CLASSIFICATION AND CLUSTERING OF RURAL AREA ELECTRICITY CONSUMERS USING MACHINE LEARNING AND GEOGRAPHICAL INFORMATION SYSTEMS

Abstract

Sustainable developments and reliable power system operations require effective controls on the electricity consumption behavior which is very essential to achieve the saving target and control. In the energy sector, practically machine learning (ML) techniques are being used to study the consumption patterns of the customers which are helpful in effective demand response management, energy efficiency policy. In this article, study carried out on cluster analysis for classification and evaluation of energy consumption (EC) patterns of domestic consumers with single phase low voltage level in Sinnar tehsil under Maharashtra State. Dataset consists of consumption data, geographical data and weather conditions. There are many clustering techniques available but here K-means (KM) clustering technique used to classify collected information into two and three classes. Basically, in this article clustering form based on the Energy Consumption pattern as well as spatial location of consumers using QGIS tool. To achieve optimal number of clusters, we apply elbow method and to cross check Silhouette score technique used to distinguish the clusters based on average monthly consumption. Different classification models are developed by the results obtained from K-means clustering to assess the consumers EC patterns. GIS is used to visualize, analyze and interpret these clusters spatially. The results analyzed that the K-means clustering reveals minimum overlap of 3.13 % and 1.48% for two classes and three classes respectively mostly living in gavthan colonies and wadi vasti domestic area having kutcha, pucca houses as per GIS visualization and analysis. The classification techniques exhibit that the XGBoost technique achieves a higher accuracy for cluster classification of this kind of information data.