kWh/100 100 100 Prius (hybrid) Energy consumption consumption Energy Prius (electric) Volt Twizy Leaf Model S 0 1625 3250 4875 6500 0 1625 3250 4875 6500 Curb weights (lbs) 3 Limitation of EVs

EVs still cannot reach the territory of ICEVs Short driving range 27 kWh battery pack (Soul EV) versus 514 kWh fuel tank (Soul) Long charging time 5 hours with level 2 and 30 minutes with level 3 battery charging systems

Driving ranges from Los Angeles

Tesla Model S (350 km)

Toyota RAV4 (166 km)

tia.ucsb.edu

4 3

Table 3

Battery Motor power Gear ratio Energy Battery size Energy Energy Driving Driving Note 220 Issue Target driving profile pack (kW) (times) consumption (kJ) (kWh) consumption (kWh) consumption (%) Drivetraincycle range (%) - Velocity/acceleration Optimal 700 10 1 2479 190 17.64 0.688 100%components26 library100.0% over time - Road slope by distance ×0.5 battery 300 10 1 2256 7.56 0.627 91% 12 47.1% size 160

500 10 1 2367Elevation (ft) 12.6 0.657 95% 19 74.8% Vehicle performance 130 - Maximum velocity/ 600 10 1 2424 0.015.12 0.9 1.80.673 2.8 98% 22 87.6% Energy-Efficient EV Vehicle simulator acceleration 1000 10 1 2618 30 25.2 0.727 106%Electric 35vehicle135.3% Energy consumption ×2 battery 1400 10 1 277322.5 35.28 0.770 112%new candidate46 178.8% - Battery SoC size generation 15 Energy consumption is reduced by 50% ×with3 battery optimal2100 driving10 and1 by 2909 52.92 0.808 117%- Motor size65 255.6% size - Battery capacity 20% with optimal vehicle design 7.5 Pareto set update

×0.5 motor 700 5 1 2429Velocity (mph) 17.64 0.675 98% 26 102.1% size with 2 0 gear ratio Start/end 0.0 0.9 1.8 2.8 ×0.8 motor 700 8 Mail1 drop 2372 17.64 0.659 96% 27 104.5% No End of size Distance (mile) Genetic algorithm Then, how to find the optimal vehicle driving and design? generation? ×2 motor 700 (a) 20 1 2564 17.64 (b) 0.712 103% 25 96.7% size Yes ×3 motor 700 30 1 2617 17.64 0.727 106% 24 94.7% sizeFig. 2. (a) A campus mail delivery route and (b) road elevation change and Solution set 0 0.000 0% driving cycle. - Vehicle performance 0 0.000 0% - Driving range - MSRP Energy consumption (%) Driving range (%) Iterative search loop 120% 300% 110 110% Fig. 4. Existing electric vehicle design-time optimization dedicated to a 100 200% driving mission. 100% 90 100% 90% 80 Build Your Own Electric Vehicle Battery energy (Wh) Irvine, CA 92602 70 80% 0% Home 10 Work 900 University Ave, Riverside, CA 92521 0.25 5 0.2 Primary purpose Work Leisure Optimal Optimal 0.15 Velocity (m/s) To work 7:30 AM From work 5:50 PM 0 0.1 size 2 motor size 3 motor size 2 motor size 3 motor

2 size 2 battery size 3 battery size 2 battery size 3 battery × × × ×

Acceleration (m/s ) × × × × 0.5 battery size 0.5 battery size 0.5 battery Driving range 3 driving cycles × × 5 Reset Submit

Fig. 3. Example electric vehicle setups and energy consumption comparison. (a) Build Your Own Electric Vehicle Motor power: 150 kW should be recharged during mail delivery, and it also makes Battery: 60 kWh battery depreciation faster due to a more charging cycles. 0-60 mph: 7.3 seconds Top speed: 90 mph Such design-time optimization shows lots of opportunities for Range: 3 commutes per charging application-specific electric vehicle optimization. MSRP: $39,000 Performance Performance Now the question is how to find the optimal setup in Back downgrade upgrade a more systematic way. Previous work finds the minimum- energy electric vehicle specifications from a set of batteries (b) and motors for a given route profile using Genetic algorithm. This is a multi-variable optimization problem that evaluates Fig. 5. An example of graphical user interface of the rapid energy-aware electric vehicle synthesis. the total energy consumption for the given route. The Genetic algorithm generates a large number of powertrain specification setups [11]. We illustrate the existing framework to find the features as well as the types of the engine via a web interface. optimal electric vehicle configuration in Fig. 4. This method However, they can just confirm the MSRP (mean suggested derives the specification of an electric vehicle that results in retail price) after they build their vehicle. The same is true for the minimum energy consumption for a given route while computer purchase. Customers can only compare the price, satisfying the vehicle performance such as the driving range but non-tech-savvy people do not have a concrete idea what as well as the maximum acceleration and velocity. However, the right configuration is for their purpose. this method should run a vehicle simulator in the inner-loop A rapid energy-aware electric vehicle synthesis helps cus- for each iterative search. This is because1 the traction force, tomers to find a right electric vehicle configuration for their curb weight, power consumption and driving range are all own purpose. The synthesis time should be shorter than a dependent among each other. second not to challenge the user’s patience as well as not to cause misperception of service interruption. Unfortunately, the V. A RAPID SYNTHESIS OF ENERGY-EFFICIENT ELECTRIC existing method cannot afford instant evaluations, which are VEHICLES commonly required in Web environments. Nowadays, customers can build a vehicle by their preference Fig. 5 shows a new electric vehicle selection function for such as the color, exterior, interior, entertainment, safety non-tech-savvy customers. Customers are supposed to pick an System-Level Approach for EV Optimization

Low-power electric vehicle system-level approach Optimal driving profile

EV driving optimization

Road/traffic info. Vehicle setup EV characterization Driving constraints Power consumption model Energy harvesting/ EV power consumption simulator charging model

Battery model EV design optimization

Road/traffic info. Driving profile Component library (cost, life, etc.)

Optimal vehicle

6 design candidates IT Technology Inspired EV Design Framework

Electric vehicle power consumption simulation

Vehicle input parameters

Battery

Motor characteristics

Mechanical specifications

Transportation input parameters EV power Battery Passenger information consumption SoC simulator Driving input parameters Driving route

Speed and acceleration

7 IT Technology Inspired EV Design Framework

ADvanced VehIcle SimulatOR (ADVISOR) works well to estimate 1) vehicle powertrain and 2) battery SoC

Vehicle input parameters

Battery

Motor characteristics

Mechanical specifications

Transportation input parameters EV power Battery Passenger information consumption SoC simulator Driving input parameters Driving route

Speed and acceleration

8 IT Technology Inspired EV Design Framework

ADvanced VehIcle SimulatOR (ADVISOR)

Vehicle type Input parameters

Powertrain operation range and efficiency map

9 EV Power Consumption Characterization

ADVISOR is a fast simulator BUT not proper to run in the inner loop of the optimization algorithms

Vehicle input parameters

Battery Not acceptable for online loop-based Motor characteristics optimization methods Mechanical specifications

Transportation input parameters EV power Battery Passenger information consumption SoC simulator Driving input parameters Driving route

Speed and acceleration

10 EV Power Consumption Characterization

Need to implement very simple but accurate power model using ADVISOR

Extract input parameters of powertrain Driving simulation 100

Powertrain spec. 50 4 (km/h) Speed 0 0 44 88 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 ADVISOR simulation result Model coefficient extraction result 100 Time (s) 30 50 0 (kW) Power 0 -30 0 Power (kW) 44 88 -50 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 0 1000 2000 3000 4000 5000 6000 Time (s) Time (s) Fig. 1: Validation of the model coefficient extraction.

TABLE I: Model coefficients for Chevrolet Bolt. Logging Summary GM Bolt (use accumulated power) a 0.06 b 9.5549 g 1.0013 25 mile driving 65 mile driving 75 mile driving 93 mile driving Constant driving 25 65 75 93 d 0.00012 C0 1000 C1 10.588 (mile) Time ess_pwr_r Energy ess_pwr_r Energy ess_pwr_ Energy ess_pwr_r Energy Sum (25 mile) 1325930.16 C2 8.11 C3 0.00032 e 0.6633 (s) (W) (Ws) (W) (Ws) r (Ws) (W) (Ws) Constant driving (km/ 40 105 121 150 z 5813.6 (W) Energy total (kWh) 0.37 h)

0 3678.94 3678.94 15706.27 15706.27 21262.37 21262.37 34795.05 34795.05 Used (%) 0.64% Driving distance (km) 4.02 10.46 12.1 14.5 20 1 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 SoC (%) 99.36% 97.26% 96.28% 93.91% 50 -6 40 Electric-4 vehicle model-0.3 Equation form 2 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (65 mile) 5669894.31 Range (km) 626.97 381.22 325.50 238.13 0 30 -2 16 3 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 1.57 Range (mile) 0.4 20 in0 degree vehicle simulator EV power model 4 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 2.74% CD = 0.200 % 0.67% -5.17% 6.45% -7.52% 12 Driving data 2 1.0 degree 5 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Reference range (km) 622.816 402 305.775 257.5 10 4 8 6 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (75 mile) 7681551.19 6 1.7 2 2 7 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 2.13 (a) Range-speed experimental data. 5 4 4.3 Ptot = Qv + C0 + C1v + C2v + C3Q Used (%) 3.72%

8 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fuel economy (km/liter) 700 11Energy efficiency (km/Wh) 9 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Simulation results 10 30 50 70 90 110 0 5 10 15 20 25 Real vehicle experiment results 10 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (91 mile) 12582801.56 Driving velocity (km/h) Driving velocity (km/h) Energy total (kWh) 3.50 525 11 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 () Ford Escort. () Custom EV. 12 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 6.09% Fig. 3: (a) Fuel economy of a Ford Escort [37] and (b) energy 13 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 350 14 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 consumption of a custom EV [2] on various road slopes and Range (km) Range 15 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 vehicle velocities. 175 16 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

17 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Vehicle specification 18 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Variable 0 work is done by energy (fuel) consumption, we compare 19 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Battery energy 57.4 40 105 121 150 energy consumption instead of power consumption for easy (kWh) 20 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Speed (km/h) comparison. 21 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 (b) Range comparison with simulation results. Fig. 3(a) shows the fuel economy (range with a liter of fuel 22 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 energy) versus ICEV velocity on various road slopes [37]. Red 23 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fig. 2: (a) Range-speed experimental data for Chevrolet stars denote the minimum-energy velocity by the average road 24 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Bolt [36] and (b) comparison with simulation results of the slope. The minimum-energy velocity is reduced as the road 25 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 proposed EV power model. slope increases. For example, the vehicle has the highest fuel 26 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 economy at 90 km/h on a 6 degree road slope and 40 km/h on 27 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 a 2 degree road slope. Fuel economy on a downhill (negative 28 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 derived power models are close enough to the real vehicle road slope) is not meaningful due to fuel cut of modern ICEVs. 29 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 measurement data [32], [33], [34], [35], [36]. We refine the Compared with the flat road (zero degree), we see the fuel 30 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 coefficients until all the available data values match well. economy drops if the road slope is steeper than 2 degree due 31 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 As a result, the derived EV power model well matches with to downshift of the transmission. 32 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the range-speed experimental data as shown in Fig. 2(b) Fig. 3 demonstrates a very interesting EV-specific energy 33 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 as well as the top speed and maximum acceleration setting behavior such that the least-energy velocity increases as the 34 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the aerodynamic resistance, rolling resistance, curb weight, road slope increases. Such energy behavior is completely 35 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 etc. with the known values. The errors in the range-speed opposite to that of ICEV. The efficiency maps of an engine 36 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 simulation is merely 1.4%, the time for 0-to-100 km/h error and an electric motor (Fig. 4) show that the motor exhibits the 37 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 is 7.8%, and the top speed error is 3.0%, respectively. 38 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 maximum torque over the wide range of RPM while the engine

39 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 exhibits the maximum fuel efficiency only at near 2,000 RPM. B. Energy consumption comparison between ICEV and EV 40 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Therefore, an engine should be incorporated with a transmis- 41 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 We compare power consumption behavior of EV and ICEV sion that makes it possible to run the engine at close to the 42 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 in this section. As most previous fuel economy analysis most efficient region, independent to the vehicle velocity. The 43 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

44 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

45 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

46 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

47 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

48 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

49 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

50 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

51 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

52 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

53 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

54 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

55 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

56 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

57 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

58 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

59 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

60 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

61 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

62 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

63 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

64 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

65 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

66 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

67 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

68 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

69 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

14 EV Power Consumption Characterization

Need to implement very simple but accurate power model using ADVISOR

Driving simulation 100

Powertrain spec. 50 4 (km/h) Speed 0 0 44 88 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 ADVISOR simulation result Model coefficient extraction result 100 Complete energy Time (s) 30 50 simulation 0 (kW) Power 0 -30 0 Power (kW) 44 88 -50 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 0 1000 2000 3000 4000 5000 6000 Time (s) Time (s) Fig. 1: Validation of the model coefficient extraction.

