Power-Mix Optimization for a Hybrid Ultracapacitor/Battery Pack in an Electric Vehicle using Real-time GPS Data

Transcription

1 Power-Mix Optimization for a Hybrid Ultracapacitor/Battery Pack in an Electric Vehicle using Real-time GPS Data Mazhar Moshirvaziri mazhar@ele.utoronto.ca Christo Malherbe Andishe Moshirvaziri Olivier Trescases ot@ele.utoronto.ca Abstract The objective of this work is to investigate the effect of an ultracapacitor/battery based hybrid energy storage system (HESS) in an electric vehicle (EV) prototype having a 2 km range. Global positioning system (GPS) data is used to enhance the HESS power-mix optimization in real-time, based on the relative position of stop signs and traffic signals. It is shown that the GPS information, which is already available in the car, is beneficial in managing the u-cap voltage and reducing the dynamic currents on the battery. Simulations comparing various power sharing algorithms show the superior performance of the GPS enhanced HESS control scheme, based on the experimental drive-cycle. It is shown that utilizing the GPS data in the power optimizer can reduce the battery s peak charge current by 38% compared to a standard HESS. Keywords Electric vehicle, hybrid energy storage system, lithium-ion battery, ultracapacitor, global positioning system, stop prediction, real-time power-mix optimization, dc-dc converter. I. INTRODUCTION Mass adoption of Electric Vehicles (EVs) has thus far been limited by high cost, despite generous government incentives, and concerns regarding the long-term performance of the battery pack. Ultracapacitors (u-caps) have symmetric input and output specific power of.5-25 kw/kg, which is at least one order of magnitude higher than typical lithium-ion (Li- Ion) based batteries []. U-caps also offer very low Equivalent Series Resistance (ESR), vastly improved cycling lifetime and thus they are complimentary to batteries in high-power automotive applications. Hybrid Energy Storage Systems (HESS) that combine batteries and u-caps intelligently have been mainly studied through system-level simulations [2] [6] with reported driving-range improvements of up to 46% [3]. The main objectives of an automotive HESS are to () minimize the battery stress during rapid acceleration in order to limit long-term capacity fading and (2) maximize the capture of the regenerative power (Regen), while reducing the wear on the mechanical brakes. Accurately predicting the battery lifetime extension due to the reduction in dynamic currents under real drive-cycle conditions is a major challenge, and is currently under testing. As the lithium battery dominates the EV system cost, extending the pack lifetime significantly with minimum incremental cost helps to increase EV adoption. The negative effect of high charge/discharge rates on the capacity fading was demonstrated in [7]. The prototype EV and chosen HESS topology is shown in Fig. and, respectively. The HESS includes a nonisolated bi-directional dc-dc converter between V uc and V bt. The chosen architecture allows () effective power-sharing control within the HESS, (2) flexible voltage swing and thus good utilization of the u-cap energy, and (3) minimal number of conversion stages from the battery to the load for high efficiency. This architecture was also used in [2] with a model predictive control to demonstrate the reduction of the discharge intensity of the battery. The main HESS control objective in [8] is to set the power-mix to operate at the optimal point of each source, minimizing the system losses and optimizing the u-cap State of Charge (SOC). Utilizing the same HESS configuration, [9] investigated the benefits of two different control approaches, namely u-cap SOC control and optimal neural network control. RAM SPI GPS I uc V uc DO = CPU Gating dc-dc AI = CRIO FPGA CAN Feedback I uc V bt BMS Vehicle CAN Bus I load = ~ Inverter PMSM Fig.. Prototype EV used in this work. The A2B weighs 75 kg with a 379 kg battery pack. HESS architecture. The goal of this project is to integrate an experimental HESS into the targeted EV prototype shown in Fig.,

