局部维度改进的教与学优化算法

针对原始教与学优化算法局部搜索能力不强和进化后期容易陷入局部最优的问题，提出基于局部维度改进和自学习扰动的教与学优化算法.将局部维度改进融入教和学2个阶段，将个体的高质量维度变量保留到下一代，不断改善低质量维度变量，提高算法的细粒度搜索能力.提出一种混合全局维度改进和局部维度改进的个体更新方式，通过2种改进权重的逐代变化实现算法早期全局搜索和后期局部探测的平衡.在新算法中增加基于个体最优位置和搜索边界信息的自学习阶段，使种群在进化后期仍能向最优解方向搜索，避免算法过早陷入局部最优.基于标准测试函数的仿真结果表明，相比于原始的教与学优化算法和当前其他优秀的改进版本，局部维度改进的教与学优化算法的收敛精度平均提高了102~105倍，收敛速度平均提高了2~3倍.
Abstract: A local-dimension-improved teaching-learning-based optimization (LDimTLBO) algorithm based on local dimension improvement and self-learning disturbance was proposed, aiming at the problem of weak local search ability and easy to fall into local optimum during the anaphase of evolution in the original teaching-learning-based optimization (TLBO) algorithm. The local dimension improvement strategy was integrated into the teaching and learning phases, which passed the high-quality dimension variables down to the next generation, improved the low-quality ones, and enhanced the fine-grained search capability of the proposed algorithm. A new individual update mode combining global and local dimension improvements was designed. Through the generational change of these two improvements' weights, the global exploration at early stage and local exploitation at late stage was balanced in the hybrid update mode. A self-learning phase based on individual so-far-best position and search boundary information was also added into the search process, which made the population search towards the optimum even in the later stage of the evolution, thus, the algorithm escaped from getting into the local optimum in the early stage. The simulation results based on testing on benchmark functions demonstrated that in contrast to results of TLBO and other improved variants, the convergence accuracy of LDimTLBO was 102 to 105 times higher, and the convergence speed was 2 to 3 times faster.