Fuzzy Based Energy Management Control of A Hybrid Fuel Cell Auxiliary Power System

Transcription

1 Fuzzy Based Energy Management Control of A Hybrid Fuel Cell Auxiliary Power System M. Godoy Simões 1, B. Blunier, A. Miraoui 1 Colorado School of Mines (United States), Université de Technologie de Belfort-Montbéliard (France) Abstract This paper presents the analysis and design of a hybrid fuel cell battery auxiliary power unit (APU) for remote applications where a fuel cell is the main energy source, operating for slow power dynamics while a battery or a supercapacitor compensates fast transient peak power requirements. A fuzzy logic based control has been implemented in the energy management performance control in order to impose that fuel cell operates most of the time in its best operating point as well as maintaining the battery state-ofcharge in its best operating range, and contributing for a longer lifetime and minimized maintenance, approaching the best fuel to electricity system efficiency. transient compensation must be accomplished with other reversible power or energy source (supercapacitors or battery, respectively) which will handle power transients [6]. Supercapacitors are preferable for loads with high power peaks, short transients and a high number of charge and discharge cycles. Batteries are preferable for loads with long transients without power peak transient cycles [7]. Index Terms fuel cell, energy management, fuzzy logic, power electronics I. INTRODUCTION Auxiliary power units are important for decentralized or distributed energy production, such as telecom, remote sites or even for military applications [1] []. Hybridization is often required, and a main fuel source is supplemented by storage. Sizing and optimization are usually performed for minimizing the weight and volume for a given application. A fuel cell powered hybrid auxiliary power unit (APU) [3] is an interesting solution as seen in Fig. 1. When refueled by hydrogen cartridges a fuel cell has nearly no noise operation, providing electricity and heat with water as by-product [4]. This paper presents an energy management performance improvement control of a combined structure (fuel cell and peaking power source), where the peaking power source (battery or supercapacitors) provide fast variation power and the fuel cell supplies the base average load [5]. A fuzzy logic algorithm maximizes the efficiency of the system and minimizes cycles of shutdown and restarting of the fuel cell, eventually contributing for system longer lifetime expectancy. It is presented in this work a topology consisting of a 300 watts fuel cell stack connected to a supercapacitors pack of a Ni-MH battery through a DC/DC buck converter. The choice between supercapacitors and/or battery depends on the user needs and the load characteristics Batteries have good energy density but low power density. On the other hand, supercapacitors have a better power density but a lower energy density. Moreover, unlike the batteries, the charge/discharge cycles are almost unlimited. Fuel cells are not bidirectional systems and they have a slow time transient response for controlling hydrogen and air inlets. Therefore, a Fig. 1: Fuel cell battery hybrid auxiliary power unit (APU) The mathematical modeling for a PEM fuel cell is very complex and requires knowledge of several electrochemistry parameters that are not easy to obtain for the commercially available fuel cell stacks [8]. The theoretical equations for expressing the best power operating point and also the fuel cell overall efficiency are very difficult to model and experimental based solutions are the only possible approach [8]. Therefore, the novelty of the proposed control is to fine tune a fuzzy based control using heuristics of the system operation, including for example the avoidance of very high current (that impose a lot of hydrogen flow) as well as the avoidance of off cycles that stress the fuel cell during turn-on and make their lifetime shorter. A rather simple and robust structure has been chosen to validate the relevance of the proposed energy management control approach, where the aim is to have the most simple and less expensive system which can be adapted for several peaking power sources such as supercapacitors or batteries, with optimized operation as a portable unit. The paper is organized as follows: initially it is presented an overview of the architecture and the components of the APU, following how the control system was designed. Finally, experimental

