Handling Insensitivity in Multi-Physics Inverse Problems Using a Complex Evolutionary Strategy

In this paper we present a complex strategy for the solution of ill posed, in-verse problems formulated as multiobjective global optimization ones.The strategy is capable of identifying the shape of objective insensitivity regions around connected components of Pareto set.The goal is reached in two phases.In the first, global one, the connected components of the Pareto set are localized and separated in course of the multi-deme, hierarchic memetic strategy HMS.In the second, local phase, the random sample uniformly spread over each Pareto component and its close neighborhood is obtained in the specially profiled evolutionary process using multiwinner selection.Finally, each local sample forms a base for the local approximation of a dominance function.Insensitivity region surrounding each connected component of the Pareto set is estimated by a sufficiently low level set of this approximation.Capabilities of the whole procedure was verified using specially-designed two-criterion benchmarks.