(PDF) Integrating Power Systems, Transport Systems and Vehicle Technology for Electric Mobility Impact Assessment and Efficient Control

Electric mobility is considered as a promising option for future individual transportation in terms of lower CO2-emissions and reduced dependence on fossil fuels. In order to analyze its impacts effectively, an agent based model is proposed. It integrates three domains which are mainly affected by electric mobility. Vehicle fleet evolution and vehicle energy demand simulations are combined with a transportation simulation, thus determining the daily behavior of electric vehicles and providing individual battery energy levels at the different locations of the vehicles during the day. Further, a power system model combined with a charging control algorithm is included in order to study general effects in electricity networks and to provide insights into new electric vehicle load patterns, as well as into changes in transport behavior. It is shown that network congestion can be mitigated using control signals. The paper describes the method and the integration of the three different domains and shows results of the integrated analysis tool.

©2012 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. 934 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 Integrating Power Systems, Transport Systems and Vehicle Technology for Electric Mobility Impact Assessment and Efﬁcient Control Matthias D. Galus, Rashid A. Waraich, Fabrizio Noembrini, Karel Steurs, Gil Georges, Konstantinos Boulouchos, Kay W. Axhausen, and Göran Andersson Abstract—Electric mobility is considered as a promising option for future individual transportation in terms of lower -emissions and reduced dependence on fossil fuels. In order to analyze its impacts effectively, an agent based model is proposed. It integrates three domains which are mainly affected by electric mobility. Vehicle ﬂeet evolution and vehicle energy demand simulations are combined with a transportation simulation, thus determining the daily behavior of electric vehicles and providing individual battery energy levels at the different locations of the vehicles during the day. Further, a power system model combined with a charging control algorithm is included in order to study general effects in electricity networks and to provide insights into new electric vehicle load patterns, as well as into changes in transport behavior. It is shown that network congestion can be mitigated using control signals. The paper describes the method and the integration of the three different domains and shows results of the integrated analysis tool. Index Terms—Agent based modelling, aggregator, electric vehicle load curves, grid to vehicle (G2V), plug-in electric vehicle (PEV), smart grid, transportation simulation, vehicle to grid (V2G). I. INTRODUCTION LUG-IN electric vehicles (PEVs), which comprise plug-in hybrid hybrid electric vehicles (PHEVs), and battery electric vehicles (BEVs) are of great promise for increasing efﬁciency in the private mobility sector. Equipped with a large battery and an internal combustion engine (ICE), PHEVs are able to replace a substantial fraction of mileage using electricity for propulsion. The ICE is only utilized as an auxiliary power source in its high efﬁciency operation region to overcome limited battery range [1]. In order to reﬁll the batteries energy can be drawn from the electricity network. Obviously, electric mobility will introduce additional load to the power system. The temporal behavior of PEVs is hard to predict. Furthermore, and contrary to current loads, the spatial P Manuscript received November 23, 2010; revised May 26, 2011, January 13, 2012, and February 22, 2012; accepted March 01, 2012. Date of current version May 21, 2012. The work was supported by the Swiss Federal Institute of Technology (ETH) Zurich under Research Grant TH 2207-3, by the Swiss Federal Ofﬁce of Energy, ETH Zurich and by the municipal utility company of Zurich, . Paper no. TSG-00271-2010. M. D. Galus and G. Andersson are with the Power Systems Laboratory (PSL) ETH Zurich, 8092 Zurich, Switzerland (e-mail: galus@eeh.ee.ethz.ch). R. A. Waraich, K. W. Axhausen are with the Institute for Transport Planning and Systems (IVT), ETH Zurich, 8092 Zurich, Switzerland. F. Noembrini, K. Steurs, G. Georges and K. Boulouchos are with the Aerothermochemistry and Combustion Systems Laboratory (LAV), ETH Zurich, 8092 Zurich, Switzerland. Color versions of one or more of the ﬁgures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identiﬁer 10.1109/TSG.2012.2190628 distribution of this new load is hard to anticipate as PEVs are inherently mobile. At ﬁrst sight, this does not seem impedient, but becomes crucial as the power system can be regarded as somewhat static compared with the transportation system. A large number of PEVs connecting in one area might cause transformer and line overloading as well as excessively low voltages since the electricity distribution infrastructure is currently designed for nonmobile loads. These effects could cause deterioration of power quality and security of supply [2]. Recently, different management schemes are developed to avoid such challenging impacts. Some of them pursue an approach which is based on a distributed and hierarchical management scheme possibly used by an aggregating unit [2]–[4] often referred to as an aggregator [5], [6]. This entity can schedule the behavior of the PEVs according to its objectives, possibly incorporating electricity prices, etc. The paper at hand follows this approach. However, other schemes adapt fully distributed approaches for the control of individual vehicles. The approaches often take exogenously communicated electricity prices as an input [7]–[9]. Fleet test can be performed in order to attain some knowledge on possible electric vehicle behavior [10]. However, efﬁcient forecast and analysis tools still need to be developed. Transport simulation models could be used for the latter purposes [11] but are limited by their lack of ability to incorporate speciﬁc energy demands of the simulated vehicles. Energy demands are crucial knowledge for analyzing the impact of PEVs on the power grid as they determine the additional load and its evolution. A model, which can offer accurate insights into temporal and spatial load evolution and possible environmental footprints of electriﬁed individual transportation, needs to integrate power and transportation systems as well as the technologies employed by the simulated vehicles. The detailed results from such a model would be useful to all of the three domains. In power systems they could be utilized for general load management (e.g., valley ﬁlling [12]), for PEV scheduling in order to balance ﬂuctuating power infeed from renewable energy sources [13], [14] or for investigating the potential of ancillary services [15], [16] whose provision through electric cars was shown to be proﬁtable [17], [18]. For the transport sector the results are useful as conclusions can be drawn about the effect of increased electriﬁcation on future transport behavior, e.g., altered driving habits due to charging restrictions. Future tail pipe emissions, primary energy use and emissions complete the conclusions. In this paper, such an integrated method for electric mobility assessment is described and developed, combining power system models, agent based transport simulations and modeling of speciﬁc vehicle technologies as illustrated in Fig. 1. The next section is dedicated to the description of the method 1949-3053/$31.00 © 2012 IEEE GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY 935 The PMPSS integrates the electric vehicles into the power system and offers an intelligent management of PEVs, minimizing charging costs while avoiding system constraint violations such as overloads or voltage limit violations. It ensures efﬁciency in the distribution of, potentially scarce, power to the vehicles. Furthermore, it allows the analysis of potentially available, aggregated storage. This functional block will be described in detail in Section V. B. Functionality Fig. 1. Integrating three domains affected by electric mobility for effective analysis: power systems, transportation systems, and vehicle technology. Fig. 2. The integrated method comprising the vehicle technology assessment model (VTAM), the multiagent transportation simulation (MATSim), and the PEV management and power system simulation (PMPSS). architecture and its functional principal. Section III, IV, and V describe the main parts of the framework, comprising vehicle technology modeling, agent based transportation modeling and power system modeling. The sixth section illustrates results and is followed by a conclusion. II. THE INTEGRATED METHOD, ITS ARCHITECTURE, AND FUNCTIONALITY A. Architecture Fig. 2 visualizes the architecture of the method. It consists of the combination of three independent models, the vehicle technology assessment model (VTAM), the multiagent transportation simulation (MATSim), and the PEV management and power system simulation (PMPSS). The models are linked and rely on outputs of each other. The VTAM comprises two features. Firstly, the evolution of the future vehicle ﬂeet can be simulated to determine the fractions of the different powertrain technologies within the ﬂeet. Secondly, it constitutes the models of the respective powertrains, which, when fed with real world drive cycles, provide energy demands of the vehicles within the ﬂeet. In the case of elecemissions (as a functric vehicles the primary energy and tion of the electricity supply system under considerations) can be calculated. For conventional vehicles the petrol consumption can be determined. This functional block will be described in detail in Section III. MATSim simulates vast amounts of agents utilizing different traveling modes, such as mass transport and individual vehicles. MATSim is able to determine and analyze transportation ﬂows, road congestion and urban planning needs. This functional block will be described in