Font Size: a A A

Recognition Of Industrial Electricity Consumption Patterns Based On Time Series Clustering

Posted on:2021-10-20Degree:MasterType:Thesis
Country:ChinaCandidate:D N LiFull Text:PDF
GTID:2510306302974669Subject:Applied Statistics
Abstract/Summary:PDF Full Text Request
China is a large industrial country,and industrial electricity has always occupied a very large proportion of electricity in the secondary industry.In recent years,with the rapid development of the national economy and the advancement of various industrial technologies,industry has occupied a more important position in various industries.At the same time,China's industrial electricity consumption has also shown a steadily rising trend.Under the huge industrial power data,many different industrial power modes are hidden.As the development of industry is more diversified,the power consumption pattern of users also shows a diversified development trend.Different business backgrounds,job contents,working hours,and schedules have different power consumption behaviors.Therefore,these power consumption behaviors cannot be identified solely by the user's industry.However,due to the lack of prior knowledge of the user's electricity consumption behavior,such recognition can be considered as unsupervised,that is,a time series clustering problem.Therefore,based on the time series clustering method,this paper conducts a cluster analysis of industrial electricity data,identifies the electricity consumption pattern of the user,and analyzes the characteristics of the electricity consumption behavior of the user.This paper mainly focuses on four aspects: introduction,clustering of power load data,extraction of typical daily load curve for single user,and clustering of multi-user load.In the introduction,this article gives a detailed introduction to the current development trend and characteristics of industrial power in China,and illustrates the current status of the diversified development of industrial power and the importance of pattern recognition for industrial power.Subsequently,this article introduces the current literature on clustering in time series,and elaborates from the perspectives of clustering methods and clustering algorithms.Among them,the clustering method is further divided into clustering based on the original sequence,clustering based on feature extraction,and clustering based on the model.In the introduction of the clustering method,this article not only analyzes the clustering method about power load data,It also introduces some time series methods on other data for more detailed reference.In terms of clustering algorithms,this paper divides clustering algorithms into hierarchical-based,partition-based,density-based,model-based,and grid-based.It mainly introduces the typical algorithms used in time series clustering research in each category.In the problem analysis part,this article analyzes the power load data in terms of data characteristics,clustering method selection and other aspects.First of all,this paper analyzes the characteristics of power load data such as time variability,complexity and randomness.The pattern recognition of power load data can not be performed by factor analysis,but should be performed by extracting information from the data itself.At the same time,this paper also analyzes the massive growth and highdimensional characteristics of power load data in recent years to pave the way for subsequent method selection.In addition,a comparative analysis of industrial power consumption data and household power consumption data in this article shows that industrial power consumption patterns are often more uniform and so on.Then,this paper analyzes the problems of the power load data during clustering according to the characteristics of industrial power load data.It mainly focuses on the selection of clustering methods,similarity measures and clustering algorithm selection.The massive and high-dimensional characteristics of the load data may cause the clustering effect to be unsatisfactory.Finally,based on the characteristics of the data and the current research,this paper proposes a two-stage clustering method of single user followed by multiple users,and attempts to use the statistical test method proposed by Michael Vogt et al.In 2017 to test the number of clusters.In the load curve clustering section,this paper first performs preprocessing such as abnormal value checking on the user's load data,and extracts the typical daily load curve of a single user,and divides the clustering curve of each user in units of days.Aiming at the relatively uniform characteristics of industrial power consumption mode,the trend characteristics that can reflect the global characteristics of the load curve are used as clustering objects.In addition,this article describes in detail the statistical test method used,including the basic theory of the method,model assumptions and statistical construction.Subsequently,this paper uses the above method to extract the typical daily load curve of each user,and briefly analyzes some of the more characteristic curves.In the multi-user clustering stage,this paper extracts the characteristics of each user's typical load curve,including 8 features such as load rate,skewness,and chaos,and uses traditional traversal search method and statistical test method for clustering.3.Determine the number of clusters and compare the pros and cons of the two methods through the three aspects of clustering time,cluster evaluation index and clustering results.It is found that the statistical inspection method is superior to the traditional traversal search method in three aspects.Finally,in this paper,the clustering results obtained by the statistical test method are used to identify and analyze industrial power load patterns,and some of the more typical industrial power patterns are introduced in detail,including the user's power consumption time,the load curve global and local The characteristics of power consumption and the speculation of electricity consumption,etc.,complete the pattern recognition of industrial users.
Keywords/Search Tags:pattern recognition, industrial electricity, time series clustering, statistical test
PDF Full Text Request
Related items