基于BP神经网络的FAST馈源舱融合测量预测研究

Research on Fusion Measurement Prediction of FAST Feed Cabin Based on BP Neural Network

500 m口径球面射电望远镜(Five-hundred-meter Aperture Spherical Radio Telescope,FAST)的跟踪观测需要馈源的空间运动配合,馈源舱主要用于实现馈源的精调定位,因此馈源舱位置的高精度测量对FAST望远镜的高效运行意义重大。但当全站仪设备失效时,无法对采用Kalman算法的GPS/IMU融合测量结果进行修正,导致馈源舱测量精度下降。为了解决这个问题,设计了基于BP(back propagation)神经网络的预测模型,包括数据预处理、模型设计和模型训练验证。模型训练数据为FAST真实测量数据,数据量为40 GB左右。为了验证模型的泛化能力,选取三种运动轨迹数据对模型预测精度进行测试,结果显示,三种运动轨迹下精度都满足15 mm要求。

When the Five-hundred-meter Aperture Spherical Radio Telescope(FAST)performs the tracking observation task,for cooperating with this task,the feed has got spatial motion.The fine-tuning positioning of the feed is realized by the feed cabin,so the highprecision measurement of the position of the feed cabin is great significance.However,when the total station equipment fails,it is unable to correct the GPS/IMU fusion measurements with the Kalman algorithm,it causes the accuracy of the feed cabin measurements decreasing.In order to solve this problem,this paper designs a prediction model based on BP neural network,which is composed of three parts,the data preprocessing,the model design and the model training validation.And the model training data is the real measurement data of FAST with a data volume of about 40 GB.In order to verify the generalization ability of the model,three kinds of motion trajectory data are selected to test the model prediction accuracy,and the results show that the accuracy meets the 15 mm requirement under three kinds of motion trajectories.