计及匹配偏差分值的用电量周期标签分类方法

Classification Method of Electricity Consumption Cycle Labels Considering Matching Deviation Score

用户用电量周期规律复杂多变,各标签之间的相关性不稳定,企业用电量分类精度较低,时间复杂度较大。为提高用电量周期标签分类精度,提出计及匹配偏差分值的用电量周期标签分类方法。根据用电量预测的偏移量收集用电用户的各个维度数据,通过计算用电用户的身份特征属性权重预测用户的用电量。采用匹配偏差分值的计算方式获取用电量的时间序列,根据电力用户的初始特征初步匹配用电用户的行业标签,结合用电量时间序列匹配流程完成用电量时间序列的匹配。基于用户特征的目标函数评估标签位置在排序过程中的移动情况,结合用电量时间序列的匹配偏差分值计算结果设计用电量周期标签算法,实现用电量的周期标签。实验结果表明,所提方法能够根据行业用电量为其标记标签,将行业分类精度和标签与行业之间的匹配度提高到了90%和95%以上,迭代次数达到30时,时间复杂度为2.94 s,提高了用电量周期标签分类精度。

The law of user power consumption cycle consuming is complex and changeable,the correlation between labels is unstable,the classification accuracy of enterprise power consumption is low,and the time complexity is large.In order to improve the classification accuracy of power consumption cycle labels,a power consumption cycle label classification method considering matching deviation score is proposed.According to the offset of power consumption prediction,the data of each dimension of power users are collected,and the power consumption of users is predicted by calculating the identity attribute weight of power users.The calculation method of matching deviation score is adopted to obtain the time series of power consumption.According to the initial characteristics of power users,the industry labels of power users are preliminarily matched.Combined with the power consumption time series matching process,the matching of power consumption time series is completed.Based on the objective function of user characteristics,the movement of label position in the sorting process is evaluated.Combined with the matching deviation score calculation results of power consumption time series,a power consumption cycle label algorithm is designed to achieve the power consumption cycle label.The experimental results show that the proposed method can label the industry according to its power consumption,and improve the industry classification accuracy and the matching degree between labels and industries to more than 90%and 95%.When the number of iterations reaches 30,the time complexity is 2.94 s,which improves the classification accuracy of power consumption cycle labels.