Clustering Electricity Big Data for Consumption Modeling Using Comparative Strainer Method for High Accuracy Attainment and Dimensionality Reduction

Abstract

In smart grid, the relation between grid and customer is bidirectional. Therefore, analyzing load consumption patterns is essential for optimal and efficient operation and planning of smart grid in addition to precise load forecasting. However, emergence of the advanced metering infrastructure, which enables a two-way flow of data and power consumption between consumers and suppliers, has resulted in data explosion in smart grid applications. Because of the volume and velocity of data generation in recent years, traditional data analysis methods are inefficient. Therefore new methods of analyzing such as “data mining", which segments data before analyzing and manipulating, are recommended. Clustering, as a well-known method in data mining, has extensively been employed in recent electricity industry. This article argues that even though clustering methods can be directly applied to raw data of electricity consumption, this approach is inefficient since it requires storage and processing of high-dimensional and high-volume data. Hence, it would be more beneficial to cluster consumption data in a space of reduced dimension. In this paper, we propose a new structure for dimension reduction to refine the electricity consumption data. The results are compared with the famous method of dimension reduction, principal component analysis (PCA). We evaluate our technique using datasets from Kaveh, an industrial area in Iran.