TABLE I: Model coefficients for Chevrolet Bolt. Logging Summary GM Bolt (use accumulated power) a 0.06 b 9.5549 g 1.0013 25 mile driving 65 mile driving 75 mile driving 93 mile driving Constant driving 25 65 75 93 d 0.00012 C0 1000 C1 10.588 (mile) Time ess_pwr_r Energy ess_pwr_r Energy ess_pwr_ Energy ess_pwr_r Energy Sum (25 mile) 1325930.16 C2 8.11 C3 0.00032 e 0.6633 (s) (W) (Ws) (W) (Ws) r (Ws) (W) (Ws) Constant driving (km/ 40 105 121 150 z 5813.6 (W) Energy total (kWh) 0.37 h)

0 3678.94 3678.94 15706.27 15706.27 21262.37 21262.37 34795.05 34795.05 Used (%) 0.64% Driving distance (km) 4.02 10.46 12.1 14.5 20 1 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 SoC (%) 99.36% 97.26% 96.28% 93.91% 50 -6 40 Electric-4 vehicle model-0.3 Equation form 2 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (65 mile) 5669894.31 Range (km) 626.97 381.22 325.50 238.13 0 30 -2 16 3 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 1.57 Range (mile) 0.4 20 in0 degree vehicle simulator EV power model 4 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 2.74% CD = 0.200 % 0.67% -5.17% 6.45% -7.52% 12 Driving data 2 1.0 degree 5 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Reference range (km) 622.816 402 305.775 257.5 10 4 8 6 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (75 mile) 7681551.19 6 1.7 2 2 7 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 2.13 (a) Range-speed experimental data. 5 4 4.3 Ptot = Qv + C0 + C1v + C2v + C3Q Used (%) 3.72%

8 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fuel economy (km/liter) 700 12Energy efficiency (km/Wh) 9 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Simulation results 10 30 50 70 90 110 0 5 10 15 20 25 Real vehicle experiment results 10 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (91 mile) 12582801.56 Driving velocity (km/h) Driving velocity (km/h) Energy total (kWh) 3.50 525 11 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 () Ford Escort. () Custom EV. 12 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 6.09% Fig. 3: (a) Fuel economy of a Ford Escort [37] and (b) energy 13 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 350 14 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 consumption of a custom EV [2] on various road slopes and Range (km) Range 15 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 vehicle velocities. 175 16 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

17 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Vehicle specification 18 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Variable 0 work is done by energy (fuel) consumption, we compare 19 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Battery energy 57.4 40 105 121 150 energy consumption instead of power consumption for easy (kWh) 20 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Speed (km/h) comparison. 21 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 (b) Range comparison with simulation results. Fig. 3(a) shows the fuel economy (range with a liter of fuel 22 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 energy) versus ICEV velocity on various road slopes [37]. Red 23 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fig. 2: (a) Range-speed experimental data for Chevrolet stars denote the minimum-energy velocity by the average road 24 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Bolt [36] and (b) comparison with simulation results of the slope. The minimum-energy velocity is reduced as the road 25 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 proposed EV power model. slope increases. For example, the vehicle has the highest fuel 26 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 economy at 90 km/h on a 6 degree road slope and 40 km/h on 27 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 a 2 degree road slope. Fuel economy on a downhill (negative 28 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 derived power models are close enough to the real vehicle road slope) is not meaningful due to fuel cut of modern ICEVs. 29 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 measurement data [32], [33], [34], [35], [36]. We refine the Compared with the flat road (zero degree), we see the fuel 30 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 coefficients until all the available data values match well. economy drops if the road slope is steeper than 2 degree due 31 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 As a result, the derived EV power model well matches with to downshift of the transmission. 32 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the range-speed experimental data as shown in Fig. 2(b) Fig. 3 demonstrates a very interesting EV-specific energy 33 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 as well as the top speed and maximum acceleration setting behavior such that the least-energy velocity increases as the 34 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the aerodynamic resistance, rolling resistance, curb weight, road slope increases. Such energy behavior is completely 35 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 etc. with the known values. The errors in the range-speed opposite to that of ICEV. The efficiency maps of an engine 36 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 simulation is merely 1.4%, the time for 0-to-100 km/h error and an electric motor (Fig. 4) show that the motor exhibits the 37 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 is 7.8%, and the top speed error is 3.0%, respectively. 38 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 maximum torque over the wide range of RPM while the engine

39 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 exhibits the maximum fuel efficiency only at near 2,000 RPM. B. Energy consumption comparison between ICEV and EV 40 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Therefore, an engine should be incorporated with a transmis- 41 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 We compare power consumption behavior of EV and ICEV sion that makes it possible to run the engine at close to the 42 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 in this section. As most previous fuel economy analysis most efficient region, independent to the vehicle velocity. The 43 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

44 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

45 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

46 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

47 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

48 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

49 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

50 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

51 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

52 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

53 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

54 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

55 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

56 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

57 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

58 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

59 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

60 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

61 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

62 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

63 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

64 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

65 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

66 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

67 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

68 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

69 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

14 Trial 1 Trial 2 Trial 3

Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (mile/hour) 4.35 17.4 31.7 39.8

Speed (km/h) 7.0 28.0 51.0 64.1 Const. Power (W) 6773.68 23231.45 47509.01 66711.57 Speed (km/h) 7.0 28.0 51.0 64.1 Const. Power (W) 8151.28 27852.43 55299.21 77990.54 VehicleDistance/energy (mile/Wh) Simulator0.00064 0.00075 0.00067 Model0.00060 Const. Power (W) 4963.89 16804.69 36566.19 52112.01 Distance/energy (mile/Wh) 0 0 0 0 Range (sim) (mile) 208 243 216 193 Distance/energy (mile/Wh) 0 0 0 0

Range (sim) (mile) 173 202 186 165 Range (exp.) (mile) 403 434 387 348 Range (sim) (mile) 284 335 281 247 Range (exp.) (mile) 403 434 387 348 DiBYDfference (%) K9 bus of48% 18 tons44% 44%curb 44%weight Range (exp.) (mile) 403 434 387 348 Difference (%) 57% 53% 52% 53% Difference (%) 29% 23% 28% 29% Range (sim) (km) 278 326 299 266 We implemented a new sub-model ofRange each (sim) (km)vehicle component457 540 452 398 Vehicle specification-1 Range (exp.) (km) 648 698 624 560 Range (exp.) (km) 648 698 624 560 BatteryEstimate energy (kWh) motor324 efficiency based on driving experimental results wh_1st_rr 0.015

Vehicle specification Maximum current (A) 250 Vehicle specification-1-1

Battery energy (kWh) 324 Total weight (kg) 14465 Battery energy (kWh) 324 wh_1st_rr 0.015 wh_1st_rr ADVISOR0.010 300 250 Maximum current (A) Range (sim) (mile) Range (exp.) (mile) Maximum current (A) Total weight (kg) 18000 We do not consider the acceleration/deceleration occurring Total weight (kg) 14465 in correspondence to speed500 changes. Such approximation is not critical however, because the acceleration energy is small Range (sim) (mile) Range (exp.) (mile) Range (sim) (km) Range (exp.) (km) 500 compared with the total energy375 consumption of each segment: acceleration/deceleration seldom last more than a few seconds. 700

375 250 IV. SIMULATION RESULTS 525 Range (mile) 250 We implemented a powertrain model of a BYD K9 bus 125 from the vehicle specification and experiment results [4], [22], 350 Range (mile) [23]. The gross vehicle weight of BYD K9 is 18,000 kg, and 125 0 Range (mile) the facial area of K9 is 2.55 by0 3.365 m.10 K915 includes20 25 two elec-30 35 40 175 tric motors that the maximum motor torque andSpeed power (mph) are 350 MC PM 100kW motor in Advisor library 0 0 5 10 15 20 Nm25 and30 9035 kW.40 The motor type is in-wheel BYD-TYC90A mc_map_spd [RPM] 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 17000 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 4000 4100 4200 4300 4400 Speed (mph)Brushless Permanent Magnet Synchronous Motor. Maximum 0.0 8.8 17.5 26.3 35.0 43.8 52.5 61.3 70.0 mc_map_spd [rad/s] 0 10 21 31 42 52 63 73 84 94 105 115 126 136 147 157 168 178 188 199 209 220 230 241 251 262 272 283 293 304 314 325 335 346 356 367 377 387 398 408 419 429 440 450 461 RPM is 7,500, and gear box ratio is 17.7:1. Manufacturers Speed (mph) mc_max_trqunveiled [Nm] driving551 551 range551 551 of551 K9551 as551 average551 551 250551 km551 based551 550 on549 550 551 13543 529 Figure512 497 6. ADVISOR482 458 simulation439 421 404 setup.385 364 344 326 309 295 286 276 265 254 243 236 231 225 220 212 203 201 194 190 Powertheir [kW] experience.0 Battery6 12 17 size23 is29 32035 kWh,40 46 and52 the58 maximum63 69 75 81 86 91 94 97 99 101 101 101 101 102 101 99 97 96 94 93 93 92 92 91 89 89 89 90 90 89 87 89 87 88 road slope to climb is 15%. 1000 Simulation results MC PM 180kW BYD K9Experimental motor results A. Vehicle Parameter Extraction mc_map_spd [RPM] 0 170 341 511 682 852 1023 1193 1364 1534 1705 1875 2045 2216 2386 2557 2727 7502898 3068 3239 3409 3580 3750 3920 4091 4261 4432 4602 4773 4943 5114 5284 5455 5625 5795 5966 6136 6307 6477 6648 6818 6989 7159 7330 7500 mc_map_spdWe use[rad/s] [24]0 to18 set36 parameters54 71 89 of107 the125 electric143 161 motor178 196 that214 the232 250 268 286 303 321 339 357 375 393 411 428 446 464 482 500 518 535 553 571 589 607 625 643 660 678 696 714 732 750 768 785 mc_max_trqinstitute [Nm] tested701 BYD701 electric701 701 701 bus701 more701 than700 699 eight700 months701 691 (from674 652 633 613 582 500559 536 514 490 463 438 415 393 375 364 351 337 324 310 301 294 287 280 270 259 256 247 242 236 230 225 219 214 PowerAugust [kW] 29, 20130 to13 May25 13,38 50 2014).63 75 They87 reported100 112 125 acceleration136 144 151 158 164 166 170 172 174 175 174 172 170 168 167 169 169 169 168 166 166 168 169 170 168 166 169 168 168 168 168 168 168 168 test including top speed (43 mph), zero to 10 mph, 20 mph, Range (km) 250 30 mph, 40 mph and top speed. We extract (a) the maximum motor torque map and (b) maximum motor power by RPM as shown in Figure 5 with repeated ADVISOR simulations. 0 mc_max_trq [Nm] Power [kW] 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0

800 180 Speed (km/h) Figure 7. Driving range validation under efficiency parameter setting. 600 135

400 90 between the estimation of power consumption by the vehicle simulator and the powertrain models; the normalized root- mean-square error is 9.12%.

200 Motor power [kW] 45 Motor torque [Nm] Motor torque

0 0 Table I. MODEL COEFFICIENTS FOR BYD K9. 0 1875 3750 5625 7500 0 1875 3750 5625 7500 RPM RPM ↵ 0.098 9.7562 1.2016 0.0001 C0 1000.0 C1 1378.2 C2 0.00001 C3 0.000015 (a) Maximum motor torque map. (b) Maximum motor power. ✏ 0.4095 ⇣ 2178.5 Figure 5. Motor parameter extraction results. (a) is the maximum motor torque map and (b) is maximum power map. C. Vehicle Simulation Setup In our experiment, we followed the battery pack configura- We specified the detailed vehicle parameters as shown in tion provided by BYD to build our battery model. The battery Figure 6, which shows ADVISOR user interface for parameter is LiFePO4 (Lithium Iron Phosphate) with 540 V battery pack setting. The motor, efficiency and battery models are imported voltage. The battery pack consists of three battery modules, into ADVISOR, and based on the simulation results, we which has 108 kWh capacity. We assumed that each battery set the parameters so that the simulation results follow the cell in the pack is ideally balanced in the following experi- experimental results. Figure 7 shows the difference between ments, then built battery pack model as section III-A indicated. experimental results and simulation results of driving range. Concerning the regenerative braking phase, we assume that We set the parameters for drivatrain efficiency to follow the charging efficiency is 20%, which means that 20% of the trend of driving range by vehicle speed. There are about 200 kinetic energy is converted to electric energy and transferred km range difference between two lines. However, the range into the pack. trend by the speed is similar enough. Also, the range difference is resolved by updating battery model in the following section. D. Case study 1: Fast Driving Energy Estimation We extract a route going to Technische Universitat B. Vehicle powertrain modeling Munchen (TUM) and another one returning from TUM as Then, we extracted the coefficients of (1), (2) and (3) with shown in Figure 9. Table II summarize information of the a number of ADVISOR simulations. Table I summarizes the routes: distance, average slope along the route and average model coefficients of BYD K9. Figure 8 shows the difference vehicle speed. Trial 1 Trial 2 Trial 3

Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (mile/hour) 4.35 17.4 31.7 39.8

Speed (km/h) 7.0 28.0 51.0 64.1 Const. Power (W) 6773.68 23231.45 47509.01 66711.57 Speed (km/h) 7.0 28.0 51.0 64.1 Const. Power (W) 8151.28 27852.43 55299.21 77990.54 Distance/energy (mile/Wh) 0.00064 0.00075 0.00067 0.00060 Const. Power (W) 4963.89 16804.69 36566.19 52112.01

Distance/energy (mile/Wh) 0 0 0 0 Range (sim) (mile) 208 243 216 193 Distance/energy (mile/Wh) 0 0 0 0

Range (sim) (mile) 173 202 186 165 Range (exp.) (mile) 403 434 387 348 Range (sim) (mile) 284 335 281 247

Range (exp.) (mile) 403 434 387 348 Difference (%) 48% 44% 44% 44% Range (exp.) (mile) 403 434 387 348

Difference (%) 57% 53% 52% 53% Difference (%) 29% 23% 28% 29%

Range (sim) (km) 278 326 299 266 Range (sim) (km) 457 540 452 398 Vehicle specification-1 Range (exp.) (km) 648 698 624 560 Range (exp.) (km) 648 698 624 560 Battery energy (kWh) 324

wh_1st_rr 0.015

Vehicle specification Maximum current (A) 250 Vehicle specification-1-1

Battery energy (kWh) 324 Total weight (kg) 14465 Battery energy (kWh) 324

wh_1st_rr 0.015 wh_1st_rr 0.010 300 250 Maximum current (A) Range (sim) (mile) Range (exp.) (mile) Maximum current (A) Total weight (kg) 18000 We do not consider the acceleration/deceleration occurring Total weight (kg) 14465 in correspondence to speed500 changes. Such approximation is not critical however, because the acceleration energy is small Range (sim) (mile) Range (exp.) (mile) Range (sim) (km) Range (exp.) (km) 500 compared with the total energy375 consumption of each segment: acceleration/deceleration seldom last more than a few seconds. 700

