Abstract: Aiming at the problem for the research on the simulation and modeling of the space information network (SIN) with high complexity, a SIN modeling approach based on simulation data is proposed. On this basis, a deep predictive auto-encoder network for SIN performance indicators and a hybrid forecasting analysis model with random forest regression are designed. Due to the characteristics of high dimensionality and wide distribution of SIN performance indicators, with using the deep learning auto-encoder network, the construction of SIN performance index is realized. Combined with the random forest regression model, the relationship model between the network design parameters of SIN and the performance indicators is built. The case study shows that the hybrid model can not only directly predict the SIN performance indicators, but also can analyze the sensitivity of the network design parameters.

Key words: simulation data driven, space information network modeling, deep auto-encoder network, random forest regression