Fast Estimation of Lambert Transfer Duration for GEO Spacecraft Rendezvous Missions Based on Residual Neural Network

To address the challenges of high computational complexity in determining the Lambert transfer duration for GEO spacecraft rendezvous missions, this study proposes a rapid estimation method based on a Residual Neural Network (ResNet). First, a Lambert transfer model for GEO spacecraft rendezvous is established. With transfer duration as the decision variable and fuel consumption as the fitness function, the correlation between these two parameters is analyzed. The Comprehensive Learning Particle Swarm Optimization (CLPSO) algorithm is adopted as the foundational method for optimizing dataset acquisition. Second, the ResNet architecture is adaptively modified to better suit data regression applications. Through structural improvements and pre-training optimization of network scale and activation functions, the ResNet is repurposed for high-precision regression tasks. The optimized network is then formally trained to achieve rapid estimation of Lambert transfer duration. Finally, simulation results demonstrate that compared to metaheuristic algorithms, the ResNet delivers highly accurate planning results within an extremely short timeframe. Compared to fully connected neural networks of the same scale, the ResNet achieves superior planning precision with competitive computational efficiency. The proposed method reduces computational resource consumption during the planning process and enhances planning accuracy.