A Monte Carlo Method to Evaluate Electric Vehicles Impacts in Distribution Networks F. J. Soares*, Student Member, IEEE, J. A. Peças Lopes*, Senior Member, IEEE, P. M. Rocha Almeida*, Student Member, IEEE * The authors are with Instituto de Engenharia de Sistemas e Computadores do Porto (INESC Porto) and with the MIT Portugal Program on Sustainable Energy Systems from Faculdade de Engenharia da Universidade do Porto (FEUP); Address: Rua Dr. Roberto Frias, 378, 4200 - 465 Porto, Portugal; Tel.: +351 222 094 000; E-mails: fsoares@inescporto.pt, jpl@fe.up.pt, pedro.almeida@fe.up.pt The example of a small grid from one of the Azores islands, Flores Island, was used for illustration purposes and two scenarios of EV integration were simulated: 25% and 50% of the current light vehicles fleet replaced by EV. After gathering all the results, an evaluation was made to identify the most problematic zones of the network, concerning voltages and lines loading violations. Abstract—This paper describes a statistical approach developed for assessing the impacts resulting from EV presence in a given electricity network was developed. The algorithm, developed for this purpose, is based on a Monte Carlo method and can be seen as a planning tool that allows obtaining average values for several system indexes, like buses voltages, branches loading and energy losses. Additionally, it also allows identifying the most critical operation scenarios and the network components that are subjected to more demanding conditions and that might need to be upgraded. The example of a small grid from one of the Azores islands, Flores Island, was used for illustration purposes and two scenarios of EV integration were considered: 25% and 50% of the current light vehicles fleet replaced by EV. 2. FLORES ISLAND NETWORK The island distribution network is presented in Figure 1. It is a very robust 15 kV Medium Voltage (MV) grid, which encloses 44 branches and 45 buses. From these, only buses 1, 19, 28 and 43 do not have any loads connected. The present network load diagram, for a typical winter weekday, is presented in Figure 6 (power demand without EV curve). The total energy demand for the selected day is 47.55 MWh and the peak load is 2.59 MW, occurring at 19:30 h. The average power factor (ratio between active and apparent power) in this island, for a typical day, is 0.77. 1. INTRODUCTION Global warming problematic is attracting growing concerns across the world, leading policy makers to seek solutions for reducing the emissions of Greenhouse Gases (GHG). The transportation sector accounts for over than half of the world’s oil consumption [1] and was responsible for 13.5% of the world’s GHG emissions in 2005 [2]. One of the most effective measures policy makers can adopt to reduce transportation sector dependency on fossil fuels is to promote the progressive replacement of the existent conventional vehicles by Electric Vehicles (EV) [3]. Yet, massive EV integration is likely to pose some problems to the distribution system operation and planning [4], [5]. Given these facts, it is of utmost importance to develop adequate tools, based in mathematical methods, capable of establishing a good representation of EV behaviour and its impacts on the electricity grids. In this sense, it was developed an algorithm in this work that allows assessing the impacts resulting from EV presence in a given electricity network. This algorithm, which can be seen as a planning tool, is based on a Monte Carlo method and allows obtaining average values and confidence intervals for several system indexes, like buses voltages, branches loading and energy losses. Figure 1 – Flores Island network (15 kV) 3. ELECTRIC VEHICLES FLEET CONSIDERED The island’s light vehicles fleet is composed by 2285 conventional cars. For both scenarios addressed in this study, 25% and 50% of the fleet replaced by EV, no specific 978-1-4244-6078-6/10/$26.00 ©2010 IEEE 365 types of EV were considered. In order to tackle the uncertainties related with the type of EV that the islands’ inhabitants would prefer to use, Gaussian probability density functions were used to initially characterize all the relevant variables related with each EV. This procedure will be thoroughly described later on this document. • 4. ELECTRIC VEHICLES MOTION SIMULATION • The EV movement along one day was simulated using a discrete-time non-Markovian process to define the states of all the EV at each 30 minutes interval (48 time instants). In this process it is assumed that each EV can be in four different states: in movement, parked in industrial area, parked in commercial area or