Online Ecological Energy Management for Plug-In HEVs Using Optimal SOC Prediction and Stochastic Optimization

Optimizing power demand and distribution in real time is the primary challenge in the energy management of plug-in hybrid electric vehicles. To address this challenge, this study proposes an online ecological energy management strategy using the optimal state of charge (SOC) prediction and stochastic optimization. Future energy-consumption features at each location are computed from the traffic and route information of the entire remaining trip and the optimal SOC value at the next location is predicted from the computed features using a recurrent neural network. By considering the predicted optimal SOC and average speed as references, the speed and SOC values at the next location are optimized using a modified estimation of distribution algorithm (EDA). The process of generating initial and new populations in EDA is modified for fast and reliable convergence. The optimal SOC prediction considering the energy consumption for the entire remaining trip from the current to final locations ensures long-term optimality. Stochastic aspects of the modified EDA can compensate for potential neural-network errors. The prediction and optimization of driving only up to the next location enables to adapt to the changes in traffic conditions. Moreover, the energy efficiency is further improved by optimizing power demand based on the characteristics of the drivetrain, as well as traffic and route conditions. Because the traffic and route information are provided on a location basis, management is performed in a distance domain. The proposed management strategy is evaluated in diverse driving scenarios, and it exhibits good performance, comparable to that of the offline optimization.