Heterogeneous teaching–learning based optimization with local search for the covering delivering problem in last mile delivery

Abstract

The generalized covering delivery problem (CDP) is a recently introduced variant of traveling salesman problem, which occurs mainly in the last mile delivery network of online shopping. To address CDP, on the basis of bi-level programming model, a new heterogeneous teaching–learning based optimization with local search is proposed in this paper. In this algorithm, to balance exploration and exploitation, the population is divided into two subpopulations: teaching subpopulation and learning subpopulation. The teaching subpopulation concentrates on local search near the global best solution by learning only from the global best solution, indicating strong exploitation. The learning subpopulation is responsible for global search by learning from each other using the comprehensive learning strategy, which can maintain diversity and strong exploration even if the teaching subpopulation converges prematurely. In addition, local search is also applied on the global best solution to further improve solutions along with the self-adaptive parameter tuning mechanism based on previous experience and evolution stage, which can be helpful to further balance exploration and exploitation. The experimental results on some benchmarks show the proposed algorithm significantly outperforms other state-of-the-art techniques in solving CDP