In order to reduce the business platform's processing of massive IoT terminals and meet the physical isolation and differential quality assurance between different services, 5G slicing technology is used to logically isolate network resources to effectively protect the differentiated service needs of different users, aiming at the problem of resource allocation during network slicing transmission in power grids. Maximize the total throughput of EMBB users and minimize the power of URLLC users by jointly optimizing the transmit power and downlink timeslot of the base station. Considering that the hybrid nonlinear integer programming problem is difficult to solve, we propose a method based on asynchronous advantaged-actor critic to solve the problem. Simulation results show that compared with DDPG algorithm, the proposed algorithm can converge quickly and has the best performance, which verifies the practicability of the proposed algorithm in the optimization of power gateway slicing resources.