解决高维优化和特征选择问题的多策略改进麻雀搜索算法

Multi-strategy Improved Sparrow Search Algorithm for Solving HighDimensional Optimization and Feature Selection Problems

为解决基本麻雀搜索算法(sparrow search algorithm,SSA)在求解高维复杂优化问题时存在收敛速度慢、容易陷入局部最优解及后期种群多样性变弱等问题,提出了一种基于海鸥优化算法算子和鲸鱼优化算法算子的改进麻雀搜索算法(improved sparrow search algorithm based on seagull optimization algorithm operator and whale optimization algorithm operator,SWSSA)。首先,该算法设计了自适应种群比例策略以增强种群在迭代过程中的多样性;其次,在局部搜索阶段融入鲸鱼优化算法气泡网捕食策略,增强麻雀搜索算法的局部搜索能力、加快收敛速度;然后,在追随者位置引入改进的海鸥优化算法算子降低算法陷入局部最优的概率。最后,选取了12个高维基准测试函数和16个UCI网站上的高维数据集进行仿真实验,将SWSSA与基本SSA、SSA变体版本、黄金正弦算法(golden sine algorithm,GSA)、蝴蝶算法(butterfly optimization algorithm,BOA)、黏菌算法(slime mold algorithm,SMA)、海鸥算法(seagull optimization algorithm,SOA),以及其他学者改进的算法进行比较。结果表明,本文提出的算法在12个测试函数上的收敛精度取得最优的比例达到了100%,在约95%的测试函数上收敛速度最快,在16个数据集中有9个数据集分类准确率最高和6个最佳特征子集数量最少。可见所提算法在处理高维函数优化和数据集特征选择问题上具有一定的优势。

In order to solve the problems of slow convergence speed,easy falling into local optima,and weakened population diversity in solving high-dimensional complex optimization problems in the sparrow search algorithm(SSA),an improved sparrow search algorithm based on seagull optimization algorithm operator and whale optimization algorithm operator(SWSSA)was proposed.Firstly,an adaptive population proportion strategy was designed to enhance the diversity of the population during the iteration process.Secondly,incorporating the whale optimization algorithm bubble net predation strategy in the local search stage,the local search ability of the sparrow search algorithm and accelerates convergence speed was enchanced.Then,an improved seagull optimization algorithm operator was introduced at the follower position to reduce the probability of the algorithm falling into local optima.Finally,12 high-dimensional benchmark test functions and 16 high-dimensional datasets from UCI websites were selected for simulation experiments to compare SWSSA with basic SSA,SSA variant versions,golden sine algorithm(GSA),butterfly optimization algorithm(BOA),slime mold algorithm(SMA),seagull optimization algorithm(SOA),and other improved algorithms proposed by scholars..The results show that the algorithm proposed achieves an optimal convergence accuracy of 100%on 12 test functions,with the fastest convergence speed on about 95%of the test functions,the highest classification accuracy on 9 out of 16 datasets,and the lowest number of 6 optimal feature subsets.It can be seen that the proposed algorithm has certain advantages in handling high-dimensional function optimization and dataset feature selection problems.