A comparative study on regression model and artificial neural network for the prediction of wall temperature in a building

Abstract

This paper demonstrates the validation and prediction of the wall temperature of a building exposed to a composite climate. Two artificial intelligence models, such as multiple linear regression and artificial neural networks, have been used to predict. The wall temperature has been predicted mainly based on the parameters like ambient temperature, wind speed and relative humidity in all four directions of the buildings. Three statistical analyses were used to validate the model's outcome: R-Squared, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). In terms of train and control data, the two models yield comparable findings. The artificial neural network model has more adaptability since it was able to adjust to unexpected changes in the input data, according to a comparison of the applied mathematics of each model. The regression model was used for this investigation because it gives constant estimate values for the factors, though the neural model isn't numerically characterized. The review infers that the neural model must be utilized as an additional way to predict the wall temperature. This study can plan and adapt buildings for future work, considering the most critical climatic conditions. It also assists architects and engineers in determining the appropriate insulation for the building envelope, which improves its thermal performance.