一种基于聚类算法的天然气负荷预测不确定性估计方法

A clustering-based uncertainty estimation method for natural gas load forecasting

对天然气负荷预测的不确定性进行准确估计是预测模型使用可靠性的关键。建立了一种基于聚类算法的天然气负荷预测不确定性估计方法。此方法首先使用敏感性分析计算模型输入的敏感性指数,在根据敏感性指数对模型的输入进行加权后,采用聚类算法得到历史数据的若干聚类簇。然后利用各个聚类簇中的残差项计算预测区间。在浙江省某天然气门站数据集上对此方法的效果进行验证。结果显示,对比传统残差聚类的方法,此方法具有更高的预测区间估计准确度,预测区间平均覆盖误差是0.46%。此方法能够量化天然气负荷预测模型的不确定性,可为天然气系统优化运行和调度控制提供更加可靠的负荷预测方法。

Accurate estimation of natural gas load prediction uncertainty is the key to the reliability of prediction models.A clustering algorithm-based uncertainty estimation method for natural gas load forecasting was established.Firstly,the sensitivity analysis was utilized to calculate the contributions of prediction model inputs to outputs.The contributions were utilized to determine weights of model inputs in residual clustering.Then,intervals of natural gas load prediction model results were calculated according to the residuals of specific clusters.This method was tested on a natural gas load dataset collected from a natural gas station in Zhejiang,China.The conclusion was that compared with residual clustering methods without weighted inputs,this method had better performance in interval estimation.The mean coverage error was 0.46%.This method can quantify uncertainties of gas load prediction models,and provide more reliable load prediction methods for management and scheduling of natural gas systems.