Research on Predictive Energy Management Algorithm for Fuel Cell Vehicles

Four different neural networks and regression algorithms are applied to the power prediction of fuel cell vehicles. The prediction results of the four different power prediction algorithms are compared using historical vehicle data. The effects of the number of nodes in the hidden layer and the parameters of the algorithm on the prediction results are investigated. Then, the relevant parameters are optimized. After the comparison, the long short-term memory algorithm (LSTM) was finally selected to be applied to the whole vehicle model, and a moving horizon algorithm was embedded to generate the moving horizon long and short-term memory algorithm (MH-LSTM). The number of hidden layer neurons, prediction time scale and moving pane scale affecting the model performance of this power prediction algorithm are analyzed and optimized. The simulation results verify that the strategy can achieve adaptive regulation of the fuel cell output power under NEDC conditions and the charge/discharge balance of the battery charge state.