Residual Life Prediction of Proton Exchange Membrane Fuel Cell Based on Improved ESN

Aiming at the problem that the current residual effective life prediction (RUL) technique for proton exchange membrane fuel cells (PEMFCs) has poor prediction effect in the medium and long term, a residual life prediction method based on the Improved Gray Wolf Optimization algorithm (IGWO) and Echo State Network (ESN) is proposed, in which the voltage of the electric stack is firstly selected as a health indicator, and the PEMFC dataset is processed by using convolutional smoothing filtering method to carry out data Smoothing and normalization are used to effectively reduce the interference of outliers on the subsequent model training. Then the reserve pool parameters of the ESN are optimized using the local and global optimization search capability of IGWO, and the IGWO-ESN network model is constructed, and the processed dataset is used for the training of the remaining life prediction model of the PEMFC, and finally it is compared with the traditional ESN for verification. The results show that the improved ESN model predicts the root mean square error and average absolute percentage error of 0.0342 and 0.9315%, respectively, and the prediction accuracy is significantly improved compared with the ordinary ESN model, and the prediction accuracy of the medium- and long-term RUL is also higher.