Abstract: Distributed energy storage system（ESS） accessed from user side is generally characterized by the decentralized controllability and unscheduled response, which increases financial risk of the electricity retailers especially under the high penetration conditions. Scientific dynamic pricing is an important strategy to avoid such risks. In order to overcome the problems of incomplete data observation, complex modeling and difficult convergence of model-based approaches, a data-driven method combined with dynamic evolution was proposed and applied to dynamic pricing strategy of the retailer considering interaction effect of user-side distributed ESS. First, state distribution of ESS was deduced by cellular automata and the real-time price was optimized with the goal of maximizing full-time profit of the retailer. Then a large amount of offline samples were extracted to fully train the improved sparse deep belief network, which took time serial number and the amount of power changed in each transformer district at the previous time slot as inputs, and directly outputs the relative change of price at current time slot. Finally, an intelligent dynamic pricing model was obtained to guide the adaptive adjustment of price and mitigate the risk of loss. By studying the examples of distributed ESS in multiple districts, it is verified that by learning simple and observable data, the proposed method is capable of adjusting price in real time, so as to effectively prevent the operational risks brought by the ESS arbitrage. Besides, the model possesses stable adaptability to the change of ESS permeability.