Statistical model for global energy consumption of an industrial site: Diagnosis and validation

Abstract

Industrial energy consumption presents a significant challenge for costs and environmental sustainability. This study developed and validated a macroscopic linear regression model to predict global industrial energy consumption. Using 48 months of data from a pharmaceutical facility in Morocco, the linear regression model incorporated operational (production units, worked hours) and environmental (temperature, RH) factors.Meticulous validation included standard statistical tests for model reliability. The model successfully passed checks for multicollinearity (VIF), autocorrelation (Durbin-Watson), and heteroscedasticity (Breusch-Pagan), ensuring reliable coefficient estimates. It explained 60% of energy consumption variance, confirming "Units Produced" and temperature as key drivers, with August shutdowns also significantly reducing consumption. However, the model's errors are not normally distributed, as shown by the Kolmogorov-Smirnov test.