2 known as the A2B. This paper reports the first major phase of the project, where the objectives are to () create a detailed electro-mechanical system model using experimental drivecycle measurements, (2) develop a new real-time powermix optimization algorithm that leverages Global Positioning System (GPS) information and (3) simulate the system performance using measured drive-cycle data to quantify the current distribution and energy consumption. A. EV Prototype The EV shown in Fig., which weighs 75 kg, was built by Toronto Electric. The chassis and body were specifically designed to carry a 33.8 kwh, 3 V, 38 kg Lithium Iron Magnesium Phosphate battery pack. Selected vehicle specifications are listed in Table I. TABLE I. A2B VEHICLE SPECIFICATIONS Vehicle Value Unit Max. Vehicle Speed, ν max 6 km/h Car Mass (without HESS) 75 kg Estimated Range 2 km Gearbox Ratio, K G 9.5 Wheel Radius, R w.28 m Battery Pack Value Unit Number of Series modules 24 Pack Mass kg Pack Volume 24.5 L Battery Sub-Module (U24-2XP) Value Unit Nominal Voltage, V bt,nom 2.8 V Module Capacity Ahr Module ESR 6 mω Specific-Energy 89. Wh/kg Specific-Power.432 kw/kg Cycle Life (2% degradation at.8c) 2,8 Cycles Drive-train Value Unit Max. Motor Speed, ω max rpm Max. Torque (continuous), T Max 65 Nm Max. Torque (for 3 sec), T Peak 7 Nm Max. Power (continuous), P Max 37 kw Max. Power (for 3 sec), P Peak 8 kw Efficiency at Nominal Operation 93 % Operating Voltage 22-4 V Motor Weight 36 kg Inverter Weight 34 kg The battery pack consists of 24 Valence U24-2XP Lithium Iron Magnesium Phosphate (LiFeMgPO4) battery modules [], connected in series to generate a 37 V bus. The distributed Battery Management System (BMS) performs cell and module balancing. The chosen battery chemistry provides a good tradeoff between cycle-life, Equivalent Series Resistance (ESR) and specific energy. The pack has a total mass and energy of 38 kg (5% of the vehicle mass) and 33.8 kwh, respectively. According to the module manufacturer, the capacity drops by 2% after 28 cycles with 2 A charge and discharge rate, which corresponds to.8 C. It is well known that the capacity-fade under real-world conditions is a complex function of temperature, time-varying current profile and depth-ofdischarge []. II. DRIVE-CYCLE DATA ACQUISITION AND ANALYSIS The EV is equipped with two CAN buses that are fed into the data acquisition system. At this stage in the testing, Regen braking is electronically disabled as a precaution to protect the battery pack from high charge currents, increasing the pack s lifetime in the Canadian climate. A 3 hour, 6 km (round trip) typical urban drive-cycle was performed in downtown Toronto in April 22. All internal sub-system parameters, as well as the GPS trajectory were recorded to develop a complete electro-mechanical model of the vehicle. The available mechanical parameters are, the vehicle speed (km/h), motor speed (rpm), motor torque (Nm), gas pedal position (%) and brake pedal status ( or ). Electrical parameters available on the CAN bus includes the battery voltage, current and SOC. The measured vehicle speed, motor speed and motor torque are shown in Fig. 2. The motor torque is always positive, since Regen is disabled. The battery voltage and current measured by the Battery Management System (BMS), and the calculated load power are shown in Fig. 3. This data, which is critical for creating the system model, is generally not available in mass production EVs. Note that the peak battery current is near 5 A (.36 C), which is more than higher than the discharge rate used in the capacity fade specification from the manufacturer. The battery pack SOC is shown in Fig. 4. The EV consumed.97 kwh net electrical energy over 6 km. The results show an average energy consumption of 9.62 kwh per km (76 kj/km), which compares favorably with the Chevy Volt [], with an official EPA measurement of 22.4 kwh per km (8 kj/km) and a much smaller 6 kwh battery pack. Of course, this is only a basic comparison, since the drive-cycles and EV payloads differ. Vehicle speed (km/h) Motor speed (rpm) Motor Torque (Nm) (c) Fig. 2. Measured vehicle speed, motor speed and (c) motor torque for a segment of the urban drive-cycle. III. HESS DESCRIPTION The designed HESS specifications are listed in Table II. TheHESSisloadedbyaninverteranda37kWrated(9kW peak) three-phase permanent magnet electric motor. Three u- cap modules (BMOD65) weighing a total of 4.5 kg ( %

