基于LSTM模型的飞行器智能制导技术研究

人工智能技术的突破性进展为飞行器再入制导技术的研究提供了新的技术途径, 本文针对预测校正制导中两方面的问题: 一是纵向“预测环节”积分计算量大和“校正环节”割线法迭代求解难以满足实时性的问题, 二是纵向制导和横向制导都需要对动力学方程进行积分, 存在明显的冗余计算问题, 提出基于长短期记忆网络(long short-term memory, LSTM) 的飞行器智能制导技术. 一方面, 在纵向制导中不需要对动力学方程进行积分来预测待飞射程, 即去除“预测环节”; 另一方面, 不再基于割线法迭代求解倾侧角的幅值, 即去除倾侧角的“校正环节”, 大大减少积分计算量, 提高计算速度. 利用深度学习在神经网络映射能力和实时性方面的双重天然优势, 基于再入飞行器的实时状态信息, 采用LSTM模型实时生成倾侧角指令. 同时, 将纵向和横向制导环节的制导周期统一为一个周期, 进一步确保制导系统满足在线制导的实时性要求. 蒙特·卡罗仿真分析表明, 本文所提的方法在飞行器再入初始状态和气动参数拉偏情况下具有精度和速度上的优势. 

RESEARCH OF LSTM MODEL-BASED INTELLIGENT GUIDANCE OF FLIGHT AIRCRAFT

The breakthrough of artificial intelligence provides a new technical approach for the research of aircraft reentry guidance. Aiming at the disadvantage in predictor-corrector guidance, where a large amount of integrations need to be calculated in the prediction step and the bank angle amplitude is iteratively solved based on secant method in the correction step. All the above calculation are difficult to meet real-time demand. Moreover, the dynamic equations need to be integrated both in the longitudinal and lateral guidance, which exist an obvious redundant calculation. In this paper, we propose LSTM (long short term memory)-based intelligent guidance technology. On the one hand, the integration of the dynamic equations in longitudinal guidance is no longer required to predict the range, that is, the prediction step is eliminated, which will greatly reduce the amount of integral calculation and increase the calculation speed. On the other hand, the amplitude of the bank angle will be no longer iteratively solved based on the secant method, that is, the correction step is eliminated. Based on the natural advantages of deep learning both in neural network mapping capabilities and real-time performance, bank angle command is generated by the output of a trained LSTM model based on the real-time state information of the gliding vehicle. At the same time, the longitudinal and lateral guidance periods in the traditional predictor-corrector guidance will be merged into one period, which has the advantage to ensure that the guidance system meets real-time requirements for online using. Monte Carlo simulation analysis show that the proposed method has the advantages both in accuracy and calculation speed under the condition of initial state error and aerodynamic parameter perturbation. The interdisciplinary fusion of guidance technology and artificial intelligence is a hot research direction, which will greatly promote the development of guidance and control of aircraft.