Research on wheel–rail equivalent conicity prediction method based on random forest algorithm

Abstract

The equivalent conicity, a core parameter for assessing wheel–rail interaction, plays a pivotal role in determining the stability and vibration characteristics of an EMU (Electric Multiple Unit). Presently, the methodologies employed for precise parameter prediction are not optimal. This research introduces a prediction model utilizing the random forest algorithm, which combines both historical and real-time EMU equivalent conicity data to forecast the equivalent conicity for a specified future period. By employing the hierarchical clustering algorithm, historical data is classified and integrated, constructing a subset of wheel–rail equivalent conicity data upon which the random forest prediction model is established. The efficacy of this model is underscored by its high accuracy, with 85% of residuals falling within the ±0.01 interval and a maximum mean squared error of approximately 0.0012. This model offers robust technical support for enhancing the precision and predictive capabilities of railway departments in wheel and rail maintenance, facilitating the transition from reactive to predictive maintenance in wheel and rail operations.