parked in residential area. Initially, a state is drawn for each EV, based on the probabilities for time instant = (hour 0 of Figure 3). The EV states for the following time instants, between 1 and 47, are defined according to the probabilities specified for each one of those time instants (see Figure 3) and according to the referred discrete-time non-Markovian process, which is represented in detail in Figure 2. As it is denoted in this figure, there are some restrictions when defining the EV states for the following time instant. While EV in movement can keep their state or change it for one of the others, parked EV can only remain in the same state or change to movement. This process is classified as non-Markovian for the reason that the state transition probabilities are not a constant function, varying with time. = = → = → = → = → = → = • • • The state transition probabilities applied in this study were determined by analyzing the common traffic patterns of Portuguese drivers. From [6] it was gathered information about the number of car journeys made per each 30 minutes interval, along a typical weekday, as well as the journey purpose and its average duration. With this data, it was possible to define the probabilities of an EV reside in a given state at a given time instant. The overall probabilities obtained are presented in Figure 3. = 1 "Movement" State "Parked in industrial area" State "Parked in commercial area" State = → = "Parked in residential area" State 0.6 0.4 0.2 = = = • 0.8 = → = → = = = • State transition probabilities = time instant t; – probability of changing from “parked in → industrial area” to “in movement” state, at time instant t; – probability of changing from “in → movement” to “parked in commercial area” state, at time instant t; – probability of changing from “parked in → commercial area” to “in movement” state, at time instant t; – probability of remaining in “parked in industrial area” state, at time instant t; – probability of remaining in “parked in commercial area” state, at time instant t; = – probability of remaining in “parked in residential area” state, at time instant t; – represents the time instants between 1 and 47 (varies between 2 and 46); – represents the last time instant of the day (n = 7). • = = 0 0 5 10 15 20 Hour Figure 2 – Discrete-time non-Markovian process Figure 3 – State transition probabilities along a typical weekday The acronyms of Figure 2 are defined next: • – probability of remaining in “in movement” state, at time instant ; – probability of changing from “in • → movement” to “parked in residential area” state, at time instant t; • – probability of changing from “parked in → residential area” to “in movement” state, at time instant t; • – probability of changing from “in → movement” to “parked in industrial area” state, at After defining the EV states, it was needed to determine where in the network each EV was located. If an EV was in movement, there was no need to define its location. However, if parked and connected to the grid for charging purposes, it was crucial to know the EV location to allocate its load to a specific network bus. The procedure adopted to solve this issue was to consider the real nature of the loads connected to each network bus. Thus, all the existing loads were classified as industrial, commercial or residential. As examples, the 400 kW load 366 are defined according with the probabilities presented in Figure 3 and Figure 4, respectively. installed at bus 25 is 35% commercial and 65% residential, while at bus 10 there is a load of 315 kW, 100% industrial. Having defined the type of loads located at each bus, using equations (1), (2) and (3), it was calculated the probability of an EV be located at a specific bus. For instance, if an EV state at time instant is “parked in a residential area”, a bus location will be drawn and assigned to it, according with a probability distribution proportional to the residential load installed in each bus. The same happens for the “parked in industrial area” and “parked in commercial area” states. = where: • • • = Draw EV states and the buses where “parked” EV are located, for the next time instant Update EV batteries SOC EV charge only when it needs (2) ∑ EV battery SOC < 30% ? EV charge whenever possible No (3) EV is parked in residential area ? – probability of an EV be located in bus i, if parked in residential/industrial/commercial area; / / – residential/industrial/commercial load installed in bus i; / / ∑ – network total residential/industrial/ commercial load. No EV arrived home from the last journey of the day ? Yes Yes No Yes / / No EV