多目标进化算法理论、算法设计与应用研究
发布时间：2021-08-04 18:07
多目标优化问题普遍存在于社会生活、生产的各个领域,但是用传统的数学方法来求解多目标优化问题并不能取得很好的效果。近年来,随着计算机技术的发展和人工智能的兴起,基于计算机技术的计算智能得到了快速的发展。作为计算智能的一个重要分支,进化计算求解多目标优化问题已经成为计算智能领域的一个研究热点。进化计算最初是指受生物进化启发而设计的基于种群的优化算法,现在己经发展成为各种受自然启发的算法和技术的统称。如今,多目标进化算法已经被广泛地应用在社会的各个领域,正在深刻的改变科学研究和生产实践应用等的方方面面。然而,多目标进化算法薄弱而滞后的数学理论研究己经严重阻碍了其在计算智能领域的进一步应用与发展。本文对多目标进化算法的理论、算法设计以及相应的实际应用进行了深入地研究。首先,针对多目标优化问题中的一类搜索不均衡问题,从理论上分析了造成搜索不均衡问题的原因,分析并定义了三类主要的搜索不均衡问题。针对这类搜索不均衡的问题设计出了一类基于种群分解的算法,并通过一系列的数值仿真实验验证了所提出算法的有效性。其次,本文研究了基于分解的进化多目标优化算法中内在并行性的外在控制理论。在此理论的基础上,着重研究...
【文章来源】：广东工业大学广东省
【文章页数】：303 页
【学位级别】：博士
【文章目录】：
ABSTRACT
摘要
INDEX OF ABBREVIATIONS
Chapter 1 Introduction
1.1 Background
1.2 Fundamental Concepts
1.3 Motivation
1.4 Contributions
1.5 Outline of the Thesis
Chapter 2 Investigating the Effect of Imbalance Between Convergence and Di-versity in Evolutionary Multi-objective Algorithm
2.1 Overview
2.2 Introduction
2.3 Related Work
2.3.1 Convergence-first EMO Algorithms
2.3.2 MOEA/D-M2M
2.3.3 Related Theoretical Studies
2.4 Imbalanced Problems in Multi-objective Optimization
2.4.1 Definition of an Imbalanced Problem
2.4.2 Illustrative Problems
2.5 A Critical Review of Convergence-first EMO Methods for Imbalanced Problems
2.6 Numerical Computations on Imbalanced Problems
2.6.1 EMO Algorithms in the MOEA/D-M2M Framework
2.6.2 Proposed Imbalanced Multi-Objective Test Suite
2.7 Results of EMO and EMO-MOEA/D-M2M Methods
2.8 EMO-M2M for Balanced Problems
2.9 Conclusions
Chapter 3 Explicit Control of Implicit Parallelism in Decomposition Based Evo-lutionary Many-Objective Optimization Algorithms
3.1 Overview
3.2 Introduction
3.3 Preliminaries
3.3.1 MOEA/D Framework
3.3.2 NSGA-Ⅲ Framework
3.4 Explicit Control of Implicit Parallelism
3.5 Variants of N2M and Results
3.5.1 DTLZ and WFG Test Problems
3.5.2 Parameter Settings
3.5.3 Simulation Results on DTLZ Problems
3.5.4 Simulation Results on WFG Problems
3.6 Extended MOEA/D-M2M and MOEA/D Algorithms with Normalization
3.6.1 Simple Normalization Procedure
3.6.2 NSGA-Ⅲ Normalization Procedure on M2M and MOEA/D
3.7 NSGA-Ⅲ and MOEA/D Variants and Results
3.7.1 NSGA-Ⅲ Variants
3.7.2 MOEA/D Variants
3.8 Discussion on Explicit Control of Implicit Parallelism on EMO Algorithms
3.9 Conclusions
Chapter 4 Effect of Objective Normalization and Penalty Parameter on PBI De-composition Based Evolutionary Many-objective Optimization Algo-rithms
4.1 Overview
4.2 Introduction
4.3 Decomposition-based EMO Algorithms
4.3.1 PBI Fitness
4.3.2 NSGA-Ⅲ's Niching Fitness
4.4 Sensitivity of Fitness Assignment Due to Normalization Instability
4.4.1 Sensitivity Ratio
4.4.2 Validation
4.5 Experimental Studies
4.5.1 Test Problems
4.5.2 Parameter Settings
4.5.3 Experimental Studies on NSGA-Ⅲ
4.5.4 Experimental Studies on MOEA/D
4.5.5 Problems with a Convex Pareto-optimal Front
4.6 Conclusions
Chapter 5 Study on the Effect of Non-dominated Sorting in DecompositionBased Evolutionary Many-objective Optimization Algorithms
