Accurate and fast clustering of residential customer load curves based on bid-vote integration algorithm

Cluster analysis as a new branch of statistics is receiving more and more attention. The number of studies using cluster analysis is also increasing, and its application in power systems is becoming more and more valuable. One of the examples is the clustering analysis of customer load curves. The c...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2022 9th International Forum on Electrical Engineering and Automation (IFEEA) s. 579 - 584
Hlavní autoři: Liu, Xiangxiang, Wang, Zongyao, Ding, Guili, Xu, Zhihao, Han, Wei, Kang, Bing
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 04.11.2022
Témata:
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Cluster analysis as a new branch of statistics is receiving more and more attention. The number of studies using cluster analysis is also increasing, and its application in power systems is becoming more and more valuable. One of the examples is the clustering analysis of customer load curves. The clustering analysis of customer load curves can yield two important pieces of information: one is to obtain typical load distribution: the second is to become important data for classifying customers by load characteristics. This method has theoretical and practical significance for demand response, load forecasting, load control, electricity consumption anomaly detection and even tariff cataloging and developing marketing strategies. At present, there are many researches on cluster analysis and clustering methods of load curve, but each has its own advantages and disadvantages. In this paper, the advantages and disadvantages of traditional k-means algorithm, Gaussian fuzzy algorithm, SOM algorithm and GWO-FCM algorithm are deeply analyzed, and and combining the clustering stability and clustering effect of different algorithms. A voting integrated clustering algorithm is proposed, which realizes the dimensionality reduction of high-dimensional data by integrated tree fitting, uses the Marxian distance to overcome the correlation of clustering index dimensions, and then determines the effective number of clusters; determines the baseline clustering algorithm by DB criterion, and finally unifies the integrated clustering results by the consistency function matrix. Finally, the effectiveness of the proposed voting integration algorithm for clustering results is verified by comparing the actual arithmetic data based on a residential demand response experiment in a province in southeast China, which increases the effectiveness and superiority of the proposed algorithm for clustering residential demand response user data by 31% on average.
DOI:10.1109/IFEEA57288.2022.10038189