考虑细分行业及多影响因素的月度电量预测

Monthly Electricity Forecasting Considering Subdivided Industries and Multiple Influencing Factors

从细分行业的电量数据出发,通过挖掘不同的行业用电规律以及引入气温、春节影响因素,实现特大都市地区月度电量的精准预测。首先,利用数形相似性距离,将各细分行业划分为电量周期性行业与电量非周期性行业;其次,通过k-means算法按照用电特性相似度对电量周期性行业进行聚类,得到不同用电规律的各类行业;接着,分析气温指标与各行业类别电量的相关性,在与气温指标相关程度较大的行业类别电量预测中加入气温协变量;然后,采用Prophet算法对各行业类别电量建立预测模型;最后,对1、2月份的预测结果针对春节效应提供联合修正方法。采用南方某市的用电量数据作算例分析,考虑细分行业后预测精度提升2.63百分点,整体平均预测准确率达到97.71%。算例分析结果表明,所建立的考虑细分行业及多影响因素的月度电量预测框架能够挖掘各行业类别的用电规律,捕捉地区产业转移和升级特点,并有效引入多影响因素指标,进而提高特大都市地区月度电量预测的准确度。

Starting from the electricity consumption data of subdivided industries,this paper excavates electricity consumption laws of different industries and introduces the influencing factors of temperature and the Spring Festival to accurately estimate the monthly electricity of the megalopolis.Firstly,it uses numerical and morphological similarity distance to divide each subdivided industry into an industry category with cyclical electricity consumption or an industry category with non-cyclical electricity consumption.Secondly,it adopts k-means algorithm to cluster the electricity-cyclical industries according to the similarities in their features of electricity consumption and create distinct industry groups with various electricity consumption patterns.Then the paper analyzes the correlation between the temperature index and electricity consumption of each industry,and adds the weather covariate for electricity forecasting of the industry groups with a strong correlation coefficient of the electricity and weather.By using the Prophet algorithm,it establishes a forecasting model for the electricity of each industry group.Finally,it offers a joint correction approach for the Spring Festival effect for the forecasting results of electricity in January and February.Taking the electricity consumption data of a southern city as an example,the paper points out the forecast accuracy is improved by 2.63 percentage points after considering the subdivided industries,and the overall average forecast accuracy reaches 97.71%.The analysis results show that the framework of monthly electricity forecasting established can discover the electricity consumption pattern of each industry group,capture the characteristics of regional industrial transfer and upgrading,and effectively introduce multiple influencing factors,thereby improving the accuracy of monthly electricity forecasting.