The Research Of Modeling For Solid Oxide Fuel Cell Based On Neural Network

Because of energy shortages and environmental pollution, as a new form of energy, solid oxide fuel cell (SOFC) technology has the advantages of efficient clean、low noise、load capacity and with no leakage、electrolyte corrosion、comprehensive utilization of high efficiency and long life, so it is getting more and more attention. All countries are very optimistic about the prospects for the development of SOFC.Some models of SOFC are too too complex to design the control system, bacause the equations are too many, In this paper, particle swarm (PSO) algorithm and neural networks are combined, all used in SOFC modeling study. The neural network is a mathematical model, by simulating biological neural networks so as to process information. It can approximate any complex nonlinear function. Radial basis function (RBF) neural network with the best approximation property, and there is no local minimum, faster convergence in the learning process suitable for online real-time control. First, establish the model of the SOFC by using RBF neural network, and then use particle swarm algorithm to optimize the parameters of the RBF neural network, to solve the problem of inaccurate initial parameters make the model accuracy reduced, make the model more precise. And compare with genetic algorithm, from the results of convergence speed and the size of the prediction error, particle swarm optimization is better than genetic algorithm, and then illustrate the superiority of this method, this method does not use complex analytic modeling or complex non-linear differential equations to describe the battery, so we can get the results quickly, lay the foundation for the online control of the SOFC.Here is the paper work and innovations:1. Analysis and summary the output characteristic of the SOFC, First, explain the effect of three kinds of polarization phenomena of the battery voltage, then from the structural parameters and operating parameters analysis the impact of various factors on the output characteristics of SOFC. This paper focuses on the SOFC temperature and hydrogen flow rate during operation of SOFC output characteristics.2. Building a model of SOFC by RBF neural network. Against the problem of building model, use RBF neural network to establish the model of SOFC.Simulation results show that at different hydrogen flow rate and temperature, it can basically be able to dynamically simulate the changes in the current density-voltage of SOFC, the prediction error of RBF neural network is very small, it shows that this method if feasible. 3. Use particle swarm algorithm to optimize the RBF neural network. The choose of RBF neural network parameters is good or bad, would have a direct impact on the SOFC modeling, so against the connection weights、the center of the basis function and variance of the RBF neural network, use PSO algorithm to train parameters, the simulation results indicate that model has a higher precision, then, compare with genetic algorithm, the results show that model improved by particle swarm optimization has a faster convergence speed, smaller prediction error and more reliable, verified the validity of this modeling approach. To achieve the goal of controlling the SOFC system, in particular in-line control create the conditions.