An adaptive energy management method based on Pontryagin's minimum principle with real-time traffic information for plug-in hybrid electric vehicles

Abstract

This study addresses the key challenge of velocity prediction for hybrid vehicle in intelligent network environment, proposing an innovative solution that incorporates real-time traffic information. Based on Long Short-Term Memory neural network and real-time traffic data (LSTMRT), a novel velocity prediction model is developed, which uniquely integrates three key real-time traffic elements: the ego-vehicle status, the traffic signal states, and the dynamics of surrounding vehicles. By analyzing the impact of structural parameters on prediction performance, a pioneering application of transfer learning is made to resolve the problem of model retraining due to driver differences, thereby enhancing the generalization capability of the model. Furthermore, an Adaptive Pontryagin's Minimum Principle (APMP) control strategy based on LSTMRT (LSTMRT-APMP) is proposed, which achieves online optimal power distribution for plug-in hybrid powertrains. Simulation results demonstrate that compared with the state-of-the-art prediction models, LSTMRT improves prediction accuracy by at least 68 % in various driving cycles, and after transfer learning, the prediction accuracy for new drivers can be further increased by 44.83 %. Comprehensive validation under four different driving cycles and SOC intervals confirms the exceptional performance of LSTMRT-APMP in fuel economy (at least 8.77 % improvement), real-time capability (maximum computing time is only 4.67 ms), charge sustainability, and robustness.