detail in Section IV. The data necessary for proper analysis consists of, in the case of power systems, the anticipation of future electricity demands, detailed load and distribution network data of the network/region under investigation as well as assumptions on possible charging locations and prospective powers. MATSim requires information about the transport and activity infrastructure (e.g., public transport, business, industry areas, etc.) including its geographical information system (GIS) data. Such information includes parking areas, which need to be identiﬁed, and their charging capacity. Assumptions concerning the initial composition of the vehicle ﬂeet and their powertrains are crucial for the VTAM. Subsequently, data on efﬁciency maps of the vehicles’ energy converters (ICE, electric motors, etc.) needs to be incorporated. In order to calculate the energy demands while driving, appropriate drive cycles need to be deﬁned [19]. These should give a realistic picture of the driving behavior within the simulated area (e.g., city driving) and should incorporate the maximal possible and average speeds in the transportation network of MATSim. The drive cycles are used in individual vehicle simulations to determine the energy consumption values of the different vehicles on the roads for different average speeds. The consumption value is obviously dependent on the chosen vehicle type. Based on these values, energy consumption functions are determined using regressions models. The functions, together with the technology ﬂeet shares, are exported to MATSim as shown in Fig. 2. Each agent is assigned a consumption function depending on its parameters, e.g., powertrain type. The consumption function is subsequently used by MATSim to calculate energy deployment and battery depletion levels while incorporating the individually faced trafﬁc situations. MATSim performs a reward based optimization where the individual agents pursue activities closely related to real life behavior. Using the detailed transportation demand of all agents, which is related to their activity schedules (work, leisure, shopping, etc.), the available parking lots, the charging areas and the individual energy consumption values, MATSim generates results which consist of • arrival and departure times; • location of the connection; • power rating at connection location; • state of charge (SOC) at connection; • desired SOC at departure; • start- and end charging times. This data is transmitted to the PMPSS. It simulates the underlying power system infrastructure, typically consisting of the distribution network. Here, the PEVs are allocated to aggregating platforms which can control the PEV charging [3], [20]. These PEV management tools distribute potentially scarce power efﬁciently between the connected vehicles, and temporally shift excessive PEV load. The area fed by one distribution transformer is assigned to an aggregating platform. This platform can be seen as one part of a future grid structure [21]. 936 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 At ﬁrst, it is assumed that PEVs connect and start charging as soon as they arrive at a destination, which features charging infrastructure. In case of system congestion due to transformer overloading, line overloading or excessively low voltages, these entities can shed PEV load and generate nodal control price signals for each simulated time interval. The time interval is chosen to be 15 min in accordance with the balancing group concept [6]. The control price signals incorporate an encoding of the power system state [20] and are returned to MATSim. The reward based optimization of the agents’ schedules can take these signals into account and change the agents’ charging and/or transport behavior accordingly until the iterative system of MATSim and PMPSS converges, i.e., all behavioral and power system constraints are fulﬁlled. The new charging schedule does not necessarily exhibit a continuous charging pattern for the individual vehicles. However, it considers the desired ﬁnal SOC and the owners constraints. Drawbacks of the discontinuous charging schedule on the battery are neglected herein. Note that real time electricity prices are not considered in this paper. The electricity prices are an exogenous input to the model and are assumed to be constant throughout the day. However, variable prices can be integrated as will be shown. Fig. 3. Vehicle ﬂeet composition by powertrains in the future. TABLE I DESIGN PARAMETER DIFFERENTIATION FOR INVESTIGATED VEHICLES III. VEHICLE SIMULATIONS AND THE ENERGY NAVIGATOR The following section describes the VTAM, the determination of the vehicle ﬂeet composition and subsequently the derivation of the energy consumption for each vehicle type’s powertrain. A. Fleet Simulation The dynamics of the car ﬂeet are simulated using a bottom-up approach with a high disaggregation level. This allows to consider the market penetration of vehicles with alternative powertrains. The following powertrain categories are considered: • internal combustion engines (ICE); • ICE-electric full hybrids (P-HEV); • ICE-electric plug-in hybrids (PHEV); • electric motors with batteries (BEV). The two latter are commonly referred to as PEVs. Regarding the energy source stored on board, gasoline, diesel, and electricity are taken into account. The differentiation includes 8 engine/motor power categories, 10 mass categories, and the cohort of the vehicle’s construction year (yearly, starting from 1971). The car ﬂeet composition is simulated using the survival probability of the cars which is calculated on the basis of dynamics observed between two subsequent years. Each year a number of vehicles is replaced. The survival probability calculated for the VTAM is (1) is the survival probability, is the year of ﬁrst mawhere triculation and is the actual year. No differentiation of depending on the fuel type, curb weight and power category is taken into account due to insensitivities to these variables [22]. The survival probability assumed is derived from past statistical values of the Swiss ﬂeet. It is assumed to remain constant. A dependency on exogenous parameters, e.g., oil prices, customer trends, is not considered. The database of the motor vehicles information system (MOFIS) of the swiss federal vehicle control bureau (EFKO) provides the input data regarding the number of vehicles in each category for the particular starting year in order to calculate . The penetration of new vehicles is based on an assumed market share development of the corresponding powertrain options. The observed trends towards hybridization, as well as growing market shares for diesel and gas vehicles, in the last years are incorporated in the model according to [23]. Shares for both, battery electric (BEV) and PEVs, are introduced using the S-shape penetration model for new products [22]–[25]. The market share of the speciﬁc technology is set to an exogenously in the year 2035. given, scenario dependent value Here, for simplicity, instead of the market share, the ﬁnal vehicle ﬂeet share is used. The S-shape penetration model is exemplarily given by (2) The penetration model results in a vehicle ﬂeet consisting of propulsion systems as shown in Fig. 3. Hence, a 65% share of PEVs refers to the vehicle ﬂeet apparent on the streets. In fact, the underlying market penetration of new sold PEV in 2035 is assumed here to be 100%. Vehicles with an alternative powertrain (other than pure ICE) may have a different mass than an equivalent conventional car. Therefore, the distribution of new cars with alternative drive trains over the weight categories has to be adjusted by taking into account the different weight of the powertrain and the batteries, etc. so that the structure of the cars’ size remains similar. Table I shows the design parameters used. B. Vehicle Analysis In order to include the vehicle ﬂeet results in the trafﬁc simulation, the actual energy consumption of the vehicles is of im- GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY portance. Its calculation depends on the vehicle design, which is part of the ﬂeet conﬁguration, and on the driving behavior, modeled through driving cycles. A library of vehicle models is built up according to possible combinations of the design parameters. The parameters considered are powertrain-, fuel- and power types as well as weight; see Table I. It yields a combinatoric maximum of 1200 vehicle types. These vehicle types are used in vehicle simulations involving converter efﬁciency maps. Hybrid and PEVs within the ﬂeet are modeled with an additional energy management control strategy [26]. A ‘technology improvement factor’ is applied to incorporate predictions on future mobility sector energy consumption. The technology improvement factor models only advances in technology, which improve the efﬁciency of the vehicle. The technology improvement factor is assumed highest for combustion vehicles because they offer more improvement possibilities than electric vehicles. Vehicle energy consumption is strongly related to the driving behavior, the boundary conditions set by legal frameworks and the trafﬁc situation on the road. Roads are considered to allow a maximum speed equal to one of the following values: 30, 50, 60, 90, 120 km/h. The average speed on any road is a measure for road congestion. The closer the average speed is to the maximum speed, the lower road congestion is. Representative driving cycles (speed proﬁles over time), compatible with the maximum speeds while differing in their average values are chosen. The difference between the ICE vehicle, illustrated in Fig. 4(a), and the battery electric vehicle, shown in Fig. 4(b), is obvious. The energy demand curves for different roads vary for the conventional vehicle because of the varying efﬁciency of the gasoline engine. For the battery vehicle the energy demand curves approach each other as the energy recuperation affects the energy demand strongly. With perfect recuperation efﬁciency, only the variability of the electric motor