陳相兵 陳晨 閔心暢
為解決混合(等式和不等式)約束的多峰優(yōu)化問題(MOPs),本文在粒子群算法框架下提出了粒子優(yōu)度比較準則和局部協(xié)同與共軛進退尋優(yōu)兩種迭代進化策略. 優(yōu)度比較準則在適應(yīng)度和約束違反度的雙重限制下指導(dǎo)粒子高效地執(zhí)行進化策略,局部協(xié)同策略可使粒子能通過局部抱團收斂到多個全局最優(yōu)解,而共軛進退尋優(yōu)策略則提升了尋優(yōu)的速度和精度. 基于優(yōu)度比較準則與兩種進化策略的有效結(jié)合,本文設(shè)計了一個協(xié)同共軛進退粒子群(CCARPSO)算法,以充分融合粒子群算法的全局搜索能力和共軛進退法的局部快速尋優(yōu)能力. 數(shù)值仿真表明, 該算法能有效解決復(fù)雜約束MOPs和非線性方程組的多根問題,在廣義Logistic分布的參數(shù)估計中有全局優(yōu)化能力和較高的計算精度.
多峰優(yōu)化; 優(yōu)度比較; 局部協(xié)同; 共軛方向; 進退法; 粒子群
O29A2023.011006
收稿日期: 2022-01-22
基金項目: 四川省科技計劃(2022JDRC0068, 2021JDRC0080); 四川省教育廳項目(18ZB0363); 中國民用航空飛行學(xué)院校級項目(J2021-058)
作者簡介: 陳相兵(1985-), 男, 安徽樅陽人, 博士, 副教授, 主要研究方向為應(yīng)用數(shù)學(xué). E-mail: chenxb85@sina.com
通訊作者: 陳晨.E-mail:chenchen_uni@foxmail.com
A cooperative conjugate advance-retreat particle swarm optimization algorithm for hybrid constrained multimodal optimization problems
CHEN Xiang-Bing1, CHEN Chen2,? MIN Xin-Chang3
(1.Division of Mathematics, Sichuan University Jinjiang College, Meishan 620860, China; 2. College of Science, Civil Aviation Flight University of China, Guanghan 618307, China;3. School of Mathematics, Sichuan University, Chengdu 610044, China)
This paper aims at the multimodal optimization problems (MOPs) with equality and inequality constraints. A new algorithm is proposed following the particle swarm optimization idea. This algorithm consists of a superiority comparison criterion and two iterative evolutionary strategies. The superiority comparison criterion guides the particles on how to evolute according to the constructed constraint violation degree and the fitness (i.e., the objective function value). The local cooperation strategy ensures that all particles can converge to multiple global optimal solutions through local clustering. The conjugate advance-retreat optimization strategy improves the speed and precision of optimization. Our algorithm, named cooperative conjugate advance-retreat particle swarm optimization (CCARPSO) algorithm, integrates the global searching ability of PSO and the local fast optimization capability of conjugate advance-retreat method. In numerical simulations, the algorithm effectively solves MOPs with complex constraints and nonlinear equations with multiple solutions, and has high global optimization ability and calculation accuracy in estimating parameters of the generalized Logistic distribution.
Multimodal optimization; Superiority comparison; Local cooperation; Conjugate direction; Advance-retreat method; Particle swarm
1 引 言
隨著大數(shù)據(jù)時代的到來,實時數(shù)據(jù)的數(shù)量和種類急劇增加,海量數(shù)據(jù)出現(xiàn)在多個領(lǐng)域,例如醫(yī)療診斷[1]、市場決策[2]、路徑規(guī)劃[3]和光伏陣列[4]等. 隨之,多變量、多約束的多峰值優(yōu)化問題[5](MOPs:Multimodal Optimization Problems)時常出現(xiàn)而且亟待解決.
