- Email: [email protected]
Development of a multi-agent management system for an intelligent charging network of electric vehicles
Proceedings of the 18th World Congress The International Federation of Automatic Control Milano (Italy) August 28 - September 2, 2011
Development of a multi-agent management system for an intelligent charging network of electric vehicles ? Jo˜ ao Miranda ∗ Jos´ e Borges ∗ M´ ario J. G. C. Mendes ∗∗ Duarte Val´ erio ∗ IDMEC/IST, TULisbon, Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal (e-mail: [email protected], [email protected], [email protected]). ∗∗ IDMEC/ISEL-Instituto Superior de Engenharia de Lisboa, Rua Conselheiro Em´ıdio Navarro 1, 1959-007 Lisboa, Portugal (e-mail: [email protected]) ∗
Abstract: This paper addresses the modelling and simulation of a battery charging infrastructure for electric vehicles, with the objective of pro-actively scheduling the charging of up to fifty vehicles so as not to overcharge the electrical network. Benefits of having the charging stations differ (as much as possible while satisfying end-user requirements) battery charging for those hours when electricity consumption is otherwise low include rendering electricity consumption more uniform along the day. A multi-agent system was used to design a distributed, modular, coordinated and collaborative multi-agent management system for this infrastructure. Simulation results show the effectiveness of this approach under the conditions of four real-life scenarios. Keywords: Modeling and simulation of power systems, Intelligent control of power systems, Smart grids, Energy Management Systems, Multi-Agent Systems, Electric Vehicles 1. INTRODUCTION Electric Vehicles (EV) provide an alternative technology allowing diversity in energy sources for transportation, therefore reducing oil dependency, while providing mobility at competitive costs. Independent market analysis by Gartner and Wheelock (2010) forecasts a steep expansion of EV market in the coming 10 years. The charging infrastructure for EV will increase stress on the power grid and, therefore, needs to be managed in order to comply with both charging needs and end-client satisfaction, while avoiding disruptions in the electricity distribution. The management and control of electricity distribution is a complex problem for both the utility grid and small scale distribution networks. The dynamics within these networks are the result of interactions between several individual components and objectives: on one hand we have the grid infrastructure, which includes hardware and software, on the other hand we have the objectives of all stakeholders associated with the distribution chain, which include electricity producers, owners of transmission lines, retailers and end-clients. In order to comply with all these, sometimes conflicting, interests, a major research effort has been put towards the development of smart grids that can provide management and control at the network infrastructure level, while delivering high quality services in the electricity distribution chain. ? This work was supported by FCT, through IDMEC, under LAETA.
978-3-902661-93-7/11/$20.00 © 2011 IFAC
Therefore, it is imperative to implement autonomous and decentralized management platforms that are capable of coordinating charging operations while ensuring safety, preserve efficiency and smartly manage the distribution of electricity for the EV batteries. Additionally, there is a need of suitable management software for distribution that can deliver extended services and support new paradigms of business models for the electricity retailers. The importance of this problem will increase as the market share of EV increases. In order to implement an autonomous and decentralized management system to charge the batteries of EV, new strategies are needed for this implementation and a good option is to use Distributed Artificial Intelligence (DAI) techniques. In particular, methodologies based on multiagent systems (MAS) are a good choice to design a distributed, modular, coordinated and collaborative multiagent management system for an intelligent charging network of electric vehicles. This paper is organised as follows: section 2 presents the layout of the system, section 3 its modelling employing a MAS, section 4 its implementation in Matlab, and section 5 the simulation results obtained. Conclusions are drawn in section 6. 2. SYSTEM LAYOUT This paper addresses a battery charging station for electric vehicles (Borges et al., 2009), assumed to have fifty charging points for as many vehicles. Such stations will be
12267
10.3182/20110828-6-IT-1002.03492
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
implemented in parking lots, and so parked cars may or may not be charging in each instant (because the battery is already full, or because too many cars are trying to charge their batteries at the same time). Different neighbourhoods (e.g. residential, commercial or office areas) are expected to lead to very different electricity demands. Users will pay electricity according to a previously chosen profile, that dictates the car’s priority in obtaining electricity: cheaper prices will correspond to lower priorities and hence to longer charging times; to ensure shorter charging times and thus higher priorities the price paid for the electricity will be higher. The objective is to satisfy charging requests, respecting the hierarchy of priorities defined by what each client is paying, and all this without overcharging the electrical network. Indeed, the idea is that charging stations will, to the maximum extent possible, differ battery charging for those hours when electricity consumption is otherwise low. In this manner, peak hours when electricity consumption is higher (typically during the day) will not be further strained, while those periods of the day (typically, these are rather periods of the night) when electricity could be cheaply available will be put to good use. The benefits of thus rendering electricity consumption more uniform along the day could be enormous. To achieve this a two level decision system was conceived: 1) charging stations must communicate with each other and decide, taking into account the expected number of cars in the next few hours and their charging priorities, how the available electricity will be split among them; 2) each charging station must then distribute the electricity it has been allocated among the cars therein parked. Trying to sort things out taking into account every single individual car in all the charging stations of a large geographical region would likely prove to be impossible from the computational point of view. This would also be very liable to fail because of communication failures. In this way, a high level distribution will first be performed, for some large time interval, and then each charging station will be left with some autonomy to deal with its local conditions, allocating electricity to particular vehicles, with the ability of changing the allocation in shorter time intervals, and being thus flexible to deal with unforeseeable fluctuations in demand. This paper only considers this second level, assuming that the first has already been carried out, and that the electricity available in each instant has been determined. Thus the layout corresponds to that of figure 1, where the Local Power Manager (LPM) is responsible for the charging station under study. Further assumptions are as follows: • the base price of energy was taken from the regulated market defined by legislation; • vehicle-to-grid and vehicle-to-home concepts (whereby cars may, if that is found profitable, provide energy to the grid or to the owner’s home (Botsford and Szczepanek, 2009)) were not considered; • no fast charge of batteries (whereby batteries can reach 80% of their maximum capacity in only 30 minutes (Botsford and Szczepanek, 2009)) was considered;