375 250 IV. SIMULATION RESULTS 525 Range (mile) 250 We implemented a powertrain model of a BYD K9 bus 125 from the vehicle specification and experiment results [4], [22], 350 Range (mile) [23]. The gross vehicle weight of BYD K9 is 18,000 kg, and 125 0 Range (mile) the facial area of K9 is 2.55 by0 3.365 m.10 K915 includes20 25 two elec-30 35 40 175 tric motors that the maximum motor torque andSpeed power (mph) are 350 MC PM 100kW motor in Advisor library 0 0 5 10 15 20 Nm25 and30 9035 kW.40 The motor type is in-wheel BYD-TYC90A mc_map_spd [RPM] 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 17000 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 4000 4100 4200 4300 4400 Speed (mph)Brushless Permanent Magnet Synchronous Motor. Maximum 0.0 8.8 17.5 26.3 35.0 43.8 52.5 61.3 70.0 mc_map_spd [rad/s] 0 10 21 31 42 52 63 73 84 94 105 115 126 136 147 157 168 178 188 199 209 220 230 241 251 262 272 283 293 304 314 325 335 346 356 367 377 387 398 408 419 429 440 450 461 RPM is 7,500, and gear box ratio is 17.7:1. Manufacturers Speed (mph) mc_max_trqunveiled [Nm] driving551 551 range551 551 of551 K9551 as551 average551 551 250551 km551 based551 550 on549 550 551 543 529 Figure512 497 6. ADVISOR482 458 simulation439 421 404 setup.385 364 344 326 309 295 286 276 265 254 243 236 231 225 220 212 203 201 194 190 Vehicle SimulatorPowertheir [kW]Model experience.0 Battery6 12 17 size23 is29 32035 kWh,40 46 and52 the58 maximum63 69 75 81 86 91 94 97 99 101 101 101 101 102 101 99 97 96 94 93 93 92 92 91 89 89 89 90 90 89 87 89 87 88 road slope to climb is 15%. 1000 Simulation results Motor MC PM 180kW BYD K9Experimental motor results A. Vehicle Parameter Extraction 750 Includes two brushlessmc_map_spd permanent [RPM] 0 170 magnet341 511 682 synchronous852 1023 1193 1364 1534 motors1705 1875 2045 2216 2386 2557 2727 2898 3068 3239 3409 3580 3750 3920 4091 4261 4432 4602 4773 4943 5114 5284 5455 5625 5795 5966 6136 6307 6477 6648 6818 6989 7159 7330 7500 mc_map_spdWe use [rad/s] [24]0 to18 set36 parameters54 71 89 of107 the125 electric143 161 motor178 196 that214 the232 250 268 286 303 321 339 357 375 393 411 428 446 464 482 500 518 535 553 571 589 607 625 643 660 678 696 714 732 750 768 785 Maximum 350 Nm mc_max_trqandinstitute 90 [Nm] testedkW performance701 BYD701 electric701 701 701 bus 701 more701 than700 699 eight700 months701 691 (from674 652 633 613 582 500559 536 514 490 463 438 415 393 375 364 351 337 324 310 301 294 287 280 270 259 256 247 242 236 230 225 219 214 PowerAugust [kW] 29, 20130 to13 May25 13,38 50 2014).63 75 They87 reported100 112 125 acceleration136 144 151 158 164 166 170 172 174 175 174 172 170 168 167 169 169 169 168 166 166 168 169 170 168 166 169 168 168 168 168 168 168 168 Performance (by ref. experimentaltest including topresults speed) (43 mph), zero to 10 mph, 20 mph, Range (km) 250 43 mile per hour top30 speed mph, 40 mph and top speed. We extract (a) the maximum motor torque map andExtract (b) maximum detailed motor motor power torque by RPM map as 6.5 second 0–100 km/hshown time in Figure 5 with repeated ADVISOR simulations. 0 mc_max_trqbased [Nm] on performance experimentalPower [kW] results 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 Gearbox 800 180 Speed (km/h) 17.7:1 Figure 7. Driving range validation under efficiency parameter setting. 600 135

400 90 between the estimation of power consumption by the vehicle simulator and the powertrain models; the normalized root- mean-square error is 9.12%.

200 Motor power [kW] 45 Motor torque [Nm] Motor torque

0 0 Table I. MODEL COEFFICIENTS FOR BYD K9. 0 1875 3750 5625 7500 0 1875 3750 5625 7500 “Federal Transit Bus Test,” The Larson RPM RPM ↵ 0.098 9.7562 1.2016 0.0001 Institute, Tech. Rep., 2014. C0 1000.0 C1 1378.2 C2 0.00001 C3 0.000015 (a) Maximum motor torque map. (b) Maximum motor power. ✏ 0.4095 ⇣ 2178.5 14 Figure 5. Motor parameter extraction results. (a) is the maximum motor torque map and (b) is maximum power map. C. Vehicle Simulation Setup In our experiment, we followed the battery pack configura- We specified the detailed vehicle parameters as shown in tion provided by BYD to build our battery model. The battery Figure 6, which shows ADVISOR user interface for parameter is LiFePO4 (Lithium Iron Phosphate) with 540 V battery pack setting. The motor, efficiency and battery models are imported voltage. The battery pack consists of three battery modules, into ADVISOR, and based on the simulation results, we which has 108 kWh capacity. We assumed that each battery set the parameters so that the simulation results follow the cell in the pack is ideally balanced in the following experi- experimental results. Figure 7 shows the difference between ments, then built battery pack model as section III-A indicated. experimental results and simulation results of driving range. Concerning the regenerative braking phase, we assume that We set the parameters for drivatrain efficiency to follow the charging efficiency is 20%, which means that 20% of the trend of driving range by vehicle speed. There are about 200 kinetic energy is converted to electric energy and transferred km range difference between two lines. However, the range into the pack. trend by the speed is similar enough. Also, the range difference is resolved by updating battery model in the following section. D. Case study 1: Fast Driving Energy Estimation We extract a route going to Technische Universitat B. Vehicle powertrain modeling Munchen (TUM) and another one returning from TUM as Then, we extracted the coefficients of (1), (2) and (3) with shown in Figure 9. Table II summarize information of the a number of ADVISOR simulations. Table I summarizes the routes: distance, average slope along the route and average model coefficients of BYD K9. Figure 8 shows the difference vehicle speed. Trial 1 Trial 2 Trial 3

Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (mile/hour) 4.35 17.4 31.7 39.8 Speed (km/h) 7.0 28.0 51.0 64.1 Const. Power (W) 6773.68 23231.45 47509.01 66711.57 Speed (km/h) 7.0 28.0 51.0 64.1 Const. Power (W) 8151.28 27852.43 55299.21 77990.54 Distance/energy (mile/Wh) 0.00064 0.00075 0.00067 0.00060 Const. Power (W) 4963.89 16804.69 36566.19 52112.01

Distance/energy (mile/Wh) 0 0 0 0 Range (sim) (mile) 208 243 216 193 Distance/energy (mile/Wh) 0 0 0 0

Range (sim) (mile) 173 202 186 165 Range (exp.) (mile) 403 434 387 348 Range (sim) (mile) 284 335 281 247

Range (exp.) (mile) 403 434 387 348 Difference (%) 48% 44% 44% 44% Range (exp.) (mile) 403 434 387 348

Difference (%) 57% 53% 52% 53% Difference (%) 29% 23% 28% 29%

Range (sim) (km) 278 326 299 266 Range (sim) (km) 457 540 452 398 Vehicle specification-1 Range (exp.) (km) 648 698 624 560 Range (exp.) (km) 648 698 624 560 Battery energy (kWh) 324 wh_1st_rr 0.015

Vehicle specification Maximum current (A) 250 Vehicle specification-1-1

Battery energy (kWh) 324 Total weight (kg) 14465 Battery energy (kWh) 324 wh_1st_rr 0.015 wh_1st_rr 0.010 300 250 Maximum current (A) Range (sim) (mile) Range (exp.) (mile) Maximum current (A) Total weight (kg) 18000 We do not consider the acceleration/deceleration occurring Total weight (kg) 14465 in correspondence to speed500 changes. Such approximation is not critical however, because the acceleration energy is small Range (sim) (mile) Range (exp.) (mile) Range (sim) (km) Range (exp.) (km) 500 compared with the total energy375 consumption of each segment: acceleration/deceleration seldom last more than a few seconds. 700

375 250 IV. SIMULATION RESULTS 525 Range (mile) 250 We implemented a powertrain model of a BYD K9 bus 125 from the vehicle specification and experiment results [4], [22], 350 Range (mile) [23]. The gross vehicle weight of BYD K9 is 18,000 kg, and

Vehicle SimulatorRange (mile) Model 125 0 the facial area of K9 is 2.55 by0 3.365 m.10 K915 includes20 25 two elec-30 35 40 175 tric motors that the maximum motor torque andSpeed power (mph) are 350 MC PM 100kW motor in Advisor library 0 320 kWh battery capacity 0 5 10 15 20 Nm25 and30 9035 kW.40 The motor type is in-wheel BYD-TYC90A mc_map_spd [RPM] 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400Average1500 1600 2501700 km0 1800 driving1900 2000range2100 (by2200 ref.2300 experimental2400 2500 2600 results2700 2800) 2900 3000 3100 3200 3300 3400 3500 3600 3700 3800 3900 4000 4100 4200 4300 4400 Speed (mph)Brushless Permanent Magnet Synchronous Motor. Maximum 0.0 8.8 17.5 26.3 35.0 43.8 52.5 61.3 70.0 mc_map_spd [rad/s] 0 10 21 31 42 52 63 73 84 94 105 115 126 136 147 157 168 178 188 199 209 220 230 241 251 262 272 283 293 304 314 325 335 346 356 367 377 387 398 408 419 429 440 450 461 RPM is 7,500, and gear box ratio is 17.7:1. Manufacturers Speed (mph) mc_max_trqunveiled [Nm] driving551 551 range551 551 of551 K9551 as551 average551 551 250551 km551 based551 550 on549 550 551 543 529 Figure512 497 6. ADVISOR482 458 simulation439 421 404 setup.385 364 344 326 309 295 286 276 265 254 243 236 231 225 220 212 203 201 194 190 Powertheir [kW] experience.0 Battery6 12 17 size23 is29 32035 kWh,40 46 and52 the58 maximum63 69 75 81 86 91 94 The97 simulation99 101 101 result101 follows101 102 the101 trend99 97 96 94 93 93 92 92 91 89 89 89 90 90 89 87 89 87 88 of the experimental result road slope to climb is 15%. 1000 Simulation results MC PM 180kW BYD K9Experimental motor results A. Vehicle Parameter Extraction mc_map_spd [RPM] 0 170 341 511 682 852 1023 1193 1364 1534 1705 1875 2045 2216 2386 2557 2727 7502898 3068 3239 3409 3580 3750 3920 4091 4261 4432 4602 4773 4943 5114 5284 5455 5625 5795 5966 6136 6307 6477 6648 6818 6989 7159 7330 7500 mc_map_spdWe use [rad/s] [24]0 to18 set36 parameters54 71 89 of107 the125 electric143 161 motor178 196 that214 the232 250 268 286 303 321 339 357 375 393 411 428 446 464 482 500 518 535 553 571 589 607 625 643 660 678 696 714 732 750 768 785 mc_max_trqinstitute [Nm] tested701 BYD701 electric701 701 701 bus701 more701 than700 699 eight700 months701 691 (from674 652 633 613 582 500559 536 514 490 463 438 415 393 375 364 351 337 324 310 301 294 287 280 270 259 256 247 242 236 230 225 219 214

PowerAugust [kW] 29, 20130 to13 May25 13,38 50 2014).63 75 They87 reported100 112 125 acceleration136 144 151 158 164 166 170 172 174 175 174 172 170 168 167 169 169 169 168 166 166 168 169 170 168 166 169 168 168 168 168 168 168 168 test including top speed (43 mph), zero to 10 mph, 20 mph, Range (km) 250 30 mph, 40 mph and top speed. We extract (a) the maximum motor torque map and (b) maximum motor power by RPM as shown in Figure 5 with repeated ADVISOR simulations. 0 mc_max_trq [Nm] Power [kW] 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0

800 180 Speed (km/h) 15 Figure 7. Driving range validation under efficiency parameter setting. 600 135

400 90 between the estimation of power consumption by the vehicle simulator and the powertrain models; the normalized root- mean-square error is 9.12%.

200 Motor power [kW] 45 Motor torque [Nm] Motor torque

0 0 Table I. MODEL COEFFICIENTS FOR BYD K9. 0 1875 3750 5625 7500 0 1875 3750 5625 7500 RPM RPM ↵ 0.098 9.7562 1.2016 0.0001 C0 1000.0 C1 1378.2 C2 0.00001 C3 0.000015 (a) Maximum motor torque map. (b) Maximum motor power. ✏ 0.4095 ⇣ 2178.5 Figure 5. Motor parameter extraction results. (a) is the maximum motor torque map and (b) is maximum power map. C. Vehicle Simulation Setup In our experiment, we followed the battery pack configura- We specified the detailed vehicle parameters as shown in tion provided by BYD to build our battery model. The battery Figure 6, which shows ADVISOR user interface for parameter is LiFePO4 (Lithium Iron Phosphate) with 540 V battery pack setting. The motor, efficiency and battery models are imported voltage. The battery pack consists of three battery modules, into ADVISOR, and based on the simulation results, we which has 108 kWh capacity. We assumed that each battery set the parameters so that the simulation results follow the cell in the pack is ideally balanced in the following experi- experimental results. Figure 7 shows the difference between ments, then built battery pack model as section III-A indicated. experimental results and simulation results of driving range. Concerning the regenerative braking phase, we assume that We set the parameters for drivatrain efficiency to follow the charging efficiency is 20%, which means that 20% of the trend of driving range by vehicle speed. There are about 200 kinetic energy is converted to electric energy and transferred km range difference between two lines. However, the range into the pack. trend by the speed is similar enough. Also, the range difference is resolved by updating battery model in the following section. D. Case study 1: Fast Driving Energy Estimation We extract a route going to Technische Universitat B. Vehicle powertrain modeling Munchen (TUM) and another one returning from TUM as Then, we extracted the coefficients of (1), (2) and (3) with shown in Figure 9. Table II summarize information of the a number of ADVISOR simulations. Table I summarizes the routes: distance, average slope along the route and average model coefficients of BYD K9. Figure 8 shows the difference vehicle speed. EV Power Consumption Characterization

Need to implement very simple but accurate power model using ADVISOR Simulate electric bus power profile Driving simulation 100

Powertrain spec. 50 4 (km/h) Speed 0 0 44 88 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 ADVISOR simulation result Model coefficient extraction result 100 Time (s) 30 50 0 (kW) Power 0 -30 0 Power (kW) 44 88 -50 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 0 1000 2000 3000 4000 5000 6000 Time (s) Time (s) Fig. 1: Validation of the model coefficient extraction.

