基于深度神经网络的高轨最优Lambert变轨规划  

Optimal high-orbit Lambert transfer planning based on deep neural network

作　　者：徐杭 宋斌 余建慧[3] 郭延宁[1] 马广富[1] XU Hang;SONG Bin;YU Jian-hui;GUO Yan-ning;MA Guang-fu(School of Astronautics,Harbin Institute of Technology,Harbin 150001,China;Shanghai Institute of Aerospace System Engineering,Shanghai 201109,China;Beijing Institute of Tracking and Telecommunications Technology,Beijing 100094,China)

机构地区：[1]哈尔滨工业大学航天学院,哈尔滨150001 [2]上海宇航系统工程研究所,上海201109 [3]北京跟踪与通信技术研究所,北京100094

出　　处：《控制与决策》2024年第9期2969-2976,共8页Control and Decision

基　　金：国家自然科学基金项目(61973100,62273118,12150008);中国航天科技集团有限公司青年拔尖人才支持工程资助项目。

摘　　要：针对高轨在轨服务与主动式碎片清除任务中的Lambert变轨规划问题,提出一种基于深度神经网络的燃料最优快速求解方案.首先,考虑J2摄动力影响与共面圆轨道假设,建立基于Lambert变轨的问题表征;其次,针对轨迹规划问题提出考虑J2摄动力影响的两步PSO优化算法,并经过仿真实验总结变轨过程?V消耗相对转移时长的变化规律(?V-T曲线特性),定义样本形式,构建燃料最优变轨知识库,基于总结的变化规律及样本非线性函数关系的特征,引出基于深度神经网络的快速轨迹规划策略,从而将Lambert计算次数缩小为两次;最后,通过仿真实验,验证所提出高轨最优Lambert变轨规划策略的有效性,在测试集上实现了关键区域变轨时长预测0.0140%的平均绝对误差,新策略具备广泛的应用前景.A fast solution scheme based on a deep neural network is proposed to solve the Lambert transfer planning problem for high-orbit on-orbit service and active space debris removal missions.Firstly,considering the J2 perturbation and coplanar circular orbit assumption,a representation of planning based on the Lambert transfer is established.Secondly,a two-step PSO algorithm considering the J2 perturbation is proposed for trajectory planning.The variation rules ofΔV cost relative to transfer time(ΔV-T curve characteristic)are summarized through simulation experiments,the sample form is defined,and the optimal fuel transfer knowledge base is constructed.Based on the summarized rules and the characteristics of the sample’s nonlinear relationship,a fast trajectory planning strategy based on a deep neural network is proposed,which can reduce the number of calculating the Lambert transfer to two.Finally,simulation results demonstrate the effectiveness of the proposed optimal high-orbit Lambert transfer planning strategy.On the test set,the average absolute error of predicting time is 0.0140%in the key area.The proposed strategy has wide application prospects.

关 键 词：深度神经网络 在轨服务 Lambert变轨 高轨 燃料最优优化 PSO 

分 类 号：V448.2[航空宇航科学与技术—飞行器设计]