Abstract: The large-scale integration of distributed energy resources and energy storage systems has introduced significant challenges to voltage control in active distribution networks, primarily due to multi-source coupling and time-varying uncertainties. To enhance the intelligence and real-time performance of voltage regulation, this paper proposes a multi-agent cooperative voltage control strategy based on deep Q-networks. A multi-agent reinforcement learning framework is established, wherein each distributed unit is modeled as an autonomous agent capable of local state perception, policy updating, and joint reward optimization. The proposed method is validated through simulations on a modified IEEE 69-bus system. The results demonstrate that the proposed approach effectively reduces node voltage deviation and system power loss, while significantly improving voltage control accuracy and operational stability.