This paper explores the application of deep reinforcement learning in optimizing scheduling strategies for smart grids. By introducing the Deep Q-Network (DQN) algorithm, the effectiveness of this technology in improving power supply reliability, reducing power generation costs, and minimizing network losses is verified. The deep reinforcement learning scheduling strategy has shown significant advantages in various power grid scenarios. The experimental results show that the DRL scheduling strategy can improve power supply reliability by nearly 3 percentage points and significantly reduce power generation costs and network losses.