發(fā)布時間：2018-05-09 03:26

  本文選題：機(jī)電系統(tǒng)設(shè)計(jì) + 多目標(biāo)優(yōu)化�。� 參考：《浙江大學(xué)》2016年博士論文

【摘要】：隨著現(xiàn)代機(jī)電產(chǎn)品復(fù)雜程度的不斷提高,針對多域機(jī)電產(chǎn)品在系統(tǒng)概念形成階段的設(shè)計(jì)優(yōu)化逐漸成為關(guān)注的研究熱點(diǎn)。其主要困難是在產(chǎn)品系統(tǒng)的概念形成階段往往涉及到多個相互耦合的學(xué)科,且設(shè)計(jì)的目標(biāo)并不單一,而是以多個目標(biāo)互相沖突的形式存在。為得到滿意的解決方案,產(chǎn)品系統(tǒng)的概念設(shè)計(jì)必須反復(fù)迭代,為此實(shí)現(xiàn)設(shè)計(jì)與優(yōu)化間的無縫高效集成也至關(guān)重要。目前,盡管已經(jīng)有一些相關(guān)研究工作,但還存在諸多不足,如:(1)應(yīng)用于復(fù)雜機(jī)電系統(tǒng)概念設(shè)計(jì)的啟發(fā)式多目標(biāo)優(yōu)化算法效率還太低;(2)已有的多目標(biāo)優(yōu)化算法在處理帶約束問題時缺乏良好的解決策略;(3)復(fù)雜機(jī)電產(chǎn)品涉及到多個學(xué)科知識,其對應(yīng)的多學(xué)科優(yōu)化問題通常表現(xiàn)出高度的內(nèi)部耦合性,該特性大大增加了求解此類問題的計(jì)算代價,已有的一些解耦策略的求解性能不能滿足研究人員的需要;(4)已有的支持概念設(shè)計(jì)的平臺很多還是依據(jù)以往的經(jīng)驗(yàn)進(jìn)行決策,或者高度依賴外部的優(yōu)化軟件,并且存在交互困難和用戶操作復(fù)雜的特點(diǎn)。為此,本論文圍繞以上問題展開研究,主要工作包括:(1)基于幾何結(jié)構(gòu)的多目標(biāo)粒子群優(yōu)化算法。算法主要特點(diǎn)是通過利用當(dāng)前Pareto前沿的幾何結(jié)構(gòu)來對整體種群進(jìn)行牽引。先將當(dāng)前Pareto前沿看作是多維空間的一組散亂點(diǎn),進(jìn)而擬合構(gòu)造出幾何參數(shù)空間,接著計(jì)算參數(shù)空間的法線以精確定位一組對應(yīng)的牽引點(diǎn),最后基于牽引點(diǎn)將種群的非前沿點(diǎn)朝著更優(yōu)的方向進(jìn)行演化。(2)針對帶約束多目標(biāo)優(yōu)化問題的高效混合搜索模式算法。主要分為兩步:①.可行域的搜索:這里主要處理優(yōu)化模型中的約束條件,對約束條件進(jìn)行歸一化處理同時使用一種適配的差分算法;②.最優(yōu)解搜索:在第一步得到的可行解域的基礎(chǔ)上,對于每一個可行解通過其對應(yīng)的局部最優(yōu)精英集和全局最優(yōu)精英集在搜索空間進(jìn)行演化。(3)針對多學(xué)科優(yōu)化的序列化部分解耦方法。主要包含三個步驟:①.通過分析學(xué)科之間的敏感度將多學(xué)科聚合為若干個子系統(tǒng);②.對于每個子系統(tǒng)通過解耦操作保證每個子系統(tǒng)沒有耦合環(huán)存在,進(jìn)而子系統(tǒng)不用迭代反復(fù)求解;③.對于每個子系統(tǒng)進(jìn)行局部優(yōu)化處理,保證全局優(yōu)化器的規(guī)模尺度。(4)基于模式的機(jī)電系統(tǒng)設(shè)計(jì)與優(yōu)化集成方法。主要包括三個步驟:①.優(yōu)化問題的構(gòu)造。根據(jù)優(yōu)化問題擴(kuò)展版型提取優(yōu)化變量并且定義優(yōu)化目標(biāo)、約束條件和相關(guān)語義信息;②.基于語義相似的優(yōu)化方法自動選取。通過計(jì)算給定問題與模式庫之間的語義相似度,為給出的優(yōu)化問題自動選取最合適的優(yōu)化方法;③.基于優(yōu)化結(jié)果的設(shè)計(jì)更新。將優(yōu)化結(jié)果直接反饋給出設(shè)計(jì)人員,幫助設(shè)計(jì)人員作出決策同時更新模式庫。
[Abstract]:With the increasing complexity of modern electromechanical products, the design optimization of multi-domain electromechanical products in the system concept formation stage has gradually become the focus of attention. The main difficulty is that the concept of product system is often involved in a number of mutually coupled disciplines, and the design objectives are not single, but exist in the form of conflicting objectives. In order to obtain a satisfactory solution, the conceptual design of the product system must be iterated over and over again. Therefore, it is also important to realize the seamless and efficient integration between design and optimization. At present, although there have been some related research work, there are still many shortcomings. For example, the efficiency of heuristic multi-objective optimization algorithm applied to conceptual design of complex electromechanical systems is still too low. The existing multi-objective optimization algorithms lack a good solution strategy when dealing with constrained problems. Complex electromechanical products involve many disciplines. The corresponding multidisciplinary optimization problems usually exhibit a high degree of internal coupling, which