用于干涉型光纤陀螺温度漂移辨识的RBF神经网络改进算法

Improved learning rule of RBFNN for identifying temperature-introduced drift of IFOG

针对干涉型光纤陀螺(IFOG)温度漂移的辨识,推导了径向基神经网络(RBFNN)中隐含层神经元、网络的抗噪声性能和拟合精度三者之间的关系,并在此基础上提出了一种新的径向基函数神经网络辨识学习规则.该方法具有很强的抗噪声性能,网络输出不会被陀螺噪声所污染,同时能动态地确定神经元数,辨识精度高,有效地避免了传统RBF网络学习算法中事先固定网络结构可能存在的盲目性.实验结果表明,该方法能够快速、准确地辨识IFOG的温度漂移.

To identify the temperature drift of interfere fiber optic gyroscope （ IFOG）, the relationship among the nerve unit, the anti-noise performance and the fit precision in radial basis function neural network （RBFNN） is deduced, and a new learning rule of RBFNN is proposed. The im- proved neural network has strong anti-noise performance and cannot be polluted by the noise of IF- OG. The method can also determine the number of nerve cells, avoiding the blindness in choosing the parameter with traditional radial basis function （RBF） network learning rules. The experimental results prove that the proposed method can identify the temperature-introduced drift of IFOG exactly.