map is pivotal, which is rather small. Using the vehicle library and the driving cycles, speciﬁc energy consumptions for all vehicles over the different driving cycles are calculated using Newton’s second law [27] and then represented through regression models. The regression models are used in the subsequent transport simulation to determine energy consumption dependent on the individual trafﬁc situation. In the case of PEVs, which have two speciﬁc energy consumptions (charge depleting mode, hybrid mode), one regression for electricity and one for gasoline consumption are used. Fig. 4 illustrates the calculated speciﬁc energy consumption functions for several average speeds of an ICE vehicle and BEV. IV. MULTIAGENT TRANSPORT SIMULATION (MATSIM) Trafﬁc can be modeled through the behavior of individual vehicles. Simulating each car as an agent is called agent based micro simulation and allows tracking of individual vehicles over time. Assigning a utility to each agent allows for individual decision modeling such as choosing the path to drive or choosing the location for reﬁlling gasoline. MATSim [28] is an agent based transport simulation framework with focus on large scenarios. Simulations with more than seven million agents on a navigation network with around one million links have already been demonstrated [29], [30]. Fig. 5 shows a simulation result for the city of Zurich, Switzerland. The dots, each one representing an agent, are at different locations in the city, under different trafﬁc situations, including trafﬁc jams. 937 Fig. 4. Example of parameterized energy demand of vehicles in the ﬂeet. (a) Average energy consumption of a gasoline vehicle. (b) Average energy consumption of a battery electric vehicle. Fig. 5. This ﬁgure visualizes the location and speed of all agents in the middle of the city of Zurich, at arround 6:30 in the morning. Agents colored red have a low speed and compared to the free speed of the road and are stuck in a trafﬁc jam. Source: www.matsim.org. Fig. 6 illustrates the MATSim simulation process. Each agent has a daily plan of trips and activities, such as going to work, school or shopping. The daily plans, the street network and the facilities offering the activities are modeled in the initial demand block. The street network and the locations of facilities are unambiguously deﬁned via the GIS [31]. The execution of all plans is scored using a utility function. For example a person with shorter travel time has a higher utility than one, which has a longer travel time due to having been delayed by trafﬁc. Working and other activities increase the agents’ utility. The goal of each agent is to maximize the total utility of its daily plan by replanning its day. This replanning is based on a co-evolutionary algorithm [32]. Such an algorithm generally tries to ﬁnd the maximum of a ﬁtness function (in this case the utility function) using crossovers and mutations. In MATSim, the utility function has multiple degrees of freedom, 938 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 where denote the GIS coordinates of the activity. All locations of the environment are contained in the set (10) Fig. 6. Co-evolutionary simulation process of MATSim. such as the routes, the time spent at work, the transport mode, the locations visited and so on. The daily plans are then evaluated and bad daily plans (plans with low performance, respectively low utility) are deleted, which corresponds to survival of the ﬁttest in co-evolutionary algorithms. Thereafter, new plans are generated based on the previous set of plans. The execution of all plans, their scoring and replanning is called a MATSim iteration. The complete simulation is an iterative process, which approaches a point of rest corresponding to an equilibrium. This equilibrium is called relaxed demand [28]. One of the reasons for using an agent based approach is that individual agent preferences can be modeled. In order to evaluate constraints of the electricity network, detailed data on PEV location, time of connection and power demand are needed. Using high resolution road networks and including distinct buildings as well as activities allows for an accurate mapping between the transportation network and the electricity infrastructure. A. The Agent’s Utility Function in MATSim The utility function used in MATSim is described in detail in [33]. It assigns an utility to the plan of each agent (3) The utility of agent ’s plan contains the utility forming an activity , where for per(4) gives the set of all possible activities in the MATSim environment. Each agent has an individual set of activities which it would like to perform during the day. These activities are denoted by (5) can be ordered through a function according The subset to the time when the activity is performed under the plan. This of agent , where the function results in an activity plan is given by The utility of depends on the activity , the starting time of the activity , its duration but also on several other aspects such as if the agent arrived at a shop prior to , since its opening. Such a case leads to a negative utility the agent needs to wait. Furthermore, if an agent starts working or departs too early from an activity it late will also be penalized. An activity may have a minimal duration so that the utility of the activity will be reduced if the duration is . The total utility of activity is expressed too short as (11) where the dependencies are omitted for shortness. of agent ’s daily plan also includes the The utility (negative) utility of traveling from one activity to the next and is calculated according to (12) So far, the transportation model did not incorporate energy consumption costs. In the following the utility function is extended such that the energy consumption of the vehicle is a part of it. B. Extending the Agent’s Utility Function in MATSim Here, the utility function of an agent in MATSim is extended by adding a term which contains the cost of electricity at all charging locations. It is assumed that the necessary charging infrastructure is available at all working activity locations and at home. The amount of energy needed is given by the activity plan and the driving pattern, which is given by the regression models whose derivation is described in Section III. reduces the agent’s utility given by The term (12). The total charging cost during the day sums charging incurred at all activities and locations costs of agent and is expressed as (13) (6) with (7) and (8) Each activity of each agent is associated to a location within the environment of MATSim, whereby the activity home of agent can be at a different location than the one of agent . The location of agent ’s activity is expressed as (9) and denote the starting and ending of acwhere tivity in continuous time notation. Function transforms the . continuous time into discrete time intervals The individual cost terms are deﬁned as the energy consumption at a certain location multiplied by the energy price during that time as denoted in (a) (b) (14) GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY is the charging time at and in where time interval . The charging time is deﬁned by the start and end times and , respecdenotes the charging power of agent tively. Variable at electricity network node in time interval . The price is considered here as an exogenous variable. It is assumed to be constant during the day. However, it is obvious that the scheme allows also to use real time prices. Then, the price levels would strongly inﬂuence the agents charging behavior, shifting it to low price periods if possible. V. PEV POWER MANAGEMENT SYSTEM The PMPSS is developed in three steps. First, mapping between the transportation network, describing the physical location of vehicles connected to charging stations, and the electric grid is explained. Second, the charging of PEVs is modeled using an agent based approach in which PEVs compete for potentially scarce power in the electric grid. The approach maps agents active in the transportation system to agents active in the power system, i.e., PEV agents. In order to avoid a computationally expensive, iterative competition, an optimization platform, called PEV Manager, is derived. Finally, the scheme how control price signals are determined by PEV Managers is explained. A. Mapping Charging PEVs to the Electric Grid The PEV management and power systems simulation (PMPSS) employs detailed behavioral information on the vehicles for computations of the current power system state and appropriate PEV control. The PMPSS model includes an electricity network consisting of a set of nodes (15) where (16) gives the GIS coordinates of the electric node. In order to link the transportation and the electricity infrastructures, activity GIS coordinates (i.e., street coordinates) are clustered according to the proximity to the network nodes. This means that coordinates of several activity locations are assigned to the node which models the load for the particular area. This mapping is expressed as a function (17) with (18) Depending on their behavior, PEVs can, during time interval , be connected at location and hence to node . The PEVs connected at are denoted through (19) where 939 assumption is scenario dependent and will be deﬁned for the test case in the Section VII. To unambiguously map agents from MATSim to other environments, a modeling approach based on agent theory seems compulsive. Therefore, general utility theory and a mechanism design approach have been chosen to model PEV management in the power system [34]. B. Using General Utility Theory and Mechanism Design to Model PEVs as Agents in the Power System Game theory, especially the class of strategic form games [35], can be applied to model PEV charging behavior in cases when congestions occur in the electric grid. A strategic form is deﬁned as a tuple , game where (21) is a ﬁnite number of players and (22) of the players , recontains strategy sets spectively. The set is called the collection of all strategy proﬁles of