TABLE I: Model coefficients for Chevrolet Bolt. Logging Summary GM Bolt (use accumulated power) a 0.06 b 9.5549 g 1.0013 25 mile driving 65 mile driving 75 mile driving 93 mile driving Constant driving 25 65 75 93 d 0.00012 C0 1000 C1 10.588 (mile) Time ess_pwr_r Energy ess_pwr_r Energy ess_pwr_ Energy ess_pwr_r Energy Sum (25 mile) 1325930.16 C2 8.11 C3 0.00032 e 0.6633 (s) (W) (Ws) (W) (Ws) r (Ws) (W) (Ws) Constant driving (km/ 40 105 121 150 z 5813.6 (W) Energy total (kWh) 0.37 h)

0 3678.94 3678.94 15706.27 15706.27 21262.37 21262.37 34795.05 34795.05 Used (%) 0.64% Driving distance (km) 4.02 10.46 12.1 14.5 20 1 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 SoC (%) 99.36% 97.26% 96.28% 93.91% 50 -6 40 Electric-4 vehicle model-0.3 Equation form 2 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (65 mile) 5669894.31 Range (km) 626.97 381.22 325.50 238.13 0 30 -2 16 3 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 1.57 Range (mile) 0.4 20 in0 degree vehicle simulator EV power model 4 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 2.74% CD = 0.200 % 0.67% -5.17% 6.45% -7.52% 12 Driving data 2 1.0 degree 5 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Reference range (km) 622.816 402 305.775 257.5 10 4 8 6 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (75 mile) 7681551.19 6 1.7 2 2 7 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 2.13 (a) Range-speed experimental data. 5 4 4.3 Ptot = Qv + C0 + C1v + C2v + C3Q Used (%) 3.72%

8 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fuel economy (km/liter) 700 16Energy efficiency (km/Wh) 9 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Simulation results 10 30 50 70 90 110 0 5 10 15 20 25 Real vehicle experiment results 10 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (91 mile) 12582801.56 Driving velocity (km/h) Driving velocity (km/h) Energy total (kWh) 3.50 525 11 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 () Ford Escort. () Custom EV. 12 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 6.09% Fig. 3: (a) Fuel economy of a Ford Escort [37] and (b) energy 13 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 350 14 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 consumption of a custom EV [2] on various road slopes and Range (km) Range 15 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 vehicle velocities. 175 16 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

17 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Vehicle specification 18 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Variable 0 work is done by energy (fuel) consumption, we compare 19 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Battery energy 57.4 40 105 121 150 energy consumption instead of power consumption for easy (kWh) 20 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Speed (km/h) comparison. 21 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 (b) Range comparison with simulation results. Fig. 3(a) shows the fuel economy (range with a liter of fuel 22 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 energy) versus ICEV velocity on various road slopes [37]. Red 23 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fig. 2: (a) Range-speed experimental data for Chevrolet stars denote the minimum-energy velocity by the average road 24 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Bolt [36] and (b) comparison with simulation results of the slope. The minimum-energy velocity is reduced as the road 25 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 proposed EV power model. slope increases. For example, the vehicle has the highest fuel 26 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 economy at 90 km/h on a 6 degree road slope and 40 km/h on 27 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 a 2 degree road slope. Fuel economy on a downhill (negative 28 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 derived power models are close enough to the real vehicle road slope) is not meaningful due to fuel cut of modern ICEVs. 29 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 measurement data [32], [33], [34], [35], [36]. We refine the Compared with the flat road (zero degree), we see the fuel 30 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 coefficients until all the available data values match well. economy drops if the road slope is steeper than 2 degree due 31 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 As a result, the derived EV power model well matches with to downshift of the transmission. 32 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the range-speed experimental data as shown in Fig. 2(b) Fig. 3 demonstrates a very interesting EV-specific energy 33 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 as well as the top speed and maximum acceleration setting behavior such that the least-energy velocity increases as the 34 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the aerodynamic resistance, rolling resistance, curb weight, road slope increases. Such energy behavior is completely 35 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 etc. with the known values. The errors in the range-speed opposite to that of ICEV. The efficiency maps of an engine 36 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 simulation is merely 1.4%, the time for 0-to-100 km/h error and an electric motor (Fig. 4) show that the motor exhibits the 37 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 is 7.8%, and the top speed error is 3.0%, respectively. 38 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 maximum torque over the wide range of RPM while the engine

39 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 exhibits the maximum fuel efficiency only at near 2,000 RPM. B. Energy consumption comparison between ICEV and EV 40 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Therefore, an engine should be incorporated with a transmis- 41 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 We compare power consumption behavior of EV and ICEV sion that makes it possible to run the engine at close to the 42 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 in this section. As most previous fuel economy analysis most efficient region, independent to the vehicle velocity. The 43 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

44 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

45 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

46 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

47 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

48 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

49 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

50 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

51 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

52 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

53 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

54 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

55 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

56 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

57 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

58 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

59 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

60 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

61 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

62 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

63 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

64 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

65 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

66 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

67 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

68 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

69 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

14 EV Power Characterization -should be updated

Extract simulation results from ADVISOR

Constant speed driving profile Benchmark driving profile

25 mph 65 mph 75 mph 93 mph 160 100

120 75

80 50

40 25 Speed (km/h) Speed (km/h) 0 0 0

0 10 20 30 40 50 60 70 80 90 100110120 41 82 123 164 205 246 287 328 369 410 451 492 533 574 615 656 697 738 Time (s) Time (s)

17 EV Power Consumption Characterization

Need to implement very simple but accurate power model using ADVISOR

Driving simulation 100

Powertrain spec. 50 4 (km/h) Speed 0 0 44 88 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 ADVISOR simulation result Model coefficient extraction result 100 Time (s) 30 50 0 (kW) Power 0 -30 0 Power (kW) 44 88 -50 132 176 220 264 308 352 396 440 484 528 572 616 660 704 748 0 1000 2000 3000 4000 5000 6000 Time (s) Time (s) Fig. 1: Validation of the model coefficient extraction. Extract coefficients of TABLE I: Model coefficients for Chevrolet Bolt. Logging Summary GM Bolt (use accumulated power) simple the power model a 0.06 b 9.5549 g 1.0013 25 mile driving 65 mile driving 75 mile driving 93 mile driving Constant driving 25 65 75 93 d 0.00012 C0 1000 C1 10.588 (mile) Time ess_pwr_r Energy ess_pwr_r Energy ess_pwr_ Energy ess_pwr_r Energy Sum (25 mile) 1325930.16 C2 8.11 C3 0.00032 e 0.6633 (s) (W) (Ws) (W) (Ws) r (Ws) (W) (Ws) Constant driving (km/ 40 105 121 150 z 5813.6 (W) Energy total (kWh) 0.37 h)

0 3678.94 3678.94 15706.27 15706.27 21262.37 21262.37 34795.05 34795.05 Used (%) 0.64% Driving distance (km) 4.02 10.46 12.1 14.5 20 1 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 SoC (%) 99.36% 97.26% 96.28% 93.91% 50 -6 40 Electric-4 vehicle model-0.3 Equation form 2 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (65 mile) 5669894.31 Range (km) 626.97 381.22 325.50 238.13 0 30 -2 16 3 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 1.57 Range (mile) 0.4 20 in0 degree vehicle simulator EV power model 4 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 2.74% CD = 0.200 % 0.67% -5.17% 6.45% -7.52% 12 Driving data 2 1.0 degree 5 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Reference range (km) 622.816 402 305.775 257.5 10 4 8 6 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (75 mile) 7681551.19 6 1.7 2 2 7 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Energy total (kWh) 2.13 (a) Range-speed experimental data. 5 4 4.3 Ptot = Qv + C0 + C1v + C2v + C3Q Used (%) 3.72%

8 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fuel economy (km/liter) 700 18Energy efficiency (km/Wh) 9 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Simulation results 10 30 50 70 90 110 0 5 10 15 20 25 Real vehicle experiment results 10 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Sum (91 mile) 12582801.56 Driving velocity (km/h) Driving velocity (km/h) Energy total (kWh) 3.50 525 11 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 () Ford Escort. () Custom EV. 12 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Used (%) 6.09% Fig. 3: (a) Fuel economy of a Ford Escort [37] and (b) energy 13 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 350 14 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 consumption of a custom EV [2] on various road slopes and Range (km) Range 15 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 vehicle velocities. 175 16 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

17 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Vehicle specification 18 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Variable 0 work is done by energy (fuel) consumption, we compare 19 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Battery energy 57.4 40 105 121 150 energy consumption instead of power consumption for easy (kWh) 20 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Speed (km/h) comparison. 21 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 (b) Range comparison with simulation results. Fig. 3(a) shows the fuel economy (range with a liter of fuel 22 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 energy) versus ICEV velocity on various road slopes [37]. Red 23 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Fig. 2: (a) Range-speed experimental data for Chevrolet stars denote the minimum-energy velocity by the average road 24 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Bolt [36] and (b) comparison with simulation results of the slope. The minimum-energy velocity is reduced as the road 25 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 proposed EV power model. slope increases. For example, the vehicle has the highest fuel 26 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 economy at 90 km/h on a 6 degree road slope and 40 km/h on 27 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 a 2 degree road slope. Fuel economy on a downhill (negative 28 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 derived power models are close enough to the real vehicle road slope) is not meaningful due to fuel cut of modern ICEVs. 29 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 measurement data [32], [33], [34], [35], [36]. We refine the Compared with the flat road (zero degree), we see the fuel 30 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 coefficients until all the available data values match well. economy drops if the road slope is steeper than 2 degree due 31 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 As a result, the derived EV power model well matches with to downshift of the transmission. 32 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the range-speed experimental data as shown in Fig. 2(b) Fig. 3 demonstrates a very interesting EV-specific energy 33 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 as well as the top speed and maximum acceleration setting behavior such that the least-energy velocity increases as the 34 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 the aerodynamic resistance, rolling resistance, curb weight, road slope increases. Such energy behavior is completely 35 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 etc. with the known values. The errors in the range-speed opposite to that of ICEV. The efficiency maps of an engine 36 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 simulation is merely 1.4%, the time for 0-to-100 km/h error and an electric motor (Fig. 4) show that the motor exhibits the 37 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 is 7.8%, and the top speed error is 3.0%, respectively. 38 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 maximum torque over the wide range of RPM while the engine

39 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 exhibits the maximum fuel efficiency only at near 2,000 RPM. B. Energy consumption comparison between ICEV and EV 40 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 Therefore, an engine should be incorporated with a transmis- 41 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 We compare power consumption behavior of EV and ICEV sion that makes it possible to run the engine at close to the 42 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66 in this section. As most previous fuel economy analysis most efficient region, independent to the vehicle velocity. The 43 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

