Artificial Neural Network based smart forecast models: High-performance close-control and monitoring in art gallery buildings

This paper presents the application of Artificial Neural Network (ANN) algorithms to develop forecast models to predict future energy consumption, outdoor weather and indoor microclimatic conditions in a historical art gallery. Each of these prediction models were implemented on two separate cases of sampling frequencies – daily and hourly sampling; providing a case of day-ahead and a case of hour-ahead predictions, respectively. The ANN models were trained with historical real-data obtained from the various sources, such as building sensors, building management information, and MetOffice. Excellent accuracy in the prediction results were observed through the statistical platform of coefficient of correlation (R) between the real-data and the ANN-predicted counterpart. It was observed that the prediction models for hour-ahead forecasting performed stronger compared to the same for day-ahead forecasting for all the cases of outdoor weather parameters, indoor microclimatic parameters, and NGS energy consumption parameters. The study further reinstates that the ANN-based forecast models can prove to be an ideal platform to investigate various optimisation strategies of the building operation in future, especially in the case of restrictive traditional building types where any retrofit solution needs a strong scientific backing before practical implementation.