Abstract: With the promotion of 5G, network slicing technology is also widely used. It virtualizes physical resources so that the network can carry multiple types of services, adapting to the new situation and the future development of ubiquitous perception and intelligent interconnection of the power grid. The control, acquisition, and application services in the power network have produced a large number of 5G network slices in multiple scenarios, and have also generated a large number of slice optimization management requirements. Based on the traditional 5G slicing and combining the characteristics of different services in the power grid, this paper proposes a solution for traffic prediction using deep learning, and generates a slicing resource strategy based on the prediction results. In particular, a hybrid isolation strategy is proposed for many key service slices in power 5G slices. Through the demonstration and evaluation of the experimental platform, the network utilization efficiency of this method has increased from 46.3% of the traditional slice to 71.5%.