KERNEL: Enabler to build smart surrogates for online optimization and knowledge discovery

Abstract

KERNEL – A novel parameter-free surrogate building algorithm using Adaptive Neuro Fuzzy Inference System (ANFIS) is presented to provide an intelligent and robust technology to optimally estimate the configuration of ANFIS along with Sobol-based fast sample size determination (SSD) methodology. The proposed algorithm is capable of fine-tuning the existing knowledge base about the physics of the process in terms of human experience. It also enables knowledge discovery through a multi-objective optimization problem (MOOP) solved by non-dominated sorting genetic algorithm, NSGA-II, thus presenting machine-invented physics of the process. Experimentally validated polymerization reaction network model is considered and ANFIS surrogates are built using KERNEL. Surrogate-based optimization was found to be nine times faster than conventional optimization using the time expensive model, thus enabling its online implementation. Comparison of ANFIS with Kriging is also included.