Adaptive differential evolution algorithm with fitness based selection of parameters and mutation strategies / Rawaa Dawoud Hassan Al-Dabbagh

Al-Dabbagh, Rawaa Dawoud Hassan (2015) Adaptive differential evolution algorithm with fitness based selection of parameters and mutation strategies / Rawaa Dawoud Hassan Al-Dabbagh. PhD thesis, University of Malaya.

[img] PDF (Full Text)
Restricted to Repository staff only until 01 January 2018.

Download (3608Kb)


    Differential evolution (DE) is a simple yet powerful evolutionary algorithm (EA). It has demonstrated good convergence, and its principles are easy to understand. DE has effectively solved various global optimization problems, including benchmark functions. These problems have shown different challenging characteristics such as non-convexity, non-linearity, and/or multi-modality which became difficult for traditional non-linear programming to deal with. The performance of DE algorithm depends heavily on the selected mutation strategy and its associated control parameters. The sensitivity of the DE algorithm to its mutation strategy and to the corresponding control parameters can significantly deteriorate its performance if the strategy is improperly selected. Hence, the process of choosing a suitable DE strategy and setting its control parameters is difficult and requires much user experience. In this thesis, the fundamental contributions include the analysis, design, and evaluation of the adaptive DE algorithms. Firstly, a comprehensive procedural analysis is conducted to investigate the various adaptive schemes that have been utilized to automatically control the DE parameters and/or its mutation strategies. In the same analysis, two taxonomies are proposed for the purpose. The first one is proposed to eliminate any ambiguity related to classify any adaptive EA. The new classification comprises three levels of categories instead of two regarding the parameter control type (deterministic, adaptive, self-adaptive) and the evidence (absolute, relative) used for determining the change of the parameter. The second taxonomy is a new taxonomy proposed to classify the adaptive DE algorithms in particular into two categories (DE with adaptive parameters and DE with adaptive parameters and strategies) considering the adaptive components used in this algorithm. Secondly, a new DE algorithm (ARDE-SPX) is introduced that automatically adapts a repository of DE strategies and parameters control schemes to avoid the problems of stagnation and make DE respond to a wide range of function characteristics at different stages of evolutionary search. ARDE algorithm makes use of JADE strategy and the MDE_pBX parameters adaptive schemes as frameworks. Then a new adaptive procedure called adaptive repository (퐴푅) is developed to select the appropriate combinations of the JADE strategies and the parameter control schemes of the MDE_pBX to generate the next population based on their fitness values. The adaptive repository mechanism is a general scheme and can be embedded with high flexibility into any population-based evolutionary algorithm. Moreover, this work is extended to integrate the SPX crossover operator with the adaptive ARDE algorithm in a new way of implementation in order to make the adaptive ARDE algorithm satisfy both the global and local search requirements. Thirdly, experimental results are presented to confirm the reliability of the proposed ARDE-SPX over several existing adaptive DE variants. These comparisons are conducted in terms of the solution precision, successful rate and robustness over thirty-three standard and transformed benchmark functions. ARDE is also used to develop a new dynamic parameter identification framework to estimate the barycentric parameters of the CRS A456 robot manipulator. The simulation results show the effectiveness of the ARDE method over other conventional techniques, transcending the limits of the existing state-of-the-art algorithms in estimating the parameters of robot.

    Item Type: Thesis (PhD)
    Additional Information: Thesis (Ph.D.) -- Faculty of Computer Science and Information Technology, University of Malaya, 2015
    Uncontrolled Keywords: Adaptive; Differential evolution; Algorithm; Fitness based selection; Parametersand; Mutation; Strategies
    Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
    Q Science > QA Mathematics > QA76 Computer software
    Divisions: Faculty of Computer Science & Information Technology
    Depositing User: Miss Dashini Harikrishnan
    Date Deposited: 19 Oct 2015 13:30
    Last Modified: 22 Dec 2015 09:21

    Actions (For repository staff only : Login required)

    View Item