Integrating ACO and Constraint Propagation

Abstract

Ant Colony Optimisation algorithms perform competitively with other meta-heuristics for many types of optimisation problems, but unfortunately their performance does not always degrade gracefully when the problem contains hard constraints. Many industrially relevant problems, such as fleet routing, rostering and timetabling, are typically subject to hard constraints. A complementary technique for solving combinatorial optimisation problems is Constraint Programming (CP). CP techniques are specialized for solving hard constraints, but they may be inefficient as an optimisation method if the feasible space is very large. A hybrid approach combining both techniques therefore holds the hope to combine these complementary advantages. The paper explores how such an integration can be achieved and presents a hybrid search method CPACS derived by embedding CP into ACS. We have tested CPACS on job scheduling problems. Initial benchmark results are encouraging and suggest that CPACS has the biggest advantage over the individual methods for problems of medium tightness, where the constraints cause a highly fragmented but still very large search space.