Fig. 1. System layout • batteries were considered to allow interrupted charging without damaging performance. 3. MULTI-AGENT MODELLING In face of the above, the concept of agent (without access to data from all the system, but only partial information from a small part of the system) comes as a solution to this problem. MAS are an “artificial social system” (Wooldridge, 2002), composed by individual agents able to decide autonomously and with “social” capabilities (communicating and collaborating with each other), in order to achieve their own design objectives. MAS are used in several different research, operation and business areas. Macal and North (2006) suggest its use in areas as business and organisation, economics, society and culture, military and biology. Applications based on MAS are also being introduced to fault tolerant control (Hines and Talukdar, 2003; Mendes, 2008) and largely used in power networks (Negenborn, 2007) and its robustness is proved with the applications in military services (Phillips et al., 2006). In the particular plant addressed in this paper, MAS provide a very efficient tool, since we are dealing with a complex system, with many different ramifications (the cars) that do not depend on each other. MAS allow us to develop a managing platform that would otherwise deal with too much information, by decentralising it, and gaining advantage of DAI flexibility. A two layer MAS was elaborated, where two different design objectives are implicit: one layer is responsible for managing the park, decide on which car to charge and ensure the power limit will not be exceeded; the second layer will be composed by agents, each one of them responsible by a car with the objective of charging it according to the costumer needs. The communication between those two layers is established by an auction, described below in section 4.3. 3.1 Macro-level architecture Due to the specific characteristics of a distributed electric charging system, the MAS macro-level architecture proposed in this work is a federated architecture (shown in figure 2), since it allows the natural common relations and communications between the different charging agents, where the different charging point (CP) agents only need to communicate with a Facilitator, and not between them. The LPM agent takes on the role of Facilitator, while the CP agents have no direct communication between them, being coordinated by the LPM Agent, which will conduct an auction process, by sending data to the CP
12268
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
Fig. 2. Multi-Agent System: Macro-level architecture
Fig. 4. Charging Point Agent Architecture
Fig. 3. Local Power Manager Agent Architecture Agents, receiving back the bids and assigning the charging confirmations according to the Auction inference system (IS) rules. 3.2 Micro-level architectures
Fig. 5. Evolution of the battery’s SOC during the charge process
The design of micro-level agent architectures should represent and define individual agents’ behaviour and interactions.In this system, both the LPM agent and the CP agents should be intelligent agents to provide the necessary performance in the charging process. To achieve some different degree of intelligence and to correctly design the two different behaviours of these agents, two different types of micro-level architectures have been designed. The LPM agent will be responsible for managing the car parking lot and in particular the communications with the CP agents, providing data for the auction from the electricity stake-holders such as energy availability and prices. The architecture for the LPM agent is shown in figure 3. The communications layer is constituted by the external link (responsible for the communications with the exterior of the park) and the CP link (responsible for the communications with the CP agents). In the operations layer, the data base conceptually represents an information buffer and connections. The decision making module will perform the negotiations with the other LPM agents (not addressed in this paper). The auction IS will manage the auction process and assign the charging confirmations to the respective CP agents. Both will be conditioned by the goals module, that represents the overall objectives of the LPM Agent. The CP agent will be responsible for ensuring maximum battery charge, respecting the client profile. Decisions on bids offered during the auctions will be supported by data information provided by the car in real-time, by the LPM agent, and by a model of the battery. CP agent architecture is presented in figure 4. The communications layer is constituted by the LPM link (responsible for exchanging information with the LPM agent), the sensor (to receive information from the EV battery) and the actuator (that sends the charge confirmation). The operations Layer con-
tains a data base (with the same function as that in the LPM agent), the bid decider (to perform the auction bid), and the battery model evaluation (to evaluate battery charging behaviour). 4. IMPLEMENTATION 4.1 Battery The battery charging process is non-linear and was modelled as shown in figure 5, that gives the battery’s state of charge (SOC, the ratio between energy stored and maximum capacity) as a function of time for an uninterrupted charge. This behaviour was implemented with a TakagiSugeno fuzzy logic model (Michels et al., 2007), so that in the future it will be possible to make use of past data on the evolution of the SOC with time during the charge of each particular battery, adopting different models for different batteries. Thus it will be taken into account that not all batteries charge in the same way, because of their past history and different wear off. This adaptive modelling has not been implemented yet. 4.2 Multi-agent platform To implement the multi-agent system described in section 3, the FTNCS-MAS Designer toolbox (Mendes et al., 2009) for Matlab was used. This toolbox allows configuring macro-level architectures simply, automatically implementing the communications between agents allocated in different computers with the User Datagram Protocol (UDP). It was chosen because, while designed for Fault Tolerant Networked Control problems as the name stands for, it is very versatile and suits very well the problem under study. The agents were distributed over five personal
12269
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
Fig. 6. Agent distribution over the five computers computers, each containing 10 CP Agents an as shown in figure 6.
