In this paper, we take the background of intelligent inspection in power plants and face the demand of minimizing the task completion rate and average energy consumption of inspection machines within their inspection scenarios. 5G MEC technology is utilized to construct an inter-node migration model for MEC to coordinate the coupling of policy formulation of multiple inspection machines. As a solution to the problem of unbalanced loading and underutilization of the edge server for deployment in inspection machine implementation, this research proposes the enhancement of Top-K algorithm. The enhanced algorithm considers base stations with the highest number of tasks for edge server deployment and ensures coverage of all the base stations by the edge server. In the offloading problem, the GT offloading decision is proposed by integrating game theory and reinforcement learning, and the GT offloading decision scheme is implemented using a decentralized algorithm to establish a distributed offloading algorithm. In addition, the SLA-based reinforcement learning method is used to promote the effective connection between inspection machines and edge servers to form the task offloading strategy of multiple inspection machines-multiple edge servers under the non-cooperative game.Under the task offloading algorithm proposed by this research, the experimental inspection system operates at the transmission delay and energy cost range of 151.19 to 1400.00 depending on the increasing number of intelligent terminals ranging between 150 and 550., and the ratio of idle CPU is only 13.40 to 67.36 during this period. The inspection system supported by the method in this paper achieves high resource utilization at a lower cost and improves the training effect while reducing the inspection cost.
