Earth - Venus Low-Thrust Optimal Transfers / Database A

A database of mass optimal trajectories of a low thrust spacecraft from Earth to Venus' orbit starting on the date 7th of May 2005 and arriving at Venus' orbit. This database was generated with a perturbation size of 0.2 and contains 429,316 trajectories with 100 samples along each trajectory. The database is in the HDF5 format with 4 dataframes included. These are the 'nominal', 'train', 'val' and 'test' dataframes each of which contains rows of entries in the following format: <pre>['t', 'p', 'f', 'g', 'h', 'k', 'L', 'm', 'lp', 'lf', 'lg', 'lh', 'lk', 'lL', 'lm', 'T', 'ux', 'uy', 'uz', 'traj_id', 'sampl_id', 'vf']</pre> which are the time, equinoctial elements (6 of them), the mass, the costates of the Optimal Control Problem Hamiltonian (7 of them), the thrust magnitude, the thrust directions (ux, uy, uz correspond to fr, ft, fn), the unique trajectory id, the sample id (nth sample from the start), and the value function. We are in the process of writing a paper titled "Real-Time Optimal Guidance for Interplanetary Transfers Using Deep Networks" that uses this dataset for the training of a neural network. The details on how we generated this data can be found in the paper, but it is essentially done using the (famous) "Backward Generation of Optimal Samples" method.本数据集收录了2005年5月7日从地球出发、目标抵达金星轨道的低推力航天器大规模最优轨迹库。该数据集以0.2的摄动尺度生成，共包含429316条轨迹，每条轨迹内含100个采样点。数据集采用HDF5格式存储，内含4个数据帧，分别为'nominal'（标称）、'train'（训练）、'val'（验证）与'test'（测试）数据帧。各数据帧内的行条目均遵循以下格式：<pre>['t', 'p', 'f', 'g', 'h', 'k', 'L', 'm', 'lp', 'lf', 'lg', 'lh', 'lk', 'lL', 'lm', 'T', 'ux', 'uy', 'uz', 'traj_id', 'sampl_id', 'vf']</pre>，各字段依次对应：时间、6个春分点轨道要素、航天器质量、最优控制问题哈密顿量的7个协态变量、推力幅值、推力方向（ux、uy、uz分别对应径向、切向、法向分量）、唯一轨迹ID、采样序号（轨迹起始处的第n个采样点）以及价值函数。本团队正在撰写一篇题为《基于深度网络的星际转移实时最优制导》的论文，该数据集将用于神经网络的训练。数据集的生成细节可参见该论文，其核心采用了知名的「最优样本逆向生成（Backward Generation of Optimal Samples）」方法。