韓旭 姜潮 陳國(guó)棟 龍湘云

摘 要：提出基于自適應(yīng)徑向基函數(shù)的多目標(biāo)優(yōu)化方法。該方法通過(guò)遺傳拉丁超立方實(shí)驗(yàn)設(shè)計(jì)、徑向基函數(shù)和隔代映射遺傳算法等技術(shù)，系統(tǒng)地評(píng)價(jià)代理模型。采用改進(jìn)的貪婪算法挑選最后迭代步中的測(cè)試點(diǎn)到最終樣本空間，獲得整個(gè)設(shè)計(jì)域上的自適應(yīng)徑向基函數(shù)模型。該方法被應(yīng)用于車(chē)身薄壁構(gòu)件耐撞性多目標(biāo)優(yōu)化設(shè)計(jì)中，快速地找到了多組設(shè)計(jì)方案，較好地平衡了薄壁構(gòu)碰撞過(guò)程中的吸能量和碰撞力。提出基于智能布點(diǎn)技術(shù)的微型多目標(biāo)遺傳算法。該算法采用加強(qiáng)徑向基函數(shù)構(gòu)建全局代理模型，再運(yùn)用高效的微型多目標(biāo)遺傳算法進(jìn)行近似優(yōu)化。并根據(jù)優(yōu)化結(jié)果信息進(jìn)行智能布點(diǎn)，反饋到設(shè)計(jì)空間進(jìn)而不斷地更新代理模型，使實(shí)驗(yàn)設(shè)計(jì)過(guò)程和近似優(yōu)化過(guò)程形成閉環(huán)的過(guò)程，提高了優(yōu)化效率。該方法被應(yīng)用于某重型商用車(chē)駕駛室動(dòng)態(tài)特性?xún)?yōu)化中，獲得大量支配優(yōu)化前的設(shè)計(jì)方案使駕駛室動(dòng)態(tài)特性更好并且質(zhì)量更輕。提出基于信賴(lài)域模型管理的優(yōu)化方法。該方法將在整個(gè)設(shè)計(jì)空間上的復(fù)雜優(yōu)化問(wèn)題，轉(zhuǎn)化為一系列信賴(lài)域上的近似多目標(biāo)優(yōu)化問(wèn)題。通過(guò)每個(gè)信賴(lài)域上的優(yōu)化結(jié)果，確定信賴(lài)度和下代域的中心、半徑。進(jìn)而不斷地縮放、平移信賴(lài)域，來(lái)保證獲得與真實(shí)模型一致的非支配解。該方法被應(yīng)用于某車(chē)門(mén)結(jié)構(gòu)優(yōu)化實(shí)際中，通過(guò)匹配關(guān)鍵部件的厚度，很好地平衡了車(chē)門(mén)的各項(xiàng)動(dòng)靜態(tài)特性指標(biāo)。結(jié)合信賴(lài)域和智能布點(diǎn)技術(shù)，用來(lái)處理信賴(lài)域模型管理需要多次重采樣導(dǎo)致效率低下的問(wèn)題。通過(guò)樣本遺傳策略，遺傳落在下代信賴(lài)域空間上的樣本，減少實(shí)驗(yàn)設(shè)計(jì)樣本個(gè)數(shù)從而提高效率。通過(guò)遺傳智能布點(diǎn)策略，根據(jù)距離比較原則從非支配解外部解集中挑選部分到信賴(lài)域空間，提高關(guān)鍵區(qū)域代理模型的精度從而加快收斂。該方法被成功應(yīng)用于基于耐撞性和模態(tài)特性的轎車(chē)車(chē)身結(jié)構(gòu)輕量化設(shè)計(jì)中，解決了汽車(chē)結(jié)構(gòu)安全中的多目標(biāo)優(yōu)化問(wèn)題。

關(guān)鍵詞：汽車(chē)結(jié)構(gòu)安全 多目標(biāo)優(yōu)化 代理模型 智能布點(diǎn) 信賴(lài)域

Multi-Objective Optimization Method Based on Metamodel for Vehicle Structural Safety

Han Xu Jiang Chao Chen Guodong Long Xiangyun

（Hunan University）

Abstract：Most vehicle structural safety optimization problems involve multiple objectives， which cannot be expressed explicitly but acquired by complex computational model， and thus it increases the difficulty of solving multi-objective optimization problems. Intelligent optimization method is able to search for multiple optimal solutions in one single simulation run， but the low efficiency limits its application to complex vehicle structural crash problems. Common multi-objective optimization methods based on metamodel can well deal with the low efficiency and become a research focus， but the solution accuracy is usually low. Therefore， this project studies the multi-objective optimization methods based on metamodel， aims to improve the efficiency and accuracy in the design of vehicle crash safety. A new multi-objective optimization algorithm is proposed based on adaptive radial basis function. This method effectively assesses metamodel by using inherit Latin hypercube design， radial basis function and intergeneration projection genetic algorithm. The proposed method is applied to the thin-walled sections for structural crashworthiness， which is beneficial to quickly find multi-group design schemes and can well balance energy absorption and collision force. A micro multi-objective genetic algorithm based on intelligent sampling technology is put forward. The algorithm adopts the extented radial basis function to build a global metamodel， and then employs the efficient micro multi-objective genetic algorithm for approximate optimization. The method has been used in the dynamic characteristic optimization of a heavy commercial vehicle cab and obtains many optimal design schemes. Optimization algorithm based on trust region model management is proposed to solve the multi-objective optimization problem in complex engineering. The method transforms the complex optimization problems in the entire design space into a series of approximation problems in trust region. The method has been applied in a door structure optimization， and well balances the static and dynamic performance by matching the thickness of key components. Based on trust region and intelligent sampling technology， an efficient multi-objective method is developed. The method has been successfully used in the lightweight design of car body based on crashworthiness and modal characteristics， and demonstrates its ability to solve multi-objective optimization problems in vehicle structural safety.

Key Words：Vehicle structural safety； Multi-objective optimization； Metamodel； Intelligent sampling； Trust region

閱讀全文鏈接（需實(shí)名注冊(cè)）：http：//www.nstrs.cn/xiangxiBG.aspx？id=50229&flag=1