Financial Time Series Prediction Based On LSTM Hybrid Model

Mining the movement rules of the financial market and accurately predicting the trend of the financial market will help financial investors to formulate low risk and high return investment strategies.Therefore,the related research of financial time series prediction has been carried out and attracted much attention.However,financial data,as a kind of time series data,has the characteristics of non-linear,non-stationary and high noise,which makes financial time series prediction recognized as the most challenging task in the world.How to accurately judge the trend of the stock market is a difficult problem that financial researchers have been studying but not completely solved.With the breakthrough of deep learning in all walks of life,more and more financial researchers have applied it to the study of financial time series prediction.Based on the deep learning method,this paper studies the prediction of financial time series,takes the stock financial data as the prediction object and the closing price as the prediction target,designs and implements two kinds of time series prediction models,singlestream prediction model and double-stream prediction model.Among them,each features stream is realized based on LSTM hybrid model.The single-stream prediction model is mainly used to predict the stock price index of the financial market.The model is composed of the self-trend stream based on the WDAE-LSTM hybrid model.In this model,wavelet noise reduction module,noise reduction self-encoder module and long and short-term memory network module are combined to extract the trend characteristics of stock price index series in financial market.The experimental results show that compared with other similar prediction models,the single-stream prediction model designed in this paper performs the best in the prediction of financial stock index series and is more valuable in the practical financial analysis.The double-stream prediction model mainly forecasts the price of a single stock.This model is realized by introducing the mutual-trend stream on the basis of the single stream prediction model.The function of the mutual-trend stream is to extract the relevant trend characteristics from the multi-branch financial time series by using the WPCA-LSTM hybrid model composed of the wavelet principal component analysis noise reduction module and the long and short term memory network module.The double-stream prediction model improves the prediction accuracy of a single stock price series by integrating its own trend characteristics with related trend characteristics.The experimental results show that the prediction performance of the double-stream model for a single stock price series is significantly improved compared with that of the single-stream network model.Finally,the double-stream prediction model is applied to the prediction of other time series in this paper.The experimental results show that the s double-stream prediction model for prediction of other time series data of the prediction performance is good.It means that the double-stream prediction model designed in this paper has certain universality and can be applied to a variety of time series prediction and analysis.