is parked in residential area ? Yes ∑ EV charge at the end of the day or whenever is convenient and the driver has time What is the EV driver behaviour ? (1) ∑ EV do not charge Sample generation and evaluation = Define EV initial conditions (initial state, bus, battery capacity, slow charging rated power, initial SOC, energy consumption and driver behaviour) EV starts charging No Determine the new load at each bus Power flow analysis The allocation probabilities of each bus, for each EV state, are presented in Figure 4. No End of the day was reached ? Yes "Parked in industrial area" State "Parked in commercial area" State "Parked in residential area" State 0.4 Indexes update 0.45 Update of grid technical indexes and vehicle usage indicators in a hourly and daily basis Bus allocation probabilities 0.35 Monte Carlo finishing criteria was met ? 0.3 Yes 0.25 Compile results: power demand, voltages, branches loading, energy losses, peak power, number of voltage and branches ratings violations 0.2 Figure 5 – Monte Carlo algorithm flowchart 0.15 The EV battery capacity, slow charging power, energy consumption and initial battery SOC are defined according with Gaussian probability density functions, whose average, standard deviation, maximum and minimum values allowed are presented in Table I. 0.1 0.05 0 5 10 15 20 25 30 35 40 45 Bus Figure 4 – Buses allocation probabilities for “parked” EV states TABLE I GAUSSIAN DISTRIBUTIONS FOR INITIAL EV CHARACTERIZATION 5. MONTE CARLO ALGORITHM 5.1. Samples Generation with Monte Carlo Battery capacity (kWh) Slow charging rated power (kW) Energy consumption (kWh/km) The flowchart of the Monte Carlo algorithm developed in this work is presented in Figure 5. The first step of the algorithm is to make the initial characterization of all the EV, concerning their initial state (in movement, parked in industrial area, parked in commercial area or parked in residential area), the bus they are initially located, battery capacity (kWh), slow charging rated power (kW), initial State Of Charge (SOC) (%), their energy consumption (kWh/km) and their owners’ behaviour. As mentioned above, the EV initial state and location bus Initial battery SOC (%) Average Standard deviation Maximum value allowed Minimum value allowed 24.73 17.19 85.00 5.00 3.54 1.48 10.00 2.00 0.18 0.12 0.85 0.09 50.00 25.00 85.00 15.00 While the initial battery SOC values were assumed for the purpose of this work, the values for battery capacity, slow charging rated power and energy consumption were obtained from the information made available on the Internet by the manufacturers of 42 different EV. In [7]-[13] 367 The fast charging was assumed to be made during 15 minutes with a power of 40 kW [15]. The fast charging station was considered to be installed in bus 12, as this is located near one of the more populated areas of the island, with a high number of potential clients. The average of the Gaussian distribution used to characterize the travelled distance in common journeys was obtained by dividing the average daily mileage in Portugal (35 km) [16] by the average number of journeys per day (3.88 journeys) [6]. The standard deviation of this Gaussian distribution was considered to be 50% of the average. The values of the Gaussian distribution used for the travelled distance to the fast charging station, were obtained by assuming that they were 50% of those used in the travelled distance in common journeys distribution. At each time instant, the EV battery SOC is updated according with the energy spent travelling or with the energy absorbed in residential slow charging or fast charging stations. It is important to stress that, in this work, EV were only allowed to charge when their state was “parked residential area”. Having the EV power consumption fully defined, together with the network initial load, it is obtained the total amount of power required from the network, decriminalized per bus and per time instant. are presented some of the Internet sites from where the EV characteristics were obtained. The maximum and minimum values allowed, presented in Table I, were used to confine the values drawn for each EV within realistic boundaries. As mentioned previously, a given driver behaviour was also assigned initially to each EV. The different behaviours considered in this study were defined according with the findings of an Internet survey made within the framework of the MERGE project [14]. The results revealed that there are four major types of behaviours regarding EV charging, as presented in Table II. TABLE II DRIVERS’ BEHAVIOURS CONSIDERED [14] Percentage of the responses