Application of an improved k‐means clustering algorithm in power user grouping

The illegal act of stealing electricity has brought serious security risks to the operation of power transmission system. The electricity consumption law of power users has obvious characteristics, which can be classified by clustering method. However, the traditional k‐means clustering method has t...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:International journal of numerical modelling Ročník 35; číslo 4
Hlavní autoři: Gang, Wu, Dongdong, Zhang, Shengrong, Fan
Médium: Journal Article
Jazyk:angličtina
Vydáno: Chichester, UK John Wiley & Sons, Inc 01.07.2022
Wiley Subscription Services, Inc
Témata:
Algorithms
Clustering
Correlation coefficients
Dk‐means clustering
Electrical loads
Electricity consumption
Electricity distribution
Euclidean geometry
Evaluation
k‐means clustering
Power consumption
power load curve
power user grouping
Residential areas
ISSN:0894-3370, 1099-1204
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The illegal act of stealing electricity has brought serious security risks to the operation of power transmission system. The electricity consumption law of power users has obvious characteristics, which can be classified by clustering method. However, the traditional k‐means clustering method has the characteristics of randomly selecting the initial cluster center, which leads to the instability of clustering results. Aiming at the shortcomings of traditional k‐means clustering algorithm, this paper proposes a density based k‐means clustering algorithm (Dk‐means clustering) to optimize the initial center selection. According to the characteristics of users' power load, the method of determining the number of clusters and the evaluation method of clustering effect are selected. Then, the traditional k‐means clustering algorithm and the improved Dk‐means clustering algorithm are used to analyze the power load data of a residential area, and six groups of characteristic curves are obtained. Based on the analysis of these curves, the power consumption characteristics of each group of users were evaluated. Finally, through the comparative analysis of Euclidean distance, Manhattan distance and correlation coefficient r, it is proved that Dk‐means algorithm has better clustering effect and is more accurate for the selection of suspicious power stealing users.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0894-3370
1099-1204
DOI:10.1002/jnm.2990