Prediction of internal temperatures during hot summer conditions with time series forecasting models

A novel application using adaptive autoregressive time series forecasting with exogenous inputs (i.e. ARX) has been developed in order to provide reliable short-term forecasts of the internal temperatures in dwellings during hot summer conditions (i.e. heatwaves). The study shows that with proper selection of the predictors, based on the Akaike Information Criterion (AIC), the forecasts provide acceptable accuracy for periods up to 72 hours. The hourly results for the analysed dwellings showed a Mean Absolute Error (MAE) below 0.63°C and 0.49°C for the two case study dwellings across the 3-day forecasting period, during the 2015 heatwave. These findings point to the potential for using time series forecasting as part of an overheating warning system in buildings, especially those housing vulnerable occupants.