Validation of the MEASURE Automobile Emissions Model: a Statistical Analysis

Validation of the MEASURE Automobile Emissions Model: a Statistical Analysis

Validation of the MEASURE Automobile Emissions Model: A Statistical Analysis IGNATIUS FOMUNUNG Clark Atlanta University SIMON WASHINGTON RANDALL GUENSLER WILLIAM BACHMAN Georgia Institute of Technology ABSTRACT This paper details the results of an external vali- dation effort for the hot stabilized option current- ly included in the Mobile Emissions Assessment System for Urban and Regional Evaluation (MEASURE). The MEASURE model is one of sev- eral new modal emissions models designed to improve predictions of CO, HC, and NOx for the on-road vehicle fleet. Mathematical algorithms within MEASURE predict hot stabilized emission rates for various motor vehicle technology groups as a function of the conditions under which the vehicles are operating, specifically various aggre- gate measures of their speed and acceleration pro- files. Validation of these algorithms is performed on an independent data set using three statistical criteria. Statistical comparisons of the predictive performance of the MEASURE and MOBILE5a models indicate that the MEASURE algorithms provide significant improvements in both average emission estimates and explanatory power over MOBILE5a for all three pollutants across almost every operating cycle tested. In addition, the MEASURE model appears to be less biased, the most critical model performance measure for point-estimate forecasts, than MOBILE5a. Ignatius W. Fomunung, Department of Engineering, Clark Atlanta University, Atlanta, GA 30314. E-mail: [email protected]. 65 INTRODUCTION and Shih 1996). Despite this recognition, few Emission rate model uncertainties in currently advances have been made in quantifying the effect employed regional emissions models arise in part of driving behavior on emissions, except for Shih et because emission rates rely primarily on average al. (1997), who used throttle position distributions speed as the dominant, continuous, independent to represent driver behavior, albeit with mixed variable in the regression analysis. However, many results. Their research provides evidence that throt- factors, both continuous and discrete, in addition tle position distributions might be used to reflect to average speed, affect the net load demanded of differences in driving behavior, but such models an engine, which in turn affects a vehicle’s resultant still need refinement. The forecasting of throttle emissions. These factors include roadway grade, position distributions, which interact with specific rolling resistance, aerodynamic drag, engine speed, driver types, facility types, and trip purposes, may engine friction losses, transmission losses, vehicle prove too difficult. mass, power consumption of accessories, and so Emerging Models forth. Numerous references identify these factors as influential in the formulation of various pollu- Efforts at improving motor vehicle emissions have tants; however, they are largely omitted in current- occupied researchers for quite some time. Cadle et ly employed emission prediction algorithms al. (1997) recently summarized advances in real- (Guensler 1993). world motor vehicle emissions modeling. The U.S. Cicero-Fernandez et al. (1997a; 1997b) demon- Environmental Protection Agency (EPA) is current- strated that emissions from an individual vehicle ly revising the MOBILE5a emissions rate model. may increase by a factor of two when driven on an MOBILE6 promises to provide significant improve- uphill grade, yet current inventory models do not ments in terms of representing modal impacts on account for grade. In addition, real-world driving emissions rates because supplemental driving cycles conditions, in terms of speed/acceleration distribu- that mimic on-road conditions under various levels tions and/or traces, are not well represented in the of congestion are being used to develop cycle-based current models. The Federal Test Procedure (FTP), speed correction factors. New certification testing appropriately used to develop baseline emissions cycles also promise to reduce the frequency of on- factors, does not capture the extremes of emission- road enrichment. The USO6 cycle represents emis- producing activities associated with aggressive dri- sions in aggressive driving, and the SCO3 cycle ving. Jimenez-Palacio (1999), using a new reflects the effects of accessory loads like air condi- definition of specific power, calculated the maxi- tioning usage, power steering, and so forth. mum specific power of the FTP to be approximate- Modal modeling approaches are also currently ly 22 kilowatts per metric ton. More telling, the under development. A modal emissions model research indicates that the onset of commanded being developed at the University of California, enrichment for many vehicles occurs at this maxi- Riverside