Improvement Of Particle Swarm Optimization Algorithm And Application On Gene Expression Data Clustering

As a new branch of evolution algorithm, particle swarm optimization algorithm is an intelligence algorithm, which simulates the swarm behavior of birds. It is simple and can be carried out easily, so it has been widely concerned since proposed. However, as the shorter development history and the less theoretical researching, particle swarm optimization algorithm still has some problems in dealing with some issues. As a result, the improvement of particle swarm optimization algorithm and new areas of its application become central issues during optimization field.Firstly, an improved particle swarm optimization algorithm based on particles'growth stages is propsed. This algorithm uses mutation mechanism to make particles jump out of the local optimal effectively, and uses the growth process of human to define the particles. Division of growth stages can make particles have different learning factors at different stages. The improved algorithm is tested by a group of Benchmark functions to verify its validity.Secondly, another improved particle swarm optimization algorithm based on double mutation is proposed. The main measures of avoiding premature convergence are keeping the population diversity or introducing mechanisms to jump out of local optima. So the improved algorithm adopts two strategies to improve the original PSO, one is adjusting inertia weight dynamically and the other is introducing mutations. The adaptive mutations make particles jump out of local optima effectively, and adjusting to the inertia weight dynamically and nonlinearly is beneficial to balance particles'ability of exploitation and exploration. This improved algorithm is tested by a group of Benchmark functions to verify its validity.Finally, the application of particle swarm optimization algorithm in gene expression data clustering is discussed. And then a hybrid clustering algorithm is proposed, which integrates the particle swarm optimization algorithm into K-means algorithm, to overcome the disadvantage of K-means algorithm. The wide-known yeast-cells gene datas are used to test the validity of the hybrid clustering algorithm.