Vehicle optimal control method based on deep reinforcement learning

The invention discloses a vehicle optimal control method based on deep reinforcement learning. The method comprises the following steps: step 1, establishing a strategy network and a mutually independent value network; 2, the vehicle is controlled to run, and samples are collected; 3, inputting the data st and at into a value network to obtain two value scores, and calculating a prediction score by taking the smaller value; inputting the state st + 1 into the strategy network to obtain an action at + 1, respectively inputting the data st + 1 and at + 1 into two value scores in the two value networks, determining a TD error according to the value scores and a prediction score, and updating the value networks; 4, updating the strategy network after the value network is updated twice; and step 5, repeating the steps 2-4 to perform network parameter tuning until the policy network achieves an expected effect, and outputting the finally updated policy network. The stability can be ensured in the process of optimiz