Intelligent operation of heavy haul train with data imbalance: A machine learning method

Abstract

Compared with high speed trains and metro subways, heavy haul train operations are more challenging for their complex dynamic characteristics and complicated running environments. When running on a continuous and steep descent, the air brake operation strategy has become the most important issue for the safety and efficiency of heavy haul train transportation. Due to the difficulty in modeling the train’s dynamics and pneumatic brake system precisely, this paper addresses the intelligent driving problem for heavy haul train based on the fusion of expert knowledge and machine learning methodologies. By considering the characteristics of manual driving on steep descent, the pneumatic brake operation problem is formulated as a multi-class classification model. To overcome the negative influences of data imbalances, EasyEnsemble for the multi-class with a KNN based Denoising (EMKD) algorithm is introduced to determine the feasible Air Pressure Reduction (APR) and the exact time for exerting and releasing the air brake. This approach utilizes the EasyEnsemble.M algorithm to moderate the class imbalanced datasets and takes advantage of the AdaBoost.M1 algorithm to ensemble weak classifiers. Specifically, the K-nearest neighbor based Denoising (KD) algorithm is elaborated to remove the possible noise data from the minority dataset. Additionally, expert knowledge is obtained by abstracting the experience of sophisticated drivers and technical specifications, which are employed as operating constraints to regulate the output of the EMKD algorithm. The operational safety in terms of the in-train forces and punctuality of the proposed algorithm are validated by a number of experiments under the real running circumstances of Shuohuang heavy haul railway.