Fig. 7. Sorting of bids and honoured requests
Because the UDP (unlike the Internet Protocol Suite (TCP/IP), for instance) does not ensure that information sent by a computer is effectively received by another (which means that the data packages send in every instant could be lost), the following strategies were followed:
client’s profile), tfull is the time needed to reach SOC= 100% under continuous charge (as computed by the battery model), and ffull is a monotonously decreasing function; • T˜50 = f50 ttleft , where t50 is the time needed 50
• a sampling time of 1 second was used, which is enough, and avoids heavy network traffic; • flag variables were sent over 20 seconds (the probability of loosing data every single second during these 20 seconds is neglectably low); • variables were fed back after being received, to ensure no communication failures.
to reach SOC= 50% under continuous charge (as computed by the battery model), and f50 is a monotonously P3decreasing function; P50 i 1 t PT +t −PT 50 i=1 left t=1 ˜ pt = fpt • H , , where PT +t PT tleft is the electricity base price in the future time instant T + t (hence the first fraction reflects how the price of electricity will evolve with time), tileft is the time the car in charging point i is expected to remain plugged (which information, according to the client’s profile, is given by the LPM, and hence the second fraction reflects the relation between the time left for each car and the average of the time left for all others), and fpt is a function that monotonously increases with both its independent variables; • and α, β and γ are coefficients that depend on the client’s profile.
Furthermore, the RT block toolbox (Daga, 2008) was used to ensure that all agents run in a constant ratio between real time and simulation time (otherwise some batteries might take so long to stop charging when ordered to, while others were already starting to charge, that there would be very short but significant energy consumption peaks, for about 1 sampling time). 4.3 Auction CP agents interact by an auction, as mentioned above in section 3, so as to sort out among them which vehicles are being charged in each instant. This auction is organised by the LPM every 15 minutes (this period was chosen because it is the frequency with which available information is updated by the Portuguese electric grid owner), and each CP agent makes a bid B for the energy C15 that it can charge during the next 15 minutes. These bids bear no relation to the price the client is actually paying, being solely values used internally by the system. The LPM sorts the bids by price as seen in figure 7, and honours as many requests as possible, by order of bid, up to the limit imposed from grid conditions. (As mentioned in section 2, this limit will in the future be determined by a negotiation between the LPMs of different charging stations.) Bids are reckoned as ˜ pt B = PT 1 + αT˜full + β T˜50 + γ H
Functions ffull, f50 and fpt are currently linear and were implemented with a Takagi-Sugeno fuzzy logic model, so that in the future it will be easy to render them non-linear, optimising the relation according to what experience shows more useful. 5. RESULTS There being as of writing no market studies to substantiate what the best commercial strategy for a charging stations might be, three different profiles that appeared reasonable were hypothesised:
(1)
where
• PT is the electricity base price (given by the LPM) for the current time instant T ; • T˜full = ffull ttleft , where tleft is the time the car full is expected to remain plugged (according to the 12270
(1) priority client, willing to pay much to fully charge the battery as soon as possible at all times (such clients will expectedly be a minority); (2) client who cannot charge at home and must thus charge during the day, when prices are higher, with the objective of reaching SOC=50% paying as less as possible; (3) client who can charge at home and thus does not need to charge during the day, as long as the battery is above SOC=50%.
