Machine learning-based modeling and fogging prevention strategies for ice rink environments

Abstract

An ice rink designed as a specialized public building for ice sports, requires a precise indoor thermal environment to function optimally. Excess moisture in this environment leads to fogging issues. This study focuses on investigating an ice rink in Taiwan, employing machine learning-based modeling for fogging prevention. Utilizing logistic regression, we develop a fogging probability model, which serves as a pivotal advanced warning system for critical operational decision-making. Subsequently, we employ Deep Neural Networks (DNN) to predict the supply air temperature and humidity of two Air Handling Units (AHUs), extending this methodology to model the air conditions within the ice rink arena. The models achieve an impressive R2 of 0.94. Leveraging this predictive capacity, our study estimates fogging occurrences through a fogging line, exhibiting high classification accuracy of 96 %. Additionally, clustering analysis utilizing the k-means algorithm is conducted, serving as the initial condition for optimization. The optimization phase, employing a genetic algorithm, results in fogging prevention rates between 72 % and 85 % across clusters. This research indicates the efficacy of a machine learning-based approach in addressing fogging challenges, providing insight into effective preventive measures to maintain an optimal ice rink environment.