This paper proposes the use of deep reinforcement learning to solve the nonlinear Electric Vehicle Routing Problem (EVRP) and redesigns the policy network.