In Korea, the temperature falls below 0 °C in winter and rises above 0 °C in spring. This change in temperature between the two seasons results in the ground alternatively freezing and thawing, which leads to road surfaces being damaged. Predicting the ground temperature becomes very important in identifying and responding to potential infrastructure damage due to the ground freezing and thawing. A simulation was conducted through numerical analysis using the Crank–Nicholson differential method to predict the temperature of each layer of a road. Moreover, the data gathered from measuring the temperature at each layer of a road over a period of 42 days in “Evaluation of Validity for Anti-frost Layer and Development of its Construction Criteria,” organized by the Ministry of Land, Transport and Maritime Affairs (2012), were used for the simulation. The training for temperature prediction of the anti-frost layer was performed using deep learning machine learning techniques with 650 days of measurement data by changing the number of hidden layers and nodes. With two hidden layers, 40 nodes, and 100 nodes, the reliability of the training result was close to 1. The reliability of the predictive model, a by-product of the training, was approximately 0.87.
Temperature Prediction of Anti-frost Layer using Machine Learning Techniques