3 V bt (V) The embedded system control is split into two controltargets, contained inside a 925 CompactRIO (CRIO) module with a 94 Chassis from National Instruments. The realtime controller features an 8 MHz processor with 4 GB of nonvolatile storage and 52 MB of DDR2 memory. The 8- slot reconfigurable embedded chassis features a Xilinx Virtex-5 reconfigurable I/O (RIO) FPGA. The high-level control, which includes the vehicle CAN bus and GPS monitoring, street map analysis and power-mix calculation are done in the CPU. The high-speed IGBT gating signals, digital average current-mode control compensator and protection functions are implemented in the FPGA for minimum latency. P Load (kw) (c) Fig. 3. Measured battery voltage, battery current and (c) load power for a segment of the urban drive-cycle. SOC bt Fig. 4. Measured battery SOC during the drive-cycle. of the battery pack) are connected in series to build a 44 V, 55 F pack having an ESR of 8.9 mω. The u-cap pack is interfaced to V bus using a 3 kw digitally controlled, nonisolated bi-directional dc-dc converter. The detailed design of this multi-phase converter is beyond the scope of this paper. TABLE II. HESS PARAMETERS U-cap Pack Value Unit Number of Series modules 3 Pack Mass 4.5 kg Pack Volume 43.5 L U-cap Module (BMOD65) Value Unit Nominal Voltage, V uc,nom 48 V Module Capacitance 65 F Module ESR 6.3 mω Specific-Energy 3.9 Wh/kg Specific-Power 6.77 kw/kg Cycle Life (2% Capacitance degradation),, Cycles 4 Phase DC-DC Converter Value Unit Input Voltage -44 V Output Voltage V Converter Mass 2 kg Converter Volume 3.5 L Maximum Power 3 kw A. GPS Based Power Optimizer Algorithm Keeping the u-cap SOC, SOC uc, near an optimal level based on vehicle speed is one of the key controller objectives. This conflicts with the short-term optimum power-mix, based solely on the immediate losses. As demonstrated in [2], the dc-dc converter efficiency varies with SOC uc,whichmakes it more lossy (or expensive from the optimization point of view) to draw energy from the u-cap as V uc drops. Predicting upcoming stops is extremely valuable to manage the trade-off between power losses and SOC uc management. The controller finds the optimal u-cap current, I uc,opt, each second, by solving I uc,opt =argminf cost, () I uc subject to the constraints I uc + = I load, (2).4 SOC bt, (3) 3, (4).6 SOC uc, (5) 26 I uc 26, (6) where I uc is the output current of the u-cap converter and f cost is the cost function defined by f cost = P bt,loss m bt +(P uc,loss + P conv,loss ) m uc (7) + P mech,loss, where P bt,loss = Ibt 2 R bt is the battery ESR loss, P uc,loss = IucR 2 uc is the u-cap ESR loss and P conv,loss is the loaddependent dc-dc converter loss. The mechanical braking loss, P mech,loss, is calculated based on the available energy from the braking event and the maximum energy that the HESS can safely absorb. The constants m uc and m bt are adaptive weights used to influence the power-mix. Another controller objective is to limit the peak battery charge and discharge current to I min and I max, respectively, as long as the u-cap has sufficient capacity. This is done by penalizing the higher battery peak currents by marginally increasing the cost of the battery contribution. The controller operates in five modes, based on SOC uc,asdefined in Fig. 5. In each mode () is solved, while m uc and m bt are adjusted as follows { A I m uc = uc < A 2 I uc <I min I +A 3 bt I m bt = min I min < I max +A 4 I max <. I 2 bt I 2 max

4 where the coefficients, A = {A,A 2,A 3,A 4 },aredefined in Table III. In all modes, the GPS data is used to adaptively adjust m uc and m bt. When a stop is predicted, the coefficients in A are adjusted towards higher efficiency and less conservative SOC management, as listed in Table III. This helps to minimize the losses during the large bursts of power in the next 2 m, since the Regen energy from braking will compensate the SOC uc. Without stop prediction, the power-mix decision based on minimum losses causes an undesired drop in SOC uc. Fig. 5. Mode : The dc-dc converter has the highest efficiency and the power-mix is chosen to minimize the system losses. Mode 2-3: The maximum V uc that allows the u-cap to fully absorb a Regen burst, SOC uc,des,h, is calculated based on the vehicle s kinetic energy. A band of % is defined according to SOC uc,des,h,inordertosetthe desired lower limit, SOC uc,des,l. Operating in Modes 2 and 3, within ±5% of SOC uc,des, is optimal from the Regen, acceleration and dc-dc converter efficiency point of