Research On Neural Network Optimizing By Particle Swarm Optimization Algorithm And Ensembles

Generalization ability, learning efficiency and ease of use are the three key challenges in the process of machine learning and its application. Through training multiple neural networks and the results into the synthesis, neural network ensemble learning significantly improve the generalization ability of the system, and the field of machine learning in recent years become an important research direction.In this paper, the method of selecting a number of the particle swarm algorithm and BP neural network algorithm and the algorithm of neural network ensemble are selected, and then researched. Based on the in-depth analysis and research on the existing algorithms and combined with the characteristics of the data, get the improved algorithm. A comparison test is done between the coronary heart disease data and the UCI data using the improved algorithm, achieved good results. The main work is divided into the following areas:(1)An effective way to determine the number of hidden layer nodes of BP network is proposed. The number of nodes in hidden layer of BP neural network has been a problem so far, and there is not a clear formula can calculate it till now. In this paper, through a special encoding and using a particle swarm algorithm of global search ability to optimize the hidden layer of BP network and the number of nodes in the network weights and threshold. Experimental results show that the network training methods can not only determine the number of hidden nodes in the weights and threshold value adjustment process, but also to further improve the network's learning ability.(2)It's the first time to use the optimal number of hidden nodes neural network for particle swarm neural network ensemble learning. One of the most important conditions for the integration of the individual networks is have enough difference. As the number of hidden nodes with the optimization of particle swarm neural network training process in the network to find the dynamic of the number of nodes in the network hidden layer weights and the best combination of thresholds, different initial state weights and threshold values corresponding to different number of hidden nodes, which makes more diversity between the individual networks. With the optimization of the number of hidden nodes neural network to integrate PSO will further improve the network generalization ability.(3)The existing methods of coronary heart disease clinics was analyzed, and with the optimization of the number of hidden nodes used in neural networks particle swarm was integrated into the treatment of coronary heart disease. Analyzing and mining the clinical data collected from Beijing Xiyuan Hospital and other three medical institutions, proved that with the optimization of the number of nodes of hidden layer neural network ensemble particle swarm can be used in diagnosis and treatment of coronary heart disease.