Pattern extraction and structured characterization for electricity consumption profiles in different types of buildings

Abstract

Accurate prediction of electricity consumption patterns in buildings is essential for optimizing energy management and integrating renewable energy sources. This study provides a practical method for predicting electricity consumption and emphasizes the impact of key factors. First, this study collected data from 196 buildings in China and applied various models to improve prediction accuracy. R2 values of the prediction results range from 0.85 to 0.91 for known buildings and from 0.48 to 0.70 for unknown buildings. Second, an information gain-based approach is applied to assess the impact of independent variables. The analysis revealed that intra-day fluctuations have the highest information gain of 0.45, highlighting their dominant influence. Third, a method is proposed to divide a building’s load into time-dependent, weather-dependent, and random components. The time-dependent component captures intra-day and annual fluctuations, while the weather-dependent load reflects cooling and heating demands. This study also indicates that a reliable load prediction can be achieved only based on a small dataset from representative buildings, and combining data from multiple buildings significantly improves regional electricity consumption forecasts. This research improves prediction accuracy for various building types and offers insights into optimizing data collection for load prediction.