粗铜冶炼中铜锍品位的动态预测模式—AR(p)与指数平滑最优组合法

来源期刊：中南大学学报(自然科学版)2000年第1期

论文作者：梅炽 姚俊峰 胡军 石明 赵伟

文章页码：34 - 36

关键词：AR(p)模型;指数平滑模型;组合加权系数;组合预测

Key words：Auto Regressive Model; Exponential SmoothingModel; combined weight number; combined forecasting

摘    要：提出了一种对铜锍品位进行预测的新方法,即以采集的现场数据为基础,采用系统辨识动态地建立了AR(p)模型与三次指数平滑模型.AR(p)模型要求数据对象是平稳时间序列,而三次指数平滑模型的数据对象具有随机性,考虑到铜锍品位的波动性,将2种模型按最小二乘原理,以组合预测误差平方和为目标函数,通过使误差平方和极小化来确定2种预测方法的最优加权系数,建立了一种新的组合模型,其预测误差最小.结果表明,在当时数据条件下,AR(p)与指数平滑组合模型比AR(p)与指数平滑模型单独使用时精确度都要高,这对指导生产具有实用意义.

Abstract: This paper proposed a newmethod for forecasting the grade of copper matte based on the collected data from the factory and it established the dynamic Auto-Regressive and Exponential Smooth Model by the system identification. The data target of the Auto-Regressive Model requires time series is smooth, while the data target of Exponential Smoothing is random .The grade fluctuation of copper matte being considered, this paper established a new kind of combined model, in which the forecasting error could become minimum among the three models. The square of the forecasting error was regarded as the target function and the best weight numbers were gotten according to the principle of the minimal least square. The result showed that the degree of accuracy was higher by using the combined model than by using the AutoReqressive Model and Exponcntial Smoth Model independently in current condition. The combined model is of momentous current significance to manufacture.