DISCRETE OPTIMIZATION-BASED ON THE COMBINED USE OF REINFORCEMENT AND CONSTRAINT SATISFACTION SCHEMES

A new approach is presented for finding near-optimal solutions to disc rete optimisation problems that is based on the cooperation of two mod ules: an optimisation module and a constraint satisfaction module, The optimisation module must be able to search the problem state space th rough an iterative process of sampling and evaluating the generated sa mples. To evaluate a generated point, first a constraint satisfaction module is employed to map that point to another one satisfying the pro blem constraints, and then the cost of the new point is used as the ev aluation of the original one, The scheme that we have adopted for test ing the effectiveness of the method uses a reinforcement learning algo rithm in the optimisation module and a general deterministic constrain t satisfaction algorithm in the constraint satisfaction module. Experi ments using this scheme for the solution of two optimisation problems indicate that the proposed approach is very effective in providing fea sible solutions of acceptable quality.