傳統(tǒng)的優(yōu)化方法,例如牛頓法、共軛梯度法、單純形法以及分支定界法等,通常要求目標和約束函數(shù)可導(dǎo),且容易陷入局部極值. 智能進化算法選擇則利用群體智慧,能夠并行處理超大規(guī)模優(yōu)化問題. 智能進化算法包含差分進化(Differential Evolution, DE)[6]算法、粒子群優(yōu)化(Particle Swarm Optimization, PSO)算法和遺傳算法(Genetic Algorithm, GA)[7]等. 在MOPs的優(yōu)化算法中,智能進化算法是當(dāng)前的熱點算法之一,如基于DE的小生境方法[8-9]. 為了降低參數(shù)的影響,一些新的進化算子融入了小生境策略[10-14].基于DE的小生境方法已經(jīng)成功地應(yīng)用于無約束或帶簡單約束的MOPs. 然而,對于帶復(fù)雜約束(如非線性約束)的MOPs,相關(guān)的研究工作還極為少見,適用的智能進化算法有待研究.
PSO算法模擬了自然群體生命現(xiàn)象的自組織、自學(xué)習(xí)和自適應(yīng)性,依據(jù)適應(yīng)度和優(yōu)勝劣汰法則迭代地搜索解空間的最優(yōu)個體[15]. 它不僅不需要目標函數(shù)的梯度信息,而且具有操作簡單、可并行計算和模型參數(shù)少等優(yōu)點,已在眾多領(lǐng)域發(fā)揮了重要作用,如基數(shù)約束的投資組合優(yōu)化[16]和多元線性回歸參數(shù)估計[17]等.
針對帶混合(等式和不等式)約束的MOPs,本文設(shè)計了新的PSO算法,其中的粒子基于優(yōu)度比較準則在局部協(xié)同和共軛進退尋優(yōu)迭代兩種進化策略中選擇合適的進化策略,我們稱之為協(xié)同共軛進退PSO(Cooperative Conjugate Advance-Retreat PSO,CCARPSO)算法. 其中,優(yōu)度比較準則在約束違反度和適應(yīng)度(目標函數(shù)值)的雙重限制下指導(dǎo)粒子有效地迭代進化,協(xié)同策略則解決了一般PSO算法容易早熟和難以同時尋找多最優(yōu)解的問題,共軛進退尋優(yōu)策略則提升了尋優(yōu)的速度和精度. 該算法充分融合了PSO的全局搜索能力和共軛進退法的局部快速尋優(yōu)能力. 數(shù)值實驗表明,在求解帶混合約束的MOPs和多根的非線性方程組時本文提出的CCARPSO算法具有優(yōu)良性能.
6 結(jié) 論
本文構(gòu)建了帶混合約束的MPOs的CCARPSO算法. 該算法是一種基于優(yōu)度比較準則來選擇迭代策略的粒子群算法. 共軛進退尋優(yōu)策略保證粒子能朝著可行域方向快速進化,局部協(xié)同策略使所有粒子能通過局部抱團收斂到多個全局最優(yōu)解. 數(shù)值仿真實驗表明,CCARPSO算法有效地解決了非線性約束MOP和多根非線性方程組的求解問題. 另外,基于碳素纖維硬度的實際數(shù)據(jù),我們利用CCARPSO算法求解參數(shù)估計優(yōu)化問題,給出了穩(wěn)健、精確的碳素纖維硬度模型參數(shù)估計.
參考文獻:
[1] Tavard F, Simon A, Leclercq C, et al. Multimodal registration and data fusion for cardiac resynchronization therapy optimization [J]. IEEE T Med Imaging, 2014, 33: 1363.
[2] Zaman F, Elsayed S M, Ray T, et al. Evolutionary algorithms for finding Nash equilibria in electricity markets [J]. IEEE T Evolut Comput, 2018, 22: 536.
[3] Zhao Y, Ioannou P A, Dessouky M M. Dynamic multimodal freight routing using a co-simulation optimization approach [J]. IEEE T Intell Transp, 2019, 20: 2657.
