The Research And Application On BP Neural Network Improvement

As the important subfield and the quintessence of Artificial Neural Network,BP Neural Network accelerated the development in this field.In 1985,Rumelhart and some other scholars advanced the Error Back Propagation theory that was improved as BP Neural Network theory today.BP Neural Network has integrated system,explicit algorithmic process,data identification and simulation function.BP algorithm also owns the excellent ability to solve nonlinear problem,therefore,the value of practical application is outstanding.Along with researching deeply,the defects of BP Neural Network have been found,such as low convergence speed,long training time,falling into local minimum easily,bad generalization ability,few principle to build network structure.These defects can depress the accuracy of BP Neural Network and damage the practical effect.So,improving BP Neural Network step by step is significant not only for theory,but also for practical application.According to the result of BP function experiment,this paper explored the flaws in training network and found the root of various flaws through studying BP theory.In order to conquer the flaws in training,this paper consulted foreign document,then advanced a new method(new adaptive learning rate algorithm)to improve BP training process.The mathematic approaches of this method had been given in this thesis clearly,and the paper realize the whole algorithm with the help of mathematic software.For confirming the practicability,contrasting the new method with the other algorithms is necessary.The comparison indicated that the new way had superiority to modify BP algorithm.This superiority include :simple algorithm process,fast convergence speed,get out local minimum easily,small oscillation and so on.In brief,this new algorithms can make the whole BP training process fast and stable.This paper used new adaptive learning rate algorithm to improve BP Neural Network,then built the total power model of agricultural machinery and the electrical load model for Heilongjiang province.Through computer programming and comparative analysis of other improvement methods,this thesis verified the reliability of new algorithm.The results of application certified:because of the small fitting error and quick training speed ,new adaptive learning rate algorithm has perfect function.At the same time the prediction of this algorithm can provided guides for trade development,so people can use it at many the other domain.Dialectic ideology in this paper is from finding problems to concluding them,finally solving problems.At the same time,this thesis linked BP Neural Network theory ,systems engineering,numerical optimization and computer simulation technology together.Then getting the support from MATLAB7.1 software,this paper finished all kinds of comparison experiment through computer programming and satisfied the request of experimental design.In order to support consultation for deep improvement to BP Neural Network,this thesis tried to discuss the flaws in BP training process and advanced a new algorithm to improve this process.While,because author's knowledge is limited, the research to the BP Neural Network is not sufficient.So,the improvement research to BP Neural Network need to be done constantly.