44 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

45 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

46 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

47 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

48 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

49 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

50 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

51 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

52 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

53 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

54 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

55 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

56 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

57 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

58 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

59 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

60 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

61 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

62 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

63 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

64 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

65 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

66 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

67 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

68 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

69 3683.15 3683.15 15749.83 15749.83 21337.85 21337.85 34952.66 34952.66

14 EV Power Characterization battery simulation working with EV driving simulation [16], B. Powertrain Modeling Process [17]. EV power model There are several powertrain simulators in academia and industry. ADvanced VehIcle SimulatOR (ADVISOR) is one III.Dynamics SYSTEM M ODELby resistances AND ESTIMATION acting on a vehicleof well-organized vehicle simulators that takes into account A. Powertrain Model Motor energy loss includes copper loss, ironvarious loss, factors friction, of vehicles and including engines, electric traction Power consumption of an EV depends on body shape motors, types of drivetrains, shape of chassis, etc. [18]. It is including facialwindage area, curbweight, loss road slope and types of tires possible to implement a certain type of vehicle in ADVISOR as well as onEV speed physical and acceleration power consumption of the EV. Figure 1 showsMotorby energy setting powertainloss parameters and simulate various vehicle the dynamic powerds by a motor rotating with torque T and driving environments by changing its powertrain or driving angular velocityP = F!. Four= Fv resistances acting on a vehicle, whereQ = P/profile.! =( Power↵ + consumption,sin✓ + )mr battery state of charge and emis- dt sion over time are simulated for a given driving cycle and FR, FG, FI , and FA are respectively the rolling, gradient, 2 2 inertia and aerodynamic=(FR + FA resistances.+ FG + FI + FB)v P = Cvehicle0 + C1v setup.+ C2v + C3Q (F + F + F )v ADVISOR, however, is not suitable to simulate power ⇡ R G I (↵ + sin✓ + a)mv, consumption in on-line manner because of its relatively long ⇡ runtime. ADVISOR considers overall vehicle dynamics and energy flow from torque on the wheels to the engine or battery pack. Overall vehicle simulation results show energy flow in detail, and this is important for energy analysis. However, it is not efficient to estimate current load in the battery point of view and make a decision to find the optimal route or vehicle velocity. So, instead of using ADVISOR itself, we adopt the vehicle powertrain models from (1) to (3) and use ADVISOR for the 19 Figure 1. Forces on an EV. extraction of model coefficients. Figure 2 shows overall pro- cess for the electric bus characterization. The process consists The power consumption at the EV powertrain is the fol- of three phases: i) parameter setting phase, ii) modeling phase lowing [17], [16]: and iii) simulation phase. ds 1) Parameter extraction phase: First we choose a vehicle P = T ! = F = Fv =(F + F + F + F )v dyna dt R G I A for the ADVISOR simulation. ADVISOR requires several 1 2 parameters and models for the simulation (e.g. motor model, FR CrrW, FG Wsin✓,FI ma, FA ⇢CdAv / / / / 2 vehicle chassis model and battery model). Vehicle manu- 2 Pdyna (↵ + sin✓ + a + v )mv facturers officially unveil their vehicle specifications on the ⇡ (1) website, such as the maximum motor power and torque and the time to reach 100 km/h. This information is used to where Crr, W , ✓, m, v, a, Cd, A are the rolling coefficient, weight, road slope, vehicle mass, vehicle speed, acceleration, implement detailed parameters and models for ADVISOR. In drag coefficient, and vehicle facial area, respectively. This this parameter setting phase, we implement an electric motor relation is simplified as a function of four coefficients ↵, , model and a drivetrain efficiency model, and a battery model and . for the ADVISOR simulation. We implement a motor torque map from the maximum motor torque/RPM, the time to reach The powertrain efficiency of electric motor and drivetrain 100 km/h and vehicle curbweight. We then implement a motor is less than 100%. The efficiency depends on the operating efficiency map using the torque map, battery size and driving RPM (revolutions per minute) and torque when the EV drives. range. The drivetrain efficiency model is obtained from driving On top of that, the drivetrain mechanical movement causes a range and resistances acting on a vehicle where we calculate power loss while delivering power to the wheels. The following resistances using vehicle body shape and type of tires. The EV specific power model considers power losses by the motor battery model is easily obtained from battery architecture and and drivetrain [17]: the battery cell specification. These models are imported into 2 2 ADVISOR system and used to simulate complex energy flow. PEV = Pdyna + C0 + C1v + C2v + C3T (2) 2) Modeling phase: ADVISOR simulates energy flow with where C0, C1, C2, and C3 are the coefficients for constant an electric vehicle model obtained from Section III-B1 and loss, iron and friction losses, drivetrain loss, and copper loss, a driving cycle. We performs simulations to obtain plentiful respectively. driving data with various vehicle speed and road slope. The Unlike ICEVs, the electric motor works like power gen- simulation results include power consumption by vehicle speed erator when when EV reduces its speed. This is done by and road slope over time. The driving cycles include driving on regenerative braking system, which converts kinetic energy on flat road with various vehicle speed and acceleration on various the wheel to electric energy and sends to the battery. The power road slope. Test driving on various road slope by distance generation by regenerative braking is modeled by a function is also performed. We use a multi-variable linear regression of the negative motor torque and vehicle speed, as follows: method to extract coefficients of the powertrain models from (1) to (3) [17]. Pregen = ✏Tv+ ⇣ (3) 3) Simulation phase: The equation form powertrain where ✏ and ⇣ are regenerative braking coefficients. model from (1) to (3) is used to extract power consumption 3

brake force provided by hydraulic brakes, respectively [27]. respectively. Even if (5) has the same form as (3), the actual The coefficients a, b, g, and d correspond to the rolling resis- coefficients become different if it reflects the characteristics of tance, gradient resistance, inertia resistance, and aerodynamic electric powertrain. resistance, respectively. ADVISOR is a vehicle simulator that takes into account The actual power consumption evaluation of ICEV, i.e., fuel various factors of vehicles including engines, electric trac- consumption rate, should take into account the powertrain tion motors, types of drivetrains, shape of chassis, etc. [29]. efficiency. A simplified model of ICEV fuel consumption ADVISOR is a fast simulator as it simulates a 700 second in [4] is written as follows: vehicle driving in one second. DP is commonly used to derive

Ptrac + Pstart the energy-aware-velocity planning of vehicles [10], [12]. The PICEV = (2) DP optimization repeatedly computes the vehicle energy in hICEV each distance, velocity and acceleration step, and thus it is where Pstart is power consumption of the engine during idling. not proper to run ADVISOR in the inner loop of the DP The engine efficiency, hICEV , is determined by the gear ratio, optimization. For instance, it takes 15 minutes of computation engine speed and engine torque. The gear ratio generally dy- to derive a 10 minute driving velocity planning profile with a namically changes by transmission shift. The engine efficiency 30-step velocity resolution. This is absolutely not acceptable is generally not simple enough to explain by analytic functions for online recomputation that can be caused by unexpected and often described as efficiency maps [4]. interrupts while driving. So, instead of using ADVISOR for Adding a traction motor power model in (2) makes (1) and (2) the speed optimization, we use ADVISOR for the model reflect more realistic EV power consumption characteristics. coefficient extraction. In other words, we use the same form Ptrac of polynomial as (5) but derive the coefficients by running PEV = (3) hEV ADVISOR. The proposed methods allows us to produce an updated Ptrac hEV = solution in a reasonable amount of time if an unexpected P + k T 2 + k w + k w3 +C trac c i w interrupt occurs. For example, if a vehicle accident occurs, where kc, T, ki, w, kw, C represent a copper loss coefficient, a GPS navigator may perform rerouting to detour the accident a motor torque, a coefficient by iron and friction loss, an area. The proposed velocity planning should be “recomputed” angular speed, a windage coefficient, and constant loss, re- for the modified route. After detouring, the new route will spectively [1]. EV Power Characterization be merged to the original route, and we can still reuse EVs mostly use regenerative braking during deceleration, the previously computed velocity planning for the rest of which converts kinetic energy to electric energy except low- travel. While the proposed methods can perform such online EV power model cost golf carts. The harvested energy from regenerative braking “recomputation,” the previous method that uses ADVISOR in is closely related to the electromagnetic flux inside of the the inner loop can hardly meet such requirement. EV power consumption motor,model and the fluxEV power is proportional harvesting to model the motor RPM. The We choose Chevrolet Bolt (1,616 kg curb weight) for the regenerative braking model of an EV can be simplified as (4) target vehicle, but our method does not restrict a particular Q = P/! =(↵ + sin✓ + )mr type of vehicles. Chevrolet Bolt has been produced since Pregen = eTv+ z (4) 2 2 Ptot = Qv + C0 + C1v + C2v + C3Q November 2016. Total 26,000 cars have been delivered, and where e and z are regenerative braking coefficients. Bolt is positioned as the third best-selling EV in the United The power consumption model (3) is commonly used in States. The top speed of Chevrolet Bolt is a 150 km/h. It is various analytical EV power managements [12], [15], [28]. capable of achieving a 6.5 second 0–100 km/h time with a 150 However, thereFRB( areregen. still brake considerable resistance)= sourcesconstant of power loss kW electric permanent magnet motor [30]. Chevrolet Bolt is from the drivetrain. To make a long story short, (1) – (3) do not equipped with a 350 V and 57.4 kWh Lithium-ion battery pack accurately explain EV-specific power consumption in various with 96-series and 3-parallel of 3.65 V cells [31]. Chevrolet driving conditions according to our intensive measurement Bolt battery chemistry is Lithium Nickel Manganese Cobalt with real EVs. Oxide from LG. In this paper, therefore, we borrow an EV-specific power (EV ⇤) We use ADVISOR to extract the model coefficients of new modeling method considering the loss in EV drivetrain using a vehicles starting from known vehicle specifications such as the multivariable20 regression method from the measured power data aerodynamic resistance, rolling resistance, curb weight, and so and achieve a polynomial fitting [1]. However, we should not on. Such data is often published in the users’ manual, website directly use this power model because the model covers only and so forth. We further use the vehicle performance data such low-speed EVs, i.e., slower than 35 km/h. Thus, we derive our as the top speed, maximum acceleration, and so on, which are own model coefficients in this paper. also available on the manuals, commercials and websites. We

Ptrac iteratively run ADVISOR and derive the model coefficients so PEV⇤ = (5) that all the known data values match among each other. The hEV ⇤ model coefficient extraction result shows as in Fig. 1. The P = trac normalized root-mean-square (RMS) deviation is 5.1%. Table hEV⇤ 2 2 Ptrac +C0 +C1v +C2v +C3T I summarizes the model coefficients of (5) for Chevrolet Bolt. where C0, C1, C2, and C3 mean coefficients for constant We further evaluate the model accuracy with the range-speed loss, iron and friction losses, drivetrain loss, and copper loss, real vehicle experiments as shown in Fig. 2(a) to confirm the 2 new performance metrics for EV, energy-delay product (EDP), consumption model with several parameters [12]. However, energy-square-delay product (E2DP) and energy-cubic-delay this power model still has important parameters missing that product (E3DP), which demonstrates a practically applicable significantly affect the model fidelity such as friction loss, energy-aware-velocity planning. The proposed method results iron loss and windage loss in the motor, loss in the drivetrain in up to a 43.4% extended driving range without increasing and regenerative braking. Inaccurate EV power consumption Trial 1 Trial 2 Trial 3 battery capacity, and a 51.7% improvement of EDP compared model misleads the energy-aware-velocity planning. A more Speed (mile/hour) 4.35 17.4with the31.7 least-energy-constant-velocity39.8 Speed (mile/hour) driving. 4.35 17.4 recent31.7 work39.8 fabricates a low-speedSpeed (mile/hour) custom EV and4.35 derived17.4 31.7 39.8 Speed (km/h) 7.0 28.0We summarize51.0 64.1 the paperConst. structure Power as(W) follows.6773.68 Section23231.45 II an47509.01 accurate66711.57 power model. TheSpeed power (km/h) model reflects7.0 various28.0 51.0 64.1 Const. Power (W) 8151.28 27852.43describes55299.21 related77990.54 work for low-energyDistance/energy driving (mile/Wh) for0.00064 both ICEV0.00075 EV-specific0.00067 0.00060 factors includingConst. regenerative Power (W) braking4963.89 and even16804.69 36566.19 52112.01 Distance/energy (mile/Wh) 0 0and EV.0 Section0 III introducesRange analysis (sim) (mile) of power consumption208 243 for drivetrain216 193 loss by regressionDistance/energy analysis (mile/Wh) of a large amount0 0 0 0

