本文關(guān)鍵詞：差分進(jìn)化算法和群集蜘蛛優(yōu)化算法的研究　出處：《安徽大學(xué)》2017年碩士論文　論文類型：學(xué)位論文

  更多相關(guān)文章： 差分進(jìn)化算法 群集蜘蛛優(yōu)化算法 協(xié)方差矩陣 自適應(yīng)權(quán)重系數(shù) 變異策略

【摘要】：差分進(jìn)化(DE)算法已經(jīng)成為解決連續(xù)型數(shù)值優(yōu)化問(wèn)題的經(jīng)典方法。本文的第一部分,把簡(jiǎn)化群優(yōu)化算法的交叉策略、協(xié)方差矩陣學(xué)習(xí)策略與傳統(tǒng)的差分進(jìn)化算法結(jié)合,提出一個(gè)新的DE算法的變種,我們把它稱作SCDE算法。正如我們所知,DE算法的變異策略在DE算法中占據(jù)了非常重要的位置,然而,傳統(tǒng)的DE算法的變異策略都是用相對(duì)位置來(lái)產(chǎn)生候選解,在本文中嘗試?yán)脗�(gè)體歷史最優(yōu)解的絕對(duì)位置來(lái)誘導(dǎo)變異產(chǎn)生候選解,這將大大的提高種群跳出局部最優(yōu)的能力。此外,我們將算法的變異和交叉操作放在由種群的協(xié)方差矩陣的所有特征向量組成的坐標(biāo)系中執(zhí)行,這將使算法的交叉和變異操作具有旋轉(zhuǎn)不變性。實(shí)驗(yàn)結(jié)果表明,本文提出的新的交叉和變異策略可以大大提高DE算法在CEC 2013中28個(gè)測(cè)試函數(shù)的結(jié)果。并且將SCDE算法應(yīng)用在解決組合優(yōu)化問(wèn)題之TSP問(wèn)題后也取得了較優(yōu)的結(jié)果。群集蜘蛛優(yōu)化算法是由Cuevas首次提出模擬群集蜘蛛相互協(xié)作的一種新型的群智能優(yōu)化算法。從數(shù)值模擬的結(jié)果顯示,相比較對(duì)比算法粒子群算法、人工蜂群算法,群集蜘蛛算法在全局尋優(yōu)能力方面的性能更強(qiáng)。然而,平衡算法的全局搜索能力和勘探能力是對(duì)一個(gè)群智能算法至關(guān)重要的一點(diǎn),它直接影響算法是否會(huì)過(guò)早收斂或精確度不足,這也是傳統(tǒng)的群集蜘蛛優(yōu)化算法所存在的問(wèn)題。受到粒子群算法和差分進(jìn)化算法啟發(fā),在本文的第二部分提出一種新的基于差分進(jìn)化變異策略和自適應(yīng)權(quán)重系數(shù)的群集蜘蛛優(yōu)化算法(表示為wDESSO)。在新算法中我們主要工作有以下幾點(diǎn):1.一個(gè)隨著種群迭代次數(shù)動(dòng)態(tài)變化的權(quán)重系數(shù)將被提出,用于自適應(yīng)群集優(yōu)化算法的搜索范圍;2.在算法結(jié)束了婚配操作之后,兩種差分進(jìn)化算法的變異策略將被應(yīng)用在新的算法中,用于增強(qiáng)算法的全局搜索能力和跳出局部最優(yōu)的能力。根據(jù)不同的變異策略,新提出的算法可以被分為兩類:wDESSO-Ⅰ算法和wDESSO-Ⅱ算法。隨后,幾組實(shí)驗(yàn)將用來(lái)檢驗(yàn)新的群集蜘蛛算法的性能,其中一個(gè)實(shí)驗(yàn)是將新型的群集蜘蛛優(yōu)化算法與傳統(tǒng)的群集蜘蛛優(yōu)化算法、粒子群算法、人工蜂群算法在15個(gè)標(biāo)準(zhǔn)測(cè)試集上做比較,并對(duì)結(jié)果做了威爾科克森符號(hào)秩檢驗(yàn);另外一組實(shí)驗(yàn)是與一些提高的優(yōu)化算法比較。結(jié)果表明,在解決復(fù)雜的數(shù)值問(wèn)題上,基于差分進(jìn)化變異策略的群集蜘蛛算法(wDESSO)的表現(xiàn)要明顯好于其他的對(duì)比算法。
[Abstract]:Differential evolution (DED) algorithm has become a classical method for solving continuous numerical optimization problems. In the first part of this paper, the crossover strategy of simplified group optimization algorithm is proposed. The covariance matrix learning strategy is combined with the traditional differential evolution algorithm, and a new DE algorithm is proposed, which we call the SCDE algorithm, as we know it. The mutation strategy of DE algorithm occupies a very important position in DE algorithm. However, the traditional mutation strategy of DE algorithm is to generate candidate solution by using relative position. In this paper, we try to use the absolute position of individual historical optimal solution to induce mutation to produce candidate solution, which will greatly improve the ability of population to jump out of local optimum. The mutation and crossover operations of the algorithm are carried out in a coordinate system composed of all the eigenvectors of the covariance matrix of the population, which will make the crossover and mutation operations of the algorithm rotation-invariant. The new crossover and mutation strategies proposed in this paper can greatly improve the DE algorithm in CEC. The results of 28 test functions in 2013. The SCDE algorithm is applied to solve the TSP problem of combinatorial optimization problem. The cluster spider optimization algorithm is led by Cuevas. A new swarm intelligence optimization algorithm is proposed to simulate the collaboration of cluster spiders. Compared with particle swarm optimization algorithm, artificial bee swarm algorithm, swarm spider algorithm has better performance in global optimization. The global search ability and exploration ability of the balanced algorithm are very important to a swarm intelligence algorithm, which directly affects whether the algorithm will converge prematurely or not. This is also the problem of the traditional swarm spider optimization algorithm, which is inspired by particle swarm optimization and differential evolution algorithm. In the second part of this paper, a new algorithm of swarm spider optimization based on differential evolution mutation strategy and adaptive weight coefficient is proposed. In the new algorithm, we mainly work on the following points: 1.A weight coefficient which varies dynamically with the number of iterations of the population will be proposed. Search range for adaptive cluster optimization algorithm; 2. After the conclusion of the matching operation, the mutation strategies of the two differential evolution algorithms will be applied to the new algorithm. Used to enhance the global search ability of the algorithm and the ability to jump out of the local optimum. According to different mutation strategies. The proposed algorithms can be divided into two categories: wDESSO- 鈪，

本文編號(hào)：1436158