2 results are discussed in order to show the performance of the algorithm. II. FUEL CELL HYBRID SYSTEM A complete electrical model has been analyzed developed and used to develop the control laws and energy management strategies before their implementation on the real system. The modeling and simulation results have been presented in a previous paper where a Ni-MH battery supplies the peak power [9]. The auxiliary power unity (APU) is depicted in Fig.. The fuel cell used is a PEM FC. It operates at low temperature and at room atmospheric pressure on the cathode side. Its rated power is 300 watts for a voltage ranging from 30 V to 60 V. A dc/dc buck converter is connected to the stack and the output dc-voltage is controlled over the dc-link. Other very interesting converter topologies with better performance and also lower costs have been proposed in the literature [10, 11, 1]. Table I: Fuel cell auxiliary power unit specifications rated power fuel cell voltage range dc output voltage dc output current (max) embedded storage 300 W 30 V 60 V 10 V 36 V 30 A battery or supercapacitor virtual prototyping an implementation methodology validated the whole system, including control and components design (inductance and capacitance sizing, battery capacity, power electronics, fuel cell power and so on, using simulation tools). The system model and the multiloop control, including the fuzzy logic based energy management systems were implemented in Matlab-Simulink as indicated in Fig. 3, then validated in simulation by direct compilation and target downloading into the dspace microautobox. For the final industrial version, the control system will be implemented in a compact low-cost microcontroller-based board. Low power fuel cells like the one utilized for this APU operates at atmospheric pressure where air is supplied with three fans at a high stoichiometric ratio. The fuel cell is aircooled using the same fans than the air supply. The high stoichiometric ratio ensures that there is nearly no oxygen depletion and no electrodes flooding. As the fuel cell system (energy source) is hybridized with peaking power source (PPS) its dynamic does not need to be high. In order to increase the lifetime of the fuel cell, the strategy aims at maintaining the fuel cell current dynamics as low as possible. Ideally, the fuel cell should deliver a constant power matching the average load power corresponding to the power at its best efficiency. Fig. : APU system and components These converters could be used in a new design of such a system. The peaking power source (PPS) imposes the value of the DC bus voltage (around 4 volts) which is dependent on its state of charge. The PPS can be changed without changing the architecture of the system as long as its voltage remains below the operating voltage of the fuel cell. This system has been designed to accommodate a voltage range from 10V to 36V. Table I shows the fuel cell auxiliary power unit specifications. The experimental work has been verified with an embedded dspace microautobox integrated with the APU. Such rapid prototyping and real-time implementation allowed the control structured designed in Matlab/Simulink to be fully developed for an industrial application. Using Fig. 3: APU control structure Therefore, it can be assumed that the fuel cell is always in steady state and for this reason a static model of the fuel cell is used in the simulation based design. For higher power applications, the static characteristic cannot be used to describe the fuel cell system; the compressor and the air dynamic (air pressure inside the channels), as well as the water content of the membrane will affect the membrane resistance have to be considered, and for large power fuel

3 systems readers are referred to more complex models presented in [13]. A. Control Strategy Optimization problems can be solved based on a successive approximation solution of the generalized Hamilton Jacobi Bellman equation, and neural network solutions have been used, where they are tuned a priori in offline mode, and a closed loop control, based on recursive least squares or other methods, may bring the updated control laws to converge to the optimal control. Usually, for an optimization based control strategy it is required to find a solution for the following equation: J * ( 0 u( t) * X ( t)) min{ J ( X ( t 1)) g( X ( t), X ( t 1)) U } (1) where g( X ( t), X ( t 1)) is the immediate cost incurred by u(t), the control action, at time t, and U 0 is a defined heuristic term. In order to adapt J ( X ( t)) the right hand side of (1) must be known a priori. So, to get that, one may have to wait for a time step until the next input becomes available Consequently, J ( X ( t 1)) can be calculated by using a performance index at time (t+1). However, when the problem temporal