中文摘要：

直接内部重整固体氧化物燃料电池（DIR—SOFC）是一种直接以碳氢化合物作为燃料，将化学能转化为电能的发电装置。小波网络（WNN）结合了小波分解的多分辨率逼近（MRA）的优点和神经网络非线性过程学习的能力。为了避免考虑DIR—SOFC内部反应复杂、参数耦合严重的情况，采用了WNN作为辨识工具，并结合免疫算法进行优化，对DIR—SOFC的动态特性进行建模。选取入口燃料流速、空气流速和负载电流作为输入量，对燃料电池的输出电压和温度的动态响应进行预测，并与采用RBF神经网络（RBFNN）方法得到的建模结果进行比较。仿真结果表明，通过该动态模型，电压和温度的预测值与验证数据间的相对误差绝对值分别小于0．0007和0．0004，平均相对误差分别为0．0003和0．0002，能够获得较高的精度并有效地对DIR—SOFC进行动态仿真。

英文摘要：

Direct inter-hal reforming solid oxide thel cell （ DIR - SOFC ） is an electric energy generation device which is directly fueled with hydrocarbons and converts the chemical energy of the fuel directly into electrical energy. Wavelet neural network （WNN） combines the advantage of multi -resolution approximation （MRA） of the wavelet decomposition and the capability of neural networks in learning from nonlinear process. In order to avoid considering the complexity of reactions and the coupling parameters inside the DIR - SOFC, the WNN, optimized by the immune algorithm, is used as an identification tool to establish a dynamic model of the DIR - SOFC with the inlet fuel flow rate, the inlet air flow rate and load current as inputs and the voltage and temperature as predictive outputs. Furthermore, the comparisons of the results between the immune - optimized WNN model and a RBF neural network （RBFNN） model are presented. The simulation results show that, by using the presented model, the absolute values of relative errors of the predictive voltage and temperature are less than 0. 0007 and 0. 0004, respectively, and the mean absolute values of relative errors are 0. 0003 and 0. 0002, respectively. The obtained dynamic model can effectively simulate the dynamics of the DIR - SOFC with relatively high accuracy.