An improved teaching–learning-based optimization algorithm for solving global optimization problem

Teaching–learning-based optimization (TLBO) is a recently proposed population-based algorithm that simulates the process of teaching and learning. Compared with other evolutionary algorithms, TLBO has fewer parameters that must be determined during the renewal process, and is very efficient for certain optimization problems. However, as a population-based algorithm, certain complex problems cause TLBO to exhibit local convergence phenomena. Therefore, to improve the global performance of TLBO, we have designed local learning and self-learning methods to enhance the search ability of TLBO. In the learner phase, every individual learns from both the teacher of the current generation and other individuals. Whether these individuals are neighbours or random individuals from the whole class is determined probabilistically. In the self-learning phase, individuals either renew their positions according to their own gradient information, or randomly exploit new positions according to a design based on the means and variances. To maintain local diversity, all individuals are rearranged after a set number of iterations. The proposed algorithm is tested on a number of functions, and its performance is compared with that of other well-known optimization algorithms. The results indicate that the improved TLBO attains good performance.