Abstract: The optimization of the cargo location of an automated warehouse is vital to improve warehouse efficiency. Aiming at the optimization of power warehouse cargo location, the method based on multi-agent reinforcement learning is adopted to improve the optimization. First, the deficiencies of DDPG algorithm and MADDPG algorithm are analyzed, and on this basis an improved algorithm ECS-MADDPG and its model proposed. In this algorithm, both the immediate reward at the current time point and the future reward factors are considered. Finally, using the historical incoming and outgoing data of electric power materials, the reinforcement learning algorithm is applied to train the cargo location optimization model. Experiments show that ECS-MADDPG has higher stability and rewards compared with algorithms such as MADDPG and DDPG.