Optimised building energy and indoor microclimatic predictions using knowledge-based system identification in a historical art gallery

This paper presents a system identification (SID) model for an historical art gallery of great cultural significance. These buildings require tight indoor temperature and moisture controls that demand significant energy from air handling units. Complex dynamic building systems, stringent conservation restrictions, and lack of detailed monitoring make diagnosing and optimising their energy use difficult. Building simulation software programmes have proven to be effective, but have tended to rely on data generated by simulation models. This study shows how artificial neural network (ANN) models trained with historical real data can predict a building’s energy use and the optimal indoor microclimate necessary for conservation. Four ANN target-data scenarios were designed for optimised model predictions, and 12 ANN training algorithms were tested with six architectural scenarios collecting daily and hourly data. The ANN models used a randomised 80% sample of the database, with the remainder (20%) validating the models. The model displayed a high coefficient of correlation (0.99), with the mean square error and mean absolute error less than 0.1% and 2%, respectively. This ANN-based SID tool efficiently represents a complex building system and could be an ideal method for investigating optimisation strategies prior to their implementation.