부산대학교 도서관

Monte-Carlo(MC) Methods and Temporal-Difference(TD) modeling techniques are often employed for predicting the charging load of conventional electric cars. In solving discrete and low-dimensional state and motion space variables, they have obtained outstanding results. However, when confronted with the travel characteristics of a plug-in electric taxi (PET) in a more complex environment, the absence of commercial competition between taxis (including commercial electric vehicles with the same travel characteristics, such as online ride-hailing) frequently leads to poor convergence and inaccurate prediction. To address the aforementioned issues, a Multi-Agent Deep Deterministic Policy Gradient (MADDGP) strategy reinforcement learning(RL) method was presented to mimic PET charging decision. The simulation results demonstrate that the approach is more accurate at predicting load and can converge to the Nash equilibrium point with full information even in an environment with imperfect information.