A connectionist method to solve job shop problems

We propose a novel framework to solve job shop scheduling problems based on connectionist ideas of distributed information processing. In our approach each operation of a given job shop problem is considered to be a simple agent looking for a position in time such that all its time and resource constraints are satisfied. Each agent considers the current time position of its constraint neighbors to gradually change its own position to reach this goal. All agents together form a recurrent dynamical system which either self-organizes after some iterations to a feasible schedule or fails to do so depending on the constraints of the problem. By gradually increasing the constraints through decreasing the allowable overall processing time for a valid schedule, better and better solutions are found up to the point where no further improvements can be made.