基于分层多子群的教与学优化算法

Improved teaching-learning-based optimization algorithm based on hierarchical subpopulation

文章针对教与学优化(teaching-learning-based optimization,TLBO)算法在求解高维函数优化问题时易陷入局部最优与"早熟"现象、迭代后期收敛速度慢、求解精度低的缺点,提出了一种基于分层多子群的教与学优化算法(hierarchical subpopulation TLBO,HSTLBO),对平均学生水平进行重新定义,并根据适应度值对教学因子动态取值;通过预设的一个等级层次结构,将种群划分为若干个子群,构成层次结构的底层;底层子群粒子相对独立进化,保证种群多样性,每个子群的最优粒子则构成层次结构的上一层,提升算法的全局收敛能力,子群自下而上更新。通过10个Benchmark函数将提出的算法与其他算法进行对比实验,结果表明,HSTLBO在求解精度和收敛速度方面均优于其他算法。

To overcome the weakness of teaching-learning-based optimization(TLBO) in solving higher dimension function optimization problems including premature, low solution precision and slow convergence speed, an improved hierarchical subpopulation TLBO(HSTLBO) is proposed. The average level of students is redefined and the learning factor is adopted dynamically according to the fitness value of the particle. Through a preset level hierarchy, the population is divided into several subpopulations constituting the bottom of the hierarchy, the underlying subpopulation particle evolves relatively independently to keep the population diversity, and the optimal particle of each subpopulation constitutes the upper level of the hierarchy to enhance the ability of global convergence. The subpopulations are updated from bottom to top. The HSTLBO algorithm and the other four types of TLBO algorithms are compared and tested through ten Benchmark functions, and the results show that the HSTLBO algorithm is significantly better with regard to the solution precision and convergence speed.