Abstract
In order to automate rescheduling, the railway system must be properly modelled to perform evaluation and optimization. The authors' research group have proposed the Mimic Panel State Model (MPSM), which is a model that can couple visual intuitiveness with underlying mathematical foundation. MPSM was first considered as a mathematical expression of the information shown on the mimic panel. MPSM is expressed by Petri nets (PTN). The state (Mimic Panel state (MPS)) is the marking of PTN expressing the MPSM. It will change according to the movement of trains. The set of all possible MPS is called a state space. Using state queues as chromosomes, Genetic Algorithms (GA) can be applied to the optimization of rescheduling problems using the MPSM. However, in Simple GA (SGA) shown in many textbook explanations of GA, chromosomes are fixed-length binaries, and the possible genes are binary 0 and 1 only. Using state queues as chromosomes mean that they are variable-length and all MPS in the state space can be genes. Therefore, careful design of crossover and mutation operations is necessary for the application of GA to MPSM, as most of the next-generation chromosomes generated from the current generation chromosomes using carelessly defined operations may become "not valid", resulting in poor calculating performance. Simple experiment has been performed on a simplified railway line model with five stations, in which two trains make return travels between the terminals at both ends of the line. Calculations over a 100 generations have shown that the fitness (both average and best) values decrease as the GA process goes on, proving that MPSM can be utilised to optimization of train rescheduling problems.
