The star earth collaboration network private network can make up for the problems of poor signal coverage, limited bandwidth, low security and high cost that exist in the existing wireless public network. It can provide reliable and stable power system communications. However, the star-earth collaborative network suffers from resource constraints due to limited on-planet spectrum resources, power, computation and storage resources. Therefore, how to allocate these limited channel resources reasonably and efficiently, it is especially important to improve the resource utilization and system performance. Based on this, we establish a star-earth collaborative network architecture to meet the application of power scenarios. To improve the resource allocation efficiency of this network architecture, we model the channel allocation process by sensing the channel allocation state and beam user service request state in the environment through the satellite intelligences of the star-earth collaboration network. And then, we propose a dynamic channel allocation algorithm based on deep learning. We design state, action and reward functions for feedback optimization of channel resource allocation strategy based on deep learning, and a deep learning based channel resource allocation method is proposed to realize the optimal allocation of system channel resources. In the proposed method, the corresponding channel resources for users is allocated based on the channel allocation policy, and environment-based reward gain information is given to optimally update the channel allocation policy. The related simulation results show that our proposed channel resource allocation algorithm for star-terrestrial collaborative network has lower service blocking rate and higher channel utilization than other assistant allocation algorithms.