遗忘因子最小二乘支持向量机及在陀螺仪漂移预测中的应用研究

以陀螺仪漂移误差系数时间序列预测为对象，研究并提出了遗忘因子最小二乘支持向量机算法。构造了以多项式、径向基、小波函数为核函数的支持向量机（SVM）、最小二乘支持向量机（LSSVM）、遗忘因子最小二乘支持向量机（FFLSSVM），比较了它们用于强非线性测试集的泛化性能。实验结果表明：FFLSSVM比由相应核函数构成的SVM、LSSVM自适应性强、预测精度高；三种核函数生成L2（R）子空间上完备基的能力不同，导致三个FFLSSVM逼近任意目标函数的精度有差异；遗忘因子最小二乘小波核支持向量机可有效地用于陀螺漂移误差动态补偿、可靠性辅助决策、故障预测。

Forgetting-Factor Least Square Support Vector Machine and Application on Drift Forecasting of Gyro

A forgetting-factor least square support vector machine (FFLSSVM) algorithm was given in the paper.Employing Polynomial,RBF and wavelet function as the kernel functions respectively,we constructed three kind of support vector machine(SVM),which are SVM,least square support vector machine(LSSVM) and FFLSSVM.Applying the three kind of SVM to build the forecasting model of the drift-error-parameter of gyro based on the non-linear time series measurement data.The experiment results show that,when employing the same kernel function,the self-adaptability and forecasting accuracy of FFLSSVM is superior to SVM and LSSVM.Due to the different ability producing complete radix of the three kinds of kernel functions,there is different approximate ability of the three FFLSSVM.The FFLSSVM,which using the wavelet function as the kernel function,can be used to compensate the dynamic drift-error of the gyro,provide the information for the reliability decision-making,forecast the fault.