Constraint-handling for Optimization with Support Vector Surrogate Models - A Novel Decoder Approach

A new application for support vector machines is their use for meta-modeling feasible regions in constrained optimization problems. We here describe a solution for the still unsolved problem of a standardized integration of such models into (evolutionary) optimization algorithms with the help of a new decoder based approach. This goal is achieved by constructing a mapping function that maps the whole unconstrained domain of a given problem to the region of feasible solutions with the help of the the support vector model. The applicability to real world problems is demonstrated using the load balancing problem from the smart grid domain.