Derivative Free Surrogate Optimization for Mixed-Integer Nonlinear Black Box Problems in Engineering

Optimization problems based on black boxes arise in engineering applications every day. Such black boxes typically represent the simulated or experimentally obtained behavior of systems for which almost no internal, structural or analytical knowledge can be provided on a relevant level for the optimization's objective. These resulting non-relaxable mixed-integer nonlinear black box-based optimization problems cannot be carried out efficiently by today's optimization methods. This work provides a new general applicable derivative free optimization approach. The performance of this method will be demonstrated for several benchmark and real world problems from electrical engineering, environmental sciences, and robotics. It will be shown that huge improvements of the optimization's objectives can be achieved for all applications, simply by applying a reasonable number of black box evaluations.