Shanghai Stock Index Prediction Based On Improved BP Neural Network

As an investment region of high risk and high profit,stock market always attracts many attentions. How to obtain profit through analyzing and forecasting the share price accurately keeps being a hot topic in both academic field and practice.With the developing of Fractal Market Hypothesis (FMH) theory,people have tried to forecast the change of the Stock market. The neural network is an important technology in the field of AI,which has been great achievement in two academic fields with the help of neural network,the approximating ability of random continuous maps by neural network and the analysis of the stability of a dynamic network. In addition , the application of neutral network has been expended into many areas,for example, pattern identification,processing,control and improvement of pictures, identification of received calls in the ATM network and so on.BP (Back Propagation) is a neural network which is adopted widely. The core is the BP arithmetic a strict and effective method to derivative problem for system based on multi-subsystem,such a method has simple configuration and mature arithmetic. Compared with the traditional statistical regress method,BP network can be used not only to study the example of training set,but also abstract some general theories and rules. It has strong characteristics of approximation of non-linear functions, which is much appropriate for short-range forecast of stock index.But BP algorithm's astringency is not good enough and easily trapped in local minimum. The result of some forecasting base on ANN is inaccurate and lost in a long-range easily. This article tries to seek networks ensemble based on BP algorithm,associating with GA algorithm,for Stock Index Forecasting. Meanwhile this improved BP network does make some improvement to the original forecast method according to the limitationand disadvantage of the BP network. Compared with the forecasting made by other ANN models,it turns out that such improvements result in better precision and a long-range forecasting ability.This article introduces the fundamental of ANN-forecasting,BP-based neural networks aided by GA algorithm and cooperating network ensemble. Then there is some analysis and consideration with three times forecasting by the new model. At the last,the author would point out the limitation of the ANN method which can not be improved.