This overview will introduce simulation-based optimization problems, challenges to their efficient solution, and algorithms best suited to various formulations and problem haracteristics. In simulation-based optimization the mapping from decision variables to objectives and constraints is at least partially implicit, requiring execution of a computational model. Algorithms typically treat this model as a “black box,” iteratively setting parameter values, running the simulation, and adapting based on objective and constraint information returned. While this approach is powerfully flexible, it is often computationally costly due to the simulation’s run time and/or its characteristics, e.g., nonlinearity, multimodality, or hidden constraints/failed evaluations.
We will tour solution techniques for simulation-based optimization and parameter estimation problems. The survey will include gradient-based local algorithms, with special considerations for least-squares calibration problems, derivative-free local optimization, and global/heuristic techniques. Potential advanced topics include surrogate-based optimization, optimization under uncertainty, and simulation-based MINLP. Challenges will be illustrated with examples from science and engineering design and calibration.