A Machine Learning-Based Prediction Model for High-Speed Railway Track Geometry

This work is devoted to developing a prediction scheme of the high-speed railway (HSR) track geometry state using machine learning technology to improve the efficiency of track maintenance. Many HSR orbit geometric data are systematically collected, and an efficient prediction framework is successfully established by using the maximum likelihood estimation method. Firstly, the research problems are clarified on the technical route, and a perfect data acquisition system is established to ensure the representativeness and reliability of data samples. Then, with the help of systematic feature engineering, the original dataset is deeply mined and preprocessed, and the most predictive feature subset is selected. In the modeling stage, many machine learning algorithms are compared horizontally, and the random forest (RF) algorithm is finally determined and optimized as the core prediction model. After rigorous experimental verification and performance evaluation, the constructed prediction system shows excellent performance in multiple evaluation dimensions, in which the accuracy rate is improved to 95%, and it also shows excellent generalization performance and robustness. This research result confirms the value of machine learning in the HSR track geometry state prediction field and provides reliable technical support for track maintenance decisions.