A Power Management Strategy for Parallel PHEV Using Deep Q-Networks

Fuel economy of a plug-in hybrid electric vehicle depends on a power management according to the driving conditions of the PHEV. A number of power management strategies on PHEV have been studied based on dynamic programming, equivalent consumption minimization strategy and reinforcement learning. Since reinforcement learning provides a methodology for mapping actions from states, reinforcement learning has the advantage that it is easier to apply to real vehicles than other methodologies. In this study, we developed the power management strategy of PHEVs which is efficient and could be implemented on actual PHEVs using deep Q-networks.