BP神经网络在多位置捷联寻北系统中的应用

Application of BP neural network to multi-position strap-down north seeking system

为了精确拟合多位置捷联寻北系统采集的数据曲线,计算陀螺初始位置和真北方向的夹角,介绍了多位置捷联寻北系统的工作原理,推导了寻北测量的数学模型,并分析了影响测量精度的因素。分别采用最小二乘法和BP神经网络法,对两种方法的拟合精度和最终计算得到的寻北结果进行了比较。实验结果表明:与最小二乘法相比,BP神经网络拟合精度较高,拟合残差和较小,达到0.23μV,残差的均方差达到1.3mV;在计算相位角时,多次寻北结果的均值基本一致,但均方差明显优于最小二乘法,达到8″,满足寻北系统对数据拟合精度的要求。

With the aim to fit the sin wave curve of a strap-down north seeking system and to calculate the angle between the start position and the real north accurately, the principles of the system are introduced, and the factors influencing the measuring precision are analyzed. Then, the least square method and the Back Propagation（BP） neural network are adopted to fit curves separately. Experimental results indicate that the BP method is preciser than the least square method, for its error sum is 0.23 uV and the squared error of residual error is 1.3 mV. Moreover,for phase angle measurement, both methods can get the same average values in six measurements, hut the squared error of BP neural network is 8″, which is more accurate than that of the least square method. These results show that the proposed BP method can satisfy the precision requirements of curve-fitting north seeking systems.