Study On The Prediction Of Stock Time Series Based On The Multivalue Associated Model Of LSTM

Stock price changes affect the investors' investment decisions.Accurate analysis of the law of stock change and prediction of its development trend has always been a problem for shareholders and securities companies to solve.At present,in view of the stock price changes have put forward all kinds of forecasting methods.Methods of forecasting stocks can be divided into two categories: mathematical statistical methods and machine learning methods.Mathematical statistics methods in general use Logistic regression model and ARCH model,etc.;machine learning methods include multi-layer Perceptron,convolutional neural network,Naive Bayes network,back propagation network,single-layer LSTM,support vector machine,RNN,etc.But these studies predict a single value,in order to forecast multiple values at the same time need to design a can handle multiple linked at the same time the stock price and output at the same time the model of stock price--multiple inputs multiple outputs neural network model.The paper does the following work:1.The paper proposes a neural network structure of deep cycle neural network with multiple inputs and outputs based on long short-term memory network(LSTM)to predict the maximum price,minimum price and opening price of stocks.The model was compared with LSTM network model and LSTM deep circulation neural network model.Experiments show that the accuracy of the multi-input and multi-output network model based on LSTM deep loop network is not only superior to the other two models,but also able to predict multiple values at the same time while ensuring the accuracy,and the prediction accuracy of each prediction value is above 95%.2.The paper modified the multi-valued associated neural network model into A multi-valued no-associated neural network model,On the condition that other conditions remain unchanged,the data of Shanghai stock exchange a-share is modeled.Experimental results found associated neural network each loss values are higher than no-associated with neural network of the each loss value about 0.07,The prediction accuracy of each sub-branch of multi-valued associated neural network is 4% higher than that of each sub-branch of multi-valued noassociated neural network.It further illustrates that there is a certain relationship between the maximum price,the minimum price and the opening price of stock data.3.The paper uses the multi-value Associated model to predict the highest,lowest and opening prices of stocks in Shanghai and Shenzhen indices.