The Research And Applications Of Improved Artificial Fish Swarm Algorithm

Swarm intelligence gradually becomes popular on the optimization problems because of the self-organized ability,and the advanced functionality based on swarm social activities.Artificial Fish Swarm algorithm(AFSA)is one of the swarm intelligence algorithms,it is capable of escaping local optimum and having better exploration ability.However,the historical optimum record cannot be properly used,and the exploitation ability is not good enough.As the result,this thesis aims at the improvements for AFSA,and implement the improved algorithms to the application of function optimization,data clustering and multithreshold image segmentation.The main works of this thesis are as follow:1.For the function optimization problems,a teaching-learning and differential communication based artificial fish swarm algorithm(TDAFSA)is proposed to overcome the drawbacks of AFSA.With the differential communication,individuals' interactions among the swarm strengthened stronger,and the teaching-learning process properly changes individuals with improvements relying on the historical optimum record.Through the simulation on 12 functions with other 2 kinds of swarm algorithms,it's obvious that the proposed algorithm is capable with better ability for exploitation and escaping local optimum,as well as faster convergence rate.2.The thesis proposes a hybrid algorithm based on an improved artificial fish swarm algorithm and Kmeans algorithm for data clustering.This hybrid algorithm combines TDAFSA and Kmeans algorithm dynamically,which overcomes the disadvantages of Kmeans algorithm that are sensitive to the initial clusters and easy to be stuck in the local optimum.This hybrid algorithm is tested on 6 datasets and compared with several other algorithms,and the results show that it has higher correct rate and performs stably in data clustering.3.The thesis proposes an improved AFSA based on crossover and mutate behavior for image segmentation using multilevel thresholding.The crossover behavior in the proposed algorithm improves the exploitation ability,and the mutate behavior increases the diversity of individual among the swarm.As the result,the proposed algorithm makes better balance between exploration and exploitation ability.Compared with other two kinds of swarm algorithms,the simulations results on 4 grey images show that the multilevel thresholding selection based on the proposed algorithm loses less image pixels and has better quality of segmented images.