基于NGO-CNN-BiLSTM神经网络的动态质子交换膜燃料电池剩余使用寿命预测

Prediction of Remaining Useful Life for Proton Exchange Membrane Fuel Cell Based on NGO-CNN-BiLSTM Neural Network

为解决质子交换膜燃料电池(PEMFC)剩余使用寿命(RUL)预测精度不高的问题,提出了一种基于北方苍鹰优化(NGO)、卷积神经网络(CNN)和双向长短时记忆(BiLSTM)神经网络的动态燃料电池RUL预测模型。首先,利用NGO对CNN-BiLSTM模型的学习率、隐藏节点及正则化系数进行寻优,然后,通过CNN-BiLSTM模型的卷积层对输入数据进行特征提取,输入到BiLSTM层进行时序建模和预测。同时,利用小波阈值去噪算法对原始数据进行平滑处理,采用皮尔逊相关系数提取模型输入变量,并搭建NGO-CNN-BiLSTM神经网络功率预测模型。仿真验证结果表明,该方法预测精度达99.49%,高于其他对比模型的预测精度。

In order to solve the problem of low accuracy in predicting the remaining service life of proton exchange membrane fuel cells,this paper proposed a dynamic fuel cell Remaining Useful Life(RUL)prediction model based on Northern Goshawk Optimization(NGO),Convolutional Neural Network(CNN)and Bi-directional Long Short-Term Memory(BiLSTM)neutral network.Firstly,NGO optimized the learning rate,hidden nodes and regularization coefficient of the CNN-BiLSTM model,and then the CNN-BiLSTM model extracted the features of the input data through the convolutional layer,and input it into the BiLSTM layer for timing modeling and prediction.In addition,wavelet threshold de-noising algorithm was used to smoothen the original data.Pearson correlation coefficient was used to extract model input variables,and NGO-CNN-BiLSTM network power prediction model was built.The simulation and verification results show that this method can effectively improve the prediction accuracy of the remaining service life of fuel cells up to 99.49%,which is higher than that of other comparative models.