|
Document Type |
: |
Thesis |
|
Document Title |
: |
GLOBAL OPTIMIZATION TECHNIQUE FOR BIG DATA PROCESSING التقنية المثالية العامة لمعالجة البيانات الكبيرة |
|
Subject |
: |
Faculty of Computing and Information Technology |
|
Document Language |
: |
Arabic |
|
Abstract |
: |
Proposing new mutation strategies and adjusting control parameters to improve the optimization performance of differential evolution (DE) is an important research study. Therefore, the main contribution of this thesis goes in the following six directions: The first direction is introducing a less greedy mutation strategy with enhanced exploration capability, named DE/current-to-ord_best/1 (ord stands for ordered) or ord_best for short.
In the second direction, we introduce a more greedy mutation strategy with enhanced exploitation capability, named DE/current-to-ord_pbest/1 (ord_pbest for short). Both of the proposed mutation strategies are based on ordering three selected vectors from the current generation to perturb the target vector, where the directed differences are used to mimic the gradient descent behavior to direct the search toward better solutions. In ord_best, the three vectors are selected randomly to enhance the exploration capability of the algorithm. On the other hand, ord_pbest is designed to enhance the exploitation capability where two vectors are selected randomly and the third is selected from the global p best vectors. Based on the proposed mutation strategies, ord_best and ord_pbest, two DE variants are introduced as EDE and EBDE, respectively.
The third direction is introducing a new semi-parameter adaptation approach (SPA) as an alternative adaptation approach for the selection of control parameters. The proposed approach consists of two different settings for the two control parameters: the scaling factor (F) and the crossover rate (Cr). The benefit of this approach is to prove that the semi-adaptive algorithm is better than the pure random algorithm or the fully adaptive or self-adaptive algorithm.
The fourth direction is a low-level hybridization framework. The proposed mutations can be combined with other mutations to enhance their search capabilities on difficult and complicated optimization problems.
The fifth direction is proposing two high-level hybridization frameworks between the proposed algorithms and other optimization algorithms. The first is a hybridization framework between the proposed algorithms and a modified version of the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The modified version of CMA-ES undergoes the crossover operation to improve the exploration capability of the proposed framework. Both algorithms will work simultaneously on the same population, but more populations will be assigned gradually to the better performing algorithm. The second hybridization framework is a new memetic hybridization framework, where the proposed algorithms will be the population-based algorithms, while a modified version of Multiple Trajectory Search (MTS) will be the local search algorithm.
The sixth direction is introducing an improved divide and conquer concept for solving large-scale global optimization problems, where the dimensions are randomly divided into groups, and each group is solved separately.
Finally, in order to verify and analyze the performance of this work, numerical experiments were conducted using CEC2013 and CEC2017 benchmarks. The performance was also evaluated using CEC2010 and CEC2013 benchmarks designed for Large Scale Global Optimization. |
|
Supervisor |
: |
Prof. Kamal M. Jambi |
|
Thesis Type |
: |
Doctorate Thesis |
|
Publishing Year |
: |
1440 AH
2019 AD |
|
Added Date |
: |
Monday, June 17, 2019 |
|
Researchers
| أنس عبدالقادر هادي | Hadi, Anas Abdulqader | Researcher | Doctorate | |
|