EV charge at the end of the day 33% EV charge only when it needs 23% EV charge whenever possible 20% EV charge whenever is convenient and the driver has time 24% For the purpose of this work, regarding the behaviours modelling and simulation, there was no relevant differences between the drivers that “charge at the end of the day” and those who “charge whenever is convenient and they have time”. Therefore, the EV to which one of these drivers’ behaviours was assigned, were assumed to behave equally along the simulations. For the drivers who charge their EV only when it needs, it was assumed that the minimum battery SOC that triggers the need for charging was 30%. The next step of the Monte Carlo algorithm was to simulate EV movement along one typical weekday. To start with, the EV states for each time instant were defined according with the discrete-time non-Marvovian process described in section 4. and with the probabilities presented in Figure 3. After, for each time instant, a bus location was attributed to parked EV, as explained in section 4. and according with the probabilities presented in Figure 4. For the EV in movement, a procedure was developed to account their energy consumption and the respective reduction in the battery SOC. First a Gaussian probability density function was used to draw the travelled distances for all the EV in movement. Therefore, if an EV was in movement in time instant and its battery SOC went below a predefined threshold (assumed to be 15%) in time instant , it was considered that the EV would make a short detour to a fast charging station for recharging purposes. The travelled distance during the detour was obtained using also a Gaussian probability density function, whose parameters are presented in Table III. 5.2. Samples Evaluation The evaluation of the samples is made by running a power flow for each time instant, using the PSS/E software, being gathered information about the voltage profiles, the power flows in the lines and the global value of the energy losses in the network. During the simulation, the average, maximum and minimum power demand and voltage value is recorded for each bus of the system. A similar procedure is adopted for the power flows in the lines. The scenario where the highest peak load occurs is also recorded, in order to provide an idea of the worst situation that might occur when 25% and 50% of the conventional vehicles are replaced by EV. In order to keep track of the most problematic buses and lines within the grid, the number of out of limit voltages and lines loading occurrences are recorded along the simulations. According with [17], voltages must be kept with the interval 0.90 – 1.10 p.u. during 95% of the time, in a weekly basis. A security margin of 0.03 p.u. was assumed in this work, being considered voltage violations all the values outside the interval 0.93 – 1.07 p.u.. 5.3. Terminating the Monte Carlo Process To terminate the Monte Carlo process, two criteria were used: number of iterations and the variances of the aggregated network load of each one of the 48 time instants. The process was set to perform 10000 iterations and check, in the end, if the variation of the 48 variances in the last 10 iterations was lower than . If at least one of the 48 variances did not meet this convergence criterion, the process was kept running more iterations until all the TABLE III GAUSSIAN DISTRIBUTIONS FOR EV MOVEMENT CHARACTERIZATION Travelled distance in common journeys (km) Travelled distance to fast charging station (km) Average Standard deviation Maximum value allowed Minimum value allowed 9.01 4.51 27.03 0.90 4.51 2.25 13.52 0.45 368 to average voltage values of 10000 iterations, and so they mask worst scenarios results that, as Figure 10 shows, violate by far the voltage lower limit of 0.93 p.u.. variances variations were lower than the predefined value. 6. RESULTS ANALYSIS To establish a proper comparison between the scenarios of EV integration studied, the results regarding network impact assessment were compiled into tables and figures presented along the next subsections. The power demand, voltage profiles, branches loading and the daily energy losses were analyzed, as well as the sample variance. 0.99 0.99 Voltages without EV (p.u.) 0.98 6.1. Power Demand Figure 6 shows the power demand, in MW, for both EV integration scenarios studied (25% and 50%), as well as for the network in its initial conditions, where no EV were considered. 