sign in sign up
<<
Home , Cognitive science

PROC. 26th INTERNATIONAL CONFERENCE ON MICROELECTRONICS (MIEL 2008), NIŠ, SERBIA , 11-14 MAY, 2008

Cognitive Sensor Networks K. Shenai and S. Mukhopadhyay

Abstract - A smart sensor network that integrates latest e.g., robotics, intelligent networks, smart buildings, advances in cognitive and artificial with low- intelligent traffic , etc. We investigate power wireless sensors is presented for emerging applications in in this paper what the advances in cognitive can energy, agriculture, environment, healthcare, medicine, do to improve the sorry state of affairs that is the current transportation, and defense and space. power grid. By a cognitive power grid we mean an electric power I. INTRODUCTION in which generation, transmission, and distribution are controlled intelligently and adaptively by smart Wall street guru George Gilder says “ distributed communication networks whose elements may quality power is one of the greatest business opportunities consist of wireless or wired sensors (e.g., power factor of our time”. However, over the last thirty years, neglect as meters, thermometers, etc.), actuators (e.g., static VAR well as lack of investment has resulted in the sorry state of compensators, variable voltage transformers, etc.), our current infrastructure for generating and delivering simulators, and other services orchestrated by a network power. Currently, over 60% of the equipments need software layer. Such grids are modulated by a distributed replacement [1]. Despite significant advances in electronics closed loop control system; apart from transforming our and during this time, the North obsolete power grid into a -managing, self-protecting, American power grid (as well as power grids in other self-healing, and survivable one, such control systems continents) still continue to use technology from the attempt to optimize the operation of a power grid with nineteenth and the early twentieth century. objectives including but not limited to minimization of line Electromechanical relays (e.g., Buchholz relays) and loss, voltage fluctuations, illegal usage, peak time load equipments are still the order of the day. On one hand, such imbalance, etc. Today cognitive power grids are still at an obsolete grid presents significant challenges in their infancy. According to Morgan Stanley [1], “Silicon transmitting power reliably and efficiently, on the other will reconfigure the grid. … will make the electric grid a hand the demands of the consumers have kept on smart, efficient, and adaptive system – just as Web and increasing for providing larger amount of energy more telecommunication systems are”. In North America alone, efficiently and reliably at a lower cost. It is to be noted that the power system market is $275 billion [1]; out of that the reliability is a main concern in managing a expansive market for electronics, communication networks, and power grid (consider the 2003 blackout). Today, energy is software for intelligently controlling the grid amounts to one of the principal driving factors of the global $81 billion. Power grids today are characterized by information economy; failure to deliver power reliably and distributed generation; one of the main for this efficiently can have disastrous consequences. With the being the increasing incorporation of renewable sources of proliferation of digital and analog devices being used in energy into the grid such as (photovoltaic, biogas, etc.). mission-critical environments, the need for delivery of high Today’s power system is thus a hybrid one (see Fig. 1) quality power becomes more critical. consisting of different types of sources of energy, different Cognitive sciences will play the same role in the types of transmission systems, along with different types of future that and information sciences have played customers. for the last fifty years. Just as an computer system Merely replacing vintage equipment with new ones processes information, a cognitive system processes without embracing modern technology will not help meet and makes decisions based on it. A cognitive the energy challenge of today. Of course, advances in system is one that can perceive the environment and adapts power electronics over the last thirty years have resulted in to it, can make intelligent decisions based on its knowledge smart equipments that are cheaper, smaller, and more that effect changes in the environment, can self-manage, efficient than previous generation counterparts. For and self-heal. During the past twenty years the artificial example, power electronics-based transformers today are intelligence community has done a tremendous amount of not only more efficient than their previous generation research in providing cognitive capabilities to computer counterparts with copper winding, they can be programmed and communication systems. The results of that research to step up or step down with different ratios. However, are bearing fruit today in diverse areas of human endeavor, installing such next generation equipments is not only K. Shenai is with EECS Department, University of Toledo, extremely capital intensive, it also solves only a part of the OH 43606-3390, USA, E-mail: [email protected] problem. Even with such costly state-of-the-art equipment, S. Mukhopadhyay is with Utah State University, Logan, UT the grid will continue to be inefficient, unless it is 84322-4205, USA intelligently managed. Managing smartly a grid spatially as

