Nomadic people optimizer (NPO) for large-scale optimization problems

• Main Author: Mohamd Salih, Sinan Qahtan
• Format: Thesis
• Language: English
• Published: 2019
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://umpir.ump.edu.my/id/eprint/29265/1/Nomadic%20people%20optimizer%20%28NPO%29%20for%20large-scale%20optimization%20problems.wm.pdf; http://umpir.ump.edu.my/id/eprint/29265/

Researchers have in the past few decades resorted to several methods that are inspired from complex optimization problems. The classical deterministic search methods are known to often get trapped in local minimum and do perform poorly on high dimensional problems. A metaheuristic is defined as an iterative generation process which guides a subordinate heuristic through a combination of different intelligent concepts for exploring and exploiting the solution space; they employ learning strategies to structure information in order to establish efficient near-optimal solutions. Three major problems are encountered when designing metaheuristics; the first problem is balancing exploration with exploitation capabilities (which leads to premature convergence or trapping in the local minima), while the second problem is the dependency of the algorithm on the controlling parameters, which are parameters with unknown optimal values. The final problem is the ability of the algorithm to solve large-scale problems, which mostly are the real world problems. In this thesis, a novel nature-inspired metaheuristic called “Nomadic People Optimizer (NPO)” was designed. The NPO is inspired by the lifestyle of the nomads. The proposed algorithm simulates the behavior of the nomads when they are searching for life sources (water or grazing fields). The basic component of the algorithm consists of several clans and each clan searches for the best place (or best solution) based on the position of their leader. The interaction between these clans is inspired by the concept of a group(s) of people controlled by their leader(s). The leaders of the clans periodically meet in a room to select an overall best leader who has control over all the other leaders. This “Meeting Room Approach (MRA)” ensures a balance between the exploration and exploitation capabilities of the proposed NPO. NPO provides two steps for exploitation part, while the exploration is performed using another step. The local search of NPO is implemented using a unique distribution formula, while the global search ability contains a levy flight equation which generates a step for moving the families towards the new positions. The NPO was tested and evaluated based on sixty unconstrained benchmark test functions. Additionally, the scalability of the NPO was evaluated by solving eighteen large-scale problems. The experimental results confirmed that the proposed NPO performed better than some of the recent metaheuristics in terms of achieving the best solutions, scalability, time complexity, and convergence rate. The NPO successfully solved 52 out of 60 (86.6%) normal sized test functions while 16 out of 18 (88.8%) large-scale problems were equally solved. Good performances were also achieved with the NPO with respect to noise and limited information problems. A Wilcoxon Signed-Rank Test was performed to measure the pair-wise statistical performances of the algorithms and from the results, NPO recorded a better statistical performance compared to the other benchmarking algorithms. Conclusively, the experimental and statistical evaluations performed in this study proved the capability of the developed NPO in solving real-world optimization problems.