Plants flower pollination algorithm(FPA) is a novel meta-heuristic algorithm inspired by the pollination behavior of plant flowers in nature. The local search and global search process of the algorithm corresponding to the self-pollination and cross-pollination process respectively, and the strength of these two form of search balanced by stochastic disturbance. Because of its simple structure, strong robustness, excellent search ability and some other characters, it has been researched by many scholars to solve various complex combinational optimization problems. While it also has drawbacks, like poor performance in precision and weak ability to jump out of the local optimal, those all limit the application range of plants flower pollination algorithm.Aiming at improving deficiencies existed in FPA, this thesis will conduct deep research to FPA with the purpose of expanding its theory and application range, and several classical optimization problems are solved by proposed methods. In summary, the works in this thesis can be categorized into the following four aspects.1. The mechanism named elite opposition based learning is introduced to expand the search space of the algorithm, and the diversity of the pollen population has also been enhanced. At the same time, local self-adaptive operator is used to strengthen its exploitation, which may improve the optimization accuracy. This variant algorithm named elite opposition-based flower pollination algorithm, which improves the solution in precision, convergence speed and could avoid premature. In other word, it improves the performance of basic FPA comprehensively.2. Aiming at solving the problem, those generated by conventional optimization methods for cluster analysis, like trapping into local optimal and low execution efficiency. Here the discard behavior which selected from artificial bee optimization is combined with simplex method to jump out of the local optimal. Simultaneously, proposed method omits the operation of computing the cluster center coordinates in k-means algorithm and improves its execution efficiency. These two improvements enhance the clustering ability of the basic FPA.3. For eliminating the conflicts of different dimensions, the idea named dimension by dimension evolution is introduced. And the lesson from particle swarm optimization is drawn in its self-pollination process to promote the information exchange between pollens. In other words, this mechanism makes full use of the surrounding information. Furthermore, a novel mathematical modal for UUV path planning for 3D space is proposed. At last, the proposed method is chosen to solve the path planning problem in 2D and 3D spaces, simulation results validate that proposed method improved the precision and convergence speed of path planning significantly.4. Modified randomized location strategy is selected to disperse the pollens in search space widely. The global exploring ability of the method is enhanced and the balance between self-pollination and cross-pollination is improved. And our improvement also include cross operator which help to increase the diversity of population. A variant method named MRLFPA based these two strategies to solve medical image segmentation problem. Simulation results indicate the promotion of MRLFPA in segmentation efficiency. |