Applying stochastic algorithms to a locomotive scheduling problem

This paper addresses a problem common to all railway networks. Given a fixed train timetable and locomotives (or other forms of traction) of various types, each train must be allocated a locomotive. This paper examines the use of stochastic algorithms for such a problem. Two types of algorithm are used-a simple ‘local improvement’ method, performed successively from randomly chosen starting points, and a ‘simulated annealing’ approach. Both are found to give considerably better results than a deterministic method in current use, and the annealing approach is probably the better stochastic method.