Research On The Improvement Of BP Neural Network Key Technology Based On Self-adaptive Approaches And Its Application

With the research on BP neural network deepening gradually,how to optimize the structure of network,increase the training of looking for the optimization and improve the ability of generalization in BP neural network have largely attracted people's attention.Many scholars have tried to use the genetic algorithm,cloud computing and other methods to make a lot of theoretical practice for many issues,such as BP neural network sample collection,the determine of network structure,falling into the local optimal easily,slow convergence speed,and the stability of the neural network.They also have made some improvements on BP neural network so that they can do accurate and fast prediction and analysis,which improve the BP neural network optimization training performance.In this paper,we study the BP neural network optimization training ability mainly from two aspects.On the one hand,we study the learning rule of BP neural network.We propose a variable step gradient algorithm based on self-adaptive method,which puts forward the self-adaptive mathematical model and solves the problem of slow convergence speed near the minimum point and also improves the generalization ability of the network learning.On the other hand,we study the feature selection of the input sample and the change of the weight between the network nodes.A KD-BP neural network feature selection and optimization algorithm based on the self-adaptive method is proposed.This algorithm uses the k-nearest neighbor method to classify the samples and optimize the structure of the input sample.It establishes the weight impulse model,which solves the problem of falling into the local minimum easily.The research shows that the algorithm proposed in this paper effectively improves the generalization ability of the node selection and the reasoning efficiency of the classification feature.The proposed algorithm in this paper is applied to the fault diagnosis of the transformer,which effectively solves the problem of the slow speed of figuring out the causes of different faults of the transformer,improves the analysis speed of the faults' causes.And it narrows the scale of fault diagnosis quickly and proposes reasonable maintenance plans for the transformer timely and accurately.