There are problems of complex power line channel impedance variation and poor load impedance mismatch. Thus a multi-parameter optimal solution method for a Π-type impedance matching network based on deep reinforcement learning is proposed, and the feasibility of deep reinforcement learning for finding the optimal matching parameters is verified and analyzed. First, the Π-type network structure is established to derive the objective function for the optimal matching in the narrowband matching and broadband matching scenarios. Secondly, deep reinforcement learning is used to use the movement of the agent to simulate the component parameters of the actual matching network, and set the formula containing the theoretical value and the optimal matching value of the parameters as a reward to build the optimal matching model. Then, this paper separately verifies the network parameters of narrowband matching and broadband matching application scenarios and optimizes the network parameters of the model. Finally, the simulation results prove that the trained optimal model has short running time and high accuracy. It can better automatically match the load impedance change of power line carrier communication, and improve the quality of power line carrier communication.