the players. Furthermore, mappings, called utility functions, are deﬁned as (23) Note that in the context of PMPSS the utility does not directly relate to the utility function used in MATSim.1 The general utility theory [36] enables to express the preferences of players in terms of payoffs in some utility scale, i.e., it allows mapping numbers to the preferences. Key assumptions are that agents incorporate rationality, intelligence and common knowledge as deﬁned in [35]. The utility of an agent depends not only on its own strategy but also on the strategies of the rest of the agents. Every proﬁle of strategies induces its own outcome. In strategic form games, agents are able to play the game several times and learn from the outcomes. Depending on their experience, they will adapt their behavior and act differently, as dictated by common knowledge, their intelligence, their utilities and rationalities. The goal of mechanism design is to design a set of rules for which the game (e.g., auction) constantly delivers the same result. Within a properly designed mechanism, the agents reveal their true types and no strategic behavior is observed [35], [37]. The approach can be used to model charging PEVs as agents. They compete for a potentially scarce resource (power/energy) can in a game at a certain node of an electric grid. Therefore, according to be translated to the set (24) with (25) (20) and denotes a PEV at node in time interval . Note that not necessarily every agent in the transportation system needs to utilize a PEV . Furthermore, only the parked and connected PEVs are mapped to the power system. Therefore charging facilities are assumed at some parking, i.e., activity, locations. The Furthermore, the agents are considered to be rational and intelligent, attributes predeﬁned by MATSim. 1The utility of the agents in the context of the PMPSS should be regarded as a modeling approach and a tool that allows the integration of the power system simulation and MATSim. The integration of the two similar concepts will be developed in Sections V-C, VI. 940 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 at the time interval, when the PEV has to start charging. Then, the PEV can be fully scheduled to avoid violation of its desires. was This is, however, not performed here. In the past, found to be a good option [3]. is used to denote the complete set of private The variable for strategy set . The set of all agent values of agent in type proﬁles in one time interval is given by (29) Fig. 7. Plot of the agents’ type . The set of outcomes from such a competition (game ) is denoted . The PEV agents are required to make a collective choice from the set of outcomes , i.e., deciding the outcome. Prior to that, each agent privately observes his preferences over the alternative outcomes in . This is modeled by supposing . that PEV agent at node privately observes a value is called private It determines his preferences. The value value or agent type. It is only known by agent . The private value, also called personal energy valuation or aggressiveness in [3], [20], can be mathematically deﬁned as (26) The rationale behind this function is that it should rise with energy left to be charged and with less time left for charging. The parameters and are tunable. The agent type is dependent on the actual time interval, on the anticipated departure time, dein continuous time notation, and on the time interval noted length in seconds. The latter is calculated according to (27) where is the number of time intervals during the day. Further, the agent type is dependent on the desired state of , on the SOC in the current time incharge at departure , and also on the potential amount of charging terval which can be done in time interval T. The latter is expressed by in (26). It is multiplied by the hourly fraction of the time interval, given as (28) and weighted by the charging efﬁciency . The chargeable energy amount in is a linear function and only dependent on the connection capacity. Nonlinear behavior of the battery during charging is not considered. The agent’s type determines the value of energy with respect is low to the remaining charging opportunities. The value of when the SOC is high and/or close to the desired SOC, when the time to departure is long, or both. It is high when the difference between the actual and desired SOC is high and the time to defor different exparture is relatively short. Fig. 7 illustrates , ponents . The parameters of (26) are set to , , , (power of 3.5 with kW and a battery of 20 kWh) and ﬁnally . The parameter value can be limited to a maximum . A typical type proﬁle is represented as Individual PEV agents have preferences over outcomes. They can be represented by an utility function . Given and , the value denotes the payoff that PEV agent , having type , . In the more general case, receives from an outcome depends not only on the outcome and the type of , but on the . types of the other PEVs as well; so In order to deﬁne an utility function for PMPSS that incorporates the agent’s type, it is assumed that a PEV agent gains beneﬁt by having a certain energy in its battery. Intuitively, the beneﬁt should be low when the battery is empty. It should be high when the battery is full. Further, the marginal beneﬁt, i.e., the inﬁnitesimal change in beneﬁt due to the inﬁnitesimal change of SOC, should decrease with higher SOC. That means, the more energy is already in the battery, the lower the sensitivity to additional energy; hence the lower the ﬁnal increase in beneﬁt. This is consistent with well known economic theory in electricity markets [38]. These latter characteristics of the beneﬁt function can be achieved by deﬁning it according to (30) with where the parameter denotes the battery capacity of PEV and deﬁne the maximal marginal beneﬁt and agent , to the the slope of the marginal beneﬁt. It is realistic to set value of the current gasoline price. The rationale is that charging will not occur if electricity costs exceed gasoline costs weighted by the converter path efﬁciencies.2 determines the bidding behavior of the The value of agent, i.e., the slope of the beneﬁt function, and is a tuning parameter. Choosing a value or price at which the PEV is . One way of charged to a certain SOC allows to determine is to use the lowest forecasted electricity price determining and are given by for the next day. Hence, (31) Fig. 8 displays two beneﬁt functions for two different batand chosen as discussed above. tery capacities with The vertical, red lines denote the area of operation (the battery 2This model assumes that electricity prices are lower than gasoline prices, which for is realistic most countries; especially for day ahead electricity prices. The inclusion of real time electricity pricing in the model is, as mentioned, possible and could result in a situation where PEVs do not charge when electricity becomes more expensive than gasoline. This seems, however, to be an exception from normal situations and reasonable. Ancillary service markets and their inﬂuence are not considered in this paper. GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY Fig. 8. Plot of the agents’ beneﬁt function for two different battery capacities. energy level boundaries). The beneﬁt is higher for larger battery capacities. Within the battery operation range, the beneﬁt , and hence confunction is monotonically increasing in tinuously differentiable. The utility function of the PEV agent is derived by multiplying the beneﬁt with the individual agent type and subtracting the incurred energy costs according to Fig. 9. Plot of the agents’ marginal utility as a function of SOC and time interval . are assumed to be common is private knowledge among all agents. The value of information of agent . The revelation theorem [34], [35], [37] can be used to avert the bidding stage of the game. The theorem states that if the mechanism is designed appropriately, the dominant strategy for the agents is to reveal their personal parameters. In such a case, the bidding stage is reduced to an optimization. In order to take advantage of the revelation theorem the incentive rationality theorem and the incentive compatibility theorem [35], [37] have to be proven for the designed utility functions. The incentive rationality theorem is stated as (32) In the utility function, the price signal, announced for the time interval, is multiplied by the amount of energy which is assigned to the agent. This is subtracted from the beneﬁt function multiplied by the agent’s is initially exogenously type. The price signal given, e.g., by the real time energy prices. It can grow above this price depending on the outcome of the game. The variable is the acquired energy, rated in per unit of battery capacity, at a given time interval . It is dependent on the outcome of the game, hence on the and the type proﬁle of the competing agents. It is given by (33) where denotes the battery capacity of PEV at and is the power assigned to the PEV. Fig. 9 illustrates the increase in the agent’s weighted marginal beneﬁt function over time, assuming that the agent is not able to attain any energy during the simulated time frame. The exof 70%. ogenous price signal is constant. The agent has a , the weighted marginal beneﬁt is increasing due Until to the growing agent’s type parameter. It remains lower than the . After , the agent’s weighted marboundary price . As the agent does not receive ginal beneﬁt is larger than any energy, the weighted marginal beneﬁt grows until it reaches its maximum, given by the weighted gasoline price. This is inlevel of dicated by the dots on the respective graphs at the 70%. It can be interpreted as the increase of the desire to acquire energy. and at , the set of outcomes For the actual game in , the set of agents at node , the agent type sets and the payoff (utility) functions 941 (34) whereas the incentive compatibility theorem is stated as (35) The incentive rationality theorem states that the agent needs to obtain a nonnegative utility in order to participate in the game (i.e., the auction for power). This can easily be proven using and is omitted here for brevity. The incentive compatibility theorem requires the PEV agent to consume the amount of power intended for his type. In these terms, it is essential to offer the right incentive so that revealing the true type is the best response to any situation, i.e., any type distribution and behavior of the other apparent PEV agents. This . Hence, revealing the true PEV is expressed by agent type is called a dominant strategy. It can be proven for uncongested cases (not shown here) but in congested cases, the utility function might give rise to strategic behavior in certain situations. However, it has to be assumed that the PEV agents cannot know the state of electricity network. Only the system operator is able to determine it. Hence, the PEV agents are not aware of whether the resource they are gaming for is scarce or not. It should be noted that the infrastructure for determining the state of the distribution network is currently not installed. However, it is assumed that in the not so far future, distribution network operators will be able to operate such necessary infrastructure. 