面向高维复杂多模态问题的教与学优化求解算法

摘要

针对教与学优化算法(teaching-learning-based optimization,TLBO)在求解一些高维多模态复杂优化问题时,存在种群容易过早陷入局部搜索,导致丢失全局最优解的问题,提出一种改进的TLBO优化算法(MTLBO).该算法以更接近人类的学习方式,对标准TLBO中的"教"和"学"过程进行了改进,并引入了新的"自学"机制来加强学员的创新学习能力,从而有效提高了算法的全局探索能力.通过10个复杂的多模态优化问题测试表明,在求解复杂多模态问题方面,与五种具有优异性能的TLBO算法和三种经典的群智能计算方法(如SaDE、CLPSO、NGHS)相比,MTLBO算法具有全局搜索能力强、稳定性好等明显优势.%Aiming at the shortcomings for solving complex multimodal optimization problems,such as searching at local optimum and premature convergence,this paper presented a modified teaching-learning-based optimization (MTLBO) algorithm to enhance the capability of global exploration power.In MTLBO,it modified the strategies of "teaching" phase and "learning" phase,and it proposed a new "self-learning" strategy to balance the global exploration power and the local exploitation power,which could enhance the innovation ability of learners for improving the global exploration power of MTLBO.Finally,it employed ten complex benchmark functions to investigate the performance of MTLBO,the experimental results indicate that,compared with other five state-of-the-art TLBO variants and three other type of algorithms (SaDE,CLPSO,NGHS),the MTLBO algorithm has stronger global exploration power and more robustness on performance.