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
Table 1. Coefficients for (1) for each profile
α β γ
1 1.0 0.0 0.0
Profile 2 3 0.3 0.4 0.3 0.6 0.4 0.0
These profiles, corresponding to the parameters of (1) given in table 1, were used to carry out tests in four different scenarios: (1) an office zone, where the affluence is significant from 8 AM to 19 PM, and most clients chose profile 2; (2) a residential zone, where the affluence is the opposite of that of scenario 1, and most clients chose profile 3; (3) a commercial zone, with high rotation of cars in the 12 AM to 12 PM period, and mixed profiles; (4) a car silo, with high affluence at all hours, and mixed profiles, the mix depending on the hour (at night there will be more profile 3 clients, during day hours profile 2 will be more significant). Car affluence is shown for the four scenarios in figure 8, which also shows how many cars are effectively able to charge at each instant. Figure 9 shows power availability assumed for each scenario (after negotiation with other LPMs). Figure 10 shows how many clients of each profile were assumed for each scenario, how their batteries were charged when arriving at the park, and how they were charged when leaving. These results show that the objectives of each profile are being satisfied in each case; the ease with which that is done depends, of course, on the constraints of each scenario (e.g. in scenario 3, profile 1 cars cannot often be fully charged because they seldom remain long enough). It seems thus reasonable to assume that the modelled system satisfies its requirements. When actual data on car affluence, available energy, clients’ expectations, etc. becomes available, the flexibility of the implementation is expected to allow adapting the system promptly. 6. CONCLUSIONS Distributed charging systems for EV can be modelled using MAS. We have discussed the main features of such model, namely both its macro- and micro-level architectures, and implementation issues, such as an auction mechanism. The MAS model that has resulted complies with load constraints imposed by the grid while attempting to maximise the number and charge of plugged-in EV in the system. Three different profiles, which characterise groups of typical EV owners, were defined for the CP Agents. The model was simulated under conditions of four reallife scenarios in order to understand both the distribution network dynamics and the EV affluence in the parking lot, therefore providing a customisable tool to support the design and dimensioning, as well as assessing the economic viability, of such type of charging infrastructures. REFERENCES Borges, J., Borges, A., Campos, J., and Nina, M. (2009). Sistema de distribui¸ca ˜o e disponibiliza¸ca ˜o de energia el´ectrica para carregamento de baterias de ve´ıculos
Fig. 8. Car affluence and batteries charging for scenarios 1 to 4(from to top to bottom) el´ectricos (in Portuguese). 104395. Portuguese Institute of Industrial Property. Botsford, C. and Szczepanek, A. (2009). Fast charging vs slow charging : Pros and cons for the new age of electric vehicles. EVS24 International Battery, Hybrid and Fuel
12271
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
100
SOC / %
80 60 40 20 0 at when at when at when arrival departing arrival departing arrival departing profile 1 (7%) profile 2 (62%) profile 3 (32%) 100
SOC / %
80 60 40 20
Fig. 9. Power availability assumed for scenarios 1, 3 and 4 (top) and for scenario 2 (bottom)
at when at when at when arrival departing arrival departing arrival departing profile 1 (22%) profile 2 (0%) profile 3 (78%) 100
SOC / %
80 60 40 20 0 at when at when at when arrival departing arrival departing arrival departing profile 1 (13%) profile 2 (87%) profile 3 (0%) 100 80 SOC / %
Cell Electric Vehicle Symposium, 1–9, Norway. Daga, L. (2008). URL http://leonardodaga.insyde. it/Simulink/RTBlockset.htm. Gartner, J. and Wheelock, C. (2010). Electric vehicle charging equipment. Technical report, Pike Research. Hines, P. and Talukdar, S. (2003). Autonomous agents and cooperation for the control of cascading failures in electric grids. Proceedings. 2005 IEEE Networking, Sensing and Control, 2005., 273–278. Macal, C. and North, M. (2006). Tutorial on agentbased modeling and simulation part 2: how to model with agents. Proceedings of the 2006 Winter Simulation Conference, 73–83. Mendes, M.J.G.C. (2008). Multi-Agent Approach to Fault Tolerant Control Systems. Ph.D. thesis, Instituto Superior T´ecnico - Universidade T´ecnica de Lisboa. Mendes, M.J.G.C., Santos, B.M.S., and S´ a da Costa, J. (2009). A Matlab/Simulink multi-agent toolkit for distributed networked fault tolerant control systems. In SAFEPROCESS’2009, Preprints of the 7th IFAC Symposium on Fault Detection, Supervision and Safety of Technical Processes, 1073–1078, Barcelona, Spain. Michels, K., Klawonn, F., Kruse, R., and N¨ urnberger, A. (2007). Fuzzy Control: Fundamentals, Stability and Design of Fuzzy Controllers. Springer, Berlin. Negenborn, R.R. (2007). Multi-Agent Model Predictive Control with Applications to Power Networks. Ph.D. thesis, Technische Universiteit Delft. Phillips, L., Link, H., Smith, R., and Weiland, L. (2006). Agent-based control of distributed infrastructure resources. Technical Report January, U.S. Department of Commerce: National Technical Information Service. Wooldridge, M. (2002). An Introduction to Multi-Agent Systems. John Wiley & Sons, Ltd. Chichester, England.
