Evaluating the descriptive and predictive ability of the grey-box modelling technique during summertime in a Passivhaus office building in the UK

Grey-box modelling is a well-established technique. However, few studies have focused on applying grey-box modelling to low-energy buildings, with a clear gap in the applicability of the technique to predicting indoor air temperatures in summertime. Therefore, this paper investigates the descriptive and predictive ability of grey-box models, using the RC modelling technique, applied to a Passivhaus-certified case-study office building in summer in the UK. Different training periods and parameter search spaces were used during calibration to explore prior knowledge needed to train the model. Using search spaces based on the calculated values of the model parameters, and informed by uncertainty considerations, the parameters maintained physically interpretable values, with no more than 22% deviation from calculated values. The accuracy of the indoor air temperature predictions was validated against measured in-situ data. The models fitted 10-minutely values of the measured data over 72 hours with an RMSE between 0.632 to 0.307. The model performed well in predicting peak temperatures. The error in predicting the peak temperature for the succeeding days of the prediction period were 0.3%, 1.4% and 2.1%, respectively.