The Research And Application On Bp Neural Network

The BP neural network as one of the most popular models of artificial neural network, its model is simple, strong plasticity and other characteristics, which makes the BP neural network has been widely used in many fields.BP neural network is able to for any nonlinear mapping approximation, with learning and adaptive to arbitrary precision, can be connected through modifying the network value changes occur in response to the system, the multiple input multiple output structure model, can adapt to the characteristics of multi variable control system, which is widely used in data fitting and prediction of BP neural network, but there are still many shortcomings.This paper mainly carries on the detailed research to the BP neural network,the analysis of excellent performance of BP neural network and some inherent defects, such as: network operation speed is slow, easy to produce the fluctuation and when there is a big gap between the number of level input data prediction accuracy of inaccurate etc.This paper is aiming at the above problems,to improve the BP neural network.In this paper,the main research content of the BP neural network are as follows:(1)Details of BP neural network model and its working principle and the BP algorithm is derived in detail,and the performance of BP neural network indicates the excellent performance of BP algorithm and some defects analysis.(2)The shortcomings of the standard BP algorithm is analyzed,and points out that the causes of the defects existing in the standard BP neural network.(3)Aiming at the shortcoming of BP neural network,research for improvement of the standard BP algorithm,the improved method,especially put forward an improved BP algorithm with a relative error as network error transfer signal.(4)Application of MATLAB software to the standard BP algorithm,BP algorithm and the improved simulation application respectively and make comparative analysis,the experimental results show that,the improved BP algorithm the existence of large magnitude relationship in the processing of input data than the standard BP algorithm has obvious advantage.