A thermal control methodology based on a machine learning forecasting model for indoor heating

Archive ouverte : Article de revue

Abdellatif, Makram | Chamoin, Julien | Nianga, Jean-Marie | Defer, Didier

Edité par HAL CCSD ; Elsevier

International audience. To take advantage of the data generated in buildings, this document proposes a methodology based on a machine learning model to improve thermal comfort and energy efficiency. This methodology uses measured data (e.g., indoor/outdoor temperature, relative humidity, etc.) and forecast data (e.g., meteorological data) to train a multiple linear regression model to forecast the indoor temperature of the space under study. Using the genetic algorithm optimization method, this model is then used to evaluate the different heating strategies generated. For each strategy, a score is assigned according to user-defined criteria in order to prioritize them and select the best one. By studying an office building simulated under the TRNSYS software, a multiple linear regression model was implemented with errors less than 1% and an adjusted R2 coefficient close to 0.9. Compared to a conventional heating strategy, this methodology can improve thermal comfort by up to 43%.

• Contient
• Sujet
• Description
• Oeuvre