greatly increases the computational cost of solving such problems. The performance of some existing decoupling strategies can not meet the needs of researchers. Many of the existing platforms supporting conceptual design are based on previous experience or rely heavily on external optimization software. And it has the characteristics of difficult interaction and complicated user operation. For this reason, this thesis focuses on the above problems. The main work includes: 1) Multi-objective particle swarm optimization algorithm based on geometric structure. The main feature of the algorithm is to use the geometric structure of the current Pareto frontier to pull the whole population. First, the current Pareto front is regarded as a group of scattered points in multidimensional space, then the geometric parameter space is constructed by fitting, and then the normal line of the parameter space is calculated to accurately locate a set of corresponding traction points. Finally, an efficient hybrid search pattern algorithm for constrained multi-objective optimization problems is proposed based on the traction point, which evolves the non-frontier points of the population towards a more optimal direction. Mainly divided into two steps: 1. The search of feasible region: this paper mainly deals with the constraints in the optimization model, and normalizes the constraints and uses a suitable difference algorithm. Optimal solution search: based on the feasible solution domain obtained in the first step, For each feasible solution, a serialization partial decoupling method for multidisciplinary optimization is proposed, which evolves in search space by its corresponding local optimal elite set and global optimal elite set. It consists of three steps: one. By analyzing the sensitivity between disciplines, the multi-discipline is aggregated into several subsystems. For each subsystem, decoupling operation ensures that there is no coupling loop in each subsystem, and then the sub-system is solved repeatedly without iteration. For each subsystem, local optimization is performed to ensure the scale of the global optimizer. (4) Mode-based electromechanical system design and optimization integration method. It consists of three steps: one. The structure of the optimization problem. The optimization variables are extracted according to the extended layout of the optimization problem and the optimization objectives, constraints and relevant semantic information are defined. The optimization method based on semantic similarity is selected automatically. By calculating the semantic similarity between the given problem and the schema library, the most suitable optimization method is automatically selected for the given optimization problem. Design update based on optimization results. The optimization results are fed back directly to the designers to help them make decisions and update the pattern library.
【學(xué)位授予單位】：浙江大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2016
【分類號】：TH-39;TP18

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 烏蘭木其,鄧家，

本文編號：1864368