nature does not allow to wait for the subsequent time steps to infer incremental costs, other solutions must be used to calculate x(t+1), and it has been found that a fuzzy logic based system would be the simplest and best solution for such optimization strategy [14]. The battery state of charge (SoC) can be considered as a dynamic system; the system can be written as: x k xk P b bts 1 () 0.95 if Pb 0 b 1.0 if Pb 0 where Pb is the battery power level defined by I bat of Fig. (1), b the battery charge acceptance (0.95 for charge and 1 for discharge) and Ts is the sampling time. The chosen criterion for N samples can be written as: N 1 k0 J m P, k T (3) H where m P k power H FC, P FC FC s ) is the hydrogen mass consumed for the during the sampling time T s. The fuel cell power is obviously limited by P FC MIN P FC P FC MAX, where in this application 0 P and FC 300W FC MIN P MAX, where the hydrogen taken for the output power will be variable in accordance to the measured cell efficiency indicated by Fig. 5(a). Ravey et al. [14] have demonstrated that such a fuzzy controller based on the SoC of the peaking power source can give optimal results for a given load profile family. They compared the results (e.g., fuel cell power, SoC profiles and hydrogen consumption) to the optimal results found with an offline dynamic programming algorithm which is able to find optimal results based on a priori known load profile. Therefore, such previous fuzzy logic control study supports to be used in this APU as a low power adaptation of the structure for fuel cell based hybrid electric vehicles [14], where the load is taken by any ouput circuit and the system should aim to have a long lifetime and minimization of cycles of turn-on and turn-off, in addition to avoid charging the battery with maximum fuel cell current operation point, which impress a very low efficiency and high utilization of hydrogen. B. Fuzzy Logic Control The main motivation and novelty in using a fuzzy logic based control for the fuel cell APU is because there is no simple model for State of Charge for batteries, and electrochemistry based modeling is heavily dependent on knowledge of the battery internal parameters. In addition, this APU can be used with either a battery or a supercapacitor storage, and the fuzzy logic controller works for both options. A PI based control has been used in [9] for generating the reference voltage for a dc-link system in order to improve the transient response of a fuel cell. However, such control does not impose the best State-of-Charge and the PI design is ad-hoc made dependent on the fine tuning of the fuel cell slow dynamic response. The controller proposed in this paper has been defined heuristically based on experimental work, where initially the fuel cell power is plotted against the output current (Fig. 5) and then, using a mass flow meter, the system efficiency is plotted against the fuel cell power. Therefore, there is an inverse mapping of system efficiency in terms of impressed output current, by using those experimental data, which can be expressed by rules. The authors made an extensive state of arte literature survey and firmly believe that nobody ever proposed a control system such as the one described in this paper, aiming to impress the fuel cell output current for achieving the best efficiency and improved state of charge of a hybrid storage (battery or supercapacitor). Most fuzzy systems are based on rules, such as indicated in Fig. 4. The rules are defined on the system heuristics, and for this controller the inverse function of what is the set-point for the fuel cell current in order to have a prescribed Stateof-Charge has been designed. The input information is fed back from the device as it operates and actuates on the operation, i.e. crisp input information from the device is converted into fuzzy values for each input fuzzy set by the

4 fuzzification block. The universe of discourse of the input variables determines the required scaling for correct per-unit operation. The following steps must be conducted to design a fuzzy logic control structure: 1. Identify the inputs and their ranges and name them.. Identify the outputs and their ranges and name them. 3. Create the fuzzy partitions (degree of fuzzy membership function) for each input and output. 