Modeling And Simulation For Dynamic Behavior Of PEM Fuel Cells

Among various types of fuel cells, Proton exchange membrane (PEM) fuel cell has become the major one for its high-energy efficiency, pollution-free features, which allows fast startup and low operation temperature. The previous research work had mainly focused on steady-state behavior, instead of the dynamic behavior. Because research on fuel cell involves in many subjects, such as fluid dynamics, heat transfer, mass transfer, thermodynamics and electrochemistry, it is very important to understand the operating mechanism, optimization of the cell structure and the systematic integration, so we must characterize the PEM fuel cell dynamic characteric. The author described the study of the dynamic characteristic with three ways, based on distributed-parameter model, lumped-parameter model and artificial neural network model, and compared each other.First, the study of the dynamic characteristic was based on the distributed-parameter model. The simulation will complete with Fluent, a computer software dealing with the PEM module of computer fluid dynamics. This paper describes the lumped-parameter model, including the quality, momentum, energy, species conservation equation in PEM cell, the electrical chemistry equation in the catalyst layer, on which the models and the dynamic simulation described in the followings are based. A 3D structure model for single cell with a single channel is established. The dynamic response and performance were calculated with the non-steadycalculator in Fluent. Those dynamic responses included the current density step-up, step-down and upstairs, dynamic responses of voltage.Secondly, the study of the dynamic characteristic was based on the lumpedparameter model. The simulation will complete with Simulink, as a software package of Matlab, it widely used for modeling and simulation of dynamic system, does not need programming, only needs linking models, so that the modeling is simple and quick. The electrochemical model and temperature model of PEM fuel cell have been introduced. Based on the models, the modeling and simulation have been developed with Matlab/Simulink. Dynamic analysis has focused on the responses of output voltage, output power, efficiency under different conditions of start-up, shut-down or load step-up of the single PEM fuel cell or stack, the study is helpful for the optimization and control of PEM fuel cell.Since both lumped-parameter and distributed-parameter models can be used to study the PEM fuel cell dynamic characteristic, the dynamic curve simulating results obtained from Fluent and Matlab/Simulink were compared to each other for the same case scenario with same working conditions and simulating parameters. Lumped-parameter model only has relation with time, but model has relation with time and space. It has been concluded that distributed-parameter and lumpedparameter models have different advantages even though both can be used to studying the dynamic simulation of PEM fuel cell. Distributed-parameter model is more helpful in optimization of the cell structure and improving the cell's dynamic characteristics because it considers the property of the physical parameters. However, modeling is complexed, and simulation will need more time. Based on lumpedparameter model, the modeling is simpler and the simulating is more rapidly if Matlab/Simulink is used. It is very useful in the modeling of single fuel cell, stack and the whole system and the optimization of control because its use is based on the operating parameters and is independent of the property parameters.PEM fuel cell is a nonlinear system with time delay and indeterminacy. The research on neural network (NN) has been developing in recent years. Since the neural network have the advantages of parallel processing and self learning, it has been widely used for modeling and identifying of nonlinear system. We have attempted modeling and simulation to study the dynamic characteristics of PEM fuel cell with the method of neural network. There was systemic description in the paper about model classification, learning method and learning rules of the artificial neural network, and the method of the NN identification of the nonlinear system. The fuel cell was modeled with error back propagation (BP) network with time delay. The sampling data included the pressure and flow rate of the H₂ and air, current density of the fuel cell for input variables, and the output voltage for output variable. After the number of input layer, hidden layer, output layer and the nodes were determined, modeling and simulating were performed with the NN toolbox of Matlab. The results of simulation were in accordance with the experimental data. This method is simple for it is based on experimental data, looking the fuel cell as 'black box'. This study will be used to optimize the control of fuel cell.