0
60 40 20 0 at when at when at when arrival departing arrival departing arrival departing profile 1 (6%) profile 2 (39%) profile 3 (55%)
Fig. 10. SOC of batteries when arriving and when leaving the charging station for scenarios 1 to 4 (from to top to bottom)
12272
Development of a multi-agent management system for an intelligent charging network of electric vehicles ? Jo˜ ao Miranda ∗ Jos´ e Borges ∗ M´ ario J. G. C. Mendes ∗∗ Duarte Val´ erio ∗ IDMEC/IST, TULisbon, Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal (e-mail: [email protected], [email protected], [email protected]). ∗∗ IDMEC/ISEL-Instituto Superior de Engenharia de Lisboa, Rua Conselheiro Em´ıdio Navarro 1, 1959-007 Lisboa, Portugal (e-mail: [email protected]) ∗
Abstract: This paper addresses the modelling and simulation of a battery charging infrastructure for electric vehicles, with the objective of pro-actively scheduling the charging of up to fifty vehicles so as not to overcharge the electrical network. Benefits of having the charging stations differ (as much as possible while satisfying end-user requirements) battery charging for those hours when electricity consumption is otherwise low include rendering electricity consumption more uniform along the day. A multi-agent system was used to design a distributed, modular, coordinated and collaborative multi-agent management system for this infrastructure. Simulation results show the effectiveness of this approach under the conditions of four real-life scenarios. Keywords: Modeling and simulation of power systems, Intelligent control of power systems, Smart grids, Energy Management Systems, Multi-Agent Systems, Electric Vehicles 1. INTRODUCTION Electric Vehicles (EV) provide an alternative technology allowing diversity in energy sources for transportation, therefore reducing oil dependency, while providing mobility at competitive costs. Independent market analysis by Gartner and Wheelock (2010) forecasts a steep expansion of EV market in the coming 10 years. The charging infrastructure for EV will increase stress on the power grid and, therefore, needs to be managed in order to comply with both charging needs and end-client satisfaction, while avoiding disruptions in the electricity distribution. The management and control of electricity distribution is a complex problem for both the utility grid and small scale distribution networks. The dynamics within these networks are the result of interactions between several individual components and objectives: on one hand we have the grid infrastructure, which includes hardware and software, on the other hand we have the objectives of all stakeholders associated with the distribution chain, which include electricity producers, owners of transmission lines, retailers and end-clients. In order to comply with all these, sometimes conflicting, interests, a major research effort has been put towards the development of smart grids that can provide management and control at the network infrastructure level, while delivering high quality services in the electricity distribution chain. ? This work was supported by FCT, through IDMEC, under LAETA.
978-3-902661-93-7/11/$20.00 © 2011 IFAC
Therefore, it is imperative to implement autonomous and decentralized management platforms that are capable of coordinating charging operations while ensuring safety, preserve efficiency and smartly manage the distribution of electricity for the EV batteries. Additionally, there is a need of suitable management software for distribution that can deliver extended services and support new paradigms of business models for the electricity retailers. The importance of this problem will increase as the market share of EV increases. In order to implement an autonomous and decentralized management system to charge the batteries of EV, new strategies are needed for this implementation and a good option is to use Distributed Artificial Intelligence (DAI) techniques. In particular, methodologies based on multiagent systems (MAS) are a good choice to design a distributed, modular, coordinated and collaborative multiagent management system for an intelligent charging network of electric vehicles. This paper is organised as follows: section 2 presents the layout of the system, section 3 its modelling employing a MAS, section 4 its implementation in Matlab, and section 5 the simulation results obtained. Conclusions are drawn in section 6. 2. SYSTEM LAYOUT This paper addresses a battery charging station for electric vehicles (Borges et al., 2009), assumed to have fifty charging points for as many vehicles. Such stations will be
12267
10.3182/20110828-6-IT-1002.03492
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
implemented in parking lots, and so parked cars may or may not be charging in each instant (because the battery is already full, or because too many cars are trying to charge their batteries at the same time). Different neighbourhoods (e.g. residential, commercial or office areas) are expected to lead to very different electricity demands. Users will pay electricity according to a previously chosen profile, that dictates the car’s priority in obtaining electricity: cheaper prices will correspond to lower priorities and hence to longer charging times; to ensure shorter charging times and thus higher priorities the price paid for the electricity will be higher. The objective is to satisfy charging requests, respecting the hierarchy of priorities defined by what each client is paying, and all this without overcharging the electrical network. Indeed, the idea is that charging stations will, to the maximum extent possible, differ battery charging for those hours when electricity consumption is otherwise low. In this manner, peak hours when electricity consumption is higher (typically during the day) will not be further strained, while those periods of the day (typically, these are rather periods of the night) when electricity could be cheaply available will be put to good use. The benefits of thus rendering electricity consumption more uniform along the day could be enormous. To achieve this a two level decision system was conceived: 1) charging stations must communicate with each other and decide, taking into account the expected number of cars in the next few hours and their charging priorities, how the available electricity will be split among them; 2) each charging station must then distribute the electricity it has been allocated among the cars therein parked. Trying to sort things out taking into account every single individual car in all the charging stations of a large geographical region would likely prove to be impossible from the computational point of view. This would also be very liable to fail because of communication failures. In this way, a high level distribution will first be performed, for some large time interval, and then each charging station will be left with some autonomy to deal with its local conditions, allocating electricity to particular vehicles, with the ability of changing the allocation in shorter time intervals, and being thus flexible to deal with unforeseeable fluctuations in demand. This paper only considers this second level, assuming that the first has already been carried out, and that the electricity available in each instant has been determined. Thus the layout corresponds to that of figure 1, where the Local Power Manager (LPM) is responsible for the charging station under study. Further assumptions are as follows: • the base price of energy was taken from the regulated market defined by legislation; • vehicle-to-grid and vehicle-to-home concepts (whereby cars may, if that is found profitable, provide energy to the grid or to the owner’s home (Botsford and Szczepanek, 2009)) were not considered; • no fast charge of batteries (whereby batteries can reach 80% of their maximum capacity in only 30 minutes (Botsford and Szczepanek, 2009)) was considered;
