Abstract: The power sensor network can collect and obtain information on the power grid equipment's working status and working environment in real time, which plays an important role in monitoring and quick response of the grid facilities. Aiming at the special requirements of system such as data queuing time delay and packet loss rate, this paper proposes a resource allocation scheme based on reinforcement learning (RL) for power sensing networks. Under the resource constraints, the scheme optimizes the queuing time and packet loss rate of sensor nodes through the resource allocation algorithms, and the optimization problem is modeled as a Markov decision process (MDP), which is solved by double deep Q-network(DDQN) algorithm. Simulation results and numerical analysis show that the proposed scheme outperforms the benchmark scheme in convergence, time delay and packet loss rate.