4. Construct the rule base that the system will operate under 5. Decide how the action will be executed by assigning strengths to the rules, 6. Combine the rules and defuzzify the output. Each rule r ij above maps the i th multivariate fuzzy input set A i to the j th univariate output set with confidence c ij. The degree to which element x is related to element y is x y defined represented by the membership function in the product space r ij A A n B 1 by: x y t i x, c j y A ij B r,,, (4) where t is a triangular norm usually chosen to be the min or the product operator. If s and t are the max and min operators the fuzzy inference becomes: y max min x, Rx y A y (5) B i, The above fuzzy output set ij must be transformed into a crisp output (with defuzzification computation) in order to get a real number variable ouput to be used in the system decision making strategy. B C. APU Energy Management Strategy The main objective of an EMS is to develop a simple and robust energy management control approach, where the objective for such EMS is to have the most simple and less expensive system which can be adapted by the user for other peaking power sources (PPS) and load depending on the application. Such a controller should be implemented in a simple microcontroller to reduce the cost of the system. The control strategy has to take into account the following components constraints: Fig. 4: Fuzzy based decision making strategy The normalization and scaling gains are very important because the fuzzy system can be retrofitted for other devices or ranges of operation by just changing the scaling of the input and output. The decision-making logic determines how the fuzzy logic operations are performed. In the fuzzy algorithm the fuzzy rules are connected together (using the or operator) to form the rule base. The fuzzy algorithm that maps p inputs onto q outputs has pq rules with the following form: r 11 : IF (x is A 1 ) THEN (y is B 1 ) r 1 : IF (x is A 1 ) THEN (y is B ) r 1q : IF (x is A 1 ) THEN (y is B q ) r 1 : IF (x is A ) THEN (y is B 1 ) r pq : IF (x is A p ) THEN (y is B q ) 1. Limits the fuel cell dynamics and voltage cycling (especially high voltages) to increase its lifetime;. Make the fuel cell working around the best working points (where its efficiency is the highest); 3. Maintain the peaking power source (battery or supercapacitors) State-of-Charge (SoC) in a prescribed window and limits the cycling in the case of a battery. The time which the fuel cell takes to reach the desired power depends on the control strategy. Advanced neural network based control systems have been proposed for implementing approximated optimal control using CMAC networks [15] [16], supporting that artificial intelligence based control is adequate for complex systems such as fuel cells. In this case, the fuel cell current is based on the stateof-charge (SoC) level of the PPS. The fuel cell current should be maintained in the best efficiency window as shown in the green box in Fig. 5(b). The operating points at higher power should be avoided but are authorized if necessary (e.g., when the SoC is too low). The operating points at lower power are avoided in this case, the fuel cell is switched off. The parameters of the fuzzy controller are tunable by the user and are represented

5 by trapezoidal functions as shown in Fig. 6(a) and Fig. 6(b) for the SoC and the fuel cell current, respectively. The rules were defined by simulation studies to be very simple: 1. IF (SoC is low) THEN (IFC is high). IF (SoC is good) THEN (IFC is opt) 3. IF (SoC is high) THEN (IFC is low) For a given load profile the membership parameters can be optimized using an offline optimization algorithm such as dynamic programming to obtain the best overall energy efficiency. (a) SoC membership functions (b) Output variable, normalized fuel cell current Fig. 6: Fuzzy controller membership functions (a) Experimental fuel cell current-power characteristic (b) Measured efficiency Fig. 5: Fuel cell system performances and working zones. The currentpower and efficiency characteristics permit the output membership function (i.e., fuel cell current) of the fuzzy controller to be defined. III. EXPERIMENTAL EVALUATION The experimental development was implemented in accordance to Fig. and Fig. 3, where the PPS is a supercapacitor bank composed of two 58 F - 15V packs. The supercapacitor state-of-charge are calculated as follows: U SoC (6) U scap, 0( t) scap,max where U scap ( ) is the instantaneous supercapacitors open, 0 t circuit voltage and U scap,max its maximum voltage. In this case U scap, max 30 V as two supercapacitors packs of 15 volts are in series. The instantaneous supercapacitor bank voltage is computed as follows: U t) U ( t) R I ( ) (7) ( scap, 0 scap s scap t where R s is the series resistance of a supercapacitors pack (given by the manufacturer) and I scap its current. The experimental setup is depicted Fig. 7; it is composed by supervision computer connected to the microautobox with a graphic user interface (ControlDesk).

