A surrogate-model-assisted evolutionary algorithm for computationally expensive design optimization problems with inequality constraints

The surrogate model-aware evolutionary search (SMAS) framework is a newly emerged model management method for surrogate-model-assisted evolutionary algorithms (SAEAs), which shows clear advantages on necessary number of exact evaluations. However, SMAS aims to solve unconstrained or bound constrained computationally expensive optimization problems. In this chapter, an SMAS-based efficient constrained optimization method is presented. Its major components include: (1) an SMAS-based SAEA framework for handling inequality constraints, (2) a ranking and diversity maintenance method for addressing complicated constraints, and (3) an adaptive surrogate model updating (ASU) method to address many constraints, which considerably reduces the computational overhead of surrogate modeling. Empirical studies on complex benchmark problems and a real-world mm-wave integrated circuit design optimization problem are reported in this chapter. The results show that to obtain comparable results, the presented method only needs 1-10% of the exact function evaluations typically used by the standard evolutionary algorithms with popular constraint handling techniques.