CBR model for predicting a building’s electricity use: On-line implementation in the absence of historical data

Abstract

This paper presents the development and on-line implementation of a case-based reasoning (CBR) model that predicts the hourly electricity consumption of an institutional building. Building operation measurements and measured and forecast weather information are used to predict the electricity use for the next 6 h. The model’s ability to efficiently deal with an initial absence of historical data and continuously learn as more data becomes available was tested by emptying the database holding historical data prior to the on-line implementation. The prediction accuracy was monitored for almost 4 months. The results show significant improvement as more data becomes available: the initial error, 1 h following the on-line implementation is close to 44%, it decreases by almost half after 16 h, and reaches 12.8% at the end of the monitored period. This shows the applicability of a CBR predictive model for new and retrofit buildings where historical data is not available.