In active distribution network, the high share of distributed photovoltaic (PV) integration raises the potential for voltage violations within (ADN), and classical model-driven voltage control methods rely heavily on precise physical parameters, which are hard to be adopted in practice. This paper presented an innovative data-driven approach to voltage control. To achieve well-managed energy storages (ESs) and static var compensators (SVCs), the optimization problem with system power loss and voltage constraint is modeled as a Markov Decision process (MDP), which is resolved utilizing a soft actor critic (SAC) reinforcement learning algorithm. The suggested voltage control mechanism is also evaluated in the IEEE 33-node distribution network for validity and feasibility.