摘要
Neural network has the abilities of self-studying, self-adapting, fault tolerance and generalization. But there are some defaults in its basic algorithm, such as low convergence speed, local extremes, and uncertain number of implied layer and implied notes. This paper presents a solution for overcoming these shortages from two aspects. One is to adopt principle component analysis to select study samples and make some of them contain sample characteristics as many as possible, the other is to train the network using Levenberg-Marquardt backward propagation algorithm. This new method was proved to be valid and practicable in site selection of practical garbage power generation plants.
Neural network has the abilities of self-studying, self-adapting, fault tolerance and generalization. But there are some defaults in its basic algorithm, such as low convergence speed, local extremes, and uncertain number of implied layer and implied notes. This paper presents a solution for overcoming these shortages from two aspects. One is to adopt principle component analysis to select study samples and make some of them contain sample characteristics as many as possible, the other is to train the network using Levenberg-Marquardt backward propagation algorithm. This new method was proved to be valid and practicable in site selection of practical garbage power generation plants.
基金
This paper is about a project financed by the Research Fund for Doctoral Program of Higher Education (No. 20040079008).
