Optimising train operation using simulation, fuzzy logic cruise control and evolutionary algorithms

Optimisation of driving strategies for safety, fatigue, energy and timeliness were investigated using a fully detailed train simulator over a train trip of 140 km. The train model used in the simulation consisted of 102 coal hopper wagons with locomotives positioned at lead and mid-train positions. The train model included detailed modelling of inter-wagon connections, locomotive performance curves, propulsion resistance, curving resistance, grade forces and train braking. Simulations were completed using track topography data selected from a coal haulage track system. The train simulator was controlled in two ways, via a time series of power and brake controls and via a fuzzy logic cruise control. In both cases an evolutionary algorithm was used to breed improved solutions, these being, power and brake control time series or fuzzy rule parameters. The fitness of solutions was evaluated using nine criteria. These criteria included: Cumulative Speed Violations, Maximum Speed, Maximum In-Train Tensile Force, Maximum In-Train Compressive Force Root Mean Square of In-Train Forces, Root Mean Square of Vehicle Accelerations, Energy Usage, Trip Time and Destination Achieved. Optimisations were attempted used various weighted single cost functions and the technique of exclusively breeding from amongst the best performers in each cost criteria. The paper presents and compares optimisation trajectories and sample simulation runs.