Research On Hybrid Explosion-based Artificial Bee Colony Algorithm And Its Applications

Optimization problems exit widely and Artificial Bee Colony(ABC) algorithm is a new idea and method for complex optimization problems. ABC is a new swarm intelligence technique inspired by the intelligent foraging behavior of honey bees. ABC is simple in concept, few in parameters, easy for implementation, more effective for continuous optimization problems and discrete optimization problems, and it has broad application prospects. Therefore, ABC has attracted widespread attention and has been successfully applied to different fields. However, the research on ABC and its applications are still at the starting stage. There are still many key problems to be solved, mainly including weak theoretical basis, imbalance between exploitation and exploration and slow convergence of ABC. Considering the disadvantages of ABC and advantages of good point set theory, Grenade Explosion Method(GEM) and Cauchy distribution, several new hybrid ABC algorithms are proposed to effectively improve the performance of classical ABC and solve more engineering problems and management problems. Concretely, based on the main clue of ABC, good point set theory is used to generate initial population in order to maintain large population diversity; GEM is embedded in the employed or onlooker bees’ phase so as to enhance exploitation ability; Cauchy distribution is introduced into the scout bees’ phase to further enhance exploration ability. Furthermore, some benchmark functions, multiobjective optimization problems, economic load dispatch problems in power systems and Traveling Salesman Problems(TSP) are used to evaluate the feasibility and effectiveness of the proposed algorithms and support decision makers to make right decisions. The main contributions of this dissertation are summarized as follows:1. The theoretical basis of bee colony algorithms is preliminary studied. It consists of essential theories and complementary theories. The essential theories include self-organization theory and division of labor theory, while the complementary theories provide supplementary explanations on the essential theories.2. The biological mechanism, computing framework, main steps, time complexity and convergence characteristic of ABC are analyzed, and then the effect of scout bees on the performance of ABC are emphatically studied. Almost all the experimental results show that ABC with multi-scout bees outperforms ABC with single scout bee and ABC with single scout bee performs better than ABC without scout bee on five classical benchmark functions. Besides, the three algorithms have almost the same execution time under the same conditions due to the same order of their search complexity. These fully demonstrate that the random exploration process adopted by scout bees has positive effect on the performance of ABC.3. To maintain large initial population diversity of ABC, the effectiveness of the two original methods generating good point set is analyzed according to the characteristics of their functions. Thereby, a new method is proposed. The three methods generating good point set and the random method are used to provide initial population for ABC on six well-known benchmark functions. The results indicate that the new method is much better than the two original methods in all the cases, and it works best on the 10-dimensional benchmarks, but performs slightly worse than the random method on the 50-dimensional and 100-dimensional functions. Furthermore, according to the concept of Pareto dominance, MABC and MABCJCA based on ABC and ABCJC A are proposed for multiobjective optimization problems, respectively. On four classical multiobjective benchmark functions, MABCJCA outperforms MABC but performs slightly worse than several popular algorithms. These suggest that ABCJCA and MABCJCA are effective.However, the good point set is not randomly selected and it will in?uence the global search capability of ABCJCA and MABCJCA. Therefore, it is very important to maintain randomness in an algorithm. In addition, different algorithms may achieve different performance under different p values, so the value of p is needed to adjust for better performance.4. Two modified versions of ABC inspired by GEM, namely GABC1 and GABC2, are firstly proposed to enhance exploitation ability of ABC. GEM is embedded in the employed bees’ phase of GABC1, whereas it is embedded in the onlooker bees’ phase of GABC2. The experiments show that GABC1 has similar or better performance than GABC2 in most cases, but GABC2 performs more robust and effective than GABC1, and they significantly outperform the competitors on two sets of benchmark functions.5. To improve exploration ability of ABC, a novel ABC combined with GEM and Cauchy mutation operator, namely ABCGC, is proposed. In ABC GC, Cauchy mutation operator is introduced into the scout bees’ phase of GABC2. The experiments confirm that ABC GC is significantly superior to the competitors on two sets of benchmark functions, especially it converges to the global optimum faster in most cases. Besides, ABCJCA, GABC2 and ABC GC are used to sovle an economic load dispatch problem in power systems. The three algorithms can provides reasonable solutions for 6-Unit system and ABCGC performs best. These results suggest that ABCGC usually achieves a good balance between exploitation and exploration and can effectively serve as an alternative for global optimization.6. Based on the mathematical model of ABC, three ABC algorithms for TSP are proposed. In each cycle of the three algorithms, an employed bee discovers or an onlooker bee selects a new candidate food source by randomly selecting two positions and interchanging them. Thereby, six hybrid explosion-based ABC algorithms for TSP are designed. In each cycle of the six hybrid algorithms, a new candidate food source is generated by GEM and updated based on minimal path operation. Six TSP instances are used to validate the much better performance of the nine proposed algorithms than the compared algorithms. Furthermore, six stochastic maps were applied to verify the universality of ABC2 G for TSP. In addition, the shortest path length for Dantzig42 is 688.31 obtained by its initial path map and signi?cantly outperforms the known shortest path length 699.