User tariff classification with multi-dimensional deep learning on electricity consumption data

Abstract

Electricity load is crucial to managing energy grids and determining energy policies. Electricity tariffs offer diverse opportunities to consumers at different prices, necessitating proper usage, detection of non-compliant users, and appropriate sanctions. This study utilizes real-life electricity load data from 310 users in Türkiye's five most common tariff classes to create a three-channel time series. An innovative method combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models is proposed to predict tariff groups based on electricity load data. After preprocessing, the time series are converted into two-dimensional matrices, and features are extracted using a 2D CNN network. These features are then utilized in an LSTM network to highlight sequential relationships. The developed model achieved 87% accuracy in training and 83% in validation, demonstrating its effectiveness in predicting tariff groups from electricity load data. This study proposes and validates a hybrid CNN-LSTM-based tariff classification approach using real-world electricity consumption data for smart grid management and energy pricing, thereby contributing to the research on analyzing and utilizing electricity load data.