针对目前的望远镜指向误差修正模型难以满足激光测距系统实现空间碎片精确探测的要求,提出采用遗传(GA)算法和列文伯格(LM)算法优化的后向传播(BP)神经网络模型修正望远镜的指向误差。按照俯仰角对方位角等间隔分区的方式在测站的半球面天区内观测了102颗恒星,并利用这102颗恒星观测数据进行建模,同时在次日观测12颗恒星验证所建模型的精度。结果表明,所建模型在方位和俯仰方向上的精度分别达到1.94″和1.12″。将GA算法和LM算法优化的BP神经网络模型应用到空间碎片和空间合作目标的探测实验中,结果表明,在利用激光测距技术探测空间合作目标和空间碎片时,望远镜的指向精度在方位和俯仰方向上分别达到1.89″和1.21″以及2.06″和1.46″,因此所建模型在提高望远镜指向精度的同时,对提高空间碎片的探测成功率也具有重要意义。

Since the existing telescope pointing error correction model cannot satisfy the needs of laser ranging systems for accurately detecting space debris, a back propagation neural network model optimized by genetic and Levenberg-Marquardt algorithms is proposed for correcting the telescope pointing error. A total of 102 stars are observed in the station's hemispherical sky area, which are divided into zones covering equal azimuth and altitude angle intervals. These observation data are used for modeling. Additionally, 12 stars are observed the following day to evaluate the model's accuracy, and the results show that its accuracies in terms of azimuth and pitch are 1.94″ and 1.12″, respectively. Finally, the proposed model is used in experiments to detect space debris and cooperative space targets. Results show that, when we use the laser ranging technique, the telescope's orientation accuracies in terms of azimuth and pitch for space debris are 1.89″ and 1.21″ respectively, and that for cooperative space targets are 2.06″ and 1.46″ respectively, showing a significant improvement over the traditional model. These results are also significant from the viewpoint of improving the detection success rate of space debris.