Design And Implementation Of Deep Learning Model For Prediction Of Price

As an significant branch in the field of machine learning,the deep learning algorithm has promoted the rapid development of information society,and that's the reason why nowadays so many domestic and foreign scholars pay more attention on it and invest a lot of money in this research.For a few specific cases in daily life,like recognition problems of images and speech,the detection method using deep learning algorithm usually performs better than artificial identification.In addition,comparing with traditional statistical and econometric models,the deep learning has good nonlinear fitting ability and comprehensive processing capability for multiple mixed information,which can provide more reasonable and reliable analysis and researchers when dealing with the prediction of stock price.The fluctuating price of stock can be regarded as a kind of time-series data which has very strong correlation.However,the cyclic neural network can have a good handle on these kind of data.On this basis,firstly several popular cyclic neural network models have been studied on the net structure and working principle.And a kind of new cyclic neural network named SULSTM is designed based on Long-Short Term Memory(LSTM)model.Furthermore,on the basis of SULSTM,a prediction system of stock price would be displayed in this paper.The main research and work can be summarized as following points.Combining with the development and application of deep learning models in different cases,the stock price prediction models could be classified as five categories according to varying functional areas: the model based on limit order,the model based on classification of price,the model based on classification of text,the model based on the wave of stock price and the model based on portfolio optimization.The stock price fluctuation model is the basis of new prediction system in this paper.This paper focuses on the structure and working principle of traditional RNN network,LSTM network and GRU network.Because of characteristic the long term,multi-variable factors and nonlinear dynamics of stock price,this paper would design a new network named SULSTM through changing the structure of LSTM model,which is the basis of prediction system later.Based on the improved SULSTM model,this paper designs and implements a multi-hidden layer price prediction neural network model.The stock price would be forecasted by using five kinds of financial data: open price,high price,low price,close price and volume.Comparing with traditional neural network models,it is founded that this improved SULSTM model can achieve a better forecasting result.