Abstract: This work proposes an efficient and intelligent optimization method for sensor array deployment in power distribution network fault monitoring, leveraging Multi-Agent Reinforcement Learning (MARL). The method integrates MARL with electromagnetic field physical simulation to achieve adaptive sensor array deployment in complex urban scenarios. A high-fidelity physical simulation engine is constructed using the electromagnetic reciprocity theorem and the shooting and bouncing ray (SBR) method, providing a reliable environment for offline agent training. Through a multi-agent Actor-Critic framework and a comprehensive reward function, agents are able to cooperatively learn and converge to a globally optimal deployment scheme. Experimental results demonstrate that the proposed method consistently and significantly improves monitoring coverage, outperforming random deployment and exhibiting the ability to complete deployment in extremely short time. Furthermore, the method shows excellent task-oriented and constraint-handling capabilities, flexibly adapting to various complex requirements. This research offers a novel solution for electromagnetic sensing-based smart grid operation and maintenance, driving the paradigm shift of sensor array deployment from "manual experience" to "autonomous intelligence.