基于改进BP神经网络辨识的DMFC建模研究  

Modeling of DMFC Based on Improved BP Neural Networks Identification

作　　者：戚志东[1] 朱新坚[1] 曹广益[1] 

机构地区：[1]上海交通大学自动化系燃料电池研究所,上海200030

出　　处：《计算机仿真》2006年第5期58-61,共4页Computer Simulation

基　　金：国家863项目(2002AA517020)

摘　　要：针对直接甲醇燃料电池(DMFC)系统过于复杂,难以建模的特点,该文试图绕开DMFC的内部复杂性,基于实验数据,利用神经网络逼近任意复杂非线性函数的能力,将神经网络辨识方法应用到DMFC这种高度非线性系统的建模。以1000组电池电压、电流密度实验数据作为训练样本,采用基于LM算法的改进BP神经网络,建立了不同温度下电池电压-电流密度动态响应模型。仿真结果表明这种方法是可行的,建立的模型精度较高,它使得设计DMFC的实时控制器成为可能。For the serious complexity of Direct Methanol Fuel Cell （DMFC）, modeling DMFC is very difficult. In this paper, we try to avoid the internal complexities of DMFC and set up a voltage and current density model with neural networks＇ability of identifying the complex non - linear function. With 1000 groups of experimental data of the cell voltage and current density as the training samples, and using the improved BP neural networks based on Levenberg - Marquardt （LM） algorithm, a cell voltage and current density dynamic response model of DMFC at different working temperatures is given. The validity and accuracy of the model are proved by the simulation results. The neural network modeling makes it possible to design on - line controller of DMFC.

关 键 词：直接甲醇燃料电池 神经网络 辨识 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]