Game Strategy Prediction for Spacecraft Orbital Pursuit–Evasion Game Based on Long Short-Term Memory

This paper presents a strategy prediction frame for multi-player orbital pursuit–evasion game that is based on discount receding horizon coevolution (DRH-CE). The proposed frame aims to enable spacecraft to indirectly characterize the target’s possible future states by predicting strategy parameters. The authors establish a game strategy model and a strategy solution model based on DRH-CE. The payoff function parameters of the DRH-CE are utilized as strategy parameters to construct the dataset by combining the strategy solutions and parameters. Furthermore, the authors establish a strategy parameter prediction model based on long short-term memory and multi-head self-attention, and combining this model with the strategy solution model allows for the prediction of the future states of targets. The numerical examples illustrate the efficacy of the proposed frame in predicting strategy parameters and the effectiveness of the future state prediction against targets.