Abstract: To improve the low precision and poor stability of Teaching-Learning-Based Optimization（TLBO） algorithm, a Multi-Classes Interaction Teaching-Learning-Based Optimization（MCITLBO） is proposed. On the basis of TLBO, a new clustering partition based on Euclidean distance is firstly introduced to facilitate the population effectively search in the feasible neighbor region which can strengthen the local search ability of the algorithm and strive a balance between exploration and exploitation. Two information sharing scheme is introduced to improve the learning styles of students so as to increase information exchange in the evolution and avoid the premature convergence and stagnation behavior of subgroups. Numerical experiments on six unconstrained functions, four constrained functions and one engineering design problem called the tension string design problem show that MCITLBO outperforms the other algorithms in precision and stability.

Key words: Teaching-Learning-Based Optimization（TLBO） algorithm, optimization precision, clustering partition method, local search ability

摘要： 针对教学优化算法（Teaching-Learning-Based Optimization，TLBO）寻优精度低、稳定性差的问题，提出多班级交互式教学优化算法（Multi-Classes Interaction TLBO，MCITLBO）。通过引入基于欧氏距离的新型聚类划分方法，实现多班级教学，加强优秀个体周围邻域的搜索，保证算法具有较好的平衡和局部搜索能力，通过引入两种新的学习方式，实现学习方式多样化，加强种群信息交互、避免子群“滞后”或“早熟”。对6个无约束、4个约束函数和优化拉压弹簧设计问题的数值实验表明，MCITLBO相比其他算法在寻优精度和稳定性上更具优势。

关键词: 教学优化算法, 寻优精度, 聚类划分方法, 局部搜索能力