Multi-Objective Individualized-Instruction Teaching-Learning-Based Optimization Algorithm

Abstract

Traditional multi-objective evolutionary algorithms (MOEAs) adopt selection and reproduction operators to find approximate solutions for multi-objective optimization problems (MOPs). The Pareto-dominance-based method is an important branch of MOEA research which exploits dominance relations information. To use dominance relations information more efficiently, this paper proposes an individualized instruction mechanism combined with the non-dominated sorting concept and the teaching-learning process of teaching-learning-based optimization (TLBO). This algorithm, with its individualized instruction mechanism (INM-TLBO), places greater emphasis on the guiding role of the non-dominated solution. INM-TLBO designates specific teachers or interactive objects to help learners improve in the individualized teaching-learning process and adopts an external archive to preserve the best solution found. In addition, the INM-TLBO needs only generic control parameters as input, such as population size, an epsilon value for the external archive, and a stop criterion (maximal generation or function evaluation). The performance of INM-TLBO was evaluated on three test problem sets, including twelve extensively used unconstrained test problems, six truly disconnected test problems, and ten complex continuous unconstrained optimization test problems originally proposed for the Congress on Evolutionary Computation 2009 (CEC 2009) competition. The numerical results are compared with those of other state-of-the-art algorithms and show that INM-TLBO has good convergence and high robustness on these test problems.