Range (sim) (mile) 173 202of ICEV186 and165 EV by the vehicleRange setup(exp.) (mile) and driving conditions.403 434 of driving387 data.348 However, theRange power (sim) (mile) model is limited284 to low-335 281 247 This section also compares power consumption characteristics speed EVs and hard to accommodate production EVs [1]. Range (exp.) (mile) 403 434 387 348 Difference (%) 48% 44% 44% 44% Range (exp.) (mile) 403 434 387 348 of EV with those of ICEV and demonstrates why ICEV eco- Some work attempts to accommodate more realistic driving Difference (%) Difference (%) 29% 23% 28% 29% 57% 53%driving52% results53% can hardly apply to EV. Section IV introduces a conditions such as traffic signals and stop signs. However, Range (sim) (km) 278 326 299 266 Range (sim) (km) 457 540 452 398 framework to derive EV-specificVehicle energy-aware-velocity specification-1 plan- the results are lack of practicality due to the assumption of Range (exp.) (km) 648 698ning. In624 this section,560 we are first to introduce new performance a constant road slope andRange a constant (exp.) (km) cruising velocity648 with698 624 560 Battery energy (kWh) 324 metrics and their solution methods to take into account both highly simplified EV power models [9], [15], [11]. wh_1st_rr driving energy and driving time: energy-delay product0.015 (EDP), We recently see a new initiative that attempts to reduce Vehicle specification Vehicle specification-1-1 energy-square-delay productMaximum (E2DP current) and (A) energy-cubic-delay250 the energy consumption of cyber-physical systems using Battery energy (kWh) 324 product (E3DP). Section VTotal demonstrates weight (kg) the14465 experimental application-aware cross-layerBattery management. energy (kWh) Such324 approaches wh_1st_rr 0.015 results. We also visualize the impacts of the model fidelity on apply algorithms, tools andwh_1st_rr methodologies relevant0.010 to de- 3 300 250 Maximum current (A) the energy-aware-velocity planning in SectionRange (sim) VI. (mile) Section VIIRangesign (exp.) automation(mile) and embeddedMaximum low-power current (A) systems [16]. Just Total weight (kg) 18000 isWe a summary do not of consider major technical the acceleration/deceleration contribution of this paper. occurringnaming several for instance,Total weight energy (kg) management14465 for EV in correspondence to speed500 changes. Such approximationalso considers is brake drivers’ force behaviors provided and related by hydraulic EV power brakes, con- respectively [27]. respectively. Even if (5) has the same form as (3), the actual not critical however, because the acceleration energy issumption small to extendThe coefficients range [17],a [18]., b, Electricg, and d vehiclecorrespond battery to the rolling resis- coefficients become different if it reflects the characteristics of Range (sim) (mile) Range (exp.) (mile) II. RELATED WORK Range (sim) (km) Range (exp.) (km) 500 compared with the total energy consumption of each segment:packs consisttance, of a large gradient number resistance, of battery inertia cells, resistance, and it is and aerodynamic electric powertrain. Low-energy driving is proposed375 for ICEVs to assists drivers in crucial to optimize the battery pack architectures in terms acceleration/deceleration seldom last more than a few seconds. resistance, respectively.700 ADVISOR is a vehicle simulator that takes into account order to save fuel consumption and reduce emissions [3], [4], of capacity, costThe and actual reliability power [19], consumption [20]. Electric evaluation vehicle of ICEV, i.e., fuel various factors of vehicles including engines, electric trac- 375 [5], [6], [7], [8]. The fuel consumption250 model is a polynomial battery management systems must be able to accommodate IMULATION ESULTS consumption rate, should take into account the powertrain tion motors, types of drivetrains, shape of chassis, etc. [29]. form based on aIV. general S vehicle dynamicsR model and takes multi-power sources, provide525 a good scalability, and mitigate into account the transmissionRange (mile) shift position. efficiency. A simplified model of ICEV fuel consumption ADVISOR is a fast simulator as it simulates a 700 second 250 We implemented a powertrain model of a BYD K9cell-to-cell bus variability [21]. Electric vehicles consume a non- There has been low-energy driving125 research for EVs as well, negligible amountin [4] of is battery written energy as follows: for non-driving power vehicle driving in one second. DP is commonly used to derive 3 from the vehicle specification and experiment results [4], [22], 350 Range (mile) but previous research relies on the vehicle dynamics models the energy-aware-velocity planning of vehicles [10], [12]. The [23]. The gross vehicle weight of BYD K9 is 18,000 kg,components. and The related study tries to optimizePtrac energy+ Pstart for 125 Range (mile) = such as rolling resistance, gradient0 resistance, aerodynamic heating, ventilation and air conditioningPICEV (HVAC) systems (2) DP optimization repeatedly computes the vehicle energy in the facial area of K9 is 2.55 by0 3.365 m.10 K915 includes20 25 two elec-30 35 40 brake forceh providedICEV by hydraulic brakes, respectively [27]. respectively. Even if (5) has the same form as (3), the actual resistance, and so forth [9], which ignores a loss of energy equipped with electric vehicles175 [22], [23]. A renewable energy each distance, velocity and acceleration step, and thus it is tric motors that the maximum motor torque andSpeed power (mph) are 350 The coefficients a, b, g, and d correspond to the rolling resis- coefficients become different if it reflects the characteristics of transformation from electricity to mechanical energy. Such aware pricingwhere schemePstart minimizesis powerMC both PM consumption 100kW the community-wide motor of in Advisor the engine library during idling. not proper to run ADVISOR in the inner loop of the DP 0 tance, gradient resistance, inertia resistance, and aerodynamic electric powertrain. 0 5 10 15 20 Nm25simplification and30 9035 kW. makes40 The it impossible motor type to count is in-wheel on the powertrain BYD-TYC90Aelectricity billThe and engine individual efficiency, electricityhICEV bill, [24]. is determined A hybrid by the gear ratio, mc_map_spd [RPM] 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 17000 1800 1900resistance,2000 2100 respectively.2200 2300 2400 2500 2600 2700 2800optimization.2900 3000 3100 ForADVISOR instance,3200 3300 3400 itis takes a3500 vehicle 153600 minutes simulator3700 3800 of computation that3900 takes4000 4100 into4200 account4300 4400 Speed (mph)Brushlessefficiency Permanent at all, regardless Magnet of ICEV Synchronous or EV. Motor. Maximumenergy storageengine systems speed improve and efficiency0.0 engine8.8 and torque.17.5 lifetime26.3 The without35.0 gear43.8 ratio52.5 generally61.3 70.0 dy- mc_map_spd [rad/s] 0 10 21 31 42 52 63 73 84 94 105 115 126 136 147 157 168 178 188 The199 actual209 220 power230 consumption241 251 262 evaluation272 283 of ICEV,293to derive304 i.e.,314 a fuel 10325 minutevarious335 driving346 factors356 velocity of367 vehicles377 planning387 including398 profile408 engines, with419 electric429 a 440 trac-450 461 RPMIt has is been 7,500, reported and that gear even box EV power ratio models is 17.7:1. that consider Manufacturerssignificant increasenamically of the changes cost [25], by [26]. transmissionSpeed shift. (mph) The engine efficiency mc_max_trq [Nm] 551 551 551 551 551 551 551 551 551 551 551 551 550 549 550 551 543 529 512 consumption497 482 458 rate,439 should421 404 take385 into364 account344 326 the30-step powertrain309 velocity295 286tion resolution.276 motors,265 254 types This243 ofis absolutely drivetrains,236 231 225 shapenot220 acceptable of212 chassis,203 etc.201 [29].194 190 unveiledthe motor driving loss [10], range [11], of [12] K9 can as hardly average reproduce 250 correct km based on is generally not simpleFigure enough 6. ADVISOR to explain simulation by analytic setup. functions PowerEV [kW] drivetrain power0 6 consumption12 17 23 [1].29 This35 paper40 demonstrates46 52 58 63 69 75 81 86 91 94 97 efficiency.99 101 101 A simplified101 101 model102 101 of ICEV99 97 fuelfor96 consumption online94 93 recomputation93ADVISOR92 92 that is91 a canfast89 simulator be89 caused89 as by90 it simulates unexpected90 89 a87 700 second89 87 88 their experience. Battery size is 320 kWh, and the maximumIII. VEHICLEandP oftenOWER described AND ENERGY as efficiencyCONSUMPTION maps [4]. how incorrect EV power model can mislead the energy-aware- in [4] is written as follows: interrupts whilevehicle driving. driving So, instead in one second. of using DP ADVISOR is commonly for used to derive road slope to climb is 15%. Adding a tractionA1000NALYSIS motor power model in (2) makes (1) and (2) the energy-aware-velocity planning of vehicles [10], [12]. The velocity planning in Section III. SimulationP resultstrac + Pstart the speed optimization, we use ADVISOR for the model A. The least-energy-constantreflect more realistic velocity EV for cruising power consumptionPICEV characteristics.= (2) DP optimization repeatedly computes the vehicle energy in There have been previous practices that demonstrate how MC PM 180kW BYD K9Experimental motor hresultsICEV coefficient extraction. In other words, we use the same form A. Vehicle Parameter Extraction each distance, velocity and acceleration step, and thus it is EVs should change the vehicle velocity to reduce energy A commonly used vehicle power750 consumption modelPtrac is from of polynomial as (5) but derive the coefficients by running mc_map_spd [RPM] 0 170 341 511 682 852 1023 1193 1364 1534 1705 1875 2045 2216 2386 2557 2727 2898 3068 where3239P 3409Pstart= 3580is power3750 3920 consumption4091 4261 of4432 the engine4602 4773 during4943 idling.5114 5284 5455 5625 5795 5966 6136 6307 6477 6648 6818 6989 7159 7330 7500 consumption under a given road and traffic conditions. Such the vehicle dynamics equation (1). ThisEV model considers (3) not proper to run ADVISOR in the inner loop of the DP mc_map_spdWe use [rad/s] [24]0 to18 set36 parameters54 71 89 of107 the125 electric143 161 motor178 196 that214 the232 250 268 286 303 321 The339 engine357 h375 efficiency,EV 393 411hICEV428, is446 determined464 482 byADVISOR.500 the gear518 ratio,535 553 571 589 607 625 643 660 678 696 714 732 750 768 785 work commonly formulates an offline optimization problem the traction force only assuming a 100% efficiency of the optimization. For instance, it takes 15 minutes of computation institute tested BYD electric bus more than eight months (from 500 engine speed and engine torque. The gear ratioThe generally proposed dy- methods allows us to produce an updated mc_max_trqand uses [Nm] a dynamic701 701 programming701 701 701 (DP)701 701 to derive700 699 the700 globally701 691powertrain674 652 (engines,633 613 electric582 559 motors536 or514 both.)490 Therefore,P463trac 438 this415 393 375 364 351 337 324 310 301to derive294 287 a 10280 minute270 driving259 256 velocity247 planning242 236 profile230 225 with219 a 214 August 29, 2013 to May 13, 2014). They reported accelerationEV Power ModelhEV = Validationnamically changes by transmission shift. The enginesolution efficiency in a reasonable amount of time if an unexpected Power [kW] 0 13 25 38 50 63 75 87 100 112 125 136 144 151 158 164 166 170 172P 174+ k175T 2 +174k w172+ k 170w3 +168C 167 169 169 169 168 166 16630-step168 velocity169 170 resolution.168 166 This169 is168 absolutely168 168 not168 acceptable168 168 168 testoptimal including solution top [10], speed [12], (43 [13], mph), [14]. However, zero to once 10 mph,again, 20model mph, is applicable to anyRange (km) sort of vehiclestracis generallyc but not noti simpleablew to enough to explain by analyticinterrupt functions occurs. For example, if a vehicle accident occurs, such work usually uses the vehicle dynamics model for the explain EV-specific power consumption.250 for online recomputation that can be caused by unexpected where kc, T, ki, w, kwand, C oftenrepresent described a copper as efficiency loss coefficient, maps [4]. a GPS navigator may perform rerouting to detour the accident 30EV mph, power 40 model. mph It and borrows top speed.the parameters We extract from commercial (a) the maximumEV power model interrupts while driving. So, instead of using ADVISOR for motor torque map and (b) maximum motor power by RPM as a motords torque, a coefficientAdding a traction by iron motor and power friction model loss, in (2) an makesarea. (1) The and proposed (2) the velocity speed optimization, planning should we use be ADVISOR “recomputed” for the model vehicles such as the vehicle curb weight, aerodynamic resis- Ptrac = F = Fv =(FR + FG + FI + FA)v shown in Figure 5 with repeated ADVISOR simulations. EV powerangular dtconsumption speed,0 a windage modelreflect coefficient, more realistic and EV constant power consumption loss, re- characteristics.for the modifiedcoefficient route. After extraction. detouring, In other the words, new we route use will the same form tance frommc_max_trq the vehicle [Nm] specifications and the rollingPower resistance [kW] 0.0 10.0 20.01 30.0 40.0 50.0 60.0 70.0 , spectively [1]., , 2 (1) Ptrac be merged toof the polynomial original route, as (5) but and derive we can the coefficients still reuse by running from the tire specifications [10]. FR µ CrrW FG µ Wsinq FI µ ma and FA µ rCdAv PEV = (3) EVs mostly use regenerative2 brakingSpeed (km/h) during deceleration,hEV ADVISOR. An800 empirical modeling has a potential180 to achieve more ac- 2 the previously computed velocity planning for the rest of Ptrac (a + bsinq + ga + dv )mv The proposed methods allows us to produce an updated curate power model of a particular target EV. Related work which⇡ converts kinetic energy to electric energy exceptPtrac low- travel. While the proposed methods can perform such online Figure 7. Driving range validation under= efficiency parameter setting. cost golf carts. The harvested energyhEV from regenerative2 braking 3 solution in a reasonable amount of time if an unexpected fabricates600 a custom EV, collects the GPS135 tracking, battery where FR, FG, FI, FA, and FB denote the rolling resistance, gra-Ptrac + kcT + kiw + kww +“recomputation,”C the previous method that uses ADVISOR in voltage and battery current and constructs an EV power dient resistance,is inertia closely resistance, related aerodynamic to the electromagnetic resistance, and flux inside of the interrupt occurs. For example, if a vehicle accident occurs, where 2kc, T, ki,2w, kw, C represent a copper lossthe coefficient, inner loop can hardly meet such requirement. Ptotmotor,= Qv and+ C the0 + fluxC1v is+ proportionalC2v + C3Q to the motor RPM. The a GPS navigator may perform rerouting to detour the accident 400 90 between the estimationa motor of power torque, consumption a coefficient by by iron the and vehicle frictionWe choose loss, an Chevroletarea. The Bolt proposed (1,616 velocity kg curb planning weight) should for be the “recomputed” EV simulatorregenerativepower harvesting and braking the model model powertrainangular of speed, an EV models; a can windage be the simplified coefficient, normalized as and (4) constant root-target loss, vehicle, re- butfor our the modified method doesroute. not After restrict detouring, a particular the new route will spectively [1]. mean-square error isP 9.12%.= eTv+ z (4) type of vehicles.be Chevrolet merged to Bolt the original has been route, produced and we since can still reuse 200 Motor power [kW] 45 regen Motor torque [Nm] Motor torque EVs mostly use regenerative braking duringNovember deceleration, 2016.the Total previously 26,000 computed cars have velocity been delivered, planning for and the rest of EV wherepowere modeland z areparameters regenerativewhich converts braking kinetic coefficients. energy to electric energyBolt except is positioned low- travel. as the While third the best-selling proposed methods EV in can the perform United such online 0 0 Table I. MODELcost golf COEFFICIENTS carts. The harvested FOR BYDenergy K9. from regenerative braking “recomputation,” the previous method that uses ADVISOR in 0 1875 3750 5625 7500 0 1875 3750 5625 7500 The power consumption model (3) is commonly used in States. The top speed of Chevrolet Bolt is a 150 km/h. It is is closely related to the electromagnetic flux inside of the the inner loop can hardly meet such requirement. various↵ analytical0.098 EV power9.7562 managements 1.2016 [12], [15], [28].0.0001 capable of achieving a 6.5 second 0–100 km/h time with a 150 RPM RPM motor, and the flux is proportional to the motor RPM. The We choose Chevrolet Bolt (1,616 kg curb weight) for the However,C0 there1000.0 areC still1 considerable1378.2 C2 sources0.00001 ofC power3 0.000015 loss kW electric permanent magnet motor [30]. Chevrolet Bolt is (a) Maximum motor torque map. (b) Maximum motor power. regenerative braking model of an EV can be simplified as (4) target vehicle, but our method does not restrict a particular from the✏ drivetrain.0.4095 To⇣ make2178.5 a long story short, (1) – (3) do not equipped with a 350 V and 57.4 kWh Lithium-ion battery pack type of vehicles. Chevrolet Bolt has been produced since accurately explain EV-specific power consumptionPregen = eTv in+ variousz (4) Figure 5. Motor parameter extraction results. (a) is the maximum motor with 96-series andNovember 3-parallel 2016. of Total 3.65 26,000 V cells cars [31]. have Chevrolet been delivered, and driving conditions according to our intensive measurement Bolt battery chemistry is Lithium Nickel Manganese Cobalt torque map and (b) is maximum power map. C. Vehicle Simulationwhere Setupe and z are regenerative braking coefficients. Bolt is positioned as the third best-selling EV in the United with real EVs. The power21 consumption model (3) is commonlyOxide used from in LG.States. The top speed of Chevrolet Bolt is a 150 km/h. It is In thisIn paper,our experiment, therefore,various we we borrow followed analytical an EV-specific the EV batterypower power managements pack (EV configura-⇤) [12],We [15], use ADVISOR [28]. to extract the model coefficients of new We specified the detailed vehicle parameters as shown in capable of achieving a 6.5 second 0–100 km/h time with a 150 tionmodeling provided method by considering BYDHowever, to build the there loss our are in battery EV still drivetrain considerable model. using The sources a batteryvehicles of power starting loss kW from electric known permanent vehicle specifications magnet motor such [30]. as Chevrolet the Bolt is Figure 6, which shows ADVISOR user interface for parameter ismultivariable LiFePO4 (Lithium regressionfrom Iron method the Phosphate) drivetrain. from the To measured with make 540 a long power V story battery data short, pack (1)aerodynamic – (3) do not resistance,equipped rolling with a 350 resistance, V and 57.4 curb kWh weight, Lithium-ion and so battery pack setting. The motor, efficiency and battery models are imported voltage.and achieve The a polynomial batteryaccurately pack fitting consists explain [1]. However, EV-specific of three we battery power should consumption not modules,on. Such in various data iswith often 96-series published and in 3-parallel the users’ of manual, 3.65 V cells website [31]. Chevrolet into ADVISOR, and based on the simulation results, we whichdirectly has use 108this power kWhdriving model capacity. conditions because We theassumed according model that tocovers our each intensive only batteryand measurement so forth. WeBolt further battery use chemistry the vehicle is performance Lithium Nickel data Manganese such Cobalt with real EVs. set the parameters so that the simulation results follow the celllow-speed in the EVs, pack i.e., is slower ideally than balanced 35 km/h. in Thus, the we following derive our experi-as the top speed,Oxide maximum from LG. acceleration, and so on, which are In this paper, therefore, we borrow an EV-specific power (EV ⇤) We use ADVISOR to extract the model coefficients of new experimental results. Figure 7 shows the difference between ments,own model then coefficients built battery in this pack paper. model as section III-A indicated.also available on the manuals, commercials and websites. We experimental results and simulation results of driving range. modeling method considering the loss in EV drivetrain using a vehicles starting from known vehicle specifications such as the Concerning the regenerativemultivariableP brakingtrac regression phase, method we from assume the measurediteratively that power data run ADVISORaerodynamic and resistance, derive the rolling model resistance, coefficients curb weight, so and so We set the parameters for drivatrain efficiency to follow the P = (5) charging efficiencyand isEV 20%,achieve⇤ h which a polynomial means fitting that [1]. 20% However, ofthat we the should all the not knownon. dataSuch values data is often match published among in each the other.users’ manual, The website trend of driving range by vehicle speed. There are about 200 EV⇤ kinetic energy is converteddirectly use to this electric power energy model because and transferred the modelmodel covers coefficient only and extraction so forth. We result further shows use the as vehicle in Fig. performance 1. The data such P km range difference between two lines. However, the range into the pack. = low-speed EVs,trac i.e., slower than 35 km/h. Thus,normalized we derive our root-mean-squareas the top speed, (RMS) maximum deviation acceleration, is 5.1%. and Table so on, which are hEV⇤ 2 2 trend by the speed is similar enough. Also, the range difference Ptracown+C model0 +C coefficients1v +C2v + inC3 thisT paper. I summarizes thealso model available coefficients on the manuals, of (5) for commercials Chevrolet and Bolt. websites. We is resolved by updating battery model in the following section. D.where CaseC0, studyC1, C 1:2, Fast and DrivingC3 mean Energy coefficients Estimation forPtrac constant We further evaluateiteratively the model run ADVISOR accuracy and with derive the the range-speed model coefficients so PEV⇤ = (5) that all the known data values match among each other. The loss,We iron extract and friction a route losses, drivetraingoing to loss, Technische and copperhEV Universitat loss, real vehicle experiments as shown in Fig. 2(a) to confirm the ⇤ model coefficient extraction result shows as in Fig. 1. The B. Vehicle powertrain modeling Munchen (TUM) and another one returningP from TUM as = trac normalized root-mean-square (RMS) deviation is 5.1%. Table hEV⇤ 2 2 Then, we extracted the coefficients of (1), (2) and (3) with shown in Figure 9. Table II summarizePtrac +C0 information+C1v +C2v + ofC3T the I summarizes the model coefficients of (5) for Chevrolet Bolt. a number of ADVISOR simulations. Table I summarizes the routes: distance, averagewhere C slope0, C1, alongC2, and theC3 routemean and coefficients average for constant We further evaluate the model accuracy with the range-speed model coefficients of BYD K9. Figure 8 shows the difference vehicle speed. loss, iron and friction losses, drivetrain loss, and copper loss, real vehicle experiments as shown in Fig. 2(a) to confirm the EV Power Model Validation

