An improved neighborhood algorithm: parameter conditions and dynamic scaling

The Neighborhood Algorithm (NA) is a popular direct search inversion technique. For dispersion curve inversion, physical conditions between parameters Vs and Vp (linked by Poisson’s ratio) may limit the parameter space with complex boundaries. Other conditions may come from prior information about the geological structure. Irregular limits are not natively handled by classical search algorithms. In this paper, we extend the NA formulation to such parameter spaces. For problems affected by non-uniqueness, the ideal solution is made of the ensemble of all models that equally fits the data and prior information. Hence, a powerful exploration tool is required. Exploiting the properties of the Voronoi cells, we show that a dynamic scaling of the parameters during the convergence to the solutions drastically improves the exploration.