Combining User Knowledge and Online Innovization for Faster Solution to Multi-objective Design Optimization Problems.

Real-world optimization problems often come with additional user knowledge that, when accommodated within an algorithm, may produce a faster convergence to acceptable solutions. Modern population-based optimization algorithms, aided by recent advances in AI and machine learning, can also learn and utilize patterns of variables from past iterations to improve convergence in subsequent iterations – an approach termed innovization. In this paper, we discuss ways to combine user-supplied heuristics and machine-learnable patterns and rule sets in developing efficient multi-objective optimization algorithms. Two practical large-scale design problems are chosen to demonstrate the usefulness of integrating human-machine information within a multi-objective optimization in finding similar quality solutions as that obtained by the original algorithm with less computational time.