Evaluating ice-melting performance of highway deicing agents: A machine learning approach

Abstract

To quantify and analyze the ice-melting performance of deicing agents on highways, indoor experiments were conducted to collect data on melting area and time for four agents (Industrial Salt, Tablet Solid Eco-Friendly Deicing Agent, Spherical Solid Eco-Friendly Deicing Agent, and Non-Chlorine Plant-Based Liquid Eco-Friendly Deicing Agent). Comparative analysis and curve fitting showed how different quantities and types affect melting area. Considering factors like agent type, time, and quantity, a real-time prediction model was developed using multiple machine learning regression algorithms. Results indicated that melting area stabilizes time over and with increasing quantity, exhibiting quadratic curve characteristics. Industrial Salt had the shortest melting time. The Gradient Boosting Decision Tree (GBDT) model achieved the highest accuracy (R² = 0.979). Time was the most important factor in the prediction model, with agent type contributing approximately 30% and quantity 16.8%. This study introduces a novel machine learning framework for real-time prediction of melting areas, addressing the lack of quantitative models in existing research.