An experimental study of modified black hole algorithms
Optimization is often required in many fields of research. Some systems are harder to optimize compared to others. For this reason, many meta-heuristic optimization methodshave been devised and improved. One of the meta-heuristic optimization methods is Black Hole (BH) algorithm, which is inspired b...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/26094/ http://umpir.ump.edu.my/id/eprint/26094/ http://umpir.ump.edu.my/id/eprint/26094/1/An%20experimental%20study%20of%20modified%20black%20hole.pdf |
| Summary: | Optimization is often required in many fields of research. Some systems are harder to optimize compared to others. For this reason, many meta-heuristic optimization methodshave been devised and improved. One of the meta-heuristic optimization methods is Black Hole (BH) algorithm, which is inspired by the black hole in cosmology. The black hole has been found theoretically in the studies of the universe as a star having massive mass and gravity. The BH algorithm is a population-based method which uses more than one agent to find a solution in a search space. In BH algorithm, an agent with the best solution forms a black hole. The biggest mass is given to the agent with the best solution and other agents update their position towards the agent with the best solution. In this thesis, it is found that the BH algorithm suffers from premature convergence. Hence, three methods to prevent the premature convergence in BH algorithm are presented. The first method is the introduction of white hole operator. The white hole operator is proposed to avoid the agents from exploring the area near the worst agent. The second method is the application of a local search. Finally, a gravitational interaction, which is the essence of the gravitational search algorithm, is also applied. Ultimately, seven variants of BH algorithms are established based on the white hole, local search, and gravitational interaction. Those algorithms are Black Hole White Hole Algorithm, Gravitational Black Hole Algorithm, Gravitational Black Hole White Hole Algorithm, Black Hole Local Search Algorithm, Black Hole White Hole Local Search Algorithm, Gravitational Black Hole Local Search Algorithm, and Gravitational Black Hole White Hole Local Search Algorithm. The algorithms are evaluated based on unimodal, multimodal, hybrid, and composite functions in CEC2014 benchmark test functions. Benchmarking with particle swarm optimization (PSO), genetic algorithm (GA), and grey wolf optimizer (GWO) is done as well. Results show that all the variants are as good as PSO and better than the original BH algorithm. Statistical analysis suggests that one variant called Black Hole White Hole algorithm is the best since the Black Hole White Hole algorithm is significantly better than GA, GWO, and BH algorithms. |
|---|