978-1-4244-1882-4/08/$25.00 © 2008 IEEE

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on February 4, 2009 at 16:49 from IEEE Xplore. Restrictions apply. Cognitive Distributed Power Grid under rapidly changing environments. It is amenable to dynamic reconfiguration in response to changing requirements without incurring any grid downtime. It is integrated with state-of-the-art provenance management techniques to prevent false triggers of actuating devices. It Biomass uses state-of-the-art data structures from distributed Electrical computing to ensure scalability over an expansive grid. Our Energy Renewable/ Load- Storage approach drastically reduces the hardware cost almost by a Alternate Regulated factor of 10. Energy Electrical Wind to Power & Figure 2 describes a sample scenario for controlling Electric Power Residential the power factor and voltage fluctuation in a power grid. Energy Management Commercial Low power factors can result in increased losses while Conversion Automotive Military voltage fluctuations can damage equipments. Voltage and Solar Space … power factor across the power grid are sampled by installing power factor meters and voltmeters distributed Unregulated Electrical Power spatially across the grid. These meters locally report to motes running cognitive agents that make intelligent Fig. 1. The Cognitive Power Grid control decisions based on the data. Motes can communicate among each other as well as with control expansive as the North American power grid calls for stations. In case the power factor is significantly below integration of state-of-the-art sensing and networking unity in a particular area, the control action might be technologies with cognitive capabilities. switching on a static VAR compensator (capacitive or We present a novel distributed wireless sensor inductive depending on whether the power factor is network-based control system to intelligently and reliably lagging or leading) to correct the power factor. This action manage the operation of large power grids. Our controller is actuated through a driver. Switching on a static VAR combines techniques from cognitive sciences with state-of- compensator may result in voltage fluctuations (with the-art distributed information fusion and networking voltage increasing for a capacitive compensator and technologies. It integrates intelligent sensor coordination decreasing for an inductive compensator). The cognitive and data fusion techniques to access, retrieve, process, and agent, in response to such data from the local voltmeter, communicate with disparate wireless sensors in an ad-hoc will order automatic switching on of a transformer (step up manner to deliver dynamic decisions and provide adequate or step down) depending on the situation. Depending on the . It provides formal guarantees stability characteristics of the grid, the transient effects will that the policies and requirements of the customer will be wear down over time resulting in stabilization of voltage met and QoS (Quality of Service) guarantees such as and power factor at the respective set points. security, fault-tolerance, timeliness, etc. be respected even

Mote

Power Factor Meter Volt Meter

Balance Load to Minimize Loss

Power Grid

Fig. 2. A Power Factor and Voltage Control Scenario

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on February 4, 2009 at 16:49 from IEEE Xplore. Restrictions apply. II. RELATED WORK magnetic induction receiver chip as shown in Figure 3. The net result is a cost reduction from $94 to $10 per mote Significant research has been performed in massively and subsequent high-level affordability. The Java Smart distributed environment-aware computing (also known as Card [7] can be of as a programmable intelligent “swarm computing” [2, 3]), in particular for creating and “plastic card” with an embedded processor on which it is reasoning about swarm programs. Most of these works possible to run a Java kilobyte virtual machine. The card have been focused on developing programming paradigms, can be programmed to perform a specific task by tools, and for swarm computing. EnviroTrack downloading and running small Java applications. It is [4], an object-based distributed system, raises possible to embed crypto coprocessors on the card thereby the level of programming abstraction for distributed sensor assuring security of the applications running on it. We networks by providing a convenient and powerful interface provide a connection between a surface magnetic induction to the application developer geared towards tracking the receiver and the Java card through a serial I/O by placing physical environment. Menezes et al. [3] study different electric contacts on the Java Card processor and connecting abstractions in the field of swarms. In [5], the authors them to the output port of the receiver. During this stage, develop an agent-based framework for simulating power the Java application downloaded on the processor will and communication systems. However, none of these receive incoming signals and will act accordingly by works are concerned with the problem of developing initiating commands to send data and status of received for building sensor-based systems that signal to a transceiver, which will transmit the signal via provide provable guarantees of meeting their requirements. RF .

