English  |  正體中文  |  简体中文  |  Items with full text/Total items : 888/888 (100%)
Visitors : 13005284      Online Users : 206
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Tips:
  • please add "double quotation mark" for query phrases to get precise results
  • please goto advance search for comprehansive author search
  • Adv. Search
    HomeLoginUploadHelpAboutAdminister Goto mobile version
    Please use this identifier to cite or link to this item: http://ccur.lib.ccu.edu.tw/handle/A095B0000Q/575


    Title: 利用適應值繼承增強演化演算法;Boosting Evolutionary Algorithm through Fitness Inheritance
    Authors: 廖容佐;LIAW, RUNG-TZUO
    Contributors: 資訊工程研究所
    Keywords: 適應值繼承;代理模型;主動式學習;演化式多工;Fitness inheritance;Surrogates;Active learning;Evolutionary multitasking
    Date: 2017
    Issue Date: 2019-07-17
    Publisher: 資訊工程研究所
    Abstract: 適應值估計已被廣為應用於現實問題,目前的適應值估計方法可被分為兩類;適應值繼承與代理模型。代理模型利用低計算資源的模型取代適應值函數;而適應值繼承基於親代之適應值估計子代之適應值。在模型的準確度與複雜度間存在折衷,由於不具有建模程序適應值繼承較代理模型有更低的計算資源消耗,相對地代理模型則有較適應值繼承更高的準確度。本研究利用適應值繼承解決設計適應值估計方法的關鍵議題,包含冷啟動、模型管理、模型選擇、演化控制、以及準確度與複雜度的折衷。具體而言,本篇研究提出三個架構,即適應值繼承輔助代理模型、基於主動式學習之適應值繼承、以及基於適應值估計的共生群落演化。首先,適應值繼承輔助代理模型提出先期建模、適應值繼承與代理模型之結合、以及基於適應值繼承之模型管理標準分別解決冷啟動、模型選擇、模型管理三個議題。實驗研究驗證適應值繼承輔助代理模型與一先進方法—代理輔助瀰因演算法—較之皆具有效度與效率。然而,適應值繼承輔助代理模型仍考慮代理模型的使用造成效率上的瓶頸。為解決此一議題,基於主動式學習之適應值繼承利用適應值繼承的集成結合基於不同距離評估之適應值繼承建立一具效度與效率之估計模型。基於主動式學習之適應值繼承並引入主動式學習的概念擴展演化控制的觀點。實驗研究驗證基於主動式學習之適應值繼承架構較代理輔助瀰因演算法以及適應值繼承輔助代理模型架構更具效度與效率。在單工最佳化之上,將適應值估計方法用於演化式多工進而產生基於適應值估計的共生群落演化。啟發自生物群落中之共生關係,基於適應值估計的共生群落同時演化多個群體作為生物群落,並以群體間的訊息交換模擬共生關係。候選解的選擇是自一連結後代經由不同工作上建立之估計模型所成之集成做為評估標準選出。此外,本研究提出一組演化多工的測試問題以驗證效能。實驗結果證明基於適應值估計的共生群落演化較先進的單工演算法—共變矩陣適應演化策略、一知名演化多工演算法—多因子演化演算法、以及基於主動式學習之適應值繼承架構,皆能有更好的效度與效率。實驗並證明基於適應值估計的共生群落演化更優於一演化多工的先進算法。
    Fitness approximation methods have been widely used to solve complex benchmark problems and real-world applications. Current approaches of fitness approximation can be categorized into two types: fitness inheritance and surrogates. Surrogate model adopts a computationally less expensive model for fitness function, while fitness inheritance approximates the fitness on the basis of parent’s fitness. There is a trade-off between model complexity and model fidelity. Though fitness inheritance is computationally cheaper than surrogates due to the lack of modeling process, surrogates can have higher fidelity than fitness inheritance. This study utilizes fitness inheritance to address the key issues in the design of fitness approximation, i.e., cold start, model management, model selection, evolution control and fidelity-complexity trade-off. Specifically, three frameworks are proposed in the present study, i.e., fitness inheritance aided surrogates (FIS) framework, active learning based fitness inheritance (ALFI), and evolution of biocoenosis through symbiosis with fitness approximation (EBSFA). First, the FIS framework presents early modeling, collaboration of fitness inheritance and surrogates, and inheritance-based model management criterion to tackle the cold start, model selection, and model management issues, respectively. Empirical studies verify that FIS framework is effective and efficient in comparison to a state-of-the-art surrogate-assisted memetic algorithm (SAMA) on CEC 2014 test problems. Nonetheless, FIS framework still considers the use of surrogates which becomes a bottleneck of efficiency. To address this issue, the proposed ALFI ponders fitness inheritance with ensemble which integrates fitness inheritance with different distance measures for generating an effective and efficient approximation model. Introducing the concept of active learning in the ALFI broadens the view of evolution control. Empirical results validate that ALFI framework has effectiveness and efficiency, in comparison to SAMA and FIS framework. Beyond single task optimization, extending the use of fitness approximation on evolutionary multitasking leads to the EBSFA. Inspiring from symbiosis in biocoenosis, the EBSFA evolves several EAs concurrently as biocoenosis and exchanges information among EAs as symbiosis. The selection of candidate individuals to be evaluated among a concatenate offspring is taken by the ensemble of multiple fitness approximation derived from different tasks. In addition, this study proposes a set of new benchmark problems for evolutionary many-tasking (MaTPs) to validate the performance. Experimental results exhibit the effectiveness of the proposed EBSFA framework over a state-of-the-art numerical optimization method, the covariance matrix adaptation evolution strategy (CMAES), a well-known evolutionary multitasking method, the multi-factorial evolutionary algorithm (MFEA), and the ALFI framework on the four MaTPs. The EBSFA also outperforms a state-of-the-art evolutionary multitasking method on the benchmark problems of CEC 2017 competition for multitask optimization.
    Appears in Collections:[資訊工程學系] 學位論文

    Files in This Item:

    File Description SizeFormat
    index.html0KbHTML338View/Open


    All items in CCUR are protected by copyright, with all rights reserved.


    版權聲明 © 國立中正大學圖書館網頁內容著作權屬國立中正大學圖書館

    隱私權及資訊安全政策

    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback