Application of clustering technique to electricity customer classification for load forecasting

With the development of smart grid and opening-up progress of the electricity market, Customer Relationship Management (CRM) plays a more and more important role in the power electric industry. Conducting medium and long term consumption pattern analysis of major customers can help the electricity providers grasp the changing trends of the future consumption, and thus better formulate the dedicated tariff offers and provide professional services according to various consumer demands. However, it´s computationally costly and impossible to conduct such analysis for every single customer. To overcome this complexity, this paper aims to provide an effective solution to group customers into certain number of categories with similar electrical behavior by utilizing clustering techniques. By combining distance and correlation, a novel clustering validity indicator is proposed to evaluate the effectiveness of clustering procedure, which is subsequently helpful for choosing algorithms and the optimum number of clusters. Eventually, a case study has been conducted with electricity market data including various electricity customers from different industrial fields. Two frequently-used clustering algorithms have been employed to illustrate the feasibility of the proposed approach.