Remaining Useful Life Prediction of Proton Exchange Membrane Fuel Cells Using Genetic Algorithm Based Nonlinear Autoregressive Exogenous Network

Abstract

The proton exchange membrane fuel cell (PEMFC) is one of the most promising clean energy sources with characteristics like high energy conversion, no electrolyte leakage, and low operational temperature. However, it is difficult to build a mathematical model because the system consists of a complex nonlinear system. In the meantime, accurate estimation of the remaining useful life (RUL) of fuel cells plays an important role in improving the safety and lifetime of fuel cells. A joint prediction method based on genetic algorithm (GA) and nonlinear autoregressive neural network with external input (NARX) is proposed. The method was designed to predict the RUL of the proton exchange membrane fuel cell. GA is used to optimize the initial weights and biases of the NARX neural network. Then, the historical voltage evolution under rated current conditions is used to train the NARX network, where the trained model is used to predict the voltage evolution under ripple conditions. Integrating the GA-NARX algorithm leads to the improvement of convergence speed and the prediction accuracy of the algorithm. This integrated algorithm obtains better estimation accuracy compared to the NARX network by itself. The proposed method was compared with the genetic algorithm-based backpropagation neural network (GA-BPNN) and genetic algorithm-based time delay neural network (GA-TDNN). The proposed method is validated with the IEEE PHM 2014 Data Challenge dataset and the results showed that the method has better prediction accuracy compared to other ANN algorithms.