5.1 Overview
5.2 Introduction
5.3 Niching Mechanism in NSGA-Ⅲ
5.3.1 Theoretical Study
5.3.2 Experimental Validation
5.4 Why does non-dominated sorting matter?
5.4.1 Experimental Studies on Modified NSGA-Ⅲ
5.4.2 Mapping
5.4.3 Experimental studies
5.5 Conclusion
Chapter 6 Dynamic Search Resource Allocation for Many-objective Optimization
6.1 Overview
6.2 Introduction
6.3 Preliminaries
6.3.1 New Solution Generation
6.3.2 Update
6.4 Adaptive Subregion Division and Weight Vector Setting
6.4.1 Adaptive Subregion Division
6.4.2 Adaptive Weight Setting
6.4.3 Main Framework of MOEA/D-AM2M
6.5 Construction of Challenging MaOPs
6.5.1 Degenerated MaOPs with disconnected PFs
6.6 Experimental Study
6.6.1 EMO Algorithms in Comparison
6.6.2 Performance Metrics
6.6.3 Experimental Setting
6.6.4 Experimental Study on Degenerated MaOPs with Disconnected PFs
6.6.5 Further Performance Study of MOEA/D-AM2M on Degenerated MaOPs with Connected PFs
6.6.6 Experimental Study on Non-degenerated MaOPs
6.6.7 Experimental Study on Imbalanced MOPs
6.6.8 The Setting of Update Parameter (G) in MOEA/D-AM2M
6.7 Conclusion
Chapter 7 Theoretical Studies on the Connection Among the Three Commonly Used Decomposition Methods
7.1 Overview
7.2 Introduction
7.3 Theoretical Study on Decomposition Methods
7.3.1 Decomposition Methods
7.3.2 Theoretical Study
7.4 Main Idea of Proposed Algorithm
7.4.1 Decomposition based Dominance Relationship
7.4.2 Properties Analysis
7.4.3 The Adaptive Setting of Parameter β
7.4.4 The novelty of D-dominance
7.4.5 Decomposition Based Crowding Measurement
7.4.6 Main Framework of Proposed Algorithm
7.5 Experimental Studies
7.5.1 EMO Algorithms in Comparison
7.5.2 Test Problems
7.5.3 General Parameter Settings
7.5.4 Experimental Studies on WFG Test Problems
7.5.5 Experimental Studies on DTLZ Test Problems
7.6 Conclusion
Chapter 8 Modelling the Tracking Area Planning Problem Using an Evolution-ary Multi-objective Algorithm
8.1 Overview
8.2 Introduction
8.3 Related Work
8.4 The TA Planning Problem
8.4.1 Problem Statement
8.4.2 Multi-objective TA Planning Model
8.5 An EMO Algorithm Based on the M2M Decomposition for the Multi-objective TA Planning Model
8.5.1 Encoding Method
8.5.2 Decoding Method
8.5.3 Initialization Based on Fuzzy Clustering
8.5.4 Crossover and Mutation
8.5.5 Constraint Handling and Repair Strategy
8.5.6 MOEA/D and M2M Decomposition Strategy
8.5.7 Main Framework of the M2M-based EMO Algorithm for Multi-objective TA Planning
8.6 Computational Experiments and Analysis
8.6.1 The Parameters of the Networks
8.6.2 Experimental Results and Analysis
8.6.3 Computational Complexity
8.7 Conclusion
Chapter 9 Multi-objective Evolutionary Triclustering with Constraints of Time-series Gene Expression Data
9.1 Overview
9.2 Introduction
9.3 Preliminaries
9.3.1 Microarray and time-series gene expression data
9.3.2 Triclustering
9.4 Multi-objective constrained triclustering
9.5 Decomposition based evolutionary algorithm for multi-objective constrained triclustering