942 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 Fig. 11. Endogenous control price signal determination scheme. Fig. 10. Integration of PEV Managers into power networks. One can further assume that large numbers of PEV agents game for the resource. This is valid because typically large numbers of PEVs are connected to the underlying network assets, e.g., transformers and lines, spanning larger geographical areas. If the number of agents is large, say hundreds or thousands, one can, with good approximation, further assume perfect competition among them. When this holds, strategic behavior can be neglected. Hence, the revelation theorem can be assumed to be fulﬁlled. The agents will announce their parameters. In order to optimally distribute scarce power between the connected PEV agents, from now on called PEVs, an optimization based on the utility function can be employed. It is performed by the PEV Manager, described next. C. The PEV Manager A typical example of the PEV Manager platform is illustrated in Fig. 10. Such a manager is presumed to be active on the interface between the 400 V and the next higher voltage network level. Note that the connection to the network does not have to be physical and integrated in the distribution management system (DMS). However, for the sake of simplicity, it is assumed here that the device, i.e., the PEV Manager, is integrated in the DMS. The information on where and when the individual vehicle connects is determined by MATSim as described in Section IV and used to assign numerous PEVs to the particular manager of the electric node. The PEV Manager optimization maximizes the sum of the within the PMPSS and distributes the utility functions available power at the node to PEVs connected in charging mode in . This avoids overloading of the transformer or excessive voltage sags which cannot be regulated by tap changers. The optimization maximizes (36) (a) (b) (c) (37) The parameters and can be derived according to (31) and need to be carefully tuned in order to achieve a desired charging behavior such as valley ﬁlling when using PMPSS on its own, i.e., with no feedback to MATSim. The constraints of the optimization feature the maximum power of the individual connection (37a), the maximum, individual SOC (37b) and the maximum, available power at network node (37c). The latter is determined as the minimum of the maximal suppliable power due to transformer limitation , violation of voltage bounds and line less the actual base load in and load bounds given by (38) with The optimization is performed for all . In the case where there is not enough power available at a certain node, the PEV Manager determines a locational control price signal. Should the network face challenges due to overloaded lines or excessively low voltages, a supervisory control relying on heuristics or optimization can be implemented in order to elect managers which decrease their PEV load level as demonstrated in [20], [39]. This will provide control signals as illustrated in [4]. Obviously, large amounts of PEV Managers are necessary to supervise a real distribution system penetrated by PEVs. On the other hand, this distributed control scheme enables an efﬁcient management of large numbers of PEVs since the individual optimization does not face computational borders due to large numbers of variables and can be parallelized. Based on the signals sent by the managers, PEVs react. Their charging is switched on or off. This control signal can be fed back to MATSim which then can take the locational congestions into account and consider them in the replanning routine. Without such a feedback to MATSim, the scheme could be used for load management assuming that people, contracted for such services as suggested in [5], would not change their daily behavior when faced with congestions. D. The Determination of Nodal Control Price Signals and Their Integration With MATSim The price signals, which occur during congested periods, are determined as illustrated in Fig. 11. It displays the nodal power capacity on the x-axis and the nodal control price signal on the y-axis. The vertical line intersecting the abscise denotes the . maximal power The graph shows the aggregated beneﬁt functions of PEVs apparent at node and weighted by their type. When the power , the current, demanded by the PEVs does not exceed GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY , which gives the base value of the exogenous price signal control price signal, is not altered. However, if demand exceeds the maximal available power, the resulting nodal control price can be derived using the Karush-Kuhnsignal Tucker (KKT) [40] equations of the optimization scheme given in (36) and (37). The KKT equations are not explicitly shown , derived from here, however, the Lagrangian multipliers the equations, are depicted in Fig. 11. The Lagrangian multipliers of the total power inequality constraint (37c) are denoted for the lower bound and for the upper bound. The others are denoted as and , accordingly. is always found to be zero as the nodal The multiplier price signal cannot be lower than the initial system price. However, as soon as the the node suffers from congestion, a nonzero appears which can be interpreted as a price premium determined by the state at the respective node. The result is the . It is dependent on the nodal control price signal . apparent PEVs, their state and their agent type and denote the individual difThe nonzero ference between the ﬁnal weighted beneﬁt of agent and the . The difference can nodal control price signal therefore be interpreted as the weighted marginal beneﬁt increase of an inﬁnitesimal enlargement of the individual power connection and/or battery capacity. To fully integrate MATSim and its utility functions with , which are so far unrelated to the agent’s PMPSS and its daily activity plan, several approaches can be envisioned. One way would be to send the information on the individually received energy in each time step to MATSim which should consider this result in its replanning stage of every iteration. This proves computationally expensive. Secondly, the utility functions used in PMPSS could be diof the rectly integrated into the agents’ utility function daily plan in the transportation system simulation. This can be referred to as the decentralized approach. Complementary to this, a centralized approach fosters a PEV charging behavior that avoids any congestions and therefore leads to the same PEV behavior in MATSim and in the PMPSS. It is not clear that both approaches lead to an equilibrium between MATSim and PMPSS. In the following only the centralized approach is pursued. 943 In non-congested times, the ﬁrst part of (39) is valid. Then, is transmitted to MATSim incorporating the ina signal superformation of the current exogenous price signal rated in relation to the imposed with the actual loading . In case the node is conmaximal suppliable power gested, the control price signal is determined by the optimization of the PEV Manager given in (36), (37). The scheduler’s objective is to avoid network congestions and to ensure that the agents are able to attain their desired SOC. Hence, the smart scheduler avoids situations where connected PEVs could potentially compete for the scarce resource (i.e., power). Thereby, the smart scheduler prevents that PMPSS and MATSim deliver differing charging schedules and thus different agent behaviors. In fact, the smart scheduler ensures compliance with the predetermined daily activity plans and desired battery energy levels. The complete, integrated method actively incorporates real time energy prices when determining the agent’s daily utility in relation to all other potential gains through activities. The scheduling is achieved by minimizing the objective function given as (40) where (41) (a) (b) (c) (d) (e) (f) VI. THE SMART CHARGING SCHEDULER One option to implement the centralized approach is to integrate a smart charging scheduler between MATSim and PMPSS. The scheduler has the possibility to react to information available from MATSim as well as from PMPSS. Therefore, the entity knows arrival and departure times of every agent as well as the SOC required at departure for the agent to be able to travel to the next location in an all electric mode. Furthermore, the scheduler receives the nodal control price signals, the daily energy prices and the base load in the particular area of each manager. The information is encoded in the form shown in the equation at the bottom of the page. (g) (42) where the dependencies are omitted for shortness. The function sums over all nodes of the system, over the connected PEVs, over the activities available at the nodes and over the time intervals during which the activities are being performed. The summands incorporate the value of the control price signal for every agent , which utilizes and connects a of node PEV as well as performs activity at location (39) 944 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 in time interval . The control price signal is multiplied by the power connection of the PEV and the charging interval in order to calculate the value of acquired energy at the node in . The charging time in time interval is determined by (41). The charging time is the difference between the charging and the charging start time end time . The constraints of the smart scheduling algorithm include the physical power connection limits of the PEVs in (42a). Further, the scheduler is designed such that the charging start time can be ﬂexibly chosen within an interval as stated by (42b). However, the charging start time must always lie before the charging end time. Both need to lie in the same interval as expressed in (42c). Clearly, the charging start times must always lie within the activity duration length (42d) just as the charging end times as denoted in (42e). The total energy consumed cannot exceed the initial depth of disat arrival as denoted in (42f). charge Finally, the total charging time must always sum up to a value as lower than or equal to the duration time of the activity stated in (42g). VII. DEMONSTRATION OF THE INTEGRATED METHOD The method and its submodules, which were described in the previous sections, offer a highly valuable tool for the analysis of future private mobility options. Insights can be obtained with reemission savings, to transportation gard to possible future pattern mutation and to effects in electricity networks such as altered load curves and V2G potentials. In the following, examples are presented which illustrate how information on PEV load curves can be attained and how the integrated tool can be used to intelligently schedule and recharge electric vehicles considering their individual, trip dependent energy demands and urban trafﬁc situations. For these illustration purposes, a test case is constructed. It consists of a power system incorporating only four nodes. Each node represents an urban area with several streets mapped to the node according to (17). An ac power ﬂow is used. The network also includes natural gas pipelines and small scale distributed generation, which offer additional generation capacity at the nodes. The nodes are modeled as energy hubs [3], [20], [41]. Generic transformer ratings of 9.4 MVA, 4.4 MVA, 8 MVA and 8.2 MVA are chosen for the hubs, respectively. The hubs are assumed to be loaded around 50% of their rating during the peak load times. The network is shown in Fig. 16 in the Appendix. The hubs feature different load curves which are displayed in Fig. 15 in the Appendix. Note that the load curves are not fully aligned with the distribution of activities in the areas as the example is generic in its nature and such detailed information was not available for the transportation network. The transformer ratings are chosen rather small for the particular areas in order to construct a case where these assets will face congestion. The sizing of other assets, such as electricity lines, is chosen such that for simplicity these assets do not face congestion. Voltage bounds are also not violated. The system sizing and the system parameters are given in Tables II and III in the Appendix, respectively. The transportation system comprises 28 000 streets. The daily activity plans of the PEV owners, i.e., agents, and their activity locations are based on a Berlin scenario [11], from which this simpliﬁed test scenario is derived. In this scenario, only car trips with home-work-home and home-education-home activity chains were considered. Charging facilities are assumed to be Fig. 12. Uncontrolled charging of PEVs. (a) Uncontrolled PEV load. (b) PMPSS signals for the uncontrolled charging approach. installed at every work and home activity location. A 1% population subsample of Berlin is used. It incorporates 16 000 agent plans. The mapping of transportation and power system results into the fact that hub 1 represents the largest area with a big number of streets and therefore apparent activities. Hub 4 is the smallest area. Hub 1 hosts about 10 000 cars, hub 2 has about 3000 cars while hub 3 has about 2000 cars and hub 4 hosts less than 1000 cars. Note that the streets, i.e., activity locations are randomly assigned to the hubs. A. Determining PEV Load Curves Utilizing the method, which is illustrated in Fig. 2, without the feedback of PMPSS control signals to MATSim, is useful to determine load patterns from electric vehicles and PEVs. These patterns take into account people’s daily behavior as well as the charging/refuelling options offered by the infrastructure of the area under investigation. The simulated load curves are valuable inputs for power system actors such as those responsible for balancing groups, distribution system operators and future PEV aggregators [6]. The energy consumption while driving has not yet been simulated using the advanced vehicle models described in Section III. The model elaborated in [27] is used for determining the energy consumption regression functions. The resulting charging schedule of MATSim in its relaxed state, i.e., in its equilibrium, is depicted in Fig. 12(a). It shows the load imposed on the test system by PEVs. In accordance with (13), (14) and (40)–(42), a constant energy price throughout the day is given to MATSim. Therefore, the agents simply recharge GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY 945 Fig. 14. Evolution of PMPSS nodal control price signals from iteration to iteration. Only the ﬁrst four iterations are plotted. Fig. 13. Controlled charging of PEVs by the integrated system comprising MATSim and PMPSS simulations. (a) Controlled charging of PEVs. (b) PMPSS Control Signals after 5th iteration. as soon as they arrive at their respective destination assuming pervasive connection possibilities, i.e., the vehicles can actually be recharged at every location. Obviously, such a scenario is highly dependent on appropriate penetration of charging infrastructure and on economical incentives for the owners to connect their cars. No attention is paid to this in this paper. In the model, the agents all act rational and charge their cars in order to reach their home at minimum energy costs. Two load peaks, of up to 30 MW, occur when the agents recharge at their work and education places as well as home locations in this uncontrolled manner. The steep increase of the load is due to an aggregation effect of the PEV load. Agents arrive at work locations as early as 07:30 and as late as 12:00. Most of them arrive between 08:00 and 09:00. As many PEVs are not fully charged in less than an hour, the load of the PEVs aggregates and causes a large load peak. Note that if charging infrastructure would not be available at work places or the electricity price during the day would change, e.g., through time of use pricing, the load curves would change and more and longer charging would occur during evening and night times; see [42] for the discussion of such scenarios. The contribution of hub 1 to the load peak is largest because its share of activity locations is larger. Note that the load at hub 1 exceeds the given transformer rating for both load peaks. This is also the case for hub 2 during the load peak in the morning. The total energy consumption during the ﬁrst peak is found to be about 4.1 MWh which gives an average energy consumption per car of about 2.5 kWh. This, according to the utilized energy consumption model, determines an average driven distance of about 13 km per car. The energy consumed in the evening is a little bit less than in the morning and is calculated to be about 3.9 MWh resulting in a marginally smaller average utilized energy and hence a marginally shorter average driven distance per car. This is because the agents chose different routes to drive back home. Their speed and hence their ﬁnal energy consumption is also different. Detours are rarely made as shopping and other activities are not included in this test case. The result of the PEV Manager optimizations within PMPSS is depicted in Fig. 12(b). The optimization scheme considers the at each hub. This load is imposed by unconbase load trollable loads. It incorporates well known load curves of household or business areas and partly loads the transformers which are feeding the respective areas. The transformer limits are used as constraints for each PEV Manager according to (38). Fig. 12(b) shows two PMPSS signal peaks for the hubs. The high peaks of the control price signal for hub 1 indicate a large load, e.g., a depleted ﬂeet trying to attain large amounts of energy thereby congesting hub 1, both in the morning and in the afternoon. This is veriﬁed when comparing with Fig. 12(a) where the PEV load of hub 1 clearly exceeds the transformer rating of 9.4 MVA. The 8:30 peak at hub 2 indicates a congestion at this hub. However, this ﬂeet is less numerous than the one at hub 1 while also having a higher average SOC. The total sensitive demand at hub 2 is lower, as can be concluded from the explanations given in Section V-D. Comparing Fig. 12(b) again with Fig. 12(a) for hub 2, it can be seen that the total PEV load in the morning is smaller than the transformer rating of hub 2. However, the congestion still occurs because the transformer is also loaded with the uncontrollable non-PEV load. It should be noted that Fig. 12(a) shows the unmanaged PEV load. After activating the PEV Managers, the excessive load is shifted to later time intervals by use of the generated control price signals. This decreases the load peaks but introduces a contradiction between MATSim and PMPSS as the charging schedules differ. The PEV Manager result is not shown. It is used in the iterative scheme of the smart grid scheduler, which is able to avoid the contradictory charging schedules of both functional blocks. It is described next. B. Employing the Smart Grid Scheduler As illustrated, uncontrolled charging can lead to large demand