Multi-parameter optimisation of train control

Optimal train control in respect to minimising energy usage is well understood and can be determined for a given train, route and time. However with the control of long freight and heavy haul trains there is also a need to consider longitudinal train dynamics. Large longitudinal train dynamics have the potential to produce wear, damage coupling components and cause derailments. Most current computational optimisation methods first produce an energy optimal driving strategy and then modify the strategy to produce acceptable longitudinal train dynamics or train handling. Good train handling is often produced by using rule based algorithms but this may not necessarily produce an optimal energy/train dynamics driving strategy. Other manual techniques to determine driving strategies are based around expert drivers, train simulation and train tests. Manual techniques can provide an acceptable driving strategy but it might not be optimal and the whole process can be time consuming. Investigation is being carried out in the use of computer searching algorithms and longitudinal train simulation software to perform multi-parameter train control optimisation. Current work with Genetic Algorithms has been successful in optimising for energy and time and this method is being expanded to include the optimisation of the longitudinal dynamics. Multi-parameter optimisation has the advantage in that individual railway operator values can be used in the optimisation. This allows cost/benefit comparisons of different ways to increase capacity, i.e. either by running faster services or by increasing the number of tracks. The determination of optimum driving strategies is a worldwide issue with railway operators around the world increasing the length of trains, developing new railway routes and increasing train speeds. This research also has possible future use in the development of locomotive cabin driver aids. The paper presents the multi-parameter optimisation method developed and some initial results.