Abstract: The increasing penetration of the distributed power sources exacerbates the voltage violations in the distribution networks and brings new challenges to the reactive voltage control. The controllable active and reactive devices in the active distribution networks play an important role in mitigating the voltage deviations. In this paper, we propose a multi-intelligent reinforcement learning-based active-reactive coordinated voltage control strategy to adjust the reactive power of the connected PV inverters and the active power of the EV charging piles with the goal of reducing the voltage deviations and the network losses without depending on the precise flow model. The reinforcement learning algorithm using the empirical enhancement technique with the attention mechanism (EAMAAC), which provide unbiased training data to improve sample efficiency. The proposed voltage control strategy is validated on the IEEE 33-bus test case. The results show that the proposed control strategy is not only effective in mitigating the voltage deviation but also has higher and more stable sampling efficiency compared to the existing reinforcement learning algorithms.