Data poisoning attack system and method for offline reinforcement learning, program, equipment and storage medium

The invention belongs to the technical field of data poisoning, and particularly relates to a data poisoning attack system and method for offline reinforcement learning, a program, equipment and a storage medium. According to the method, tiny disturbance is added to the key time step in the offline reinforcement learning process, the effect that the attacked algorithm model learns a poor target strategy is achieved with very low attack cost, and the effectiveness and feasibility of the attack are determined. According to the method, the trajectory having great influence on the learning task can be determined in the offline reinforcement learning process, poisoning is performed for key time steps, the attack cost is reduced to a great extent, and the attack efficiency is improved; according to the disturbance method provided by the invention, tiny disturbance conforming to the proportion of the data set can be dynamically added to the data set, compared with original data, the change amplitude is very small an