In order to solve the interference problem of device-to-device (D2D) communication in cellular network, this paper introduces simultaneous wireless information and power transfer technology, and proposes a distributed resource allocation algorithm based on double deep Q-network. This algorithm helps D2D link learn the optimal strategy under the constraints of meeting the minimum quality of service requirements of equipment and incomplete channel state information, so as to alleviate the interference in the system, realize distributed resource allocation and maximize the energy efficiency of D2D link. Firstly, the resource allocation problem of D2D communication is expressed as a Markov decision process. And then, the allocation problem is decomposed into two sub problems: power control and channel allocation. The problem is transformed according to reinforcement learning technology, modeled as a resource allocation problem with multiple agents, and a training algorithm is designed. Experimental results show that the proposed algorithm can effectively converge, significantly improve the energy efficiency of D2D link layer and the throughput of D2D link, and has certain feasibility and effectiveness.