From quasi-solutions to solution: An evolutionary algorithm to solve CSP

Abstract

This paper describes an Evolutionary Algorithm that repairs to solve Constraint Satisfaction Problems. Knowledge about properties of the constraints network can permit to define a fitness function which is used to improve the stochastic search. A selection mechanism which exploits this fitness function has been defined. The algorithm has been tested by running experiments on randomly generated 3-colouring graphs, with different constraints networks. We have also designed a specialized operator "permutation", which permits to improve the performance of the classic crossover operator, reducing the generations number and a faster convergence to a global optimum, when the population is staying in a local optimum. The results suggest that the technique may be successfully applied to other CSP.