基于多维变量筛选-非参数组合回归的长期负荷概率预测模型

Long Term Probabilistic Load Forecasting Based on Multivariate Selection and Nonparametric Combination Regression Model

长期负荷预测是电网规划及电力市场中长期交易的基础。针对长期负荷受多维因素驱动、不确定性强的特点,提出了非参数组合回归的长期负荷概率预测模型。通过Granger因果分析对驱动负荷长期发展的多维变量进行初步筛选;为提高预测精度,基于逐步平均组合将筛选后的变量集进行非参数组合回归建模,在实现最优组合模型的同时综合各变量对长期负荷的动态驱动;基于随机变化率对最优组合模型包含的多维变量进行不确定性建模,并应用于长期负荷概率预测,获得长期负荷10%、50%、90%分位点值。算例分析结果表明,非参数组合回归模型不仅精度较高,且结合多维变量不确定性建模能实现长期负荷概率预测。

Long term load forecast（LTLF） is the foundation of planning and mid-long term marketing transactions. Since long-term power demand is characterized with multi-variate drive and strong uncertainty, this paper adopted nonparametric combination regression model（NCRM） for long term probabilistic load forecast（LTPLF）. For multi-variate drive, Granger causality test was deployed for preliminary multi-variable selection. Then, based on stepwise simple averaging method, the variable set was embedded into NCRM, to improve accuracy with optimal combination of nonparametric models for different variables and achieve comprehensively dynamic drive of each variable. For strong uncertainty, a probabilistic modeling method based on random variance ratios for the variables embedded into the optimal NCRM was applied into LTPLF, with acquisition of 10%, 50%, and 90% quantiles and planning forecasts. Case study shows that the proposed model, with help of probabilistic modeling, has high accuracy and good performance in LTPLF.