基于数据挖掘的用电数据异常的分析与研究

发布时间：2018-01-13 14:00

本文关键词：基于数据挖掘的用电数据异常的分析与研究　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：在大数据时代的背景下,我国电力企业更加重视营销业务的信息化,十三五电力发展规划中指出,利用大数据等技术提升信息平台承载能力和业务应用水平。随着用电信息采集系统的推广,海量的用电数据得以采集,为用电环节的大数据分析提供了坚实的数据基础。但是面对海量用电数据的增加,目前大多数电力部门仅使用传统的统计方法进行异常分析,异常数据背后蕴藏的事件信息无法得到有效提炼。为此,有必要将数据挖掘技术引入到异常分析中,充分挖掘用电数据的异常信息。首先,考虑到所有异常都会在电量数据上得以体现,故采用规律性强的日负荷曲线作为异常检测的特征指标。利用具有调节参数少、准确率高、计算效率快等突出优点的孤立森林算法构建异常检测模型,并对模型中的重要参数进行了分析设置。该模型输出所有用户的异常分值及疑似概率排序。研究结果表明,利用该排序,只需要检测异常分值靠前的少数用户即可查出大部分异常用户。其次,为了突出基于孤立森林算法在用电异常检测方面的优越性,通过建立基于聚类分析、局部离群因子算法的异常检测模型并进行比较,发现本文构建的用电数据异常检测模型在计算效率方面的优势尤为显著,并且准确率保持较高,证明了孤立森林算法用于构建用电数据异常检测模型的准确性和高效性。第三,考虑到日负荷曲线受用户用电习惯的影响较大,因此需要结合疑似异常用户的电气变量作进一步分析,建立用电数据异常识别模型,减少误判率。利用决策树算法易于理解实现和效率高等特点,实现对计量点电压进行自动快速分类,并辅助电流数据进行判断,识别出电能计量装置异常。在现场异常排查中,验证了用电数据异常识别模型的有效性。最后,在实际案例中发现,由于异常的电能计量装置中存在残余电压,导致电量追补出现差错。为此,有必要对传统的更正系数计算方法进行改进,通过用电信息采集系统在故障期间冻结的96点电压数据进行分析,综合考虑故障相的残余电压,并研究电能计量装置计算电量的原理,调整更正系数,较现有的方法更全面和更具公正性。
[Abstract]:Under the background of big data era, Chinese electric power enterprises pay more attention to the informationization of marketing business, as pointed out in the 13th Five-Year Plan of Electric Power Development. Using big data and other technologies to enhance the bearing capacity of the information platform and the level of business application. With the promotion of power information acquisition system, massive power consumption data can be collected. It provides a solid data base for big data analysis of power consumption. However, in the face of the increase of massive power consumption data, most power departments only use the traditional statistical method for abnormal analysis. The event information contained behind the abnormal data can not be extracted effectively. Therefore, it is necessary to introduce the data mining technology into the anomaly analysis to fully mine the abnormal information of the electrical data. Considering that all anomalies will be reflected in the data of electric quantity, the daily load curve with strong regularity is used as the characteristic index of abnormal detection. The outlier detection model is constructed by the isolated forest algorithm which has the advantages of fast computing efficiency and so on. The important parameters in the model are analyzed and set. The model outputs outlier scores and suspected probability ranking of all users. The research results show that the ranking is used. In order to highlight the superiority of the isolated forest algorithm in the detection of electrical anomaly, the clustering analysis is established. The anomaly detection model of local outlier factor algorithm is compared and it is found that the outlier detection model constructed in this paper has significant advantages in computing efficiency and high accuracy. It is proved that the isolated forest algorithm is accurate and efficient in building the model of abnormal detection of power consumption data. Thirdly, considering that the daily load curve is greatly affected by users' power consumption habits. Therefore, it is necessary to further analyze the electrical variables of suspected abnormal users, establish a model of abnormal identification of electrical data, reduce the error rate, and use the decision tree algorithm to understand the characteristics of high efficiency and easy to understand. It realizes the automatic and fast classification of the voltage of the measuring point, and adjusts the current data to judge, and identifies the outliers of the electric energy metering device. Finally, it is found that the residual voltage exists in the abnormal electric energy metering device, which leads to the error of power compensation. It is necessary to improve the traditional calculation method of correction coefficient and analyze the 96 points of voltage data frozen by the electrical information acquisition system during the fault, and consider the residual voltage of the fault phase synthetically. The principle of electric energy metering device is studied, and the correction coefficient is adjusted, which is more comprehensive and more just than the existing method.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;TM76

【参考文献】