Fig. 1. System layout • batteries were considered to allow interrupted charging without damaging performance. 3. MULTI-AGENT MODELLING In face of the above, the concept of agent (without access to data from all the system, but only partial information from a small part of the system) comes as a solution to this problem. MAS are an “artificial social system” (Wooldridge, 2002), composed by individual agents able to decide autonomously and with “social” capabilities (communicating and collaborating with each other), in order to achieve their own design objectives. MAS are used in several different research, operation and business areas. Macal and North (2006) suggest its use in areas as business and organisation, economics, society and culture, military and biology. Applications based on MAS are also being introduced to fault tolerant control (Hines and Talukdar, 2003; Mendes, 2008) and largely used in power networks (Negenborn, 2007) and its robustness is proved with the applications in military services (Phillips et al., 2006). In the particular plant addressed in this paper, MAS provide a very efficient tool, since we are dealing with a complex system, with many different ramifications (the cars) that do not depend on each other. MAS allow us to develop a managing platform that would otherwise deal with too much information, by decentralising it, and gaining advantage of DAI flexibility. A two layer MAS was elaborated, where two different design objectives are implicit: one layer is responsible for managing the park, decide on which car to charge and ensure the power limit will not be exceeded; the second layer will be composed by agents, each one of them responsible by a car with the objective of charging it according to the costumer needs. The communication between those two layers is established by an auction, described below in section 4.3. 3.1 Macro-level architecture Due to the specific characteristics of a distributed electric charging system, the MAS macro-level architecture proposed in this work is a federated architecture (shown in figure 2), since it allows the natural common relations and communications between the different charging agents, where the different charging point (CP) agents only need to communicate with a Facilitator, and not between them. The LPM agent takes on the role of Facilitator, while the CP agents have no direct communication between them, being coordinated by the LPM Agent, which will conduct an auction process, by sending data to the CP
12268
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
Fig. 2. Multi-Agent System: Macro-level architecture
Fig. 4. Charging Point Agent Architecture
Fig. 3. Local Power Manager Agent Architecture Agents, receiving back the bids and assigning the charging confirmations according to the Auction inference system (IS) rules. 3.2 Micro-level architectures
Fig. 5. Evolution of the battery’s SOC during the charge process
The design of micro-level agent architectures should represent and define individual agents’ behaviour and interactions.In this system, both the LPM agent and the CP agents should be intelligent agents to provide the necessary performance in the charging process. To achieve some different degree of intelligence and to correctly design the two different behaviours of these agents, two different types of micro-level architectures have been designed. The LPM agent will be responsible for managing the car parking lot and in particular the communications with the CP agents, providing data for the auction from the electricity stake-holders such as energy availability and prices. The architecture for the LPM agent is shown in figure 3. The communications layer is constituted by the external link (responsible for the communications with the exterior of the park) and the CP link (responsible for the communications with the CP agents). In the operations layer, the data base conceptually represents an information buffer and connections. The decision making module will perform the negotiations with the other LPM agents (not addressed in this paper). The auction IS will manage the auction process and assign the charging confirmations to the respective CP agents. Both will be conditioned by the goals module, that represents the overall objectives of the LPM Agent. The CP agent will be responsible for ensuring maximum battery charge, respecting the client profile. Decisions on bids offered during the auctions will be supported by data information provided by the car in real-time, by the LPM agent, and by a model of the battery. CP agent architecture is presented in figure 4. The communications layer is constituted by the LPM link (responsible for exchanging information with the LPM agent), the sensor (to receive information from the EV battery) and the actuator (that sends the charge confirmation). The operations Layer con-
tains a data base (with the same function as that in the LPM agent), the bid decider (to perform the auction bid), and the battery model evaluation (to evaluate battery charging behaviour). 4. IMPLEMENTATION 4.1 Battery The battery charging process is non-linear and was modelled as shown in figure 5, that gives the battery’s state of charge (SOC, the ratio between energy stored and maximum capacity) as a function of time for an uninterrupted charge. This behaviour was implemented with a TakagiSugeno fuzzy logic model (Michels et al., 2007), so that in the future it will be possible to make use of past data on the evolution of the SOC with time during the charge of each particular battery, adopting different models for different batteries. Thus it will be taken into account that not all batteries charge in the same way, because of their past history and different wear off. This adaptive modelling has not been implemented yet. 4.2 Multi-agent platform To implement the multi-agent system described in section 3, the FTNCS-MAS Designer toolbox (Mendes et al., 2009) for Matlab was used. This toolbox allows configuring macro-level architectures simply, automatically implementing the communications between agents allocated in different computers with the User Datagram Protocol (UDP). It was chosen because, while designed for Fault Tolerant Networked Control problems as the name stands for, it is very versatile and suits very well the problem under study. The agents were distributed over five personal
12269
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
Fig. 6. Agent distribution over the five computers computers, each containing 10 CP Agents an as shown in figure 6.
