Abstract: Multi-access edge computing (MEC) and device to device (D2D) technologies can be used to improve the capability of sensing data transmission and processing in power intelligent inspection. However, the problem of optimal allocation of network resources under spectrum reuse and interference should be solved. For D2D-assisted MEC networks, this paper proposes a joint optimal resource allocation algorithm based on deep reinforcement learning. Firstly, the terminal capacity, power consumption and delay calculation methods of D2D-assisted MEC networks are analyzed under the constraints of channel multiplexing and interference, power and computation. Secondly, considering the requirements of throughput, power consumption and delay, a resource optimal allocation model based on comprehensive utility function maximization is established. Deep reinforcement learning is used to realize the joint optimization of task offloading and resource allocation. Simulation results show that the proposed algorithm can effectively improve the comprehensive performance of system capacity and task offloading.