Back Propagation Neural Network Approach for Space Objects Orbit Prediction Improvement

High accuracy orbit determination (OD) and prediction (OP) are essential for high-precision spatial applications, the avoidance of collision risks for space objects and space situational awareness. Two-line element (TLE) orbit parameter database is most comprehensive. However, when conducting orbital calculations using TLE, the spatiotemporal variation characteristics of OP errors are complex. The existing modeling methods struggle to accurately reveal the OP errors evolutionary patterns. Traditional numerical methods calculation speed is slow and the calculation efficiency is low. In this study, six datasets of OP errors are constructed based on the long-term historical TLE data for six space objects at different orbit altitude. The Back Propagation Neural Network (BPNN) algorithm is employed to establish a model for OP errors, aiming to enhance OP accuracy. The results show that compared with using the SGP4/SDP4 model directly for OP, the OP accuracy is improved by at least 50% after the errors processing model designed in this study. These findings provide theoretical and technical support for space object collision warning systems. Future research will focus on the investigation of input variable importance, model generalization capabilities, and selection of data capacities for different space objects.