Research On BiLSTM Stock Price Prediction Based On Quadratic Decomposition And Attention Mechanism

With the continuous development of Chinese economy,the development status and operating conditions of financial market have great influence on the macro-economic policy and the optimization of resource allocation,so the change of financial market is paid attention by domestic and foreign scholars.Appropriate financial market forecasting and economic situation management can offer specific data assistance for economic strategy.An essential component of the financial industry is the stock market.The research on the stock market forecast can reflect the macroeconomic operation of a certain country or industry and estimate the risks in advance.Due to the complexity and variability of stock market,the traditional time series prediction model has low accuracy and weak generalization ability.To address this issue,many scholars have applied artificial intelligence fields such as machine learning and deep learning to stock price prediction,which can make more accurate judgments under the characteristics of high noise,non-stationary and nonlinear stock price time series.This paper applies signal decomposition method and deep learning to stock price prediction to provide certain enlightenment for all parties in the stock market.Taking four representative stock closing price data as the research object,this paper proposes an EMD-AMBi LSTM stock price prediction method.First,the closing price time series was decomposed by empirical mode decomposition to transform the original time series into subseries,decreasing the non-stationary of time series and extract the features of time series.In order to improve the operational efficiency of the deep learning algorithm,the subseries was normalized.Secondly,AM-Bi LSTM prediction normalized subsequences were constructed respectively.Finally,the prediction results of the subsequence are superimposed as the final predicted value of the stock closing price.Among them,when AM-Bi LSTM prediction is made for each subsequence,it is discovered that the model for IMF1 has a poor level of prediction accuracy.In order to further improve the prediction accuracy of IMF1 and closing price,it is considered to continue the ensemble empirical mode decomposition of IMF1,and the EMD-EEMD-AMBi LSTM stock price prediction model was proposed,that is,after empirical mode decomposition of time series,ensemble empirical mode decomposition of IMF1 was carried out.AM-Bi LSTM was used to predict each sub-sequence,and the predicted results were superimposed as the final predicted value.In order to evaluate the predictive performance of the proposed model,this paper constructs a series of models for comparative experiments,including SVR,Decision Tree,LSTM,Bi LSTM,AM-Bi LSTM and EMD-LSTM,and uses the constructed models to predict four stock closing price sequences.The root mean square error(RMSE),mean absolute error(MAE),mean absolute percentage error(MAPE)and determination coefficient(R~2)were selected as evaluation indexes to compare the merits and demerits of the model.The experimental results demonstrate that EMD-EEMD-AMBi LSTM has a good performance on the four data sets,and secondary decomposition has higher accuracy than primary decomposition,which can accurately predict the stock price.