Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/15059
Authors: Dinkar, Shail Kumar
Keywords: Antlion Optimizer (ALO);Metaheuristic for Global;Opposition Based Learning;Acceleration Parameter
Issue Date: Feb-2019
Publisher: I.I.T Roorkee
Abstract: Antlion Optimizer (ALO) is a metaheuristic for global optimization problems based on life cycle and unique hunting behaviour of antlion. Though this algorithm is developed recently, yet its efficiency and effectiveness can be demonstrated to solve various real life complex applications in many areas such as feature selection problems, optimal power and load dispatch in electrical engineering, wireless sensors networks, clustering and classification problems, unit commitment problems are few of many. The objective of this Ph.D Thesis is to improve the efficiency, reliability and stability of Antlion Optimizer (ALO). To achieve this goal, Chapter 2 proposes a novel variant of ALO namely Opposition Based Laplacian Antlion Optimizer (OB-𝐿-ALO) for unconstraint global optimization. This version addresses the drawback of premature convergence and inability to avoid entrapment into local optima of ALO. For this purpose, the random walk of classical ALO is improved by applying Laplace distribution in place of uniform distribution as a first strategy to enhance exploration of search region in early iterations. The second strategy is to apply opposition based learning (OBL) model to the best (elite) candidate solution. The performance of proposed OB-𝐿-ALO is verified over a set of 31 benchmark problems of varying difficulties containing 23 state-of-the-art problems (a set of unimodal, multimodal and fixed dimension multimodal functions) and 8 IEEE CEC 2014 composition functions. This set is produced in Appendix I of this thesis. A wide analysis is performed to validate the performance of the proposed OB-𝐿-ALO such as convergence behaviour, trajectory analysis of best candidate solution, average distance analysis of search agent before and after applying the updating strategies and elite convergence curve. Statistical significance of OB-𝐿-ALO is tested using Wilcoxon ranksum test. The obtained numerical results establish that the proposed OB-𝐿-ALO outperforms classical ALO for most of the problems. Chapter 3 introduces two variants namely OB-ac-ALO and OB-SAC-ALO to accelerate the convergence to opposition based ALO as given in Chapter 2. The OBL model is hybridized with varying acceleration parameter (ac) to propose OB-ac-ALO which is useful to control the abrupt behaviour of the solutions at later stages of the generation and accelerate the convergence speed. The second modification is accomplished by hybridizing the OBL mechanism applied to best (elite) candidate solution with sine acceleration coefficient (SAC) to propose OB-SAC-ALO. The performance of both the proposed variants is investigated over same set of benchmark functions as applied in chapter 2 and reproduced in Appendix I of this thesis. The similar analysis ii metrics are performed as utilized in chapter 2. The obtained results and analysis prove that OB-ac-ALO and OB-SAC-ALO perform better than ALO in majority of the problems. Chapter 4 proposes another extended variant of opposition based ALO with inspiration to improve the random walk by taking the advantage of random jump based on step length of lévy flight distribution and utilizing the hybridization of OBL mechanism with acceleration parameter to propose opposition based lévy flight antlion optimizer (OB-LF-ALO). The performance of proposed OB-LF-ALO is tested on same set of benchmark problems as utilized in chapter 2 and 3 and compared with classical ALO which is reproduced in Appendix I of this thesis. Wilcoxon ranksum test is used to show the statistical significance with computational complexity. To keep the uniformity among all chapters, similar analysis metrics including search history behaviour are performed to investigate the performance of proposed OB-LF-ALO. The analysis of results prove that OB-LF-ALO outperforms the classical ALO. Chapter 5 is divided into two parts. In first part, the modification utilizes a different distribution namely Cauchy distribution to generate the random number in place of uniform distribution to implement the random walk as a first strategy. As a second strategy, OBL mechanism is applied around the best (elite) candidate solution and hybridized with acceleration parameter to keep proper balance between early exploration and later exploitation. The modification is proposed as opposition based antlion optimizer using Cauchy distribution (OB-C-ALO) .OB-C-ALO is verified using the same set of benchmark problems as used in previous chapters and compared with classical ALO. Similar analysis metrics are used in previous chapters for analysis and Wilcoxon ranksum test is used to show the statistical significance with computational complexity. The second part of the chapter presents the performance comparison in terms of obtained results and analysis among all the five proposed variants of classical ALO i.e. OB-𝐿-ALO, OB-ac-ALO, OB-SAC-ALO, OB-LF-ALO and OB-C-ALO. In Chapter 6, the performance of proposed variants of classical ALO is investigated over a real world complex application of model order reduction of linear time invariant system in the field of control system. The performance of these algorithms are investigated by applying on three single input single output (SISO) systems including two four and one eight order problem of different characteristics. Chapter 7 attempts to determine optimal values of heat transfer coefficient (𝐻̃) and pressure drop(Δ𝑃) parameters. This problem consist of two conflicting objective functions: first to maximize heat transfer coefficient and second to minimize pressure drop value. The problem iii is optimized using two approaches: First, single objective approach for both the objective functions. Secondly, multi-objective approach in which both the objective functions are optimized .This purpose is achieved using two different methods of multi-objective optimization: (i) Using weighted sum approach of multi-objective optimization using classical ALO and its proposed variants (ii) Pareto based multi-objective optimization using multi-objective antlion optimizer (MOALO). Chapter 8 investigates the performance of designed variants of classical ALO for optimizing the production of biodiesel from renewable energy sources. In this work, the regression equation demonstrating the relationship among three independent variables namely temperature ,methanol to oil ratio and concentration of catalyst is successfully optimized for biodiesel production using Antlion Optimizer (ALO) and its proposed modified versions OB-𝐿-ALO,OB-ac-ALO,OB-SAC-ALO,OB-LF-ALO and OB-C-ALO. Chapter 9 concludes the Thesis by deriving overall observations and concluding remarks. It also outlines the limitations and scope of the proposed variants of ALO algorithms. Later on, some suggestion to future research are provided.
URI: http://localhost:8081/xmlui/handle/123456789/15059
Research Supervisor/ Guide: Deep, Kusum
metadata.dc.type: Thesis
Appears in Collections:DOCTORAL THESES (Maths)

Files in This Item:
File Description SizeFormat 
G28808.pdf4.36 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.