Fig. 7. Sorting of bids and honoured requests
Because the UDP (unlike the Internet Protocol Suite (TCP/IP), for instance) does not ensure that information sent by a computer is effectively received by another (which means that the data packages send in every instant could be lost), the following strategies were followed:
client’s profile), tfull is the time needed to reach SOC= 100% under continuous charge (as computed by the battery model), and ffull is a monotonously decreasing function; • T˜50 = f50 ttleft , where t50 is the time needed 50
• a sampling time of 1 second was used, which is enough, and avoids heavy network traffic; • flag variables were sent over 20 seconds (the probability of loosing data every single second during these 20 seconds is neglectably low); • variables were fed back after being received, to ensure no communication failures.
to reach SOC= 50% under continuous charge (as computed by the battery model), and f50 is a monotonously P3decreasing function; P50 i 1 t PT +t −PT 50 i=1 left t=1 ˜ pt = fpt • H , , where PT +t PT tleft is the electricity base price in the future time instant T + t (hence the first fraction reflects how the price of electricity will evolve with time), tileft is the time the car in charging point i is expected to remain plugged (which information, according to the client’s profile, is given by the LPM, and hence the second fraction reflects the relation between the time left for each car and the average of the time left for all others), and fpt is a function that monotonously increases with both its independent variables; • and α, β and γ are coefficients that depend on the client’s profile.
Furthermore, the RT block toolbox (Daga, 2008) was used to ensure that all agents run in a constant ratio between real time and simulation time (otherwise some batteries might take so long to stop charging when ordered to, while others were already starting to charge, that there would be very short but significant energy consumption peaks, for about 1 sampling time). 4.3 Auction CP agents interact by an auction, as mentioned above in section 3, so as to sort out among them which vehicles are being charged in each instant. This auction is organised by the LPM every 15 minutes (this period was chosen because it is the frequency with which available information is updated by the Portuguese electric grid owner), and each CP agent makes a bid B for the energy C15 that it can charge during the next 15 minutes. These bids bear no relation to the price the client is actually paying, being solely values used internally by the system. The LPM sorts the bids by price as seen in figure 7, and honours as many requests as possible, by order of bid, up to the limit imposed from grid conditions. (As mentioned in section 2, this limit will in the future be determined by a negotiation between the LPMs of different charging stations.) Bids are reckoned as ˜ pt B = PT 1 + αT˜full + β T˜50 + γ H
Functions ffull, f50 and fpt are currently linear and were implemented with a Takagi-Sugeno fuzzy logic model, so that in the future it will be easy to render them non-linear, optimising the relation according to what experience shows more useful. 5. RESULTS There being as of writing no market studies to substantiate what the best commercial strategy for a charging stations might be, three different profiles that appeared reasonable were hypothesised:
(1)
where
• PT is the electricity base price (given by the LPM) for the current time instant T ; • T˜full = ffull ttleft , where tleft is the time the car full is expected to remain plugged (according to the 12270
(1) priority client, willing to pay much to fully charge the battery as soon as possible at all times (such clients will expectedly be a minority); (2) client who cannot charge at home and must thus charge during the day, when prices are higher, with the objective of reaching SOC=50% paying as less as possible; (3) client who can charge at home and thus does not need to charge during the day, as long as the battery is above SOC=50%.