view. Mode 4: The u-cap is rarely activated to limit the battery current, while it tries to absorb most of the Regen energy, despite the dc-dc converter losses for low V uc. Mode 5: In this mode the u-cap draws a minimum amount of power from the battery whenever the battery current is below I max to increase SOC uc.thisis a critical mode, where the u-cap is charged with lower efficiency, at the expense of higher battery current and should ideally be avoided. TABLE III. COST FUNCTION COEFFICIENTS Mode Weight Coefficients A = {A,A 2,A 3,A 4} Regular Driving Predicted {,,, 3} {,,, 3} 2 {.4,.5, 2,.8} {.5,,,.8} 3 {.2, 3, 2,.5} {.3, 2.5,,.5} 4 {., 4, 3, } {.2, 3, 2, } SOC uc.64.6 Mode: Controller modes based on u-cap SOC. SOC uc,max SOC uc,des,h SOC uc,des SOC uc,des,l SOC uc,crit SOC uc,min t B. Open Street Map and GPS Data Processing A post-processed version of the vectorized street map of Toronto [3] is stored in the vehicle controller. GPS data processing is performed to detect the presence and relative position of stop signs and traffic lights that are within 2 m of the car trajectory. There are numerous challenges in predicting when the EV will likely come to a complete stop due to the traffic conditions, the exact location of traffic stops and the fact that the state of the traffic lights is unknown to the controller. Consider the map shown in Fig. 6. Points A and C represent traffic signals at which the vehicle might stop, depending on the state of the traffic lights. Points B and F represent stop signs which can potentially be falsely detected. Point D shows the specific pedestrian crossing point which only occasionally results in a stop. Point E also falls within the 2 m detection zone, however it can be neglected as it is out of the vehicle moving direction. This is achieved by calculating the vehicle moving direction and the displacement vector for the position of the stop sign/traffic signal. Point G shows a stop sign that might be falsely picked up, like points B and F. Although at this point the vehicle is turning, the vehicle does not necessarily need to stop. A detailed analysis of the experimental drive-cycle showed that vehicle stopped at 66% of the stop signs and 49% of the traffic lights that were detected by the GPS system. Despite this uncertainty, the GPS systems bring a significant value to the HESS that would otherwise have no indication of impending stops. E G X X D F C B A 2m Fig. 6. A hypothetical driving situation with the stop signs and traffic signals shown over a map. The route is shown in blue and the possible points that the GPS data processor may pick up are marked in purple. IV. SYSTEM SIMULATIONS BASED ON EXPERIMENTAL DRIVE-CYCLE MATLAB is used for system level simulations to investigate the benefits of the proposed GPS based HESS approach, based on the experimental drive-cycle data from the EV. A. Vehicle Modeling The battery, u-cap and dc-dc converter models used in this work are identical to our prior work on Light Electric Vehicles (LEV) [2], hence this section is focused on the vehicle modeling. The equivalent force available on the wheels is given by F net = F acc + F F + F D, (8) where F acc is the needed force for acceleration, F F is the rolling friction resistance and F D is the drag resistance. Rolling friction force is modeled by: F F = C rr N, (9) where C rr is the dimensionless rolling resistance coefficient and N is the normal force to the surface that the wheels are

5 rolling on. Drag force is also modeled using: F D = 2 ρ ν2 C D A, () where ρ is the mass density of air, ν is the vehicle speed, C D is the drag coefficient and A is the orthogonal projection area of the vehicle to the direction of motion. It is assumed that the vehicle travels on a flat surface and the torque delivered to the wheels is modeled by τ net = R w F net, () where R w is the displacement vector. Based on typical values of C rr and C D for standard sedan vehicles, C rr =.5 and C D =.4 are used. The torque delivered to the wheels is a function of mechanical system efficiency: τ net = K G η m τ m, (2) where K G is the gearbox ratio, η m is the mechanical system efficiency that depends on the operating point and τ m is the torque available on the motor shaft. The linear approximation: η m = η + T m (η Tmax η ), (3) T max is used in this work, where η and η Tmax are optimized using the least square method, such that the calculated motor torque best matches the drive-cycle measurements. The load current can be calculated using the drive-train efficiency data: I load = P e, V bt (4) P e = P m, η D (5) P m = τ ω, (6) η D = f(τ,ω). (7) B. Simulation Procedure and Test Scenarios Several test scenarios are considered to simulate the impact of the HESS using the experimental drive-cycle:. Sim #: SESS with No Regen: Baseline EV with Regen disabled. 