Modelling Combinatorial Problems for CLP(FD+R)

The paper presents results of modelling and solving a large industrial combinatorial problem with Constraint Logic Programming (CLP) in cooperation with Linear Programming (LP). The problem at hand is scheduling the maintenance of the power units in the power plants of Flanders. Using finite domain solvers ROPE [VD94] and SICStus [Swe97] we were never able to derive an optimal solution. Most of constraints used in the model are linear inequations over integer variables. A number of experiments have been done at IC-Parc by running the Finite Domain solver in ECLiPSe [Eur98] using CPLEX to work on the linear constraints. Unfortunately, the solver was inefficient for the given problem. We have modified the existing model to a model with only linear constraints over boolean variables. Most of these constraints are uni-modular constraints. The same solver has derived significantly better results on the modified model. We describe both models and compare the empirical results.