A complete expected improvement criterion for Gaussian process assisted highly constrained expensive optimization

作者：Jiao, Ruwang; Zeng, Sanyou^*; Li, Changhe^*; Jiang, Yuhong; Jin, Yaochu

来源：Information Sciences, 2019, 471: 80-96.

DOI：10.1016/j.ins.2018.09.003

• Evolutionary computation
• Expensive optimization
• Expected improvement
• Constrained optimization
• Gaussian process

Expected improvement (El) is a popular infill criterion in Gaussian process assisted optimization of expensive problems for determining which candidate solution is to be assessed using the expensive evaluation method. An El criterion for constrained expensive optimization (constrained El) has also been suggested, which requires that feasible solutions exist in the candidate solutions. However, the constrained El criterion will fail to work in case there are no feasible solutions. To address the above issue, this paper proposes a new El criterion for highly constrained optimization that can work properly even when no feasible solution is available in the current population. The proposed constrained El criterion can not only exploit local feasible regions, but also explore infeasible yet promising regions, making it a complete constrained El criterion. The complete constrained El is theoretically validated and empirically verified. Simulation results demonstrate that the proposed complete constrained El is better than or comparable to five existing infill criteria.

• 出版日期2019-1
• 单位中国地质大学（北京）