peaks. MATSim generated an agent behavior resulting in load peaks of almost 30 MW. In order to avoid this unfavorable behavior and the occurrence of congestions, the smart scheduler described in (40)–(42) is used. 946 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 Fig. 13 depicts the result of the iterative version of the described method. The iterative scheme takes into account the daily load curve of the system. The curves, differing for each hub, are depicted in the Appendix in Fig. 15. The ones for hub 1 and hub 2 incorporate a relatively high load in the evening hours between 18:00 and 23:00 with a peak at 22:30. Fig. 13(a) illustrates how the PEV load is distributed more evenly throughout the day by the smart grid scheduler. In comparison to the earlier case, where charging was uncontrolled, the aggregated load peak is substantially reduced to a maximum of 4.8 MW, or 16.7% of the previous amount. Note the reduced scale of the ordinate in Fig. 13(a). Although convergence is not inherent for such a setup, a stable equilibrium of the transportation and charging behavior is reached within 7 iterations for this example. The equilibrium represents a situation in which all PEV agents attain enough energy to reach their next activity location while at the same time fulﬁlling their daily activity plan. Fig. 13(b) illustrates a typical PMPSS output on the verge of reaching a stable equilibrium. It shows the control price signals at all hubs after ﬁve iterations. For most times the signals do not exhibit peaks. However, local signal maxima are observable during the evening and night hours of the simulation. This is due to the fact that during this time the power system is subject to the highest base loading. Fig. 14 illustrates the evolution of the control price signal of the heavily stressed hub 1 from one iteration to the next. The ﬁrst four iterations are shown. It can be seen that the PMPSS control signals decrease from iteration to iteration due to the scheduling of the PEVs performed by the smart grid scheduler. The system evolves towards an equilibrium, i.e., agent behavior, where individual activity plans, desired battery energy levels and mobility constraints imposed on the large agent population are fully taken into account. The smart charging scheduler can be also employed if the transformers exhibit more ﬂexibility than in this rather conservatively chosen test case. Then, the smart charging scheduler is able to exploit the load valleys in certain areas much better. Furthermore, when using a variable tariff scheme, the smart charging scheduler will be able to schedule the vehicle charging as cost effective as possible, distributing it over several, low cost time intervals. this however is not shown in this paper due to limitations on space. The energy consumption proﬁles are input to the transportation simulation tool called MATSim. It generates detailed, individual transport behavior considering the transportation infrastructure. The output from MATSim is mapped to a physical power system including intelligent PEV management devices called PEV Managers. In case of network overloads, these devices are able to distribute scarce power efﬁciently to the connected PEVs. The functionality of the developed method is illustrated using a test case where no active PEV rescheduling is performed. It could be observed that uncontrolled charging leads to large load peaks stressing the electric infrastructure. However, the rescheduling of the vehicles charging performed only by the PEV Managers can lead to changes in the agents’ transport behavior. Therefore, the generated control price signals are fed back to a smart scheduler which takes advantage of the available information on both the power and transportation systems and alters the charging behavior in a favorable way from the system point of view. It is demonstrated that the scheduler is able to reduce the maximum load peak induced by the vehicles to less than 17% of the original value, while achieving a behavioral equilibrium which incorporates individual activity plans, the urban trafﬁc situation and the power system state. The integrated method proves to be effective for investigating the impact of electric mobility on the systems affected by this new technology path. The tremendous advantage of the model is derived from the availability of information on individual agents’ temporal and spatial behavior affecting the utilized infrastructures. However, to enable wide scale electric mobility and its efﬁcient management, the current power system infrastructure needs to be adapted. Charging facilities and a proper communication infrastructure for PEVs as well as in distribution networks is need to be build so that algorithms, such as the proposed one, can be implemented. APPENDIX TABLE II CAPACITY OF LINES AND TRANSFORMERS VIII. CONCLUSIONS AND OUTLOOK Electric mobility will heavily impact the individual transportation sector and the power sector. For power systems, well known load curves and utilization rates of network assets will be drastically changed, possibly resulting in local or system wide overloads. Trafﬁc ﬂows will be altered as charging infrastructure affects route planning. Finally, greenhouse gas emissions from an altered power generation merit order will change the overall environmental impact of private transport. The paper presents an integrated method which is able to assess the impacts of electric mobility on the domains of power and transportation systems as well as on the environment. The method comprises models for the vehicle ﬂeet evolution, for the development of transport emissions, for the spatial and temporal transportation behavior and for power systems. The method derives energy consumptions for the vehicle ﬂeet from real world drive cycles and advanced vehicle models. TABLE III LINE PARAMETERS GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY Fig. 15. Electricity base load curves of the different energy hubs. Fig. 16. 4 node test power system used in the example. ACKNOWLEDGMENT This work was carried out within the framework of the research project Vision of Future Energy Networks. The authors would like to thank Emil Iggland for substantial linguistic revision. REFERENCES [1] T. H. Bradley and A. A. Frank, “Design, demonstrations and sustainability impact assessments for plug-in hybrid electric vehicles,” Renewable Sustainable Energy Rev., vol. 13, no. 1, pp. 115–128, 2009. [2] J. A. P. Lopes, F. J. Soares, and P. M. R. Almeida, “Identifying management procedures to deal with connection of electric vehicles in the grid,” in Proc. IEEE PowerTech, Bucharest, Romania, Jun. 2009. [3] M. D. Galus and G. Andersson, “Demand management for grid connected plug-in hybrid electric vehicles (PHEV),” in Proc. IEEE Energy 2030 Conf., Atlanta, GA, Nov. 2008. 947 [4] M. D. Galus, R. A. Waraich, and G. Andersson, “Predictive, distributed, hierarchical charging control of PHEVs in the distribution system of a large urban area incorporating a multi agent transportation simulation,” in Proc. Power Syst. Comput. Conf. (PSCC), Stockholm, Sweden, Aug. 2011. [5] C. Guille and G. Gross, “A conceptual framework for the vehicle-to-grid (V2G) implementation,” Energy Policy, vol. 37, no. 11, pp. 4379–4390, 2009. [6] M. D. Galus, M. Zima, and G. Andersson, “On integration of PHEVs into existing power system structures,” Energy Policy, vol. 38, pp. 6736–6745, 2010. [7] Z. Ma, D. S. Callaway, and I. A. Hiskens, “Decentralized charging control of large populations of plug-in electric vehicles,” IEEE Trans. Control Syst. Technol., pp. 1–12, 2011 [Online]. Available: ieeexplore. ieee.org/xpls/abs_all.jsp?arnumber=6081962 [8] D. S. Callaway and I. A. Hiskens, “Achieving controllability of electric loads,” Proc. IEEE, vol. 99, no. 1, pp. 184–199, 2011. [9] N. Rotering and M. D. Ilic, “Optimal charge control of plug-in hybrid electric vehicles in deregulated electricity markets,” IEEE Trans. Power Syst., vol. 26, no. 3, pp. 1021–1029, 2011. [10] S. Midlam-Mohler, S. Ewing, V. Marano, Y. Guezennec, and G. Rizzoni, “PHEV ﬂeet data collection and analysis,” in Proc. IEEE Veh. Power Propulsion Conf. (VPPC), Dearborn, MI, Sep. 2009, pp. 1205–1210. [11] M. Rieser, K. Nagel, U. Beuck, M. Balmer, and J. Ruemenapp, “Agent oriented coupling of activity-based demand generation with multiagent trafﬁc simulation,” Transp. Res. Rec., J. Transp. Res. Board, vol. 2021, no. 2007, pp. 10–17, 2007. [12] “Comparing the beneﬁts and impacts of hybrid electric vehicle options,” EPRI, Tech. Rep. 1000349, Jul. 2001, EPRI. [13] M. D. Galus, R. La Fauci, and G. Andersson, “Investigating PHEV wind balancing capabilities using heuristics and model predictive control,” in Proc. IEEE Power Energy Soc. (PES) Gen. Meet., Minneapolis, MN, Jul. 2010. [14] M. D. Galus and G. Andersson, “Balancing renewable energy sources with vehicle to grid (V2G) services from a large ﬂeet of plug in hybrid electric vehicles controlled in a metropolitan area distribution network,” in Proc. Cigre Symp. ‘The Electric Power System of the Future’, Bologna, Italy, Sep. 2011. [15] M. D. Galus, S. Koch, and G. Andersson, “Provision of load frequency control by PHEVs, controllabel loads and a co-generation unit,” IEEE Trans. Ind. Electron., vol. 58, no. 10, pp. 4568–4582. [16] A. Ulbig, M. D. Galus, and G. Andersson, “General frequency control with aggregated control reserve capacity from time-varying sources: The case of PHEVs,” in Proc. iREP Symp.