Stock Price Prediction Model Based On EMD-PCA-GRU Neural Network

With the continuous development of China’s stock market,investors need to change their investment strategies timely and accurately according to the changes of stock prices to avoid risks.Therefore,it is very necessary to conduct research on stock price prediction.Compared with individual stocks,due to the nonlinear,chaotic and long memory characteristics of stock price index,there will inevitably be some noise in most available historical transaction data,and it will become more obvious with the increase of data frequency.In the field of previous forecasts,due to data exist excessive noise or tight forecast effect is not ideal,lead to stock prediction results,but with the continuous development of computer deep learning,more and more people begin to set his sights on forecasting field of artificial intelligence,and deep learning in risk control and quantitative performed very well in the field of investment.Based on the above reasons,combined with empirical mode decomposition(EMD),principal component analysis(PCA)and gated cyclic unit network(GRU),this paper proposes a deep learning composite prediction model EMD-PCA-GRU,which is used to predict financial time series data.Compared with the single model,the composite model can solve the problem of over-fitting the noise.Meanwhile,by integrating the advantages of EMD,PCA and deep learning methods,the multi-scale,nonlinear and complex dynamic characteristics of financial time series can be fully described,so as to enhance the generalization and prediction ability of the prediction model.In order to overcome the inadequacy of traditional time series prediction accuracy and explore the relationship between the accuracy of prediction results between diurnal data and high-frequency data.To verify the accuracy of the model in this paper,the representative daily data of The Shenzhen and Shanghai 300 index from January 1,2017 to August 31,2020 and the closing price of the 5-minute high-frequency data from October 15,2019 to August 31,2020 are selected for prediction,and compared with the LSTM model.The results show that the prediction accuracy of EMD-PCA-GRU model is better than other models.It is proved that the fitting degree of the composite model is higher than that of the single model.Under the same model,the 5-minute high frequency time series has better prediction accuracy.