An improved teaching-learning-based optimization for constrained evolutionary optimization

When extending a global optimization technique for constrained optimization, we must balance not only diversity and convergence but also constraints and objective function. Based on these two criteria, the famous teaching-learning-based optimization (TLBO) is improved for constrained optimization. To balance diversity and convergence, an efficient subpopulation based teacher phase is designed to enhance diversity, while a ranking-differential-vector-based learner phase is proposed to promote convergence. In addition, how to select the teacher in the teacher phase and how to rank two solutions in the learner phase have a significant impact on the tradeoff between constraints and objective function. To address this issue, a dynamic weighted sum is formulated. Furthermore, a simple yet effective restart strategy is proposed to settle complicated constraints. By adopting the ε constraint-handling technique as the constraint-handling technique, a constrained optimization evolutionary algorithm, i.e., improved TLBO (ITLBO), is proposed. Experiments on a broad range of benchmark test functions reveal that ITLBO shows better or at least competitive performance against other constrained TLBOs and some other constrained optimization evolutionary algorithms.