by An et al. (1997) is based on 300 vehi- mum. Commanded enrichment is responsible for cles tested under a variety of laboratory driving elevated or “super” emissions, which can be one to cycles. Two modal approaches developed at the several orders of magnitude higher than emissions Georgia Institute of Technology are included in the obtained under stoichiometric engine operation. As GIS-based modal emission model: an aggregate a result, a large proportion of commanded enrich- modal model based on statistical analysis of his- ment is not likely to appear under the FTP. toric laboratory data (Guensler et al. 1998) and a Driver behavior may also be an important load-based prediction module based on analysis of source of uncertainty and variability in motor vehi- instrumented vehicle data (Rodgers 1995). cle emissions (Bishop et al. 1996). A study of For the past six years, the Georgia Tech Re- repeated measurements on the IM240 driving cycle search Partnership has been developing a research- indicates that driver behavior may be responsible grade motor vehicle emissions model within a for potentially order-of-magnitude differences in geographic information system (GIS) framework. emissions for clean low-emitting vehicles (Webster Once validation and peer review efforts are com- 66 JOURNAL OF TRANSPORTATION AND STATISTICS SEPTEMBER 2000 plete, MEASURE may serve as an alternative or sup- option currently included in MEASURE. The perfor- plement to the current MOBILE5a model. The mances of MEASURE and MOBILE5a are compared aggregate modal model within MEASURE predicts using mean absolute prediction errors, linear corre- emissions from light-duty vehicles. The Georgia lation coefficients between observed and predicted Tech aggregate modal model predicts emissions as a emissions, and mean prediction errors. Results are function of vehicle operating mode, representing a provided for each driving cycle and for vehicle tech- spectrum of vehicle operating conditions including nology classes. cruise, acceleration, deceleration, idle, and the power demand conditions that lead to enrichment, MEASURE AGGREGATE MODAL MODELS that is, high fuel to air ratios. The model accounts In the context of this paper, the term “model” for interactions between specific vehicle fleet charac- refers to a mathematical algorithm or expression teristics and vehicle operating modes. For each tech- that relates emissions measurements to various nology group within a light-duty motor vehicle fleet, explanatory variables. The model estimation data the relationships between modal activity and emis- consisted of more than 13,000 laboratory tests sions can differ significantly. The framework allows conducted by EPA and the California Air Re- for facility-level aggregations of microscopic traffic sources Board (CARB) using standardized test simulation or disaggregation of traditional macro- cycle conditions, as well as alternative driving scopic four-step travel demand forecasting models to cycles (Fomunung et al. 1999). The aggregate develop emission-specific vehicle activity data. modal model algorithms presented below were The aggregate modal model within MEASURE estimated using the logarithm of the emission rate employs emission rates based on theoretical ratio for each pollutant as a response variable engine-emissions relationships that have been (Fomunung et al. 1999). The ratio is the emission modeled using various statistical techniques rate (in grams per second) (g/sec) for a vehicle dri- (Fomunung et al. 1999). The emissions rate mod- ven on a given cycle (or equivalently across a els have been estimated through a process that uti- specified speed/acceleration matrix), divided by lizes the best aspects of hierarchical tree-based that vehicle’s emissions rate (g/sec) obtained from regression (HTBR) (Breiman et al. 1984) and ordi- the FTP bag 2 testing cycle. MEASURE’s nary least squares (OLS) regression. The relation- Aggregate Modal Model predicts the ratio of ships are dependent on both modal and vehicle g/sec emission rates for several vehicle technology technology variables, and they are “aggregate” in groups. The following sequence of equations the sense that they rely on bag data to derive their shows the method for calculating the predicted modal activities (Washington 1994). Thus, they are emissions rates for each pollutant in units of suitable for existing aggregate approaches con- ~ either g/sec (␺) or g/mile (␺ ): tained within the travel demand modeling (TDM) ␺ ␺~ ϫ framework. i = i DIST / DUR (1), ␺ ␺~ ϫ Although much effort has been conducted and ibag2 = ibag2 3.91/866 (2), reported in the literature on the emission algorithms and within MEASURE, little has been done toward the R = P / ␺ (3). external validation of the MEASURE emissions pre- i i ibag2

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    20 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us