Teaching-learning-based optimization algorithm with dynamic neighborhood and crossover search mechanism for numerical optimization

This paper presents an improving teaching-learning-based optimization algorithm (called DRCMTLBO) combined with the dynamic ring neighborhood topology. Firstly, based on the individuals’ fitness distribution and clustered state within and beyond the ring neighborhood, two evaluations of relative neighborhood quality (RNQ) are developed to separately guide population evolution. On the one hand, a dynamic neighborhood strategy driven by fitness-based evaluation is used to adjust neighborhood radius, maintaining individual variability and neighborhood diversity. On the other hand, to utilize individuals’ information of the entire topology, a novel crossover search mechanism driven by Euclidean distance-based evaluation is used to expand the search space, determining whether individuals should enhance exploitation within the neighborhood or exploration beyond the neighborhood. Finally, the above strategies are embedded into the TLBO algorithm, assisted by improved search approaches that achieve a significant balance between exploitation and exploration. Numerical computation results on functions of CEC2014 and CEC2020 show that our proposed DRCMTLBO algorithm outperforms other ten typical algorithms significantly, and its computational performance can compete with several CEC winner algorithms.