6 Fig. 7: Experimental setup Fig. 8: Current on the DC bus. At the beginning, the fuel cell current is zero because the supercapacitors SoC is high. The fuel cell current is very slow. The industrial version of the APU has been supplied by an external metal hydride tank as shown in Fig. 1, but for the experiment, is was more convenient to use external compressed hydrogen tanks as the metal hydride tanks need to be refilled. The electrochemical fuel cell system needs a balance of plant supervisory layer that manages hydrogen purge, air for reaction and for cooling, plus humidity and temperature management. These are not in the scope of this paper and interested readers should refer to [9]. The model and the simulation permit the energy management strategy to be tested under various driving cycles. This allows several strategies and set-up parameters to be tuned in order to optimize the system response and check the system constraints (dynamics, currents, voltage, peaks, etc.). The simulations have been given by the authors in [7] and for the sake of applicability, this paper has focused only on the experimental approach. A. Experimental Results 1) Test 1, Gaussian Distribution: The load current profile (see Fig. 8) has been randomly generated (gaussian distribution) with an average current of 7A corresponding to the nominal fuel cell power, a variance of 6A and a sample time of 3s to produce a high dynamic currents with high peaks of currents. Fig. 8 shows the currents on the DC bus. It can be seen that the fuel cell current contribution on the DC bus is smooth and almost constant. On the the other hand, the supercapacitors handle the peaks and dynamics. The fuel cell current is most of the time maintained in the best efficiency region as shown in Fig. 9. The current is always between 3A and 6A corresponding to the best efficiency power window described in Fig. 5(a). The experimental powers distribution over the load profile is shown in Fig. 10 : when the fuel cell delivers some power, it is always at the best efficiency. The fuel cell is switched off at the beginning as the supercapacitors state-of-charge is high: the number of fuel cell on/off cycles is equal to one over the load profile. At the beginning of the tests, the supercapacitors SoC is high (see Fig. 11). As the ``good'' region corresponds to a SoC between 0.5 and 0.8 (see Fig. 6(a)) the fuel cell does not deliver any power at the beginning to let the SoC reach a lower SoC. Then, the SoC is maintained over the load profile in the SoC window defined by the user. ) Test 1, uniform load profile: The load current profile (see Fig. 1) has been randomly generated (uniform distribution) with currents between 0A and 10A. With this current profile, the load variation are more pronounced and the supercapacitors are more solicited. Two tests have been carried out with the same load profile: the first one with a high initial state-of-charge and the second one with a low state-of-charge. The objective is to demonstrate that the controller works independently from the initial state-ofcharge even it is low. The DC bus currents for the high and low initial state-of-charge are shown in Fig. 1 and Fig. 13, respectively.

7 Fig. 9: Fuel cell current is always around the current giving the maximum efficiency (between good and high for the fuzzy membership functions). Fig. 1: Currents on the DC bus: the SoC at t=0 is high, the fuel cell current is zero at the beginning. Fig. 10: Fuel cell power statistical distribution. The fuel cell delivers its power at its maximum efficiency according to Fig. 5(b). Fig. 13: Several currents flowing in the DC bus: the SoC at t=0 is low, the fuel cell current is high at the beginning to recover a good SoC. The fuel cell current and the SoC for both two tests are superimposed in Fig. 14 and Fig. 15, respectively. Expect at the beginning when the SoC is very low, the fuel cell current is maintained in its best operating range. For a high initial SoC, the fuel cell is switched off one time during the experiment. In the two cases the initial conditions influence the starting of the system but the control strategy permits the system to recover the best operating conditions after some minutes (less than two minutes in this case). In the two experiments, most of the working points, when the fuel cell system delivers some power, are in the best efficiency region as shown in Fig. 16 and Fig. 17. Fig. 11: Supercapacitor SoC is maintained in the good region.

8 Fig. 14: Fuel cell currents for two different initial SoC and the same current load profile. Fig. 17: Low initial state-of-charge. The fuel cell power is maintained at the best efficiency region with less stop times compared to tests with a high initial state-of-charge (only at the end when the load actually stops) because the SoC had to be recovered). Fig. 15: The final SoC is well recovered. After a transient the SoC is maintained in the best operating region defined by the user. IV. CONCLUSION A hybrid fuel cell battery auxiliary power unit has been designed, modeled, simulated and evaluated experimentally. The fuel cell in this system supplies energy for the APU, but because it has slow dynamics due the electrochemical balance of plant, a peaking power source must be used to compensate the fast transient load peak power requirements. The overall system control system has been approached in order to optimize the fuel to electricity efficiency, and maintain a rapid recovery of the battery state of charge. A fuzzy logic based decision making strategy advanced the control system achieving the best performance range. It was observed experimentally the minimization of shut-down and over-current conditions, contributing for a longer lifetime and minimal maintenance needs. Such operating conditions at low voltage, less shut-down cycles (thermal and voltage cycling) permit the fuel cell to have a longer lifetime and avoid cells degradation. It is then expected that this controller will permit to improve the system reliability (less maintenance) and lifetime. With a simple rule based system, the system performance was found to recover the state-ofcharge conditions without typical control overhead by state machine decision making procedures. The proposed control strategy is easy to retrofit to any other APU system and it is expected to improve the industry applicability of traditional APU controllers. Fig. 16: Fuel cell power is always maintained in the best efficiency region. There are two fuel cell on/off cycles see Fig. (see Fig. 14). DEDICATION We dedicate this paper to the memory of our friend and co-author, Dr. Benjamin Blunier, formerly an Associate Professor at the University of Technology of Belfort- Montbéliard (UTBM), who passed away on Feb/3/01.

