The Running Time Prediction of Spacecraft Simulation Job Based on HC-LSTM

With the increasing scale of spacecraft, the demand for simulation computing continues to increase, and the requirements for high-performance computing (HPC) resources for simulation jobs are gradually increasing. For the spacecraft designers, the estimation of computing resources and running time required by simulation computing jobs is not accurate enough. Accurate prediction of the running time of spacecraft simulation job is helpful to the effective operation of Backfilling scheduling strategy and further improves the utilization rate of resources. In this paper, according to the characteristics of spacecraft simulation computing jobs, the historical logs and simulation application parameters are analyzed, and key operation characteristics are extracted. At the same time, considering the time sequence correlation of computing jobs, we propose a running time prediction algorithm based on hierarchical clustering and LSTM. Through the comparative analysis of experiments, it is verified that the algorithm has high accuracy in predicting the running time of spacecraft simulation operation, and the prediction effect is better than other related machine learning algorithms.