PEMFC remaining useful life prediction method combined particle swarm optimization algorithm with improved echo state network

粒子群优化算法结合改进回声状态网络的PEMFC剩余使用寿命预测

¹ School of Automation and Electrical Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
² CRRC Tangshan Co., Ltd., Tangshan 063035, China

Received: 7 May 2024

Abstract

In order to improve the accuracy of degradation prediction of proton exchange membrane fuel cell(PEMFC), a PEMFC voltage prediction method based on particle swarm optimization(PSO) algorithm to optimize the revised echo state network(RESN) is proposed. The nonlinear fitting process is accelerated by improving the connection mode of each neuron in the echo state network reservoir. The PSO algorithm is used to optimize the model spectral radius, leakage rate, number of neurons, etc, to improve the prediction accuracy of the model. The SG (Savitzky-Golay) filtering algorithm is used to retain the original trend of the original data and effectively remove the peak and noise. The PSO-RESN was used to accurately predict the PEMFC voltage. Different sample data sets are used as training sets and test sets, and the proposed model is compared with extended Kalman filter and traditional echo state network under static and quasi-dynamic experimental data sets. The results show that when the proportion of the training set is 80%, for the static condition FC1, compared with ESN, the root mean square error(RMSE) and the average percentage error(MAPE) of PSO-RESN method are reduced by 17.50% and 25.53%, respectively. For the quasi-dynamic condition FC2, compared with ESN, RMSE and MAPE are reduced by 16.93% and 21.28%, respectively. The proposed method can achieve higher precision degradation trend and remaining useful life prediction of PEMFC.

摘要

为提高质子交换膜燃料电池(proton exchange membrane fuel cell, PEMFC)退化预测的精度, 提出一种基于粒子群(particle swarm optimization, PSO)算法优化改进回声状态网络(revised echo state network, RESN)的PEMFC电压预测方法。通过改进回声状态网络水库中各神经元连接方式, 加快非线性拟合过程; 利用PSO算法优化模型谱半径、泄漏率、神经元数量等, 提高模型预测精度, 采用SG(Savitzky-Golay)滤波算法对原始数据有效去峰去噪, 再利用PSO-RESN准确预测PEMFC电压; 采用不同样本数据集作为训练集和测试集, 将所提模型在静态和准动态实验数据集下与扩展卡尔曼滤波、传统回声状态网络进行对比。结果表明, 在训练集占比为80%时, 对于静态工况FC1, 相较于ESN, PSO-RESN方法的均方根误差(root mean square error, RMSE)和平均百分比误差(mean absolute percentage error, MAPE)分别降低了17.50%和25.53%;对于准动态工况FC2, 相较于ESN方法, PSO-RESN方法的均方根误差和平均百分比误差分别降低了16.93%和21.28%。所提方法能够实现PEMFC更高精度退化趋势与剩余使用寿命预测。