9 REFERENCES [1] R. Sharma and H. Gao, Low cost high efficiency dc-dc converter for fuel cell powered auxiliary power unit of a heavy vehicle, vol. 1, no. 3, pp , 006. [] S. Singhal, Solid oxide fuel cells for stationary, mobile, and military applications, Solid State Ionics, vol. 15, pp , 00. [3] T. A. Nergaard, J. F. Ferrell, L. G. Leslie, and J.-S. Lai, Design considerations for a 48 v fuel cell to split single phase inverter system with ultracapacitor energy storage, in Proc. IEEE 33rd Annual Power Electronics Specialists Conf. pesc 0, vol. 4, 00, pp [4] F. Baratto, U. Diwekar, and D. Manca, Impacts assessment and tradeoffs of fuel cell-based auxiliary power units:: Part i: System performance and cost modeling, Journal of power sources, vol. 139, no. 1-, pp , 005. [5] K. Rajashekara, Hybrid fuel-cell strategies for clean power generation, vol. 41, no. 3, pp , 005. [6] J. Duarte, M. Hendrix, and M. Godoy Simoes, A three-port bidirectional converter for hybrid fuel cell systems, IEEE Transactions on Power Electronics, vol., no., pp , March 007. [7] B. Blunier, M. G. Simoes, and A. a. Miraoui, Fuzzy logic controller development of a hybrid fuel cell-battery auxiliary power unit for remote applications, in 9th IEEE/IAS International Conference on Industry Applications (Induscon), 010. [8] Fei Gao, B. Blunier, A. Miraoui and A. Ee-Moudni, Cell Layer Level Generalized Dynamic Modeling of a PEMFC Stack Using VHDK-AMS Language, International Journal of Hydrogen Energy, vol. 34(13), pp , 009. [9] D. Paire, M.G. Sim es, J. Lagorse, A. Miraoui, A real-time sharing reference voltage for hybrid generation power system" Industry Applications Society Annual Meeting (IAS), 010 IEEE, Oct [10] A. Khaligh, A. Rahimi, Y. Lee, J. Cia, A. Emadi, and S. Andrews, Digital control of an isolated active hybrid fuel cell/li-ion battery power supply, IEEE Transactions on Vehicular technology, vol. 56, no. 6, november 007. [11] S. Wang, M. Krishnamurthy, R. Jayabalan, and B. Fahimi, Low-cost quasi-resonant dc-dc converter for fuel cells with enhanced efficiency, in Applied Power Electronics Conference and Exposition, 006. APEC 06. Twenty-First Annual IEEE, pp March 006 [1] S. Wang, M. Krishnamurthy, R. Jayabalan, and B. Fahimi, Quasiresonant dc/dc converter for high power fuel cells systems, in Power Electronics Specialists Conference, 006. PESC th IEEE, 18-, pp. 1 7, June 006 [13] F. Gao, B. Blunier, A. Miraoui, and A. Moudni, A multiphysic dynamic 1d model of a proton exchange membrane fuel cell stack for real time simulation, IEEE Transactions on Industrial Electronics, Vol 57 (6), pp , June 010 [14] A. Ravey, B. Blunier, and A. Miraoui, Control strategies of a fuel cell based hybrid electric vehicles: from offline to online, in Vehicle Power and Propulsion Conference, 011. VPPC 011. IEEE, Chicago, IL. USA, September [15] P. E. M. Almeida and M. G. Simoes, Parametric cmac networks: fundamentals and applications of a fast convergence neural structure, IEEE Transaction on Industrial Applications, vol. 39, no. 5, pp , 003. [16] P. E. M. Almeida and M. G. Simoes, Neural optimal control of pem fuel cells with parametric cmac networks, IEEE Transaction on Industrial Applications, vol. 41, no. 1, pp , 005.