基于改进LSSVM的动态测量误差实时预测方法

Dynamic measurement errors real time forecasting method based on improved least squares support vector machines

误差修正是提高动态测量精度的有效途径,其中误差的建模是关键。在分析现有动态测量误差预测技术不足的基础上,提出基于改进的最小二乘支持向量机的动态测量误差回归建模和预测方法。在最小二乘支持向量机的基础上,通过将价值函数改为最小二乘价值函数以及用等式约束代替不等式约束,将求解的二次规划问题转变为一组等式方程,减少了待定参数的个数,很大程度地缩短了支持向量机的训练时间;同时针对最小二乘支持向量机稀疏性丢失这一缺陷,采用剪枝算法改进其性能,使其具有更好的稀疏性。通过实例验证及与其他建模方法的对比,表明该方法具有优良的预测效果和动态性能,为动态测量误差预测提供了一种新的可行方法。

Error compensation is an effective approach to enhance dynamic measurement precision and error model is the key factor for the compensation. In view of the limitation of the present dynamic measurement errors prediction technologies, this paper put forward least squares support vector machines model for dynamic measurement errors. In LS-SVM, the SVM problem formulation was modified by introducing a least squares cost function and equality instead of inequality constraints, and the solution followed directly from solving a sot of linear equations instead of quadratic programming. Thus the tuning parameter of LS-SVM was less than standard SVM and also the training time becomes shorter. Also, a pruning algorithm was proposed to enhance the sparse ability. Finally, a real application example and comparison with other modeling methods is given, which indicates that the proposed method has an excellent prediction performance and is a new feasible approach for dynamic measurement errors prediction.