With the market-oriented economic reform,more and more attention has been paid to the information prediction of financial market.Stock price trend prediction can not only bring investors a lot of forward-looking information to obtain excess returns,but also provide help for asset risk management.Compared with traditional econometric models,increasingly mature machine learning and deep learning techniques have achieved surprising accuracy in many time series prediction studies.Machine learning and deep learning algorithms can quickly process large dimensional data without complicated assumptions before use,and learn relevant features from them to make predictions.At present,relevant machine learning and deep learning algorithms are gradually applied to the prediction of financial time series.Financial time series has the characteristics of nonlinear,non-stationary and low signal-to-noise ratio.The research on finding appropriate financial time series data processing methods and constructing prediction models based on machine learning and deep learning algorithms is not only related to the vital interests of financial market investors,but also related to the overall long-term development of the whole financial market.Therefore,the research on improving the accuracy of financial time series prediction has both theoretical and practical significance.In recent years,the new energy industry has been developing rapidly.As the policy of "double control" and the goal of "double carbon" are proposed,the future energy will be clean,low carbon and high efficiency as the new transformation development requirements.This paper mainly studies the data.of two new energy stocks(BYD 002594,China Energy Electric 300062).Firstly,in view of the low signal-to-noise ratio of financial time series data,the wavelet transform threshold method is selected to reduce the noise of the data,this method is developed according to the requirement of time-frequency localization and is very suitable for processing non-stationary nonlinear time series signals.Secondly,the deep features of the data generated by SAEs are applied for stock data prediction.In the prediction phase,this article will be expanded from the following contents:(1)A single deep learning model(LSTM,TCN)is used for prediction research and the prediction results are analyzed.In the part of prediction using deep learning model,the data before noise reduction and feature extraction is firstly substituted into LSTM and TCN for prediction analysis.(2)Combine the data processing algorithm and deep learning algorithm to form a multi-algorithm hybrid prediction model based on WT-SAEs,and make a comparative analysis between the predicted results and the previous results.(3)On the basis of the above,attention mechanism is introduced to improve the accuracy of the model.The empirical study selected the price data of three stocks.The empirical results show that,based on wavelet transform denoising and stack autoencoder feature extraction,the mixed deep learning model is better than the single depth model,and TCN is better than LSTM in the performance of stock price prediction. |