[4] 徐儀圓, 許祺峰. 適用于商業(yè)航天的全局MPPT優(yōu)化算法[J]. 西北工業(yè)大學(xué)學(xué)報, 2020, 38: 133.
[5] Li X, Epitropakis M G, Deb K, et al. Seeking multiple solutions: an updated survey on niching methods and their applications [J]. IEEE T Evolut Comput, 2017, 21: 518.
[6] 李汶駿, 龍偉, 曾力. 基于差分進化和核主元分析的燃氣輪機故障檢測[J]. 四川大學(xué)學(xué)報: 自然科學(xué)版, 2021, 58: 022004.
[7] 鄧希, 胡曉兵, 江代渝, 等. 基于混合遺傳算法的柔性作業(yè)車間機器和AGV規(guī)劃[J]. 四川大學(xué)學(xué)報: 自然科學(xué)版, 2021, 58: 022003.
[8] Thomsen R. Multimodal optimization using crowding-based differential evolution [C]//Proceedings of the 2004 Congress on Evolutionary Computation. Piscataway: IEEE, 2004.
[9] Gao W, Yen G G, Liu S. A cluster-based differential evolution with self-adaptive strategy for multimodal optimization [J].IEEE T Cybernetics, 2014, 44: 1314.
[10] Zhao H, Zhan Z H, Lin Y, et al. Local binary pattern-based adaptive differential evolution for multimodal optimization problems [J]. IEEE T Cybernetics, 2020, 50: 3343.
[11] Wang Z J, Zhan Z H, Lin Y, et al. Automatic niching differential evolution with contour prediction approach for multimodal optimization problems [J]. IEEE T Evolut Comput, 2020, 24: 114.
[12] Chen Z G, Zhan Z H, Wang H, et al. Distributed individuals for multiple peaks: A novel differential evolution for multimodal optimization problems [J]. IEEE T Evolut Comput, 2020, 24: 708.
[13] 陳宗淦, 詹志輝. 面向多峰優(yōu)化問題的雙層協(xié)同差分進化算法[J]. 計算機學(xué)報, 2021, 44: 1806.
[14] 廖作文. 基于差分進化算法的非線性方程組多根聯(lián)解研究[D]. 武漢: 中國地質(zhì)大學(xué), 2019.
[15] Kennedy J, Eberhart R. Particle swarm optimization [C]//A Proceedings of the Fourth IEEE International Conference on Neural Networks. Piscataway: IEEE, 1995.
[16] 朱沙, 陳臣. 一種求解基數(shù)約束投資組合優(yōu)化的混合粒子群算法[J]. 統(tǒng)計與決策, 2016, 4: 64.
[17] 李翼, 張本慧, 郭宇燕. 改進粒子群算法優(yōu)化下的Lasso-Lssvm預(yù)測模型[J]. 統(tǒng)計與決策, 2021, 37: 45.
[18] Li X,Engelbrecht A, Epitropakis M G. Benchmark functions for CEC 2013 special session and competition on niching methods for multimodal function optimization [R]. Melbourne: RMIT University,? 2013.
[19] Balakrishnan N. Handbook of the logistic distribution [M].New York: Marcel Dekker, 1991.
[20] 韓雪, 程維虎. 三參數(shù)I型廣義Logistic分布參數(shù)的一類改進估計[J]. 數(shù)理統(tǒng)計與管理, 2016, 35: 445.
[21] Kao J H K. Computer methods for estimating Weibull parameters in reliability studies [J]. IRE T Re Qual C, 1958, 13: 15.
[22] Swain J, Venkatraman S, Wilson J. Least-squares estimation of distribution functions in Johnsons translation system [J]. J Stat Comput Sim, 1988, 29: 271.
[23] 陳海清, 曾婕, 胡國治. 三參數(shù)I型廣義Logistic分布參數(shù)的改進最小二乘估計[J]. 數(shù)理統(tǒng)計與管理, 2018, 37: 835.