具有振荡搜索和自适应突变的教与学优化算法

Improved TLBO with Oscillation Search and Adaptive Mutation

摘要: 为了克服标准教与学优化(TLBO)算法在求解高维问题中表现出的后期收敛速度慢、容易早熟的问题，提出了一种具有振荡搜索和自适应突变的改进教与学优化算法(ITLBOA)。该算法首先在“教”算子中引入二阶振荡搜索策略，期望在早期抑制粒子向最优粒子的过快收敛，而在后期使其具有较强的勘探能力，提高算法的解精度。其次，引入自适应突变，随着迭代的进行逐渐提高粒子的突变概率，并随机选择部分维度变异，赋予算法摆脱局部最优约束的能力。在5个标准Benchmark函数上的测试，结果表明该算法具有较好的全局搜索能力和解精度。选择8个UCI数据集上进行特征选择问题求解，ITLBOA获得的有效特征数目比TLBO平均减少了1.35个，最优适应度值下降了15.91个百分点。一系列的实验表明，ITLBOA不仅适合求解较高维度的函数优化问题，同样也能够高效求解组合优化问题。

Abstract: To overcome the weakness of slow convergence and easy precocity of teaching and learning based optimization (TLBO) in solving higher dimensional problems, an improved teaching and learning optimization algorithm with oscillatory search and adaptive mutation (ITLBOA) was proposed. Firstly, the second-order oscillation search strategy is introduced into the “teach” operator, which is expected to restrain the rapid convergence of individuals to the optimal individuals in the early stage, and make it have strong exploration ability and improve the solution accuracy of the algorithm in the later stage. Secondly, adaptive mutation is introduced to gradually improve the mutation probability of individuals with the progress of iteration, and some dimensional mutation is randomly selected to give the algorithm the ability to escape from local optimal in the later stage. Testing on 8 standard benchmark functions showed that ITLBOA has better global search ability and higer accuracy than TLBO. It was applied to the experiment of feature selection on 11 UCI data sets. The results showed that the number of effective features obtained by ITLBOA was reduced by 1.35 compared with TLBO, and the optimal fitness value was reduced by 15.91%. A series of experiments showed that ITLBOA is not only suitable for solving higher dimensional function optimization problems, but also can efficiently solve combinatorial optimization problems.