Many problems bring into solving optimization tasks. Along with the developments of natural science, modern optimal problems become more and more complex. Because of the complexity of optimization problems, traditional optimization methods are inadequate for solving these problems. Therefore, the new methods are needed for these problems. Much more universal, about the kinds of solved problems, are the algorithms motivated by simulating biology intelligent. The only assumption needed by those algorithms is the existence of the solution. Those simulating biology intelligent algorithms can solve complex optimization problems adequately and robustly. Those algorithms are parallel algorithms. Therefore, after initial invented of these algorithms many studies have been carried out on these algorithms and applied for solving TSP problem, neural network training, image processing and others. Artificial bee colony (ABC) algorithm has been proposed in 2005, which is a simulating biology intelligent algorithm. Since its invention, ABC algorithm has been applied in many areas. However, there still an insufficiency in ABC algorithm as other intelligent optimization algorithms, such as lower convergence speed for uni-modal problems and easily trapped in local optimal for complex multimodal problems. And the theoretical analysis is lacking. In this dissertation, some improved artificial bee colony algorithms are proposed to overcome these drawbacks, and applied for numerical function optimization, constrained optimization and non-negative linear least square problem. Meanwhile, a martingale analysis method is proposed to study the almost sure convergence of ABC algorithm. The main researches and creative results of this dissertation are shown as follows.1. The convergence analysis is proposed for artificial bee colony algorithm. The most convergence analysis on artificial bee colony algorithm is based on ergodicity analysis and conducted in the sense of probabilistic convergence. Such analysis cannot infer in general that the ABC algorithm would be convergent to a global optimum in a finite number of evolution steps. In this paper, a martingale analysis method is proposed to study the almost sure convergence of ABC algorithm. It is shown that ABC algorithm can surely converge to a global optimum with probability 1 in a finite number of evolution steps. The obtained results underlies application of the ABC algorithm, and the suggested martingale analysis method provides a new technique for convergence analysis of ABC algorithm.2. A hybrid artificial bee colony algorithm is proposed for global optimization problems. To further improve the performance of artificial bee colony algorithm (ABC), a hybrid ABC (HABC) for global optimization is proposed via exploring an orthogonal initialization. Furthermore, to maintain population diversity, a novel search strategy is also developed. The algorithm is applied to benchmark functions with various dimensions to verify its performance. Numerical results demonstrate that the proposed algorithm outperforms the ABC in global optimization problems.3. An improved artificial bee colony algorithm is proposed for numerical function optimization. To further improve the performance of artificial bee colony algorithm (ABC), an improved ABC (IABC) algorithm is proposed for global optimization via employingorthogonal initialization method. Furthermore, to balance the exploration and exploitation abilities, motivated by differential evolution, a new search mechanism is also designed. The performance of this algorithm isverified by using benchmark functions. And the comparison analyses are given between the proposed algorithm and other nature-inspired algorithms. Numerical results demonstrate that the proposed algorithm outperforms the original ABC algorithm and other algorithms fornumerical function optimization problems.4. An effective hybrid artificial bee colony algorithm is proposed in this paper for non-negativelinear least squares problems. To further improve the performance of algorithm, orthogonal initialization method is employed to generate the initial swarm. Furthermore, to balance the exploration and exploitation abilities, motivated by particle swarm optimization and differential evolution algorithm, a new search mechanism is designed. The performance of this algorithm isverified by using benchmark functions and non-negative linear least squares test problems. And the comparison analyses are given between the proposed algorithm and other swarm intelligence algorithms. Numerical results demonstrate that the proposed algorithm obtained better results than other algorithms for global optimization problems and non-negativelinear least squares problems.5. In this paper, an effective chaotic artificial bee colony approach is proposed to global optimization. And the proposed approach is applied to non-negativelinear least squares problems. To overcome the insufficiency in artificial bee colony algorithm, opposition-based learning initialization method is employed to generate the initial swarm. To further improve the performance of algorithm and to balance the exploration and exploitation abilities, a new search mechanism is designed. Furthermore, a new chaotic local search operator is embedded in algorithm, which can do the local search around the best solution. The performance of this algorithm isverified by using benchmark functions and non-negative linear least squares test problems. Numerical results demonstrate that the proposed algorithm obtained better results than other algorithms for global optimization problems and non-negativelinear least squares problems, which shows the feasibility, effectiveness and robustness of proposed algorithm.6. An algorithm portfolio (AP) based on evolutionary strategy (ES) and artificial bee colony (ABC) algorithm is proposed for constrained optimization problems. Although a wide range of constraints dealing methods have been developed and studied, the performance of an algorithm with different constraints dealing methods may vary significantly from problem to problem. Instead of choosing one algorithm and one constraints dealing method and investing the entire time in it, it would be less risky to distribute computing time in different algorithms and constraint dealing methods. Based on this idea, a EAs algorithm with adaptive trade off model constraint dealing method and a swarm intelligent algorithm with Debâ€™s rules are employed in AP. Moreover, a migration phase is designed to reflect the interactions between two algorithms. The new method is tested on well-known benchmark functions, and the empirical results suggest it outperforms or performs similarly to other state-of-the-art algorithms. In addition, a parameter analysis is conducted and appropriate values for each parameter are obtained. |