Adaptive Energy Management for Fuel Cell Heavy Trucks Based on Wavelet Neural Network Speed Predictor and Real-Time Weight Distribution

In this article, an adaptive predictive energy management strategy (EMS) for fuel cell hybrid heavy trucks (FCHHTs) is proposed, including a wavelet neural network (WNN) speed predictor and a dynamic weight distribution method. In the offline session, to fully acquire the evolving tendency of upcoming vehicle speed, the recent driving state information is expanded from the time domain to time-frequency domain as the WNN input. Then, fuzzy C-means (FCMs) clustering is adopted to help train several subprediction network models based on the segmented driving states. Besides, an optimized weight distribution reference matrix specialized for each standard driving state is established using 3-D weight maps. In the online session, with real-time driving state recognition results, the vehicle's upcoming demand power sequence is obtained via the dynamic matched subprediction models. Then, the weighting coefficients for power-allocating optimization are determined by a fuzzy matching approach. Finally, hardware-in-the-loop (HIL) testing results showed that compared with benchmark EMSs, the proposed EMS could reduce the operating cost on average by 20.91%, with the economy and durability of the hybrid propulsion system being improved by 25.18% and 2.63%, respectively. Moreover, the computation time per step is less than 0.02 s, indicating its real-time practicality.