基于余弦相似度反向策略的自然计算方法

现有的基于反向策略的优化算法大多根据初始种群适应度值大小进行反向择优，没有充分考虑迭代过程中的反向且存在收敛速度慢的问题。针对此问题，提出一种基于余弦相似度反向策略的快速收敛自然计算方法，通过计算每个粒子与区域中心粒子的余弦相似度，将粒子划分为相似子群与非相似子群，对非相似子群按照相似程度进行加权反向，进而加快收敛速度，同时引入柯西扰动提高种群多样性。将该策略应用到三种不同的自然计算方法中，对收敛性进行分析，并采用12个经典测试函数验证其性能，对实验数据进行非参数检验。分析结果表明，应用余弦相似度反向策略的方法在大多数测试函数上表现优异，说明提出的方法具有很好的普适性和有效性。

Natural Computation Method Based on Cosine Similarity Opposition Strategy

Most of the existing optimization algorithms based on the opposite learning strategy are under the initial population fitness value. The problem of opposition and slow convergence is not fully considered. Thus, a fast convergent natural computation method based on a cosine similarity opposition strategy is proposed to solve the aforementioned problem. The particles are divided into similar and nonsimilar subgroups by calculating the cosine similarity between each particle and the particle in the regional center. The nonsimilar subgroups are weighted opposition according to the similarity degree, thus the convergence speed is accelerated. The Cauchy disturbance is then introduced to improve population diversity. The strategy is applied to three different natural computing methods, and 12 classical test functions are used to analyze the convergence and verify its performance. Finally, nonparametric tests are performed on the experimental data. Experiments show that the method performs well in most test functions and has good universality and effectiveness.