Data-driven Policy on Feasibility Determination for the Train Shunting Problem
Dutch Railways currently uses a local search heuristic to solve the Train Unit Shunting Problem to determine if a problem instance has a feasible solution. In this work, we use a Deep Graph Convolutional Neural Network model to predict the feasibility of solutions obtained during the run of the Local Search heuristic in order to accelerate the search process. Our approach improves on prediction accuracy and leads to computational gains.
Wan-Jui Lee is a AI researcher of Dutch Railways leading the R&D in machine learning and AI for service logistics. She holds a Ph.D. in Electrical Engineeringand was a post-doctor in machine learning and pattern recognition at TU Delft for several years. With her academic background, she also has a linking-pin role to strengthen the collaboration with universities.