III. A COGNITIVE SENSOR NETWORK V. SENSOR COORDINATION AND INFO-FUSION Intelligent monitoring and control of power grid Figure 3 shows a schematic architecture for an parameters is achieved through a software tool, called intelligent sensor network. At the lowest level, wireless AUTOMAN (Automated Management) [8,9] that runs on sensors capture power grid characteristics (e.g., voltage, the motes and the base stations providing policy-based, on- power factor, wattage, etc.) from different points of the demand co-ordination of irrigation systems associated with grid. These sensors report to local motes running cognitive different sub-regions of a given region using a network of reactive agents that generate control actions in response to wireless sensors (voltmeters, energy meters, power factor sensor data. The motes are able to communicate with those meters, etc.) and actuators (static VAR compensators, nearby as well as have a multi-hop link to a control station. circuit breakers, etc.). Figure 4 provides a schematic The agents have knowledge of the local policies as well as diagram for the AUTOMAN infrastructure deployed in a the operational requirements of the grid. Global decisions multi-zone irrigation system. The AUTOMAN system can be made at the control stations. provides event-driven cognitive capabilities to the sensor network. Distributed intelligent agents for controlling local

Control Station Control Station segments of the grid are synthesized based on a declarative behavioral specification of the global as well as the Level 2 regional irrigation management policies of the customer in the Secure Operations (SOL) using a deductive

Mote Mote Level 1 DATA SOURCE INTELLIGENT INFORMATION

Sensor Sensor Sensor Sensor Sensor Sensor 1. 1. I I N N T 2. BEHAVIORAL T E MODELING OF E 2. R SERVICE &/OR R F F 3. A INFORMATION A 3. Fig. 3. Architecture of a Cognitive Sensor Network. C C E AUTOMAN E JAVA BASED

M N Many Disparate Intelligent Many Disparate IV. LOW COSTE MOTE Sources Dynamic Receivers Reliable Several Means of Several Means of INFORMATION Information / Control Data Transmission MANAGEMENT We use novel custom-developed wireless surface Transmission motes [6] consisting of a commercial low-power RF transceiver interfaced between a Java smart-card and a Fig. 4. The AUTOMAN framework.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on February 4, 2009 at 16:49 from IEEE Xplore. Restrictions apply. approach. SOL is a synchronous predicates. Then an invariant for a SOL specification [10] for specifying distributed service-oriented systems. It comprising M1 and M2 is deduced by the following proof has capabilities of handling service invocations rule. asynchronously, provides strong typing to ensure enforcement of information flow and security policies, and inv(p1, M1) inv(p2, p1 š M2) has the ability to deal with failures (both benign and inv(p1 š p2, M1 * M2) byzantine) of components. In the synchronous where M1 * M2 is the SOL specification consisting of the , the programmer is provided with state machines M1 and M2. an abstraction that respects the synchrony hypothesis, i.e., one may assume that an external event is processed The agents synthesized from SOL specifications use completely by the system before the arrival of the next the inputs of the sensors and the management policies of event. One might wonder how a synchronous programming the customer to generate real-time control decisions for paradigm can be effective for dealing with widely managing energy flow. The control decisions can either be distributed sensor network systems where there is inherent delivered to appropriate personnel for manual intervention asynchrony. The answer may seem surprising to some, but or directly actuated through wireless actuators. The agents perfectly reasonable to others: It has been shown [11] that can be deployed automatically on a distributed network under certain sufficient conditions (which are preserved in involving sensors tracking soil