基于改进深度神经网络的火箭弹在线弹道规划方法

Rocket Online Trajectory Planning Method Based on Improved Deep Neural Network

摘要

为提升火箭弹在线弹道优化效率并缩短最优弹道生成时间,以落速最大为性能指标,提出了一种基于改进深度神经网络的在线弹道规划方法。针对制导火箭弹滑翔弹道模型,构建了一个可自适应调节状态注意力权重的深度神经网络,用于实现弹道快速规划。首先,利用Radau伪谱法对单条制导火箭弹弹道进行优化;其次,通过对初始点状态施加一定程度的扰动,生成大量典型弹道数据用于进行训练;然后,针对不同状态量学习难度差异以及积分误差累积问题,提出了带有状态注意力模块的神经网络。在此基础上,建立了完整的弹道预测模型,并开展仿真对比和蒙特卡洛实验。结果表明,所提出的改进深度神经网络能够动态调整各状态量之间的权重,在不同初始条件下快速生成合理的弹道规划方案,且各项误差相比于传统深度神经网络均有所下降,显示出更好的学习能力和适应性。此外,该方法计算速度也远远快于伪谱法,具有在线弹道规划的应用潜力。

Abstract

To reduce the optimal trajectory generation time and enhance online trajectory optimization capability in rocket guidance, an online trajectory planning method based on an improved deep neural network (IDNN) was proposed in this paper. Taking maximum impact velocity as the performance index and the glide trajectory model of guided rockets as the research object, a deep neural network capable of autonomously adjusting state attention weights was constructed for trajectory planning. First, the Radau pseudospectral method was employed to optimize a single guided rocket trajectory. Then, perturbations were introduced to the initial state to generate a large set of typical trajectory data for training. To address the varying learning difficulties associated with different state variables and the issue of integrated error accumulation, a neural network incorporating a state attention mechanism was proposed and trained on the same dataset together with a traditional neural network. Based on this, a comprehensive trajectory prediction model was established, and comparative simulations along with Monte Carlo experiments were conducted. The results show that the proposed IDNN can dynamically adjust the weights among different state variables, rapidly generate reasonable trajectory planning schemes under different initial conditions. It achieves lower errors in all aspects compared to traditional deep neural networks, demonstrating superior learning ability and adaptability. Moreover, the calculation speed is also much faster than that of the pseudo-spectral method, showing its potential for online trajectory planning.

关键词

注意力机制 / 神经网络 / 制导火箭弹 / 弹道规划

Key words

attention mechanism / neural network / guided rocket / trajectory planning

参考文献

[1] TANG X, CHEN J. Direct trajectory optimization and costate estimation of infinite-horizon optimal control problems using collocation at the flipped Legendre-Gauss-Radau points[J]. IEEE/CAA Journal of Automatica Sinica, 2016, 3(2): 174-183.

[2] MANSELL J R, GRANT M J. Adaptive continuation strategy for indirect hypersonic trajectory optimization[J]. Journal of Spacecraft and Rockets, 2018, 55(4): 818-828.

[3] KIYAK Z J, HARTFIELD R J, LEDLOW T W. Missile trajectory optimization using a modified ant colony algorithm[C]// Proceedings of 2014 IEEE Aerospace Conference. Big Sky, Montana: IEEE, 2014.

[4] 夏红伟, 翟彦斌, 马广程, 等. 基于混沌粒子群优化算法的空间机械臂轨迹规划算法[J]. 中国惯性技术学报, 2014, 22(2): 211-216.

[5] XIA Hongwei, ZHAI Yanbin, MA Guangcheng, et al. Path planning algorithm of space manipulator based on chaos particle swarm optimization algorithm[J]. Journal of Chinese Inertial Technology, 2014, 22(2): 211-216. (in Chinese)

[6] HARRIS M W, ACIKMESE B. Lossless convexification of non-convex optimal control problems for state constrained linear systems[J]. Automatica, 2014, 50(9): 2304-2311.

[7] MAO Y, SZMUK M, ACIKMESE B. Successive convexification of non-convex optimal control problems and its convergence properties[C]// Proceedings of 2016 IEEE 55th Conference on Decision and Control. Las Vegas, Nevada: IEEE, 2016: 3636-3641.