Normalized root-mean-square (RMS) deviation: 9.12%

200 ADVISOR simulation result Model coefficient extraction result 100

0 Power (kW)

-100 0 500 1000 1500 2000 2500 3000 3500 Time (s)

Figure 8. Powertrain model validation result.

550

500 Altitude (m) 0 5 10 15 20 25 30 Distance (km) 40

20 22

Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Time (s) 200 40

100 20 0

Power (kW) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) Time (s)

550

500 Altitude (m) Figure 9. Routes from home to TUM and vice versa. 0 5 10 15 20 25 30 Distance (km) 40 Table II. A ROUTE INFORMATION. 20

Route Distance (km) Avg. slope (degree) Avg. speed (km/h) Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Home to TUM 31.2 -0.1728 24.67 Time (s) TUM to Home 26.2 0.2221 22.24 200 40

100 20 The simulation results for the routes are performed in 0

Power (kW) 0 Figure 10. The first two figures show the road altitude from 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) home to TUM and the speed profile that we obtained the profile Time (s) from road speed limit and Google Maps traffic information. Figure 10. Simulation result from home to campus (first to third figures) and The third figure shows the corresponding power and energy from the campus to home (fourth and sixth figures). consumption over time. The power consumption is low in the first half compared with another half because the degree of based on the fast simulation results. negative slope along the road is high. Fourth and fifth figures show the road altitude and speed profile from TUM to home. The sixth figure shows the power and energy consumption. E. Case Study 2: Battery Size Analysis The road slope is positive in this case. Therefore, energy One important merit of the proposed power model is to consumption is higher than the energy consumption to go to estimate energy consumption in a short time, which is useful to TUM. The driving distance going to TUM is longer than the iteration-based parameter sizing. For example, we can find the route returning from TUM. However, the energy consumption optimal battery size using iterative vehicle simulation. Short to go to TUM (27.4 kWh) is nearly 10% lower than the energy simulation time is mandatory in this approach. We perform consumption to return from TUM (30.8 kWh) because of road driving simulation on flat 100 km distance with different slope. The proposed equation-form energy model helps us to battery pack size. We assume that the vehicle speed is 50 km/h, estimate the energy consumption along the road slope and which is average bus speed in suburb of the city. We assume traffic immediately and to decide which route is economic that the battery pack voltage is the same, and additional battery System-Level Approach for EV Optimization

Low-power electric vehicle system-level approach Optimal driving profile

EV driving optimization

Road/traffic info. Vehicle setup EV characterization Driving constraints Power consumption model Energy harvesting/ EV power consumption simulator charging model

Battery model EV design optimization

Road/traffic info. Driving profile Component library (cost, life, etc.)

Optimal vehicle

23 design candidates 200 ADVISOR simulation result Model coefficient extraction result

100200 ADVISOR simulation result 200 Model coefficient extraction result ADVISOR simulation result 0 Model coefficient extraction result Power (kW) 100 100

-100 0 Power (kW) 0 500 1000 1500 2000 2500 3000 35000 Power (kW) Time (s)

-100 Figure 8. Powertrain modelCase validation Study result. 1: Fast Driving-100 Energy Estimation 0 500 1000 1500 2000 2500 3000 35000 500 1000 1500 2000 2500 3000 3500 Time (s) Time (s) 550Benchmark route information Figure 8. Powertrain model validation result. Figure 8. Powertrain model validation result. 500 Altitude (m) 0 5 10 15 20 25 30 550 Distance (km) 550 50040 500 Altitude (m) Altitude (m) 20 0 5 10 15 20 25 30 0 5 10 15 20 25 30 Distance (km) Distance (km)

Speed (km/h) 0 40 31.2400 km, 500avg. -0.17 1000 degree, 1500 24 2000 km/h 2500 3000 3500 4000 Time (s) 20 20 200 40

Speed (km/h) 0 Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 100 0 500 1000 1500 2000 2500 3000 3500 400020 Time (s) 0 Time (s) 200 40 Power (kW) 200 400 26.2 0km, avg. 500 0.22 1000 degree 1500 and 2000 22.24 2500 km/h 3000 3500 4000 Energy (kWh) 100 20 100 Time (s) 20 0

0 Power (kW) 0 Energy (kWh)

Power (kW) 550 0 0 500 1000 1500 2000 2500 3000 3500 4000 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) Time (s) 500 Time (s)

Altitude (m) 550 Figure 9. Routes from home to TUM and vice versa. 5500 5 10 15 20 25 30 Distance (km) 500

50040 Altitude (m)

Altitude (m) Figure 9. Routes from home to TUM and vice versa. 0 5 10 15 20 25 30 Table II. A ROUTE INFORMATION. 24 Figure 9. Routes from home to TUM and vice versa. 20 0 5 10 15 20 25 30 Distance (km) Distance (km) 40

Route Distance (km) Avg. slope (degree) Avg. speed (km/h) Speed (km/h) 400 Table II. A ROUTE INFORMATION. 20 Table II. A ROUTE INFORMATION. 0 500 1000 1500 2000 2500 3000 3500 4000 Home to TUM 31.2 -0.1728 24.67 20 Time (s)

TUM to Home 26.2 0.2221 22.24 Route Distance (km) Avg. slope (degree) Avg. speed (km/h) Speed (km/h) 0 200 40 0 500 1000 1500 2000 2500 3000 3500 4000

Route Distance (km) Avg. slope (degree) Avg. speed (km/h) Speed (km/h) 0 Home to TUM 31.2 -0.1728 24.67 100 0 500 1000 1500 2000 2500 3000 3500 4000 Time (s) TUM to Home 20 26.2 0.2221 22.24 Home to TUM 31.2 -0.1728 24.67 Time (s) 200 40 0 TheTUM tosimulation Home results26.2 for the0.2221 routes are performed22.24 in 200 40 100 Power (kW) 0 20 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) Figure 10. The first two figures show the road altitude from 100 Time (s) The simulation20 results for the routes are performed in 0 Power (kW) 0 home to TUM and the speed profile that we obtained the profile Energy (kWh) The simulation results for the routes are performed in 0 Figure 10. The first two figures show the road altitude from 0 500 1000 1500 2000 2500 3000 3500 4000 from road speed limit and Google Maps traffic information. Power (kW) 0 Time (s) Figure 10. The first two figures show the road altitude from Figure0 10. Simulation 500 1000 result 1500from home 2000 to campus 2500 (first 3000home to third3500 to TUM figures) 4000 and andEnergy (kWh) the speed profile that we obtained the profile The third figure shows the corresponding power and energy from the campus to homeTime (fourth (s) and sixthfrom figures). road speed limit and Google Maps traffic information. consumptionhome to TUM over and time. the speed The power profile consumption that we obtained is low the profilein the Figure 10. Simulation result from home to campus (first to third figures) and from road speed limit and Google Maps traffic information. The third figure shows the corresponding power and energy from the campus to home (fourth and sixth figures). first half compared with another half because the degree of basedFigure 10. on Simulation the fast simulation result from home results. to campus (firstconsumption to third figures) over and time. The power consumption is low in the negativeThe third slope figure along shows the road the corresponding is high. Fourth power and fifth and figures energy from the campus to home (fourth and sixth figures). consumption over time. The power consumption is low in the first half compared with another half because the degree of based on the fast simulation results. show the road altitude and speed profile from TUM to home. negative slope along the road is high. Fourth and fifth figures first half compared with another half because the degree of E. Case Study 2: Battery Size Analysis The sixth figure shows the power and energy consumption. based on the fast simulation results. show the road altitude and speed profile from TUM to home. negative slope along the road is high. Fourth and fifth figures The road slope is positive in this case. Therefore, energy One important merit of the proposed powerThe sixth model figure is to shows the power and energy consumption. E. Case Study 2: Battery Size Analysis show the road altitude and speed profile from TUM to home. consumption is higher than the energy consumption to go to estimateE. Case energy Study 2: consumption Battery Size in Analysis a short time,The which road is useful slope isto positive in this case. Therefore, energy One important merit of the proposed power model is to The sixth figure shows the power and energy consumption. consumption is higher than the energy consumption to go to estimate energy consumption in a short time, which is useful to TUM.The roadThe driving slope is distance positive going in this to TUM case. is Therefore, longer than energy the iteration-basedOne important parameter merit of sizing. the proposed For example, power we model can find is the to route returning from TUM. However, the energy consumption optimal battery size using iterative vehicleTUM. simulation. The driving Short distance going to TUM is longer than the iteration-based parameter sizing. For example, we can find the consumption is higher than the energy consumption to go to estimate energy consumption in a short time,route which returning is useful from to TUM. However, the energy consumption optimal battery size using iterative vehicle simulation. Short toTUM. go to TUM The driving (27.4 kWh) distance is nearly going 10% to TUM lower is than longer the than energy the simulationiteration-based time parameter is mandatory sizing. in For this example, approach. we canWe find perform the consumption to return from TUM (30.8 kWh) because of road driving simulation on flat 100 km distanceto go with to TUM different (27.4 kWh) is nearly 10% lower than the energy simulation time is mandatory in this approach. We perform route returning from TUM. However, the energy consumption optimal battery size using iterative vehicleconsumption simulation. Short to return from TUM (30.8 kWh) because of road driving simulation on flat 100 km distance with different slope.to go The to TUM proposed (27.4 equation-form kWh) is nearly energy 10% lower model than helps the energy us to batterysimulation pack time size. is We mandatory assume that in the this vehicle approach. speed We is 50 perform km/h, estimate the energy consumption along the road slope and which is average bus speed in suburb of theslope. city. The We proposed assume equation-form energy model helps us to battery pack size. We assume that the vehicle speed is 50 km/h, consumption to return from TUM (30.8 kWh) because of road driving simulation on flat 100 km distanceestimate with the different energy consumption along the road slope and which is average bus speed in suburb of the city. We assume trafficslope. immediately The proposed and equation-form to decide which energy route model is helps economic us to thatbattery the pack battery size. pack We voltage assume is that the the same, vehicle andtraffic speed additional immediately is 50 battery km/h, and to decide which route is economic that the battery pack voltage is the same, and additional battery estimate the energy consumption along the road slope and which is average bus speed in suburb of the city. We assume traffic immediately and to decide which route is economic that the battery pack voltage is the same, and additional battery 200 ADVISOR simulation result Model coefficient extraction result 100