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
Table 1. Coefficients for (1) for each profile
α β γ
1 1.0 0.0 0.0
Profile 2 3 0.3 0.4 0.3 0.6 0.4 0.0
These profiles, corresponding to the parameters of (1) given in table 1, were used to carry out tests in four different scenarios: (1) an office zone, where the affluence is significant from 8 AM to 19 PM, and most clients chose profile 2; (2) a residential zone, where the affluence is the opposite of that of scenario 1, and most clients chose profile 3; (3) a commercial zone, with high rotation of cars in the 12 AM to 12 PM period, and mixed profiles; (4) a car silo, with high affluence at all hours, and mixed profiles, the mix depending on the hour (at night there will be more profile 3 clients, during day hours profile 2 will be more significant). Car affluence is shown for the four scenarios in figure 8, which also shows how many cars are effectively able to charge at each instant. Figure 9 shows power availability assumed for each scenario (after negotiation with other LPMs). Figure 10 shows how many clients of each profile were assumed for each scenario, how their batteries were charged when arriving at the park, and how they were charged when leaving. These results show that the objectives of each profile are being satisfied in each case; the ease with which that is done depends, of course, on the constraints of each scenario (e.g. in scenario 3, profile 1 cars cannot often be fully charged because they seldom remain long enough). It seems thus reasonable to assume that the modelled system satisfies its requirements. When actual data on car affluence, available energy, clients’ expectations, etc. becomes available, the flexibility of the implementation is expected to allow adapting the system promptly. 6. CONCLUSIONS Distributed charging systems for EV can be modelled using MAS. We have discussed the main features of such model, namely both its macro- and micro-level architectures, and implementation issues, such as an auction mechanism. The MAS model that has resulted complies with load constraints imposed by the grid while attempting to maximise the number and charge of plugged-in EV in the system. Three different profiles, which characterise groups of typical EV owners, were defined for the CP Agents. The model was simulated under conditions of four reallife scenarios in order to understand both the distribution network dynamics and the EV affluence in the parking lot, therefore providing a customisable tool to support the design and dimensioning, as well as assessing the economic viability, of such type of charging infrastructures. REFERENCES Borges, J., Borges, A., Campos, J., and Nina, M. (2009). Sistema de distribui¸ca ˜o e disponibiliza¸ca ˜o de energia el´ectrica para carregamento de baterias de ve´ıculos
Fig. 8. Car affluence and batteries charging for scenarios 1 to 4(from to top to bottom) el´ectricos (in Portuguese). 104395. Portuguese Institute of Industrial Property. Botsford, C. and Szczepanek, A. (2009). Fast charging vs slow charging : Pros and cons for the new age of electric vehicles. EVS24 International Battery, Hybrid and Fuel
12271
18th IFAC World Congress (IFAC'11) Milano (Italy) August 28 - September 2, 2011
100
SOC / %
80 60 40 20 0 at when at when at when arrival departing arrival departing arrival departing profile 1 (7%) profile 2 (62%) profile 3 (32%) 100
SOC / %
80 60 40 20
Fig. 9. Power availability assumed for scenarios 1, 3 and 4 (top) and for scenario 2 (bottom)
at when at when at when arrival departing arrival departing arrival departing profile 1 (22%) profile 2 (0%) profile 3 (78%) 100
SOC / %
80 60 40 20 0 at when at when at when arrival departing arrival departing arrival departing profile 1 (13%) profile 2 (87%) profile 3 (0%) 100 80 SOC / %
Cell Electric Vehicle Symposium, 1–9, Norway. Daga, L. (2008). URL http://leonardodaga.insyde. it/Simulink/RTBlockset.htm. Gartner, J. and Wheelock, C. (2010). Electric vehicle charging equipment. Technical report, Pike Research. Hines, P. and Talukdar, S. (2003). Autonomous agents and cooperation for the control of cascading failures in electric grids. Proceedings. 2005 IEEE Networking, Sensing and Control, 2005., 273–278. Macal, C. and North, M. (2006). Tutorial on agentbased modeling and simulation part 2: how to model with agents. Proceedings of the 2006 Winter Simulation Conference, 73–83. Mendes, M.J.G.C. (2008). Multi-Agent Approach to Fault Tolerant Control Systems. Ph.D. thesis, Instituto Superior T´ecnico - Universidade T´ecnica de Lisboa. Mendes, M.J.G.C., Santos, B.M.S., and S´ a da Costa, J. (2009). A Matlab/Simulink multi-agent toolkit for distributed networked fault tolerant control systems. In SAFEPROCESS’2009, Preprints of the 7th IFAC Symposium on Fault Detection, Supervision and Safety of Technical Processes, 1073–1078, Barcelona, Spain. Michels, K., Klawonn, F., Kruse, R., and N¨ urnberger, A. (2007). Fuzzy Control: Fundamentals, Stability and Design of Fuzzy Controllers. Springer, Berlin. Negenborn, R.R. (2007). Multi-Agent Model Predictive Control with Applications to Power Networks. Ph.D. thesis, Technische Universiteit Delft. Phillips, L., Link, H., Smith, R., and Weiland, L. (2006). Agent-based control of distributed infrastructure resources. Technical Report January, U.S. Department of Commerce: National Technical Information Service. Wooldridge, M. (2002). An Introduction to Multi-Agent Systems. John Wiley & Sons, Ltd. Chichester, England.
0
60 40 20 0 at when at when at when arrival departing arrival departing arrival departing profile 1 (6%) profile 2 (39%) profile 3 (55%)
Fig. 10. SOC of batteries when arriving and when leaving the charging station for scenarios 1 to 4 (from to top to bottom)
12272