A Study Framework of Industrial Electricity-Consumption Correlation Clustering: Taking Xiaoshan Textile Industry as Example

Abstract

Based on daily industrial electricity-consumption data, a study framework of industry clustering is proposed. It consists of qualitative analysis, clustering and forecast. In qualitative analysis, seasonal decomposition is performed and distinguishable characters, i.e., macroscopical trend, periodic/accidental disturbance-response and noise, are separated. It is found that industries in the same category show strong similarity in electricity consumption pattern, on the basis of which a clustering analysis is made with Pearson-correlation-coefficient-based distance and Minimum Spanning Tree method. Typical industries are clustered in the tree topology. Without a-priori economic hypothesis, conclusions beyond traditional cognition are drawn from discussions on Xiaoshan textile industry clustering results. With the clustered industries, Vector-Auto-Regression-based multivariate regression model is applied in electricity consumption forecast in order to take the causality between related industries into consideration. The forecast results coincide well with the observation data and are able to envelop most of the fluctuation.The aim of the study is not only to make a comprehensive forecast for short-to-medium term electricity-consumption, but also to build a cognitive method using ontology of data and logistics of clustering to release more socio-economic value from electricity-consumption data.