Research On Wavelet Neural Network Method And Its Application In Stock Index Futures Prediction

In the international market,Share Price Index Future has become the widespread means of risk management in recent years,Share Price Index Future with the stock index as the subject matter in the specified date for trading.In order to avoid some artificial controllable interference factors,the main analytic target of this essay is rela-tively mature the Shanghai Stock 50 of Share Price Index Future,SSE 50 was compiled by the Shanghai Stock Exchange,and listed all the stocks as the calculation range,the circulation as a comprehensive weights of a series of indices,the stock index reflects the stock sample as a whole price movement,SSE 50 refers to the trend of the super-large Blue Clip,is the stock index futures in the important targeted index,it reflects the overall trend of the Security exchange market,because the Shanghai stock index is a combination of a variety of factors to form,even if there is a stock is controlled by speculation,generally,as long as most of the stocks in line with the market law,The Shanghai stock index can still reflect the laws of the stock market.Therefore,the research on SSE 50J s forecasting method is of great significance to the development of national economy.At present,there are many forecasting methods of Share Price Index Future,with the development of the market of Share Price Index Future more and more complex,through a single neural network forecasting technology is not enough to achieve peo-ple's ideal results.Wavelet analysis is a new frontier research field developed in the past 30 years,which is an epoch-making development after Fourier analysis,wavelet technology can be combined with neural network to predict because of its own char-acteristics,which can make each performance more superior,and effectively avoid the shortcomings in the neural network model,its appearance provides an effective method for many fields,not only for the basic theory of mathematics,and it has had a profound impact on the scientific and technological circles and the engineering circles.In addi-tion,the research of Wavelet neural network has developed rapidly in recent years,and this kind of intelligent synthesis research combines two or more methods,which can synthetically take advantages of the benefits of multiple approaches,or can complement them to form a new method.The essay studies mainly the forecasting effect of wavelet and neural networks combined with different forms,in which the neural network se-lects BP Neural network,combines wavelet and BP neural networks through means of disconnect-type and embed-type,and uses SSE 50 for empirical analysis and com-parison.Firstly,introducing the disconnect-type neural network,it uses Multiwavelet for noise-suppressed processing,and then uses neural network to train and learn the denoising signal,in which Multiwavelet is transformed in different dimensions by mul-tiple wavelet,and a lot of information can be excavated,which is combined with neural network to make the training more adequate and more accurate.The network structure is relatively stable.Secondly,with regard to embedded type,this paper proposes that the two types of single hidden layer and double hidden layer are combined with single wavelet and double wavelet respectively,and compares the predictive effects of these four structural models,and the combination of multiwavelet and double hidden layer neural networks by introducing multi hidden layers effectively increases the prediction accuracy.The dual-implicit layer network structure is proposed,and the double-implicit layer single wavelet and Multiwavelet network model is constructed accordingly,which changes the learning mode of wavelet neural network,and can retain all the advantages of the previous small wave neural network models.In addition,the momentum factor is added to the model optimization process.In order to make the network converge faster and avoid obtaining the local optimal solution.In this paper,five models through SSE 50 for empirical analysis,of which SSE 50 selected 256 trading days of data,the article with SSE 50 refers to the opening price,maximum price,bottom price,turnover,the total amount of 5 indicators as input to analyze,through the synthetical analysis of the stock market,Closing price is the final data of a day of multi-party fighting,but also the basis of the next day's opening price,it can depict and evaluate the trend of the day,so the output choose the closing price to predict.The final test results show that among the five models,the embedded type double hidden Layer multiwavelet Neural network prediction model has the strongest generalization ability,the best prediction effect,and the learning speed has been effectively improved,so as to compare and analyze the performance and advantages and disadvantages of each model,the proposed model has played a very important application research value to the stock market forecast research,And it is of great strategic significance to promote the development of market economy in China.