With the rapid development of power digitization technology, substations, as a key component of the power system, face a large amount of computing pressure. In this paper, firstly, a three-layer cloud-edge-local computing offloading framework is proposed for the computing business requirements of substation converged terminals. Secondly, considering the mobility of terminal equipment and the limited computing resources on the edge side, we construct a cloud-edge-end system task consistency computation model considering the resource constraints and construct a Markov decision model with the goal of maximizing the total system task consistency. Finally, a dynamic task offloading optimization method is constructed based on the deep deterministic extreme gradient algorithm, which realizes the optimal scheduling of the overall computing and network resources in the cloud-edge-local three layers. The results show that compared with the edge-local two-layer offloading scheme, the consistency of the total system tasks of the proposed scheme is improved by 29%.