9.5.1 Encoding and decoding
9.5.2 Recombination operators
9.5.3 Two-step local search
9.5.4 Multi-objective triclustering algorithm
9.6 Experimental studies
9.6.1 Performance metrics
9.6.2 Parameter setting
9.6.3 Experiments on artificial datasets
9.6.4 Experiments on real-life datasets
9.7 Engineering applications
9.7.1 Key disease-related genes detection on HIV-1 progression data
9.7.2 Recommendation system for anonymous social network users
9.8 Conclusion and future work
Conclusions
References
List of Published/Submitted Papers
Acknowledgements
本文编号：3322159
【文章来源】：广东工业大学广东省
【文章页数】：303 页
【学位级别】：博士
【文章目录】：
ABSTRACT
摘要
INDEX OF ABBREVIATIONS
Chapter 1 Introduction
1.1 Background
1.2 Fundamental Concepts
1.3 Motivation
1.4 Contributions
1.5 Outline of the Thesis
Chapter 2 Investigating the Effect of Imbalance Between Convergence and Di-versity in Evolutionary Multi-objective Algorithm
2.1 Overview
2.2 Introduction
2.3 Related Work
2.3.1 Convergence-first EMO Algorithms
2.3.2 MOEA/D-M2M
2.3.3 Related Theoretical Studies
2.4 Imbalanced Problems in Multi-objective Optimization
2.4.1 Definition of an Imbalanced Problem
2.4.2 Illustrative Problems
2.5 A Critical Review of Convergence-first EMO Methods for Imbalanced Problems
2.6 Numerical Computations on Imbalanced Problems
2.6.1 EMO Algorithms in the MOEA/D-M2M Framework
2.6.2 Proposed Imbalanced Multi-Objective Test Suite
2.7 Results of EMO and EMO-MOEA/D-M2M Methods
2.8 EMO-M2M for Balanced Problems
2.9 Conclusions
Chapter 3 Explicit Control of Implicit Parallelism in Decomposition Based Evo-lutionary Many-Objective Optimization Algorithms
3.1 Overview
3.2 Introduction
3.3 Preliminaries
3.3.1 MOEA/D Framework
3.3.2 NSGA-Ⅲ Framework
3.4 Explicit Control of Implicit Parallelism
3.5 Variants of N2M and Results
3.5.1 DTLZ and WFG Test Problems
3.5.2 Parameter Settings
3.5.3 Simulation Results on DTLZ Problems
3.5.4 Simulation Results on WFG Problems
3.6 Extended MOEA/D-M2M and MOEA/D Algorithms with Normalization
3.6.1 Simple Normalization Procedure
3.6.2 NSGA-Ⅲ Normalization Procedure on M2M and MOEA/D
3.7 NSGA-Ⅲ and MOEA/D Variants and Results
3.7.1 NSGA-Ⅲ Variants
3.7.2 MOEA/D Variants
3.8 Discussion on Explicit Control of Implicit Parallelism on EMO Algorithms
3.9 Conclusions
Chapter 4 Effect of Objective Normalization and Penalty Parameter on PBI De-composition Based Evolutionary Many-objective Optimization Algo-rithms
4.1 Overview
4.2 Introduction
4.3 Decomposition-based EMO Algorithms
4.3.1 PBI Fitness
4.3.2 NSGA-Ⅲ's Niching Fitness
4.4 Sensitivity of Fitness Assignment Due to Normalization Instability
4.4.1 Sensitivity Ratio
4.4.2 Validation
4.5 Experimental Studies
4.5.1 Test Problems
4.5.2 Parameter Settings
4.5.3 Experimental Studies on NSGA-Ⅲ
4.5.4 Experimental Studies on MOEA/D
4.5.5 Problems with a Convex Pareto-optimal Front
4.6 Conclusions
Chapter 5 Study on the Effect of Non-dominated Sorting in DecompositionBased Evolutionary Many-objective Optimization Algorithms
