Abstract: Under the background of double-carbon strategy (carbon emission peak and carbon neutrality) and related energy policies, in order to stabilize the stochastic fluctuation of power flow caused by large-scale access to distributed energy, distributed energy storage will be gradually popularized and applied in the distribution network. In this paper, a distributed energy storage planning model adapted to stochastic sequential decision-making is established, and the voltage amplitude, energy storage action frequency and electricity cost are defined as immediate reward to optimize the distributed energy storage response. Energy storage parameter configuration is based on optimal combined sequential actions. The deep reinforcement learning method based on dueling deep Q network is used to carry out self-learning optimization, and the site and configuration scheme of distributed energy storage is determined with the goal of maximizing the investment income of the whole life cycle. Finally, on condition that the IEEE 33-node system is connected to distributed photovoltaic and energy storage, the rationality and effectiveness of the method are verified.