A Study of Classification of Electricity Consumers by Electricity Companies in Comparison to Dynamic Data-driven Clustering Based on Consumption Patterns

Abstract

This research study aimed at determining the impact of dynamic data-driven machine learning clustering technique on billing and revenue by the electricity companies with Kenya Power and Lighting Company (KPLC) as a case study. The study explored the customer classification techniques in use by the electricity distribution company and compared them with the K-Means clustering technique in machine learning. The electricity consumption data used for the research study comprised of 100,000 randomly selected electricity customer accounts for a 12 months consumption period starting from January 2020 to January 2021. The dataset was processed and explored using Python programming language on Jupyter Notebook. The mapping of clusters with the actual tariff classification by the clustering model placed correctly clustered customers at 52% based on their electricity consumption patterns. However, depending on the customers' consumption habits, the 48% getting clustered and mapped to different tariff categories with their actual grouping by the electricity distributor would greatly impact billing and revenue. The study recommends that electricity customer tariff classification by the electricity distribution companies, which informs billing, should be dynamic and data-driven. This, to a great extent, will ensure correct billing and attract more revenue as it is difficult and impossible for electricity distribution companies to detect the change in customer class or lifestyle. The dynamic and data-driven clustering as compared to static once-off customer classification at the point of application and connection to electricity supply will strike a balance between customers’ correct and fair billing and the company’s optimized revenue flow.