2. Sim #2: SESS with -.25 C< : Regen is enabled, however the battery is unable to absorb all the Regen energy due to its limited charge current, based on battery lifetime considerations. Regen energy for which -.25 C < is returned to the electrical system, while the rest is dissipated in the mechanical brakes. 3. Sim #3: SESS with No Limits: Similar to Sim #2, where the charge and discharge currents are not explicitly limited. 4. Sim #4: HESS without GPS Data Processing: HESS with Regen enabled. The battery charge current is limited to -.25 C and therefore having a u-cap with enough capacity for the charge helps to recover more energy. The battery discharge current is unlimited, but the power optimizer tries to limit battery discharge current to.75 C. This limit can be violated if SOC uc is too low and the u-cap cannot handle the load demand. This feature maintains the EV driving experience with improved battery lifetime. 5. Sim #5: HESS with GPS Data Processing: Similar to Sim #4, with the GPS based stop prediction enabled. C. Simulation Results and Discussion The comparison for the five scenarios outlined in Section IV-B for the measured 3 hour urban drive-cycle is presented in Table IV. The simulated energy consumption for the baseline Sim # based on the detailed EV model was within.4% of the measured experimental SESS data. Enabling Regen in the EV, with and without charging limits, results in an energy consumption reduction of 3.4% (Sim #2) and 4.5% (Sim #3), respectively, compared to the baseline SESS (Sim #). The HESS without GPS processing (Sim #4) achieves the same energy consumption as the SESS with no limits (Sim #3), while drastically reducing the battery current range. Finally the HESS with GPS processing decreases the peak battery charge and discharge rates by 76% and 47%, respectively, compared to Sim #3. Interestingly, the GPS processing reduces the peak charge current by 38% compared to the baseline HESS in Sim #4, with virtually no added system cost. The reduction in required peak battery current can translate into a different choice of battery chemistry with higher specific-energy and lower peak power requirement. The partial simulation results for Sim #3, Sim #4 and Sim #5 are shown in Fig. 7(c), (d). It is interesting to see that with the predicted up-coming stop based on the real-time GPS data processing shown in Fig. 7, the u-cap has been effectively used to further reduce the battery peak current between t = 935s and t = 9365s. Furthermore, future Regen bursts are also absorbed by the u-cap (t = 952s) to compensate for the extra charge that has been used and further limit the battery charging current. The HESS effectively limits the battery current, based on the histogram, as shown in Fig. 8. Fig. 8 shows the battery current based on the SOC bt. In this work, the charge/discharge limits of the battery current for the HESS system are fixed, but they can be decreased at lower SOC bt, improving the battery operating condition. V. CONCLUSIONS The model of the prototype EV and the proposed HESS has been developed to provide a powerful and accurate tool for system simulations. It is shown that utilizing the GPS data in the power optimizer can reduce the battery charging peak current by 38% compared to a standard HESS, without substantially increasing the cost, since the GPS is already available in the EV. In addition to reducing long-term capacity fading in the battery, the GPS enhanced HESS can relax the battery module power specifications and allow the use of chemistries optimized for high specific-energy, while the peak power demands are met by the u-cap. ACKNOWLEDGMENT The authors thank Steve Dallas from Toronto Electric and Feisal Hurzook from of Archronix for their valuable help. This work was supported by the Natural Sciences and Engineering Research Council of Canada, the Canadian Foundation for Innovation and the Ontario Research Fund.

