分位数回归方法在盛夏日用电量预测中的应用

Application of quantile regression method in prediction of daily electricity consumption in mid-summer

利用南方某市2010—2012年盛夏期间日用电量和日气温、降水、相对湿度等数据,探索了分位数回归方法在日用电量预测中的应用。均一化处理后得到的日用电量系数序列剔除了经济社会发展和双休日等因素影响,相关分析表明其变化与前日用电量和当日最高气温变化的关系最为密切;将前日用电量和当日最高气温作为预报因子建立分位数回归方程,独立样本检验结果表明预测效果良好;与常用的均值回归等方法相比,分位数回归方法能够给出预测值的条件概率分布情况,可为电力调度和风险管理提供更多参考信息,具有较好的应用前景。

The 2010—2012 midsummer daily electricity con-sumption and temperature, precipitation, relative humidity of onesouthern city were used to explore the application of quantile regres-sion method in forecasting of daily power consumption. The resultsshow that daily consumption coefficient series after homogenizationeliminated the impact of the economic and social development andweekend factors, whose variation is most closely related to change ofpower consumption on the previous day and maximum air tempera-ture of the day. Independent sample test results show that the effectof regression equation is good with predicator of tempreture electrici-ty consumption and maximum air temperature of the day. Comparedwith the commonly used least square regression method, quantile re-gression method can give out conditional probability distribution ofthe prediction, providing more reference information for the powerdispatch and risk management, and has good application prospects.