In recent years, the application of cyber-physical systems (CPSs) is becoming more and more widespread. Since the openness of CPSs and network attackers are becoming more and more intelligent, it is necessary to study the security of CPSs. This paper mainly considers the interaction process of a CPS with several remote state estimation subsystems under DoS attacks. In each system, each sensor monitors its respective system and a scheduler assigns channels to packets from each sensor and sends its local estimates to the remote state estimator with the goal of minimizing the total estimation error covariance. In order to be closer to the real application scenario, the channel signals are considered to be affected by different environments during multi-channel transmission, so that the energy consumed to transmit data in channels of different environments varies. The scheduler and the attacker need to meet the minimum energy requirements in order to transmit data and attack. For the attacker, it is considered to be more intelligent, so if there is still residual energy after attacking one channel and satisfying the remaining channel requirements, it can choose to attack other channels at the same time, thus achieving the opposite goal as the scheduler. A two-player zero-sum game is constructed and the Nash Q learning algorithm is used to solve the optimal policies of both sides, which provides research ideas for studying the secure state estimation of CPSs.