0.98 0.97 0.97 0.96 0.96 0.95 0.95 0.94 0.93 0.94 0.92 0.93 0.91 0.9 17 6 0.92 0.91 19 21 23 Bus Power demand without EV Average power demand with 25% EV Average power demand with 50% EV 25 27 0 6 12 18 24 4 1 1 0.99 3 0.99 Voltages with 25% EV (p.u.) 0.98 2 1 0 4 8 12 Hour 16 0.9 Hour Figure 7 – Voltages in buses 17 to 27, in the scenario without EV 5 Power demand (MW) 1 1 20 24 Figure 6 – Power demand along a typical winter weekday 0.97 0.96 0.96 0.95 0.95 0.94 0.93 0.94 0.92 0.93 0.91 0.92 0.9 17 It is interesting to notice that the peak hour changes, in both scenarios with EV, from 19:30 h to 00:00 h and that the demand at the end of the day is very different from the value observed at the initial time instant. This results from the values initially assumed for the Gaussian distribution used to define EV SOC at time instant = . A different Gaussian distribution would lead to different power demand values, principally during the first hours of the day. With 25% of EV, the power demand in the peak hour increases 41%, from 2.6 to 3.7 MW, whereas with 50% of EV it increases 109%, from 2.6 to 5.4 MW. The daily energy demand increases from 95.1 to 124.7 MWh, with 25% of EV, and to 152.7 MWh, with 50% of EV. 0.98 0.97 0.91 19 21 23 Bus 25 27 0 12 6 18 24 0.9 Hour Figure 8 – Voltages in buses 17 to 27, in the scenario with 25% EV 1 1 0.99 0.99 Voltages with 50% EV (p.u.) 0.98 6.2. Voltage Profiles 0.98 0.97 0.97 0.96 0.96 0.95 0.95 0.94 0.93 0.94 0.92 0.93 0.91 To provide a clear picture of the EV impact in terms of voltage profile of one feeder, along the entire day, the average voltages obtained for buses 17 to 27 were compiled and presented in Figure 7 (scenario without EV), Figure 8 (scenario with 25% EV) and Figure 9 (scenario with 50% EV). As these three charts show, the extra power demanded by EV provokes a considerable voltage drop along this feeder, namely at the beginning and at the end of the day. It should be mentioned that the results presented are referred 0.9 17 0.92 0.91 19 Bus 21 23 25 27 0 6 Hour 12 18 24 0.9 Figure 9 – Voltages in buses 17 to 27, in the scenario with 50% EV In order to assess the worst voltage conditions that these levels of EV integration might lead to, the highest peak load 369 scenarios registered along the 10000 iterations were analyzed, and the corresponding voltage values were plotted in Figure 10. The probabilities presented in this figure were obtained using equation (4): . 1 Peak hour voltage (p.u.) 0.96 0.94 0.92 0.9 0.86 Without EV With 25% EV With 50% EV System operator voltage lower limit EN 50160 voltage lower limit 5 10 15 . . (4) Given that out of limit voltages only occurred in buses 18 to 40, only these were chosen to appear in Figure 11. For the scenario with 25% of EV, only a small number of violations were record. There were only some problems in buses 38, 39 and 40.The probability of having in these buses voltages below 0.93 p.u. is lower than 2%. In what regards the scenario of 50% EV integration, the probability of having voltages below the imposed limit is rather significant in a large number of buses. The highest probabilities appear again for buses 38, 39 and 40, reaching values around 12%. 0.98 0.88 = 20 25 30 35 40 45 6.3. Branches loading Bus Figure 10 – Network voltage profiles for the highest peak load identified in both scenarios of EV integration Even though this is not the most critical aspect of this network, since the highest line loading in the scenario with 50% of EV is only 54%, branches loading is also an issue that deserves special attention. In fact, in other networks with different characteristics, the lines loading can be the limiting factor to high integration levels of EV. Figure 12, Figure 13 and Figure 14 provide an overall idea of the impact provoked by EV in the network lines loading, during the average peak load demand (average of the peak loads obtained along the 10000 iterations). Figure 12 is referred to the average peak load of the scenario without EV, while Figure 13 and Figure 14 are referred to the average peak load of the scenarios with 25% and 50% of EV, respectively. The colour grading between light green and dark red stands for increasing line loading values, ranging from 0 to 100%. The network voltage profile in the scenario without EV is also presented in Figure 10, for comparison purposes, as well as two reference voltage levels: one stipulated by EN 50160 [17] and the other with a security margin of 0.03 p.u.. While an EV integration of 25% does not decrease voltages to very problematic values, 50% of EV integration lead voltages to considerably low values, which in almost 50% of the buses are below the 0.93 p.u. threshold. Moreover, the voltages in the higher EV integration scenario violate the limit of 0.90 p.u. in more than 25% of the network buses. As mentioned in subsection 5.2, all the voltage and lines loading limit violations were recorded along the simulation, in order to keep track of the most problematic areas of the network. As this network is exceptionally robust, despite the increase in the network power flows, no lines loading above the limit were registered. Given that the EV integration only increases consumption, only voltages decrease will occur and so, as it is obvious, no voltages above the higher limit were recorded. Conversely to high voltage problems, voltages below the lower limit occurred very often, as denoted in Figure 11. 