基于PCA-RBF神经网络的航空飞行几何参数拟合

Geometric parameter fitting of air flight based on PCA-RBF neural network

采用地面异常线圈对直升机时域航空电磁探测系统进行标定时,发射-接收线圈姿态的变化将导致实测数据产生误差,影响标定的精度.本文基于时间域航空电磁系统,计算了发射-接收线圈姿态任意变化时异常线圈的电磁响应,提出了主成分分析-径向基神经网络(PCA-RBF)的拟合算法,采用主成分分析法提取飞行几何参数的贡献率,利用径向基神经网络法对电磁响应进行了测线剖面的批量数据拟合,并对理论仿真和河南桐柏直升机飞行试验数据进行拟合分析,单一异常体理论数据的绝对误差平均值小于20nV·m-2,双异常体理论数据绝对误差平均值为160nV·m-2.野外实测数据在异常线圈中心位置的拟合相对误差小于1%,整条剖面测线的拟合相对误差小于±6%,平均值为2.5%.结果表明PCA-RBF拟合算法能够较好地实现航空电磁系统飞行参数的拟合,为航空电磁系统海量实测数据的快速处理提供了新方法.

In the calibration of helicopter time-domain airborne electromagnetic detection system using the abnormal ground coils,the variations in flight speed,wind,and attitude of the transmit-receive coil can lead to errors in measured data,thus influencing the precision of calibration.This work calculated abnormal response of the electromagnetic coil to arbitrary attitude change of transmit-receive coil based on the airborne electromagnetic system.We proposed the fitting algorithm for the principal component analysis and RBF neural network（PCA-RBF）.Using the principal component analysis,we extracted the contribution of multiple parameters.Then we fitted the batch electromagnetic response of the line profile with the radial basis function（RBF）neural network method.Finally we analyzed theoretical simulation data and actual flight data from the Tongbai area,Henan province.The results show that the mean absolute value error of single wire loop is less than 20nV·m^-2 and two wire loop is about 150nV·m^-2.For field measured data at the center of the wireloop,the fitting relative error is less than 1%,and the whole section line fitting relative error is less than 6%,with average relative error 2.5%.These results show that the PCA and RBF fitting method can fit the airborne electromagnetic data better,which provide a new approach to process the flood data from airborne electromagnetic surveys.