Abstract: The mode transition of source-grid-load-storage multi-step collaboration regulation has been driven by the large-scale access of distributed energy, adjustable load, and storage devices into the distribution network. The frequent bidirectional interaction between distribution network and the distributed resources poses higher requirements on the comprehensive perception and wide-area data transmission capability of the communication network. The fusion of power internet of things (PIoT) and 5G provides crucial technical support through the deep collaboration of cloud-edge-end multi-layer resources. The application of existing cloud-edge- end collaboration technology in the fusion of PIoT and 5G is faced with challenges such as poor adaptability with the PIoT services, poor coordination of heterogeneous resource scheduling, and unguaranteed data privacy security. To solve the above challenges, this paper proposes a 5G cloud-edge-end multi-layer collaboration framework for the PIoT, which provides support for the collaboration interaction between the distributed resources and the distribution network. Then, leveraging the federated deep Q learning algorithm, the semi-distributed artificial intelligence (AI)-based cloud-edge- end collaboration resource scheduling is presented. The device-side task offloading and power control as well as the server-side computation resource allocation are jointly optimized under the ultra-reliable and low-latency communication constraint. Finally, the example analysis is provided to verify the superior performance of the proposed technology in the energy consumption, delay, and throughput. Compared with the AHP-DQN and the EMM, the average throughput is improved by 15.29% and 23.87%, and the total queuing delay is decreased by 53.35% and 62.20%. The results demonstrate the proposed framework and method is able to satisfy the differentiated communication requirements of PIoT services, supporting the bidirectional interaction between the distributed resources and the distribution network.