Experiences with stochastic algorithms for a class of constrained global optimisation problems

The solution of a variety of classes of global optimisation problems is req uired in the implementation of a framework for sensitivity analysis in mult icriteria decision analysis. These problems have linear constraints, some o f which have a particular structure, and a variety of objective functions w hich may be smooth or non-smooth. The context in which they arise implies a need for a single, robust solution method. The literature contains few exp erimental results relevant to such a need. We report on our experience with the implementation of three stochastic algorithms for global optimisation: the multi-level single linkage algorithm the topographical algorithm and t he simulated annealing algorithm. Issues relating to their implementation a nd use to solve practical problems are discussed. Computational results sug gest that, for the class of problems considered simulated annealing perform s well.