基于预测误差的用水预测模型选择方法研究

Research on Selection of Water Consumption Forecast Model

时间序列法是用水量预测的常用方法,其中预测模型的选择是提高预测精度的关键。针对目前以拟合精度作为模型选择标准的方法,提出了以预测值的置信区间最小为标准选择预测模型的方法。以山西省运城市工业用水量资料为例,分析给出了幂函数、S函数、直线、指数函数、二次和三次抛物线等6个模型95%置信水平下5年和10年两种预测期的预测区间,均以幂函数预测精度最高;对该系列近5年的用水量进行了预测,预测误差也以幂函数最小。表明以预测区间最小选择用水预测模型的方法是合理的、可靠的。

The time series method is one of common methods for forecasting water consumption. The prediction accuracy on water consumption can be guaranteed by the selection of forecast models. View of the current selection of forecasting models in order to minimize the fitting error, a method of selecting forecast model was put forward based on minimum error of forecast. Taking the industrial water consumption data of Yuncheng city for example, the prediction interval of forecast period of 5 and 10 years were given in the 95% confidence level for six models such as power function, S function, linear, exponential functions, quadratic and cubic parabola. The results show that power function has the highest prediction accuracy. The series amount of water data in the nearly 5 years was predicted. Power function has the minimum prediction error. The results show that the method of choosing models according to the minimize prediction range is reasonable and reliable.