—Bulk Power Syst. Dyn. Control VIII, Buzios, Brazil, Aug. 2010. [17] E. Larsen, D. Chandrashekhara, and J. Ostergard, “Electric vehicles for improved operation of power systems with high wind power penetration,” in Proc. IEEE Energy 2030 Conf., Atlanta, GA, Nov. 2008. [18] S. L. Andersson, A. K. Elofsson, M. D. Galus, L. Göransson, S. Karlsson, F. Johnsson, and G. Andersson, “Plug-in hybrid electric vehicles as regulating power providers: Case studies of Sweden and Germany,” Energy Policy, pp. 2751–2762, 2010. [19] P. de Haan and M. Keller, “Real-world driving cycles for emission measurements: Artemis and Swiss cycles,” INFRAS—Consulting Group for Policy Analysis and Implementation, Swiss Agency for Environment, Forests and Landscape (SAEFL), Tech. Rep. Folgearbeiten zum BUWAL-Bericht SRU Nr. 255, 2001. [20] M. D. Galus and G. Andersson, “Integration of plug-in hybrid electric vehicles into energy systems,” in Proc. IEEE PowerTech, Bucharest, Romania, Jun. 2009. [21] A. Ipakchi and F. Albuyeh, “Grid of the future,” IEEE Power Energy Mag., vol. 7, no. 2, pp. 52–62, 2009. [22] F. G. Noembrini, “Modeling and analysis of the Swiss energy system dynamics with emphasis on the interconnection between transportation and energy conversion,” Ph.D. dissertation, ETH Zurich, Zurich, Switzerland, 2010, 10.3929/ethz-a-006032819. [23] M. Keller, “Der Energieverbrauch des Verkehrs 1990–2035, Ergebnisse der Szenarien I-IV und der zugehörigen Sensitivitäten ‘BIP hoch’, ‘Preis hoch’, und ‘Klima wärmer’,” (in German) Swiss Federal Ofﬁce of Energy (BFE), Tech. Rep.. [24] P. Meyer, Integrated Market Functions (in German), 4th ed. Stuttgart, Germany: Kohlhammer, 1996. [25] A. Janssen, “Modeling the market penetration of passenger cars with new drive train technologies,” Ph.D. dissertation, ETH Zurich, Zurich, Switzerland, 2008. 948 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 2, JUNE 2012 [26] L. Guzzella and A. Sciaretta, Vehicle Propulsion Systems, 2nd ed. New York: Springer, 2007. [27] M. D. Galus and G. Andersson, “Power system considerations of plug-in hybrid electric vehicles based on a multi energy carrier model,” in Proc. IEEE Power Energy Soc. (PES) Gen. Meet., Calgary, AB, Canada, Jul. 2009. [28] MATSim-T. Multi Agent Transportation Simulation Toolkit, 2008. [29] D. Charypar, K. W. Axhausen, and K. Nagel, WCTRS, “An eventdriven parallel queue-based microsimulation for large scale trafﬁc scenarios,” in Proc. 11th World Conf. Transp. Res., Jun. 2007. [30] R. A. Waraich, D. Charypar, M. Balmer, and K. W. Axhausen, STRC, “Performance improvements for large scale trafﬁc simulation in matsim,” in Proc. 9th Swiss Transp. Res. Conf., Sep. 2009. [31] M. Balmer, “Travel demand modeling for multi-agent trafﬁc simulations: Algorithms and systems,” Ph.D. dissertation, ETH Zurich, Zurich, Switzerland, May 2007, 10.3929/ethz-a-005429518. [32] , J. H. Holland, Ed., Adaptation in Natural and Artiﬁcial Systems: An Introductory Analysis With Applications to Biology, Control, and Artiﬁcial Intelligence. Cambridge, MA: MIT Press, 1992. [33] D. Charypar and K. Nagel, “Generating complete all-day activity plans with genetic algorithms,” Transportation, vol. 32, no. 4, pp. 369–397, 2005. [34] M. Fahrioglu and F. L. Alvarado, “Designing incentive compatible contracts for effective demand management,” IEEE Trans. Power Syst., vol. 15, no. 4, pp. 1255–1260, 2000. [35] Y. Narahari, D. Garg, R. Narayanam, and H. Prakash, Game Theoretic Problems in Network Economics and Mechanism Design Solutions. London, U.K.: Springer, 2009, vol. 1. [36] J. Von Neumann and O. Morgenstern, Theory of Games and Economic Behavior. Princeton, NJ: Princeton Univ. Press, 1955, vol. 6. [37] D. Fudenberg and J. Tirole, Game Theory. Cambridge, MA: MIT Press, 1991, London, U.K. [38] D. Kirschen and G. Strbac, Fundamentals of Power System Economics. Hoboken, NJ: Wiley, 2004. [39] S. Koch, M. D. Galus, S. Chatzivasiliadis, and G. Andersson, “Emergency control concepts for future power systems,” in Proc 18th IFAC World Congr., Milan, Italy, Aug. 2011. [40] H. W. Kuhn and A. W. Tucker, “Nonlinear programming,” in Proc. 2nd Berkely Symp. Math. Stat. Probab., Berkely, CA, 1951. [41] M. Geidl and G. Andersson, “Optimal power ﬂow of multiple energy carriers,” IEEE Trans. Power Syst., vol. 22, no. 1, pp. 145–155, 2007. [42] M. D. Galus, R. Waraich, M. Balmer, K. W. Axhausen, and G. Andersson, “A framework for investigating the impacts of plug-in hybrid electric vehicles,” in Proc. Int. Adv. Mobility Forum (IAMF), Geneva, Switzerland, Mar. 2009. Matthias D. Galus (S’07) was born in Swientochlowitz, Poland. He received the Dipl.-Ing. degree in electrical engineering and the Dipl.-Ing. degree in industrial engineering from the RWTH Aachen, Germany, in 2005 and in 2007, respectively. He joined the Power Systems Laboratory of ETH Zurich, Switzerland in 2007, where he is working toward the Ph.D. degree. He spent research visits at the Pennsylvania State College (PSU), at the Massachusetts Institute of Technology (MIT), and at INESC Porto, Portugal. His research is dedicated to modeling, optimization, and efﬁcient integration of PEV into power systems. Mr. Galus is a Student Member of the VDE (German society of electrical engineers). Rashid A. Waraich was born in Sargodha, Pakistan. He studied computer science at ETH Zurich, Switzerland, with an exchange year at KTH Stockholm and his ﬁnal masters semester at University of California, Berkeley, where he wrote his master’s thesis in the area of distributed systems and information security. He received the M.Sc. degree in computer science from ETH Zurich in 2007. In 2008, he joined the Institute for Transport Planning and Systems at ETH Zurich, where he is working towards the Ph.D. degree. His research interests include large-scale parallel microsimulation, simulation of plug-in hybrid electric vehicles, and parking search behavior. Fabrizio Noembrini received the Diploma degree in mechanical and process engineering and the Ph.D. degree from the ETH Zurich, Switzerland, in 2003 in 2010. He is currently managing director of the Energy Science Center (ESC) of the ETH Zurich and in charge of the Energy Systems Group at the Aerothermochemistry and Combustion Systems Laboratory at the Institute for Energy Technology, ETH Zurich, in the group of Prof. Boulouchos, and also lecturer at ETH Zurich. His research focus is on the integrated assessment of vehicles with respect to cumulative energy demand, greenhouse gas emissions and costs, based on a life-cycle approach considering interactions with energy conversion sectors. Karel Steurs received the M.Sc. degree in electro-mechanical engineering from Ghent University, Belgium. He is currently working toward the Ph.D. degree in mechanical engineering at the Aerothermochemistry and Combustion Systems Laboratory, at the Institute for Energy Technology, ETH Zurich, Switzerland. Gil Georges received the M.S. degree in mechanical engineering from ETH Zurich, Switzerland in 2010. Currently he is working toward the Ph.D. degree in mechanical engineering with the Aerothermochemistry and Combustion Systems Laboratory at the Institute for Energy Technology, ETH Zurich. His research focus is on the technology assessment of new drivetrain technologies on a transport-system level. By means of large-scale vehicle simulations the entire current and foreseeable future ﬂeet is investigated to predict the performance, operation costs, primary energy demand, and total emission levels under various trafﬁc conditions and varying driver behavior. Konstantinos Boulouchos was born in 1955 in Greece. He received the Diploma degree in mechanical engineering from the National Technical University of Athens, Greece, in 1978 and the Ph.D. degree in thermodynamics and combustion engines from the Swiss Federal Institute of Technology (ETH), Zürich, in 1984. In 2002 he was elected Full Professor and head of the Aerothermochemistry and Combustion Systems Laboratory at ETH Zurich. His research interests focus on fundamentals of chemically reactive systems in energy conversion technology with regard both to modeling and simulation of laminar and turbulent reactive ﬂows and to non-intrusive diagnostic methods in combustion systems. Dr. Boulouchos is a member of the Steering Committee of the Swiss Competence Center for Energy and Mobility (CCEM), the Advisory Board of the Institute for Vehicle Technology of the German Aerospace Center (DLR) and the board of trustees of ProClim—the Swiss forum for Climate and Global Change issues. GALUS et al.: INTEGRATING POWER SYSTEMS, TRANSPORT SYSTEMS AND VEHICLE TECHNOLOGY Kay W. Axhausen was born in Heidelberg, Germany. He received the M.Sc. degree in civil and environmental engineering from the University of Wisconsin-Madison in 1984 and the Ph.D. degree from the University of Karlsruhe, Germany, in 1988. In 1989 he joined the University of Oxford, U.K., as Senior Research Ofﬁcer and in 1991 he was appointed Lecturer and later Senior Lecturer at the Imperial College London, U.K. He worked as a full Professor at the University of Leopold-Franzens, Innsbruck, from 1995 to 1999. Since 1999 he is full Professor at the Institute for Transport Planning and Systems at ETH Zurich, Switzerland, where he heads the Transport Planning group. His current work focuses on the agent-based micro-simulation toolkit MATSim (see www.matsim.org). He was the chair of the International Association of Travel Behavior Research (IATBR) and is an editor of Transportation and DISp, both ISI indexed journals. 949 Göran Andersson (M’86–SM’91–F’97) was born in Malmö, Sweden. He received the M.Sc. and Ph.D. degrees from the University of Lund, Sweden, in 1975 and 1980, respectively. In 1980 he joined ASEA, now ABB, HVDC division, Ludvika, Sweden, and in 1986 he was appointed full Professor in electric power systems at the Royal Institute of Technology (KTH), Stockholm, Sweden. Since 2000 he is full Professor in electric power systems at ETH Zurich, Switzerland, where he heads the Power Systems Laboratory. His research interests are in power system analysis and control, in particular power system dynamics and issues involving HVDC and other power electronics based equipment. Dr. Andersson is a member of the Royal Swedish Academy of Engineering Sciences and Royal Swedish Academy of Sciences.

RELATED PAPERS

RELATED TOPICS

Log In

Integrating Power Systems, Transport Systems and Vehicle Technology for Electric Mobility Impact Assessment and Efficient Control

Integrating Power Systems, Transport Systems and Vehicle Technology for Electric Mobility Impact Assessment and Efficient Control

Related Papers

RELATED PAPERS

RELATED TOPICS