Completing a dataset for a passive house using a Feedforward Neural Network with Adam optimization

Abstract

In the context of monitoring variables through sensor systems, potential sensor failures may lead to data gaps, negatively impacting scientific studies and overall time and effort efficiency. This study specifically addresses the challenge of filling missing data for the interior temperatures inside a passive house by utilizing artificial neural networks (ANN). So, to achieve this aim, various monitored predictors, including outdoor temperature (Tout), outdoor humidity (RHout), atmospheric pressure (Press mm Hg), wind speed, dew point temperature (Tdewpoint), visibility, and energy consumption within the house, are employed to construct our developed models. Indeed, our primary goal is to complete the mean temperature (Tmean) of the passive house, enhancing its utility as an additional predictive variable for supplementing internal temperatures in individual rooms (T1, T2, T3, T4, T5, T7, T8, and T9). Significantly, our study provides evidence of the enhanced performance of the proposed approach through evaluation metrics. Specifically, during the training phase, the root mean square error (RMSE) is reported as 0.12, with a remarkably high coefficient of determination (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) at 0.99. Similarly, for the testing phase, the achieved values are RMSE=0.25 and R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> =0.98 in the prediction of Tmean. Ultimately, these metrics underscore the effectiveness of the proposed methodology in addressing missing data and enhancing the predictive capabilities of the passive house temperature model.