Performance of Various Artificial Intelligence Models for Predicting Temperature in an Industrial Building—A Case Study

Abstract

This article presents a comparative analysis of the performance of various artificial intelligence models for predicting temperature in an industrial building. The main objective is to identify an optimal algorithm that enables efficient thermal management, which is essential for ensuring product quality, maintaining process safety, and optimizing energy consumption. Through an in-depth study, this research evaluates the accuracy, learning speed, and adaptability of the models in a dynamic industrial environment. Gradient Boosting algorithms, notably XGBoost and LightGBM, have demonstrated a capacity to give promising results, particularly for short- and mid-term predictions. The main results were obtained with XGBoost, achieving a mean absolute error of only (Formula presented.) (Formula presented.) for a 5 min prediction horizon—below the (Formula presented.) (Formula presented.) accuracy of the sensors used. The article also highlights the practical implications of these findings for Industry 4.0. By integrating high-performing predictive models into equipment control systems, it is possible to reduce energy costs, improve thermal comfort, and ensure efficient process stability. This research underscores the importance of an algorithmic approach tailored to both the thermal characteristics of the building and the computational capabilities of industrial infrastructure, with feature importance analysis confirming that external temperature and heating system states are the primary predictive factors, thereby helping to optimize industrial operations.