5.1 Overview
5.2 Introduction
5.3 Niching Mechanism in NSGA-Ⅲ
5.3.1 Theoretical Study
5.3.2 Experimental Validation
5.4 Why does non-dominated sorting matter?
5.4.1 Experimental Studies on Modified NSGA-Ⅲ
5.4.2 Mapping
5.4.3 Experimental studies
5.5 Conclusion
Chapter 6 Dynamic Search Resource Allocation for Many-objective Optimization
6.1 Overview
6.2 Introduction
6.3 Preliminaries
6.3.1 New Solution Generation
6.3.2 Update
6.4 Adaptive Subregion Division and Weight Vector Setting
6.4.1 Adaptive Subregion Division
6.4.2 Adaptive Weight Setting
6.4.3 Main Framework of MOEA/D-AM2M
6.5 Construction of Challenging MaOPs
6.5.1 Degenerated MaOPs with disconnected PFs
6.6 Experimental Study
6.6.1 EMO Algorithms in Comparison
6.6.2 Performance Metrics
6.6.3 Experimental Setting
6.6.4 Experimental Study on Degenerated MaOPs with Disconnected PFs
6.6.5 Further Performance Study of MOEA/D-AM2M on Degenerated MaOPs with Connected PFs
6.6.6 Experimental Study on Non-degenerated MaOPs
6.6.7 Experimental Study on Imbalanced MOPs
6.6.8 The Setting of Update Parameter (G) in MOEA/D-AM2M
6.7 Conclusion
Chapter 7 Theoretical Studies on the Connection Among the Three Commonly Used Decomposition Methods
7.1 Overview
7.2 Introduction
7.3 Theoretical Study on Decomposition Methods
7.3.1 Decomposition Methods
7.3.2 Theoretical Study
7.4 Main Idea of Proposed Algorithm
7.4.1 Decomposition based Dominance Relationship
7.4.2 Properties Analysis
7.4.3 The Adaptive Setting of Parameter β
7.4.4 The novelty of D-dominance
7.4.5 Decomposition Based Crowding Measurement
7.4.6 Main Framework of Proposed Algorithm
7.5 Experimental Studies
7.5.1 EMO Algorithms in Comparison
7.5.2 Test Problems
7.5.3 General Parameter Settings
7.5.4 Experimental Studies on WFG Test Problems
7.5.5 Experimental Studies on DTLZ Test Problems
7.6 Conclusion
Chapter 8 Modelling the Tracking Area Planning Problem Using an Evolution-ary Multi-objective Algorithm
8.1 Overview
8.2 Introduction
8.3 Related Work
8.4 The TA Planning Problem
8.4.1 Problem Statement
8.4.2 Multi-objective TA Planning Model
8.5 An EMO Algorithm Based on the M2M Decomposition for the Multi-objective TA Planning Model
8.5.1 Encoding Method
8.5.2 Decoding Method
8.5.3 Initialization Based on Fuzzy Clustering
8.5.4 Crossover and Mutation
8.5.5 Constraint Handling and Repair Strategy
8.5.6 MOEA/D and M2M Decomposition Strategy
8.5.7 Main Framework of the M2M-based EMO Algorithm for Multi-objective TA Planning
8.6 Computational Experiments and Analysis
8.6.1 The Parameters of the Networks
8.6.2 Experimental Results and Analysis
8.6.3 Computational Complexity
8.7 Conclusion
Chapter 9 Multi-objective Evolutionary Triclustering with Constraints of Time-series Gene Expression Data
9.1 Overview
9.2 Introduction
9.3 Preliminaries
9.3.1 Microarray and time-series gene expression data
9.3.2 Triclustering
9.4 Multi-objective constrained triclustering
9.5 Decomposition based evolutionary algorithm for multi-objective constrained triclustering
9.5.1 Encoding and decoding
9.5.2 Recombination operators
9.5.3 Two-step local search
9.5.4 Multi-objective triclustering algorithm
9.6 Experimental studies
9.6.1 Performance metrics
9.6.2 Parameter setting
9.6.3 Experiments on artificial datasets
9.6.4 Experiments on real-life datasets
9.7 Engineering applications
9.7.1 Key disease-related genes detection on HIV-1 progression data
9.7.2 Recommendation system for anonymous social network users
9.8 Conclusion and future work
Conclusions
References
List of Published/Submitted Papers
Acknowledgements
本文编号：3322159
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/3322159.html
