An Experience Information Teaching–Learning-Based Optimization for Global Optimization

Teaching-learning-based optimization (TLBO) is an intelligent optimization algorithm with relatively fewer parameters that should be determined in updating equations. For solving complex optimization problems, the local optima often appear in the evolution. To decrease the possibility of this phenomenon, a novel TLBO variant (EI-TLBO) with experience information (EI) and differential mutation is presented. In the method, neighborhood information (the best individual NTeacher and the mean individual NMean) of each learner's neighbors is introduced to improve the exploration capability. The EI before the current iteration of each learner is introduced to make him or her accurately judge the learning behavior in future. In addition, instead of duplicate elimination to maintain the diversity of population at the end of each generation in the original TLBO, differential mutation is introduced to maintain the diversity of learners during the iterative learning process. The main contribution of this paper is to improve the convergence speed and accuracy by introducing neighborhood topology structure, EI, and differential mutation. The efficiency of the proposed algorithm is evaluated on 46 benchmark functions, among which 27 functions are selected from CEC2013. Its performance is compared with those of six other reported EAs. The results indicate that EI-TLBO algorithm can achieve superior performance.

I. Introduction

Fuzzy technology [1], [2], neural networks [3]–[6], support vector machines [7], etc. have been used to deal with complex nonlinear problems, and some successful results are derived. With the rapid development of technology and science, some real-world engineering problems can be converted as global optimization problems. To solve these problems, many nature-inspired optimization algorithms have attracted growing research interest from many research fields and been widely developed over the years. These population-based optimization algorithms are mainly divided into two classes: 1) evolutionary algorithms (EAs) and 2) swarm intelligence (SI). EAs [8]–[12] have originated from the natural evolution phenomena and principles, and SIs are inspired from the social character and behavior [13]–[16] of living things. These population-based optimization algorithms have also been successfully dealt with in some kinds of real-world optimization problems for decades.