Power management of plug-in hybrid electric vehicles using neural network based trip modeling

The plug-in hybrid electric vehicles (PHEV), utilizing more battery power, has become a next-generation HEV with great promise of higher fuel economy. Global optimization charge-depletion power management would be desirable. This has so far been hampered due to the a priori nature of the trip information and the almost prohibitive computational cost of global optimization techniques such as dynamic programming (DP). Combined with the intelligent transportation systems (ITS), our previous work developed a two-scale dynamic programming approach as a nearly globally optimized charge-depletion strategy for PHEV power management. Trip model is obtained via GPS, GIS, real-time and historical traffic flow data and advanced traffic flow modeling. The gas-kinetic based model was used for the trip modeling in our previous study. The complicated partial deferential equation based model with several parameters needs to be calibrated had for implementation. In this paper, a neural network based trip model was developed for the highway portion, using the given data from WisTransPortal. The real test data was used for the training and validation of the network. The simulation results show that the obtained trip model using neural network can greatly improve the trip modeling accuracy, and thus improve the fuel economy. The potential of the advantages were indicated by the fuel economy comparison.