water content, actuators our case) the synchronous of a SOL application controlling pumps and valves, and diverse computing and are preserved when it is deployed on an asynchronous, communication elements such as PDA’s etc. Reliable distributed infrastructure. Agent behaviors are specified in communication medium between the diverse elements of SOL just as hardware is specified in hardware description the network is provided by the Secure Infrastructure for languages. The specification fragment for a small power Networked Systems (SINS), a formally verifiable factor controller is given below where the event @T(c) for middleware framework, developed as part of the SINS the constraint c is triggered if the truth value of c changes project [14]. Compared with existing techniques from false to true. [2,3,4,15,16], the proposed technique provides formal guarantees that the policies of the customer will be PowerFactor = initially Permitted then respected and the Quality of Service (QoS) goals specified case PREV(PowerFactor) { by them will be met even under changing environments [] Permitted -> like sensors failing, actuators changing their configuration if { or network nodes getting compromised. Changing [] @T(PowerFactor >= High) environments are handled by a dynamic reconfiguration of -> High the co-ordination agents by a monitoring agent. otherwise -> PREV(PowerFactor) Reconfiguration may include substituting new } services/devices for existing ones and can be used to provide new functionalities in response to changing To provide reliability, we formally verify the SOL requirements. Deployed software agents will provide specifications. SOL specifications are formally verified naming, discovery, routing, identifying, and security for using static analysis tools such as theorem provers [12] wireless motes. Location of all the individual nodes can be integrated into the system architecture. A SOL tracked using wireless communication within a range of specification describes a collection of interacting state 350 ft. SINS uses the Spread [17] group communication machines. It provides a behavioral description of an agent tool from Johns Hopkins University as a transport layer. as well as its environment. Given a state machine with state Given a set of services and devices available in a network space S, and transition relation T, we call a function p:So (in this case, underground sensors, surface relays, surface Boolean a state predicate. We say that a state predicate is sinks, base stations, and software and hardware services an invariant if s H S p(s) = true. Safety verification of a running on them), the UPnP [18] models of services and state machine involves showing that a state predicate is an devices (available as a UPnP from the invariant for that machine. To establish that a state manufacturers) are automatically compiled to formal predicate is an invariant we use induction: (1) show that interface specifications. The resulting formal specifications p(init) = true for all initial states init of S, and (2) prove that are used to populate a database called the Master Directory. s s’H S T(s,s’) š p(s)= true op(s’)=true. The Master Directory provides an interface for querying To show that a state predicate is an invariant for a and updating records. In the event of any change in the SOL specification, we use assume -guarantee reasoning configuration of the services/devices (e.g., sensor running [13]. Let us denote by inv(p,M) the fact that the state out of power), the nearest network node is wirelessly predicate p is an invariant for the state machine M. Let informed of the changes through the SINS transport layer. pšM be the restriction of a state machine M to those states The Master Directory is then updated through its interface. where the state predicate p holds. Assume that M1 and M2 The Master Directory is replicated for fault-tolerance with are two state machines and let p1 and p2 be two state standard techniques used to maintain consistency. The

