The paper illustrate the development and validation of two regression models i.e., poisson and artificial neural network employed to predict the significance of different climatic variables for an institutional building in composite climate. For prediction of outer wall temperature using five input parameters (ambient temperature, relative humidity, global radiation, wind speed and wind direction), two modeling approaches: poisson and artificial neural network were used. The results of both the model were validated through test data analysis in terms of three statistical measures namely, R-squared, root mean square error and mean absolute error. The performances of both the models are comparable in terms of train and test data. Comparative statistical analysis of both the model shows a better fit for the artificial neural network model as it was capable of adjusting themselves to the unpredicted changes in the input data. For the current study where weather parameters are continuously fluctuating poisson model was preferred as it generates the estimation coefficient values for the variables while neural model is not mathematically defined. The study demonstrates that neural model should be used as an alternative method for temperature prediction. The content and format of the paper has been widely used for designing and retrofitting of buildings by considering the most significant climatic variables for future work. Also, it helps the architects and engineers to find the relevant insulating material for the building envelope as per the requirement and subsequently improves its thermal behavior.
Read full abstract