Research On Stock Price Prediction Model Based On Deep Learning

Under the background of continuous improvement of the domestic market economic system,the influence of the stock market is increasingly large,and the position in our economic system is more and more important,and the number of people participating in investment also continues to increase.The fluctuation of stock price is closely related to the stable development of China’s market economy,and also affects every aspect of social life.Therefore,scholars all over the world have been devoted to the research of stock price prediction.However,the stock market is a nonlinear system with complicated variable relations,and the stock data has the characteristics of nonlinear,time-varying and high volatility.And the traditional time series prediction model can only find the linear relationship between the data,which has certain limitations and can not get the ideal analysis result.Therefore,how to predict its future price or trend more accurately is a topic worth studying,and can also provide investors with good investment reference suggestions.In recent years,a large number of researchers have adopted a single neural network model for stock prediction,but few examples have been studied with mixed models.Therefore,to solve the above problems,this paper proposes a CNN-ATLG hybrid stock price prediction research model based on deep learning theory.In this paper,two traditional neural network models are combined to make experimental comparison,and the basic model LSTM-GRU with the best performance is selected as the initial model.Secondly,in the input layer,the powerful feature processing ability of CNN is used to extract the deep features of the stock time series data and input them into the initial model.At the same time,the Self-Attention mechanism is used to assign different weights to the hidden layers of the model to solve the problem that the neural network model is prone to information loss.The final experimental results show that the CNN-ATLG model has low prediction error and good fitting effect,and the prediction accuracy is improved to 64.78%,which further improves the accuracy of stock price prediction.In order to verify the real validity of the model,this paper also brings the predicted results of the model into the stock market,tests its application in the actual market,and confirms its practicability.After simulating initial funds,three representative backtest indicators,namely DMI strategy,MACD strategy and KDJ strategy,were selected to verify the return of backtest.Through comparative analysis,the returns of the three strategies are 18.93%,24.81% and19.48% respectively,which are generally higher than the returns of CSI 300 Index in the same period,proving that quantitative trading based on deep learning prediction results combined with trading strategies can be regarded as an effective investment scheme.It shows that the article has certain practical significance and can provide reference for related researchers.