基于径向基函数神经网络的燃料电池温度非线性建模与预测

针对现有的熔融碳酸盐燃料电池（MCFC）模型过于复杂的弊端，本文应用RBF神经网络辨识方法建立了MCFC的温度非线性模型。简要分析了MCFC电堆的温度特性，讨论了应用RBF神经网络进行多输入／多输出非线性系统建模的主要问题，并详细给出了其辨识结构，算法和模型训练方案，应用仿真对建模的有效性和建模精度进行了检验，并与BP神经网络辨识的效果进行了对比，仿真结果证明RBF神经网络远比BP神经网络收敛得快，应用RBF神经网络辨识方法对MCFC电堆建模是可行的，它避免了用复杂的微分方程组来描述MCFC，通过神经网络可快速地得到其输入同特性，它为MCFC温度的在线预测和在线控制奠定了基础。

Nonlinear Temperature Modeling and Prediction of MCFC Based on RBF Neural Networks

According to the drawback of the models existed which are too complicated, we set up a nonlinear temperature model of MCFC using RBF neural networks identification technology. The temperature characters of MCFC stack are briefly analyzed. A summary of RBF neural networks for the multi-input and multi-output (MIMO) nonlinear system identification is discussed, and the identification structure, algorithm and model training process are given in detail. The validity and accuracy of modeling are tested by simulations, and the comparison between RBF neural networks and BP neural networks identification is given. The simulation tests reveal the modeling method is feasible, and the training of RBF networks is much faster than BP networks. The modeling process avoids using complicated differential equation groups to describe MCFC, and the input-output performance can be quickly gotten by neuro-calculation, which makes it possible to predict and control the temperature of MCFC online.