基于核超限学习机和局部加权回归散点平滑法的PEMFC剩余使用寿命预测方法

Remaining Useful Life Prediction Method of PEMFC Based on Kernel Extreme Learning Machine and Locally Weighted Scatterplot Smoothing

为解决质子交换膜燃料电池(protonexchange membranefuelcell,PEMFC)剩余使用寿命预测(remaining usefullife, RUL)问题,提出基于核超限学习机(kernel extreme learning machine,KELM)和局部加权回归散点平滑法(locally weighted scatterplot smoothing,LOESS)的PEMFC剩余使用寿命预测新方法。该方法分别采用等间隔取样和局部加权回归散点平滑法实现数据重构和数据的平滑处理。不仅可以保留原始数据的主要趋势,而且能有效地去除噪声和尖峰。利用核超限学习机对测试数据实现剩余使用寿命预测,能在保证预测精度的情况下大幅降低计算复杂度。1154h的PEMFC老化实验分析表明:该方法的预测准确率为99.23%,运算时间为0.0146s,平均绝对误差和均方误差分别为0.0028和0.0037。对比分析表明:该方法的预测准确率比BP神经网络高28.46%;运算时间、相对误差、平均绝对误差和均方误差都远远小于BP神经网络。因此,该方法可快速准确地预测PEMFC剩余使用寿命。

In order to solve the remaining useful life(RUL) problem of proton exchange membrane fuel cell(PEMFC), a novel RUL prediction method of PEMFC based on kernel extreme learning machine(KELM) and locally weighted scatterplot smoothing(LOESS) was proposed. The method used equal interval sampling and LOESS to realize data reconstruction and data smoothing. Not only can the main trend of the original data be preserved, but noise and spikes can be effectively removed. The KELM was adopted to predict the remaining life of the test data. This method can greatly reduce the computational complexity while ensuring the prediction accuracy. 1154-hour experimental analysis of PEMFC aging shows that the prediction accuracy of the novel method is 99.23%, the operation time is 0.0146 seconds, the mean absolute error and root mean square error are 0.0028 and 0.0037 respectively. The comparison analysis shows that the prediction accuracy of the novel method is 28.46% higher than that of BP neural network. The operation time, relative error, mean absolute error and root mean square error are all much less than that of BP neural network. Therefore, the novel method can quickly and accurately predict the remaining service life of PEMFC.