Research On Financial Time Series Prediction Algorithm Based On LSTM Model

Under the background of China's rapid economic development and advancement of science and technology,the influence of financial markets is also growing.The analysis and forecast of financial time series have a great impact on investors' decision-making.Due to the non-linear,high-noise and non-stationary characteristics of financial data,the prediction of financial time series has been fully studied and developed in economics,mathematics and other disciplines.The prediction model has also undergone a transition from linear to nonlinear models.And Fintech is also producedIn order to improve the accuracy of financial time series prediction,this paper proposes a combined time series prediction model combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)algorithm and Long Short Term Memory(LSTM)with attention mechanism.First,the LSTM model is improved based on the attention mechanism.It makes full use of the output information of each moment of the LSTM hidden layer,calculates the attention distribution,and weights the output information.The effectiveness of the proposed improved LSTM-ATTE model is verified by comparison with other models.Then,the Empirical Mode Decomposition(EMD)and CEEMDAN algorithms are studied.The decomposition of the simulated signal shows that the CEEMDAN algorithm can eliminate the modal aliasing phenomenon generated by the EMD algorithm.Finally,based on the above research,a time series combined prediction model based on CEEMDAN algorithm and LSTM-ATTE is proposed.Firstly,the CEEMDAN algorithm is used to decompose the financial time series to obtain a series of sub-sequences with different time scales.Then use the LSTM-ATTE model to predict them separately,and superimpose the prediction results to obtain the final predicted values.At the same time,the actual SSE 50 index closing price sequence is used for verification and prediction.The correctness of the combined model prediction is proved by linear regression analysis.Comparing with single model and other combined models,the results show that the proposed model has smaller prediction error and improves prediction performance.