Research On Stock Price Trend Prediction Of Kweichow Moutai Based On LSTM Hybrid Model

In the current environment,the government still proposed in the work report that China’s GDP growth target for 2022 is set at about 5.5%,which reflects the country’s determination and confidence to vigorously develop the Chinese market economy.Stock investment as an important means of consumption to stimulate the market economy,most retail investors will use the existing data curve combined with some of their own experience to predict the future price trend of stocks and obtain investment income,but the stock market risks and opportunities coexist,so the issue of stock rise and fall prediction has also attracted the attention of researchers at home and abroad.Compared with the recurrent neural network(RNN)model,the LSTM network adds memory gate and forget gate to the internal structure,which solves the problem of gradient explosion or gradient disappearance caused by the RNN due to the long time series.This thesis takes Kweichow Moutai and other five stocks under the liquor sector as research objects,and proposes a hybrid model based on CNN-LSTM-BILSTM three neural networks to predict the rise and fall trend of Kweichow Moutai stock price,so as to achieve the optimal prediction effect.(1)In the current research method,only a single stock data is often used as the input of the model,ignoring the correlation between adjacent stocks under the same sector,that is,similar stocks under the same sector may walk out of a similar trend in the same time period due to certain influencing factors.In order to solve this problem,in the process of building the model,the four "price" and one quantity data of Baima high-quality stocks under the same plate will be selected as the initial input characteristics of the model,and it will be increased by 1*1 convolution to increase the receptive field,and make the input fuse more information.(2)Considering that technical indicators can also reflect the price change trend of stocks in the recent period,the Python’s stockstats library is used to construct technical indicators for individual stocks.Too many indicators is not a good thing for the model,the more input features,the noise of the model may also increase,so it is necessary to reduce the dimensionality of the obtained indicators to avoid data redundancy,choose the Pearson correlation coefficient method to reduce the dimensionality of the data and finally determine the number of input features of the model.(3)The CNN-LSTM network model is constructed,the attention mechanism is introduced into the model,and the prediction effect is improved by redistributing weight information and adding residual blocks to avoid the loss of initial information features.The model has a good forecast effect in Kweichow Moutai and 5 other stocks under the liquor sector.(4)In order to further improve the prediction accuracy,a CNN-LSTM-Bi LSTM hybrid network model is proposed,that is,the prediction results of the previously constructed CNN-LSTM model are input into the Bi-LSTM model as new features and the previously constructed individual stock technical indicators,and Bi-LSTM is used to fine-tune it and continuously optimize the final prediction result.The experimental results show that the accuracy of the improved hybrid model in the Kweichow Moutai stock dataset has increased from 52.97% to 69.16%,and the prediction accuracy of the model has been greatly improved.The same method was applied to the other 5 stocks under the liquor sector,and the accuracy of the model prediction was higher than that of the initial LSTM model,indicating that the established hybrid model was indeed feasible for such stock price prediction problems.