Abstract: The increase in the number of satellites has brought about huge challenges to the traditional multi-satellite TT&C scheduling methods. Problems such as long scheduling time and low task satisfaction make these methods no longer suitable for large-scale satellite scheduling. Therefore,deep neural networks(DNN) algorithm which has the characteristics of supporting big data, parallel computing and autonomous learning was introduced, and a multisatellite TT&C resource scheduling method based on DNN was proposed. According to the characteristics of multi-satellite TT&C resource scheduling and the requirements of the DNN algorithm, the relevant entities and constraints that affect the scheduling results during the scheduling process were analyzed. Factors that have great impact on the scheduling results were selected and discretized as eigenvalues of DNN. Moreover, this method changed the full matching between TT&C tasks and resources to an effective one through the preprocessing method, which reduced the solution space, the characteristic latitude of the DNN and the difficulty of training. Then a DNN model based on the extracted feature values and scheduling characteristics was built, and the training of the DNN model was completed through a large amount of historical scheduling data. Experiments show that the task satisfaction of the method proposed reaches 99%, and that the running time is reduced by 83% after feature dimension reduction. The results verify that the method proposed is feasible and effective.

Key words: DNN, multi-satellite TT&, C, task scheduling, characteristic analysis, feature dimension reduction