Advanced Hierarchical Missile Guidance Strategies Based on Temporal Forecasting and Deep Reinforcement Learning

In response to the evolving requirements of modern combat scenarios, there is a growing emphasis on developing missile guidance systems that are more autonomous and capable of intelligently navigating and responding to challenging conditions. To address the challenges posed by increasingly complex target evasion patterns in modern warfare, this paper presents a novel hierarchical missile guidance strategy that integrates temporal forecasting with deep reinforcement learning. Our approach combines an Evasion Pattern Forecasting Network (EPFNet) with multiple reinforcement learning-based guidance policies, each tailored to a specific evasion pattern. The EPFNet, built on a bidirectional LSTM architecture, achieves a 97.5% accuracy in predicting target evasion patterns. We employ Proximal Policy Optimization (PPO) to train guidance policies, incorporating a Zero-Effort Miss (ZEM) guidance law to reduce the action space and enhance computational efficiency. Our hierarchical strategy demonstrates superior performance compared to traditional methods, achieving a 91.4% hit rate within a 10-meter tolerance across various evasion patterns. This significant improvement underscores the effectiveness of our approach in maintaining precise strike capabilities in challenging battlefield environments. The proposed method offers a promising solution for next-generation missile guidance systems capable of adapting to diverse and complex target behaviors.