High penetration of distributed photovoltaics in medium and low voltage distribution networks aggravates voltage violations and reverse power flow. Distributed energy storage systems can improve the regulation flexibility of distribution networks under the randomness and fluctuation of photovoltaic generation. However, traditional centralized optimization based on physical models and single reactive power control methods are difficult to coordinate massive resources effectively due to high-dimensional computation and active-reactive power coupling. To address these problems, this paper proposes an active-reactive power coordinated control method based on networked distributed proximal policy optimization. A fully distributed training framework based on independent learning is developed, where each agent relies only on local observations and limited neighboring information to approximate the global optimal policy, thus reducing the dimensionality burden in large scale distribution networks. A coordinated action space considering inverter capacity constraints is designed, in which the charging and discharging power of distributed energy storage systems is used for active power regulation to suppress voltage fluctuations. In addition, line losses are included in the reward function to guide power flow optimization while improving the voltage qualification rate.Simulation results on 33, 141, and 322 bus distribution networks show that the proposed method achieves high voltage qualification rates across different system scales and maintains relatively low line losses in most scenarios, demonstrating good voltage regulation performance, comprehensive optimization capability, and adaptability to different network topologies.