Abstract: Aiming at the electric vehicle charging routing problem, an optimal control model was built based on the dynamic heterogeneous traffic network, vehicle dynamics model, and predicted traffic state. The travel time of the vehicle is the optimized objective and the battery state is the hard constraint in the model. To solve the optimal control model online, a hybrid learning optimization algorithm (HLOA) was constructed based on the spatial-discrete model predictive control theory and reinforcement learning method. Relying on Dueling deep Q-networks (Dueling DQN), the algorithm has designed two solving modules, i.e., offline and online, to realize the mixed multi-step training of its parameters. After the training, the algorithm combines the predicted traffic state, dynamic heterogeneous traffic network, and vehicle dynamics model to output online driving strategies in real-time. Based on the above methods, the local road network in one city is the experimental scenario. Combined with the traffic simulation, the competitive ratio was used as the evaluation index to analyze the value of the predicted traffic information and the feasibility of HLOA. The experimental results show the value of the prediction information and the feasibility of HLOA. The competitive ratio of HLOA is 33.65% higher than the other algorithms. The algorithm can help electric vehicle users to plan their trips. It can also provide technical support for autonomous vehicles.