14 With 50% EV With 25% EV Voltage lower limit violation probability (%) 12 10 8 6 4 Figure 12 – Lines loading for the average peak load demand of the scenario without EV 2 As expected, the line loadings increase with the number of EV in the network, being the most problematic branches those located in the beginning of the heavily loaded feeders. 0 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 Bus Figure 11 – Voltage lower limit violation probability 370 The daily energy losses grow 58% from the scenario without EV to the one with 25% of EV and 140% to the one with 50% of EV. As expected, these values prove that energy losses do not follow linearly the load demand increase. Thus, for higher EV integration levels, where the network would be operated near its technical limits, losses are likely to be a rather important issue for the system operator, since it will be wasting large amounts of money when delivering energy to the final consumers. Therefore, in such circumstances, the system operator should look for efficient mechanisms to manage EV charging somehow, in order to avoid this problem. 6.5. Monte Carlo Convergence and Sample Variance As mentioned in section 5.3. the Monte Carlo process ends when two criteria are met: when 10000 iterations are performed and when the variations of the 48 variances between the last 10 iterations are lower than . The variances variation is calculated using equation (5): Figure 13 – Lines loading for the average peak load demand of the scenario with 25% of EV ∆ = (5) where represents the variance of the network load at time instant (that varies between 0 and 47), in the iteration. For all the scenarios simulated, the variances variation criterion was met before the algorithm reach iteration 10000. As an example, it is presented in Figure 16 the evolution of the variance with the highest variation rate of both scenarios of EV integration. As it is shown, the variation rate after iteration 5000 is very low, indicating that the algorithm reached, or at least is near to reach, the convergence criteria. 0.07 With 25% EV With 50% EV 0.06 Figure 14 – Lines loading for the average peak load demand of the scenario with 50% of EV Variance 0.05 6.4. Energy Losses In Figure 15, it is depicted the average value of the daily energy losses, obtained along all the iterations performed. 0.04 0.03 0.02 2500 0.01 0 0 Average daily energy losses (kWh) 2000 2000 3000 4000 5000 6000 Monte Carlo iterations 7000 8000 9000 10000 Figure 16 – Evolution of the variances with the highest variation rate in both scenarios of EV integration 1500 When iteration 10000 was reached in the simulation of the scenario with 25% of EV, the variance with the highest ∆ converged to 9. . For the scenario with 50% of EV, the variance with the highest ∆ converged . The higher value registered in this scenario to 7. indicates that the samples generated have a lower precision when compared with the samples of the scenario with 25% of EV. In simpler terms, these findings show that the 1000 500 0 1000 Without EV With 25% EV With 50% EV Figure 15 – Average daily energy losses 371 The simulation platform developed in this work proved to be very efficient in performing a realistic evaluation of the impacts that result from a massive integration of EV in distribution networks. Besides the evaluation of the steady state operating conditions of the grid, it also allows identifying the most critical operation scenarios and the network components that are subjected to more demanding conditions and that might need to be upgraded. uncertainty in the network load estimation is larger in the scenario with a higher number of EV. The variances analysis led to a further interesting finding: the values to which the 48 variances tend, when grouped together, present the same curve form of the state transition probabilities to in movement state, as shown in Figure 3. This evidence can be found in both scenarios of EV integration, as shown in Figure 17. In an ultimate