Research On Depth Feature Extraction And Prediction Of Stock Information In Multivariate Time Series

The stock market has attracted the close attention of the majority of investors because of its characteristics of high risk and high return.In order to gain profits from the stock market,a set of scientific investment methods become the urgent demand of the majority of investors.With the further research of researchers,two kinds of methods of forecasting financial series have been formed.One is the analytical method in the financial field,and the other is the analytical method using computer technology.These methods can not only help investors gain profits,but also avoid risks,thus playing a good warning role in the financial market.In recent years,with the rapid development of deep learning,deep learning has been applied in various practical scenarios,including financial market prediction research.Since stocks are typical time series data,this paper chooses LSTM(Long Short Term Memory Network)to study the prediction of some stocks in China,and introduces Convolutional Neural Network to improve and optimize the model and compare the performance of different models in the stock prediction effect.For stock prediction for ever only adopted the opening price,closing price,the highest and the lowest price and the volume of a few features as the basis of stock prediction,this paper on the basis of innovation,introduced a variety of stock features as an indicator of stock prediction,and on this basis,the LSTM is put forward to combine with CNN and apply in stock prediction.In this paper,the LSTM model and 50 features of individual stocks are firstly used to predict the rise and fall.Dropout is introduced in the prediction process to prevent overfitting of the model.Meanwhile,in order to improve the training speed of the model,Adam optimization algorithm is introduced to optimize the model parameters.In order to compare the training effect of the model,MSE loss function is introduced to calculate the model loss.In order to further improve the prediction effect,this paper makes an innovation based on the LSTM model and adds a CNN model.The advantage of the former is that it has good memory ability,can remember the previous features and infer the later results,while the latter has advantages in feature extraction.CNN can automatically extract highlevel features and handle high-dimensional data without pressure.Based on LSTM model,LSTM-CNN model is proposed,and 230 kinds of features are introduced for prediction.The prediction effects of LSTM-CNN model and LSTM model were compared.The prediction effects of LSTM-CNN model using 50 features and 230 features were compared.Through experiments,it is found that the prediction effect of LSTM-CNN model is better than that of LSTM model,while the prediction effect of 230 stock features of LSTM-CNN model is better than that of 50 stock features of LSTM-CNN model.