Research On Stock Price Prediction Based On Various Neural Networks

Since its establishment,China's stock market has been playing the role of China's economic barometer.In recent years,with the gradual maturity of China's financial market trading mechanism,the emergence of a large number of high-quality listed companies and the global trend of large volume,China's stock market tends to be stable and good.The obvious change is that the number of investors is increasing year by year,and the total market value of Shanghai and Shenzhen stock markets is growing with each passing day Increase.Therefore,the reasonable analysis and prediction of the future trend of the stock market is becoming a hot area of research.With the advent of the era of artificial intelligence,neural network algorithm is widely used in various fields of complex data processing,and has achieved better results,so more and more scholars continue to explore its application in the stock market.In this paper,the specific work of stock data research based on neural network is described as follows:1?Stock data preprocessing.Taking Guizhou Maotai(600519)as the experimental data,this paper calculates the corresponding basic index,trend index,probability index and energy index through 12 characteristics including closing price,highest price and lowest price,and introduces Shanghai stock index,Shanghai Shenzhen index and other data representing the market index.Finally,the dimension of the data is reduced by Spearman correlation coefficient and principal component analysis,In order to achieve higher calculation results.2?Build models.By constructing BP neural network,Gru gated neural network and LSTM long short memory neural network,taking t day data as input data,the closing price of t+1 day is calculated,and the model is evaluated by the difference between the real value and the test value,and the neural network model more suitable for stock data is selected.3?Through the use of dropout,adagrad,rmsprop and Adam optimization methods,each model is optimized in order to get better results and have a more accurate grasp of the short-term trend of the stock market in the future.Through the use of dropout,adagrad,rmsprop and Adam optimization methods,each model is optimized in order to get better results and have a more accurate grasp of the short-term trend of the stock market in the future.Through the experiment,it can be concluded that: it is feasible to predict the stock trend by using neural network;compared with the traditional BP neural network,the result of introducing time series of recurrent neural network is much better,among which the gating neural network is the best;the use of dropout and Adam and other optimization algorithms can effectively reduce the role of model over fitting,and also make the final error more reasonable It was significantly reduced.Hope that somebody through this study,we can provide little reference for the future stock research based on neural network,and provide new ideas for the majority of investors to control the stock trend.