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on February 4, 2009 at 16:49 from IEEE Xplore. Restrictions apply. business goals of the application, domain-specific policies may include substituting new services/devices for for security, situation-awareness, failure handling [19], and existing ones and can be used to provide new real-time response are input to a mission planner running functionalities in response to changing requirements. on a networked server that is equipped with an interface for x Cost: Our approach reduces the price of motes and querying the Master Directory. The mission planner uses thus allows affordable deployment of cognitive sensor application-specific business goals and policies as well as networks over a spatially distributed expansive power the information on available devices in the network (query grid. on Master Directory) to automatically synthesize a set of agents [8,9] and their deployment information using intelligent deduction. The synthesis of the agents is done in REFERENCES a way that guarantees that the business goals of the application as well as the QoS constraints (e.g., security, [1] http://www.globalenvironmentfund.com/Emerging_Smart_Gri fault-tolerance, etc.) are respected. The agents are then d.pdf. compiled to Java byte-code and automatically deployed on [2] D. Evans, Programming the Swarm, NSF C. A. Proposal. SINS virtual machines (SVMs) running on different hosts [3] R. Menezes, R.Tolksdorf: “A New App. to Scal. Linda- in the network (base stations, surface relays, surface sinks) systems Based on Swarms. “SAC 2003: 375-379, 2003. using the generated deployment information. SVMs can [4] T. Abdelzaher, B. Blum B, Q. Cao, Y. Chen, D. Evans, J. communicate among themselves and can interact with George, S. George, L. Gu, T. He, S. Krishnamurthy, L. Luo, physical devices through drivers for actuating reactions S. Son, J. Stankovic, R. Stoleru and A. Wood, EnviroTrack: generated by processes running in it and for receiving Towards an Environmental Computing Paradigm for Distributed Sensor Networks, The 24th International alerts. They have a small footprint and can run on Java Conference on Systems. Tokyo, Kilobyte Virtual Machines (KVMs), thus allowing Japan. March 23-26, 2004. deployment on resource-constrained surface relays and [5] K. Hopkinson et. al. EPOCHS: Integrated Cots Software For sinks. Monitoring agents running in SVMs can detect Agent-Based Electric Power And Communication possible violation of policies and take actions for , IEEE Transactions on Power Systems. immediate recovery. [6] K. Shenai, N. Jack, and S. Mukhopadhyay, “Novel Low Cost Hardware and Management Software for Wireless Sensor Our approach results in the following benefits: Networks”, IP Disclosed to TCO, USU, 2007. x Formal Guarantees: This is the uniqueness of the [7] Wikipedia – Internet Encyclopedia, 2007. [online]. Available: presented approach. It formally guarantees that http://en.wikipedia.org/wiki/Java_Card [Accessed September irrigation goals of the user are met, QoS constraints are 11, 2007]. respected and none of the policies imposed by the [8] S.S. Yau, H. Davulcu, S. Mukhopadhyay, et. al. “Automated individual services/devices or the application are Synthesis of Workflows for Service-Based Systems”, To violated even under changing environments. Appear in the International Journal on Web Services Satisfaction of the non-functional safety specifications Research, 2007. 13. S. S. Yau, H. Davulcu, S. Mukhopadhyay, D. Huang, Y.Yao: Adaptable Situation- is also guaranteed. Aware Secure Service-Based (AS3) Systems. ISORC 2005: x Automatic synthesis of agents: Our technique allows 308-315, 2005. synthesis of intelligent distributed agents from [9] R. Bharadwaj, S. Mukhopadhyay, N. Padh: Service declarative specification of the business , QoS Composition in a Secure Agent-Based Architecture. EEE goals, and formal models of available devices and 2005: 787-788, 2005. services. [10] A. Benveniste, P.Caspi, S. A. Edwards, N. Halbwachs, P. Le x Automatic deployment: Our approach allows Guernic, R. de Simone: The synchronous languages 12 years automatic deployment of the bytecode obtained by later. Proceedings of the IEEE 91(1): 64-83 (2003). [11] R. Bharadwaj, S. Mukhopadhyay, From Synchrony to SINS, compiling an agent to run on a SVM specified by the Technical Report Utah State University, 2006. deployment information. [12] R. Bharadwaj and S. Sims. Salsa: Combining constraint x Security and other QoS goals: The programming solvers with BDDs for automatic invariant checking. In Proc. model for SOL as well as the implementation of SINS 6th International Conference on Tools and for the encompasses an inbuilt security model for providing Construction and Analysis of Systems (TACAS'2000), security guarantees, which are essential for irrigation ETAPS 2000, Berlin, March 2000. control in a multi-zone implementation. [13] P. Maier, A Lattice-Theoretic Framework for Circular Assume Guarantee Reasoning, PhD Thesis, MPI, 2005. x Maintaining data provenance: Our approach [14] R. Bharadwaj, "Verifiable Middleware for Secure Agent provides a mechanism for certifying the Interoperability," Proc. 2nd IEEE W. on Formal Approaches to provenance/pedigree of the data exchanged between Agent-Based Systems (FAABS II), 2002, pp. 126-132. processes and helps in preventing generation of [15] Harold Abelson, Don Allen, Daniel Coore, Chris Hanson, spurious triggers for activating services/devices. George Homsy, Thomas F. Knight, Radhika Nagpal, Erik x Reconfigurability: It is possible in our approach to Rauch, Gerald J. Sussman, Ron Weiss: Amorphous reconfigure an agent at runtime. Such reconfiguration Computing. Commun. ACM 43(5): 74-82, 2000.

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on February 4, 2009 at 16:49 from IEEE Xplore. Restrictions apply. [16] D. Gay, P. Levis, R. von Behren, M. Welsh, E. Brewer, and Technical Report, Johns Hopkins University. D. Culler, The nesC Language: A Holistic Approach to [18] http://www.upnp.org. Networked Embedded Systems, In Proceedings of [19] S. S. Yau, S. Mukhopadhyay, R. Bharadwaj: Specification, Programming Language Design and Implementation (PLDI) Analysis and Implementation of Architectural Patterns for 2003, June 2003. Dependable Software Systems. WORDS 2005: 197-204, [17] Y. Amir, C. Danilov, M. Miskin-Amir, J. Schultz, J. Stanton, 2005. “The Spread Toolkit: Architecture and Performance”

Authorized licensed use limited to: UNIVERSITY OF NEW MEXICO. Downloaded on February 4, 2009 at 16:49 from IEEE Xplore. Restrictions apply.