6 TABLE IV. SESS VERSUS HESS ANALYSIS. SESS 2. SESS 3. SESS 4. HESS without GPS 5. HESS with GPS (No Regen) ( > C ) (No limits) (I 4 bt > C ) (I 4 bt > C ) 4 Net Electric Energy Usage(kWh) Energy Benefits (%) Battery Current Range,,min/max / / / / /82.5 Speed (km/h) GPS Flag SOC uc (c) (d) Fig. 7. Measured EV speed, GPS lstop prediction signal. (c) Simulated for SESS with no limit (blue), HESS with GPS (green) and without GPS (red).(d) U-cap SOC for HESS with (green) and without GPS (red). REFERENCES [] U-charge xp lithium iron magnesium phosphate battery, Valence Technology, datasheet, accessed in July 22, available at [2] H. A. Borhan and A. Vahidi, Model predictive control of a power-split hybrid electric vehicle with combined battery and ultracapacitor energy storage, in Proceedings of American Control Conference, ACC, 2, pp [3] J. Cao, B. Cao, Z. Bai, and W. Chen, Energy-regenerative fuzzy sliding mode controller design for ultracapacitor-battery hybrid power of electric vehicle, in Proceedings of IEEE International Conference on Mechatronics and Automation, ICMA, 27, pp [4] K. Kawashima, T. Uchida, and Y. Hori, Development of a novel ultracapacitor electric vehicle and methods to cope with voltage variation, in Proceedings of IEEE Vehicle Power and Propulsion Conference, VPPC, 29, pp [5] X. Liu, Q. Zhang, and C. Zhu, Development of a novel ultracapacitor electric vehicle and methods to cope with voltage variation, in Proceedings of IEEE Vehicle and Propulsion Conference, VPPC, 29, pp [6] V. Shah, S. Karndhar, R. Maheshwari, P. Kundu, and H. Desai, An energy management system for a battery ultracapacitor hybrid electric vehicle, in Proceedings of International Conference on Industrial and Information Systems, 29, pp No. of occurances SESS(No limits) HESS w/o GPS HESS with GPS SESS(No limits) HESS w/o GPS HESS with GPS SOC bt Fig. 8. Simulated battery current histogram and simulated battery current versus SOC bt for the 3 hour urban drive-cycle. [7] G. Ning, B. Haran, and B. N. Popov, Capacity fade study of lithiumion batteries cycled at high discharge rates, Journal of Power Sources, vol. 7, pp. 6 69, May 23. [8] T. Kohler, D. Buecherl, and H. G. Herzog, Investigation of control strategies for hybrid energy storage systems in hybrid electric vehicles, in IEEE Vehicle Power and Propulsion Conference, VPPC, Sept. 29, pp [9] M. Ortuzar, J. Moreno, and J. Dixon, Ultracapacitor-based auxiliary energy system for an electric vehicle: Implementation and evaluation, IEEE Transactions on Industrial Electronics, vol. 54, no. 4, pp , Aug. 27. [] O. Erdinc, B. Vural, and M. Uzunoglu, A dynamic lithium-ion battery model considering the effects of temperature and capacity fading, in IEEE International Conference on Clean Electrical Power, 29, pp [] Chevrolet volt brochur, General Motors, accessed in July 22, available at [2] O. Laldin, M. Moshirvaziri, and O. Trescases, Predictive algorithm for optimizing power flow in hybrid ultracapacitor/battery storage systems for light electric vehicles, Power Electronics, IEEE Transactions on, vol. 28, no. 8, pp , 23. [3] Map data OpenStreetMap contributors, CC BY-SA, June 22, available at and