Enhancing Constraint Propagation in ACO-based Schedulers for Solving the Job Shop Scheduling Problem

Abstract Increasing number of variants lead to growing complexity in planning processes in production. Not only is the initial planning a tremendous task if there is a huge variety of products but also reacting to changes becomes more frequent and more demanding. Many algorithms being able to solve the static problem need to perform a full recalculation if there is disturbance in production which makes them too time consuming for instant reactions to changes in production. Ant Colony Optimization (ACO) has proven its potentials in solving the theoretical Job Shop Scheduling Problem offering the advantage of not needing an entire recalculation in the case of changes. But when using the algorithm for calculation in real time scenarios with returning data from production plants several restrictions have to be fulfilled. The reaction to those restriction is currently not sufficiently provided by implementations of the ACO which prevents the use in practical applications. These restrictions are modelled as constraints that can for example involve the reaction to disturbances like failures or manual changes. But also considering transportation times or providing the possibility to realize batch processes is discussed. There are different possibilities to realize the reactions to restrictions in ACO, but in this paper they are modelled as constraints affecting the ACO during optimization. The constraint propagation is implemented by restricting the selection of succeeding edges, an approach that only has little impact on computational performance. In this paper the concept of constraining the Ant Colony Optimization in Job Shop Scheduling is being introduced and explained. Subsequently the demand for additional constraints is presented and enhancements to the existing approach are defined and commented theoretically.