analysis, in a not completely surprising way, these results reveal that there is a very high correlation between the EV motion patterns and the uncertainties in the network power demand estimation. REFERENCES [1] 0.07 [2] With 50% EV With 25% EV [3] 0.06 Variance 0.05 0.04 [4] 0.03 0.02 [5] 0.01 0 4 8 12 Hour 16 20 [6] 24 [7] [8] [9] [10] [11] [12] [13] [14] Figure 17 – Values obtained for the network load variances of the 48 time instants 7. CONCLUSIONS By analysing the results obtained for the case study addressed in this work, it might be concluded that the island network is very robust, being therefore capable of integrating a large number of EV without the occurrence of line loading and voltage limits violations. With 25% of EV only a small number of voltage lower limit violations were recorded along the 10000 iterations performed. However, in the scenario with 50% of EV the number of violations registered was greatly increased. These results show that while the network resists to a 25% replacement of the conventional vehicles fleet by EV, it is impossible to proceed to a 50% replacement rate without making large investments in network reinforcements, in order to tackle the low voltage problems identified. Important findings were also made regarding the energy losses. Their value grows 58% from the scenario without EV to the one with 25% of EV and 140% to the one with 50% of EV. These values show that, for higher EV integration levels, losses are likely to become a very important issue for the system operator. Therefore, in such circumstances, the system operator should look for efficient mechanisms to manage EV charging, in order to avoid wasting large amounts of money in the energy distribution process and in network reinforcements. [15] [16] [17] K. A. Baumert, T. Herzog, and J. Pershing, "Navigating the Numbers: Greenhouse Gas Data and International Climate Policy," World Resources Institute, Washington D.C., 2005. 2009, Climate Analysis Indicators Tool (CAIT) Version 6.0. Available: http://cait.wri.org/cait.php?page=sectors J. A. Peças Lopes, F. J. Soares, P. M. Rocha Almeida, Patrícia C. Baptista, Carla M. Silva, Tiago L. Farias, “Quantification of Technical Impacts and Environmental Benefits of Electric Vehicles Integration on Electricity Grids”, ELECTROMOTION 2009 - 8th International Symposium on Advanced Electromechanical Motion Systems , Lille, France, July, 2009. J. A. Peças Lopes, F. J. Soares, P. M. Rocha Almeida, A. M. Moreira da Silva, “Smart Charging Strategies for Electric Vehicles: Enhancing Grid Performance and Maximizing the Use of Variable Renewable Energy Resources”, EVS24 - The 24th International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium & Exhibition, Stavanger, Norway, May, 2009. J. A. Peças Lopes, F. J. Soares, P. M. Rocha Almeida, “Identifying Management Procedures to Deal with Connection of Electric Vehicles in the Grid”, PowerTech2009 - PowerTech 2009, Bucharest, Romania, June, 2009. INE – Instituto Nacional de Estatística, “Inquérito à mobilidade da população residente”, 2000 (in Portuguese). http://www.venturi.fr http://www.micro-vett.it/ http://www.mitsubishi-cars.co.uk/imiev/innovation http://www.chevrolet.com/pages/open/default/future/volt.do http://www.peugeot.co.uk/vehicles/peugeot-car-range/peugeot-ion/ http://www.phoenixmotorcars.com/pdf/SUV-Specifications.pdf http://www.modeczev.com/fast_facts.asp N. Downing, M. Ferdowsi, "EU MERGE Project: Mobile Energy Resources in Grids of Electricity," deliverable D1.1, "Identification of Traffic Patterns and Human Behaviours," April, 2010. Decreto-Lei n.º 39/2010, Diário da República, 1.ª série — N.º 80 — April 26th, 2010 (Decree-Law in Portuguese). C. L. Azevedo, “Métodos de estimativa de volumes anuais de tráfego rodoviário - um modelo para Portugal”, Master Thesis in Transports, IST, Universidade Técnica de Lisboa, 2008 (in Portuguese). EN 50160:2007, “Voltage characteristics of electricity supplied by public distribution systems”, European Committee for Electrotechnical Standardization – CENELEC. ACKNOWLEDGMENT This work was supported in part by Fundação para a Ciência e Tecnologia under SFRH/BD/48491/2008 and SFRH/BD/47973/2008 grants and within the framework of the Project “Green Island” with the Reference MIT-PT/SESGI/0008/2008, by Fundo de Apoio à Inovação (Ministério da Economia, da Inovação e do Desenvolvimento), within the framework of the Project REIVE – Redes Eléctricas Inteligentes com Veículos Eléctricos, and by the European Commission within the framework of the European Project MERGE – Mobile Energy Resources in Grids of Electricity, contract nr.241399 (FP7). 372