基于非线性最小二乘法指数平滑系数估计

Based on nonlinear least-squares index smoothing coefficient alpha estimation

运用指数平滑预测模型进行时序数据的预测分析时,关于指数平滑系数α最优估计是研究者们长期以来需要解决的关键性问题。本文提出基于非线性最小二乘法的指数平滑系数α选取方法,其核心思想在于根据预测值与实测值之间的拟合误差平方和最小值,利用非线性最小二乘法中具有松弛性质的搜索算法,通过高斯-牛顿迭代程序估计最优指数平滑系数α,使得指数平滑预测模型在预测过程中达到更为精准的预测精度。

Using exponential smoothing prediction model to predict the time sequence data, optimal estimation of exponential smoothing coefficient is the key problems researchers have needed to solve. In this paper, an exponential smoothing coefficient alpha selection method based on the nonlinear least squares is proposed, the core idea lies in the minimum sum of squared error between the predicted value and the measured value, using the search algorithm with relaxation property in nonlinear least square method, the optimal smoothness coefficient is determined by the Gaussian-Newton iteration procedure, the prediction accuracy of the exponential smoothing prediction model is achieved in the prediction process.