[8] 刘译蔓, 陈琦. 火箭弹中制导凸优化方法[J]. 兵器装备工程学报, 2024, 45(9): 175-181.

[9] LIU Yiman, CHEN Qi. Convex optimization method in rocket guidance[J]. Journal of Ordnance Equipment Engineering, 2024, 45(9): 175-181. (in Chinese)

[10] 尹秋霖, 陈琦, 王中原, 等. 考虑时空协同的滑翔制导炮弹单炮多发快速弹道规划[J]. 兵工学报, 2024, 45(3): 798-808.

[11] YIN Qiulin, CHEN Qi, WANG Zhongyuan, et al. Rapid trajectory planning for single-gun multi-firing gliding guided projectiles considering time-space coordination[J]. Acta Armamentarii, 2024, 45(3): 798-808. (in Chinese)

[12] 王庆海, 陈琦, 王中原, 等. 基于伪谱凸优化和L1罚函数的弹道规划方法研究[J]. 弹道学报, 2022, 34(1): 22-30.

[13] WANG Qinghai, CHEN Qi, WANG Zhongyuan, et al. Research on trajectory planning method based on pseudospectral convex optimization and L1 penalty function[J]. Journal of Ballistics, 2022, 34(1): 22-30. (in Chinese)

[14] SHI Y, WANG Z. A deep learning-based approach to real-time trajectory optimization for hypersonic vehicles: AIAA 2020-0023[R]. 2020.

[15] SANCHEZ-SANCHEZ C, IZZO D. Real-time optimal control via deep neural networks: study on landing problems[J]. Journal of Guidance, Control, and Dynamics, 2018, 41(5): 1122-1135.

[16] 伍元培, 王阳, 赵爱红, 等. 基于深度学习网络的高超声速飞行器避障轨迹规划方法[J]. 战术导弹技术, 2022(1): 53-59.

[17] WU Yuanpei, WANG Yang, ZHAO Aihong, et al. Obstacle avoidance trajectory planning method for hypersonic vehicles based on deep learning networks[J]. Tactical Missile Technology, 2022(1): 53-59. (in Chinese)

[18] ZAVOLI A, FEDERICI L. Reinforcement learning for robust trajectory design of interplanetary missions[J]. Journal of Guidance, Control, and Dynamics, 2021, 44(8): 1440-1453.

[19] 池海红, 周明鑫. 融合强化学习和进化算法的高超声速飞行器航迹规划[J]. 控制理论与应用, 2022, 39(5): 847-856.

[20] CHI Haihong, ZHOU Mingxin. Trajectory planning for hypersonic vehicles integrating reinforcement learning and evolutionary algorithms[J]. Control Theory & Applications, 2022, 39(5): 847-856. (in Chinese)

[21] 谢蕃葳, 王旭刚. 基于改进DQN方法的滑翔制导炮弹弹道规划[J]. 弹道学报, 2023, 35(4): 53-60.

[22] XIE Fanwei, WANG Xugang. Trajectory planning for guided gliding projectiles based on improved DQN method[J]. Journal of Ballistics, 2023, 35(4): 53-60. (in Chinese)

[23] 王亚洲, 佃松宜, 向国菲. 深度神经网络辅助的垂直回收火箭在线轨迹优化方法[J]. 中国空间科学技术, 2024, 44(4): 130-141.

[24] WANG Yazhou, TIAN Songyi, XIANG Guofei. Online trajectory optimization method for vertical recovery rockets assisted by deep neural networks[J]. Chinese Space Science and Technology, 2024, 44(4): 130-141. (in Chinese)

[25] ALABASSY B, SAFAR M, EL-KHARASHIM W. A high-accuracy implementation for softmax layer in deep neural networks[C]// Proceedings of 2020 15th International Conference on Design & Technology of Integrated Systems in Nanoscale Era. Marrakech, Morocco: IEEE, 2020: 1-6.

基金资助

国家自然科学基金(52202475)

江苏省自然科学基金(BK20200498)