应用累积法建立产量预测模式

An Production Forecast Model Based on Accumulative Method

在产量预测模式研究中常用的回归参数估计方法是最小二乘法,但该方法以某些统计假设为前提,计算也比较复杂,特别是在应用短序列资料建模时会受到很大的限制。而累积法是对原始数据序列按一定的规律进行相应叠加,用之建立线性模型,其估计量具有无偏、线性、有效、唯一等特点,估计效果与最小二乘法相同且计算过程更简便。因此,应用累积法,建立了安徽省桐城市一季稻产量预测模式,并介绍了累积法的建模过程和误差分析方法。应用累积法建立的桐城一季稻产量预测模式的历史回代误差率平均为3.90%,2011、2012两年试报准确率分别为95.7%和97.0%,与最小二乘法建立的预测模式误差率相近且略小,其估计精度完全符合业务要求。投入业务使用后,2013、2014年实际预报准确率分别为92.9%、98.5%。累积法的不足之处是随着自变量个数和样本容量的增大,正规方程组系数矩阵的病态性也会随之加剧,从而影响参数估计的精度。

The least-squares method in production forecast model study is commonly used in regression parameter estimation. But it is premised on some statistical assumptions with complex calculation, and will be greatly limited especially in the application of short sequence data modeling. In accumulative method, the original data sequence is superposed according to certain rules, and then is used to establish linear model. Its estimator is unbiased, linear, effective and unique, and has the same effect with the least squares method for parameters estimating, but more simple. In this paper, applying this method, the yield production forecasting model of single season rice in Tongcheng, Anhui was built, and the mod- eling process and error analysis method were described. History back substitution error rate was averaged 3.90%. The reported accuracy was 95.7% and 97.0% respectively for 2011 and 2012. Comparatively, the error rate was similar but slightly smaller with the least squares method. The estimation accuracy could be full compliance with operational requirements. After used in operations, the actual prediction accuracy rate was 92.9% , 98.5% respectively in 2013 and 2014. The deficiencies for the accumulated method are that the normal equations coefficient matrix of ill-will would be increased with increasing number of independent variables and the sample size, thus affecting the parameter estimation accuracy.