Abstract: A deep reinforcement learning based wind-photovoltaic-storage system joint dispatch model is proposed. First, a joint dispatch model that fully considers the constraints of various wind and solar storage stations is established, where tracking dispatch plans, wind and solar curtailment, and energy storage operation costs are considered in the objective function. Then, the state variables, action variables and reward function under the reinforcement learning framework are defined. Later, a deep deterministic policy gradient algorithm is introduced, using its environmental interaction and strategy exploration mechanism to learn the joint scheduling strategy, so as to achieve the dispatch strategy tracking, reduce wind and solar abandonment, and reduce energy storage charging and discharging. Finally, the historical data of wind power, photovoltaic, and dispatch plan in a certain area of northwestern China are employed to train and analyze the model. The results of the case studies show that the proposed method can adapt well to the changes in the wind power and photovoltaic power in different periods, and the joint scheduling strategy can be obtained under given data of wind and photovoltaic.