0 Power (kW)

-100 0 500 1000 1500Case Study 2000 1: Fast 2500 Driving Energy 3000 Estimation 3500 Time (s)

Figure 8. PowertrainEstimate model validation power result.consumption in online manner

550

500 Altitude (m) 0 5 10 15 20 25 30 Distance (km) 40

20 Obtain traffic information and estimate vehicle speed

Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Time (s) 200 40 Estimate energy consumption immediately 100 20 0

Power (kW) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) Time (s)

550 25 500 Altitude (m) Figure 9. Routes from home to TUM and vice versa. 0 5 10 15 20 25 30 Distance (km) 40 Table II. A ROUTE INFORMATION. 20

Route Distance (km) Avg. slope (degree) Avg. speed (km/h) Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Home to TUM 31.2 -0.1728 24.67 Time (s) TUM to Home 26.2 0.2221 22.24 200 40

100 20 The simulation results for the routes are performed in 0

Power (kW) 0 Figure 10. The first two figures show the road altitude from 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) home to TUM and the speed profile that we obtained the profile Time (s) from road speed limit and Google Maps traffic information. Figure 10. Simulation result from home to campus (first to third figures) and The third figure shows the corresponding power and energy from the campus to home (fourth and sixth figures). consumption over time. The power consumption is low in the first half compared with another half because the degree of based on the fast simulation results. negative slope along the road is high. Fourth and fifth figures show the road altitude and speed profile from TUM to home. The sixth figure shows the power and energy consumption. E. Case Study 2: Battery Size Analysis The road slope is positive in this case. Therefore, energy One important merit of the proposed power model is to consumption is higher than the energy consumption to go to estimate energy consumption in a short time, which is useful to TUM. The driving distance going to TUM is longer than the iteration-based parameter sizing. For example, we can find the route returning from TUM. However, the energy consumption optimal battery size using iterative vehicle simulation. Short to go to TUM (27.4 kWh) is nearly 10% lower than the energy simulation time is mandatory in this approach. We perform consumption to return from TUM (30.8 kWh) because of road driving simulation on flat 100 km distance with different slope. The proposed equation-form energy model helps us to battery pack size. We assume that the vehicle speed is 50 km/h, estimate the energy consumption along the road slope and which is average bus speed in suburb of the city. We assume traffic immediately and to decide which route is economic that the battery pack voltage is the same, and additional battery 200 ADVISOR simulation result Model coefficient extraction result 100

0 Power (kW)

-100 0 500 1000 1500 2000 2500 3000 3500 Time (s)

Figure 8. Powertrain model validation result.

550

500 Altitude (m) 0 5 10 15 20 25 30 Distance (km) 40

20

Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Time (s) Case Study200 1: Fast Driving Energy Estimation40 100 20 We can pick0 the best way using the up-to-date traffic information

Power (kW) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) Time (s)

550

500 Altitude (m) Figure 9. Routes from home to TUM and vice versa. 0 5 10 15 20 25 30 Distance (km) 40 Table II. A ROUTE INFORMATION. 20 Obtain traffic information and estimate vehicle speed

Route Distance (km) Avg. slope (degree) Avg. speed (km/h) Speed (km/h) 0 0 500 1000 1500 2000 2500 3000 3500 4000 Home to TUM 31.2 -0.1728 24.67 Time (s) TUM to Home 26.2 0.2221 22.24 200 40 Estimate energy consumption immediately 100 20 The simulation results for the routes are performed in 0

Power (kW) 0 Figure 10. The first two figures show the road altitude from 0 500 1000 1500 2000 2500 3000 3500 4000 Energy (kWh) home to TUM and the speed profile that we obtained the profile Time (s) from road speed limit and Google Maps traffic information. Figure 10. Simulation result from home26 to campus (first to third figures) and The third figure shows the corresponding power and energy from the campus to home (fourth and sixth figures). consumption over time. The power consumption is low in the first half compared with another half because the degree of based on the fast simulation results. negative slope along the road is high. Fourth and fifth figures show the road altitude and speed profile from TUM to home. The sixth figure shows the power and energy consumption. E. Case Study 2: Battery Size Analysis The road slope is positive in this case. Therefore, energy One important merit of the proposed power model is to consumption is higher than the energy consumption to go to estimate energy consumption in a short time, which is useful to TUM. The driving distance going to TUM is longer than the iteration-based parameter sizing. For example, we can find the route returning from TUM. However, the energy consumption optimal battery size using iterative vehicle simulation. Short to go to TUM (27.4 kWh) is nearly 10% lower than the energy simulation time is mandatory in this approach. We perform consumption to return from TUM (30.8 kWh) because of road driving simulation on flat 100 km distance with different slope. The proposed equation-form energy model helps us to battery pack size. We assume that the vehicle speed is 50 km/h, estimate the energy consumption along the road slope and which is average bus speed in suburb of the city. We assume traffic immediately and to decide which route is economic that the battery pack voltage is the same, and additional battery Case Study 2: Battery Size Analysis

The optimal battery size is derived using the iteration-based problem solver

Vehicle input parameters

Battery

Motor characteristics Acceptable for loop-based Mechanical specifications Design-time optimization

Transportation input parameters EV power Battery Passenger information consumption SoC simulator Driving input parameters Driving route

Speed and acceleration

27 modules attached in parallel. We assume that the battery pack [2] J. Wang, I. Besselink, and H. Nijmeijer, “Battery electric vehicle is ideally balanced during battery charging and discharging. energy consumption prediction for a trip based on route information,” Proceedings of the Institution of Mechanical Engineers, Part D: Journal We performed the vehicle simulations by changing the of Automobile Engineering, vol. 232, no. 11, 2017, pp. 1528–1542. battery size from 70% of nominal battery size of BYD K9 [3] Y. Zhang, W. Wang, Y. Kobayashi, and K. Shirai, “Remaining driving to to 130%. Figure 11 shows the simulation results. As we range estimation of electric vehicle,” in 2012 IEEE International Electric increase battery size, the driving range also increase. However, Vehicle Conference (IEVC). IEEE, 2012, pp. 1–7. the driving efficiency (energy consumption per unit distance) [4] “BYD Product,” BYD Europe Official Website, [Online] Available: Case Study 2: Batterydecreases because Size of theAnalysis increase in battery weight. The driv- https://www.bydeurope.com [accessed: 2020-05-01]. ing range increases nearly 28% if we increase the battery size [5] A. M. Andwari, A. Pesiridis, S. Rajoo, R. Martinez-Botas, and V. Esfa- by 30%. On the other hand, the driving efficiency decreases up hanian, “A review of battery electric vehicle technology and readiness Estimate below for the same driving profile levels,” Renewable and Sustainable Energy Reviews, vol. 78, 2017, pp. to 21%. Therefore, we should carefully decide the battery size 414–430. Driving range with the consideration of cost of electricity per unit distance [6] R. Zhang and et al., “State of the art of lithium-ion battery SOC Energy consumption(efficiency) and driving and busefficiency service time(Wh (driving per meter) range with a fully estimation for electrical vehicles.” Energies (19961073), vol. 11, no. 7, charged battery). 2018, pp. 1–36. Table 1 [7] Y. Wang and et al., “Research on driving range estimation for EVs Efficiency (Wh/m) based on corrected battery model,” in SAE 2015 World Congress & Battery Range (m) Range Energy Efficiency (Wh/m) Range (km) Exhibition, 2015. weight (km) consumption (Wh) 300 300 (%) [8] A. Bocca, A. Macii, E. Macii, and M. Poncino, “Composable battery 0.90 model templates based on manufacturers’ data,” IEEE Design & Test, 70% 161984 162 We can pick140014 the best battery0.86 size for the electric bus’s purpose vol. 35, no. 3, 2018, pp. 66–72. 80% 184146 184 140758 0.76 225 225 [9] M. Uitz and et al., “Aging of Tesla’s 18650 lithium-ion cells: Corre- 90% 206075 206 141502 0.69 0.68 lating solid-electrolyte-interphase evolution with fading in capacity and power,” Journal of The Electrochemical Society, vol. 164, no. 14, 2017, 100% 227775 228 142245 0.62 pp. A3503–A3510. 110% 249249 249 142989 0.57 150 150

ciency [10] S. Kachroudi, M. Grossard, and N. Abroug, “Predictive driving guid-

0.45 Range 120% 270501 271 143733 0.53 ffi ance of full electric vehicles using particle swarm optimization,” IEEE E Trans. on Vehicular Technology, vol. 61, no. 9, 2012, pp. 3909–3919. 130% 291534 292 144477 0.50 75 75 [11] K. Vatanparvar, J. Wan, and M. A. Al Faruque, “Battery-aware energy- 0.23 optimal electric vehicle driving management,” in International Sympo- sium on Low Power Design (ISLPED), 2015, pp. 353–358. 0 0 [12] D. Baek and N. Chang, “Runtime power management of battery electric vehicles for extended range with consideration of driving time,” IEEE 0.00 70%70% 90% 110% 130%130% Transactions on Very Large Scale Integration System, 2019, pp. 549– 559. Figure28 11. Simulation results by battery size. [13] X. Yuan, C. Zhang, G. Hong, X. Huang, and L. Li, “Method for evalu- ating the real-world driving energy consumptions of electric vehicles,” Energy, vol. 141, 2017, pp. 1955–1968. V. C ONCLUSION [14] A. Gonzalez-Castellanos, D. Pozo, and A. Bischi, “A non-ideal linear We have proposed an improved EV range estimator in- operation model for a li-ion battery,” arXiv:1901.04260, 2019. corporating (i) the parameter extraction for complex EV sim- [15] G. Liu, M. Ouyang, L. Lu, J. Li, and J. Hua, “A highly accurate ulation, (ii) equation-form powertrain modeling by coefficient predictive-adaptive method for lithium-ion battery remaining discharge extraction and (iii) fast vehicle energy simulation with a traffic- energy prediction in electric vehicle applications,” Applied Energy, vol. and altitude-aware driving cycle. We introduce two case studies 149, 2015, pp. 297–314. as application of the proposed range estimator: (i) fast energy [16] Y. Chen and et al., “A systemc-ams framework for the design and simulation of energy management in electric vehicles,” IEEE Access, consumption estimation along the route information and (ii) vol. 7, 2019, pp. 25 779–25 791. bus battery sizing considering driving efficiency and range. [17] D. Baek and et al., “Battery-aware operation range estimation for The estimator can work either offline, by estimating the terrestrial and aerial electric vehicles,” IEEE Transactions on Vehicular range upfront without intermediate updates like a traditional Technology, 2019, pp. 5471–5482. GPS navigator, or online, refining the estimate at the cost [18] T. Markel, A. Brooker, T. Hendricks, V. Johnson, and K. Kelly, “Advisor: a systems analysis tool for advanced vehicle modeling,” of a route re-calculation. Our range estimator is meant as a Journal of Power Sources, 2002. “plug-in” for traditional or traffic-aware (e.g., Google Maps) [19] Y. Chen, E. Macii, and M. Poncino, “A circuit-equivalent battery model GPS navigators, allowing route decisions besides traditional accounting for the dependency on load frequency,” in Proceedings of information based on travel time and route distance. the Conference on Design, Automation & Test in Europe. European Design and Automation Association, 2017, pp. 1177–1182. ACKNOWLEDGMENT [20] Y. Chen, D. J. Pagliari, E. Macii, and M. Poncino, “Battery-aware design exploration of scheduling policies for multi-sensor devices,” in This work was supported by the National Research Foun- ACM/IEEE Great Lakes symposium on VLSI (GLSVLSI), 2018. dation of Korea (NRF) grant funded by the Korea government [21] Google Maps, [Online] Available: https://www.google.com/maps [ac- (MSIT) (No. 2020R1F1A1074843). cessed: 2020-05-01]. [22] “BYD K9,” Land Transport Guru, [Online] Available: https://landtransportguru.net/byd-k9/ [accessed: 2020-05-01]. REFERENCES [23] K. Norregaard, B. Johnsen, and C. H. Gravesen, “Battery Degradation [1] European Automobile Manufacturers Association (ACEA), “Making in Electric Buses,” Danish Technological Institute, Tech. Rep., 2016. the transition to zero-emission mobility,” [Online] Available: [24] “Federal Transit Bus Test,” The Larson Institute, Tech. Rep., 2014. https://www.acea.be/uploads/publications/Study ECV barriers.pdf, June 2018.

1 Conclusion

Electric bus powertrain modeling Conventional vehicle estimator is accurate but slow for online estimation or optimization problem We suggest fast but accurate powertrain modeling Extract vehicle component parameters Use simulation results Extract coefficients of powertrain model

Case studies with the fast powertrain model Estimate power/energy consumption online manner Show the range/energy/efficiency by the battery size Fast powertrain model can be utilized for the various area!

29