Research On Stock Forecasting Based On The Deep Two-way LSTM-RNN Model

In today's economic prosperity and development,people's financial awareness is gradually rising,many users will invest the surplus funds in the fields of finance,education,medical care and so on.Although it is possible to gain huge profits quickly by investing in stocks,the price fluctuation of stocks will be affected by many uncertain factors,which makes the changes of stock market quite complex and full of unpredictable risks.How to valid arrest this risks and investments failure are the focus of many investors.For a long time,most of economists and mathematician have committed to the searching of the stock price fluctuation,using chart analysis,financial index,inductive reasoning and other methods to predict the stock market,and achieved good results.However,due to the continuous development of the stock market,the prediction effect and accuracy of the classical statistical analysis methods have been unable to meet the requirements of the current stock investors.How to use the historical stock data to enhance accurate the prediction based on the time series to provide investors with the optimal allocation strategy is the current problem to be solved.The main contents of the study include the following aspects.First of all,this paper studies the stock forecasting methods,application techniques and basic theories put forward by predecessors,analyzes the classic BP neural network forecasting methods in depth,and demonstrates the irrationality of the model for time series forecasting in theory,and explain RNN model which is good at processing temporal data.Analyzes the advantages and disadvantages of BP ANN and RNN,and makes particular contrast trial run.Then,based on RNN,it modifies and optimizes the model,and constructs a prediction model with BRNN as the core and multiple optimization algorithms as the guide.It has fully proved that the precision of this model is taller than that of BP and RNN model by many comparative experiments,and the errors of this model are reduced by about 2.2% on average.Secondly,considering the factors of time series and long series,this paper introduces the LSTM model,theoretically analyzes the advantages and disadvantages of LSTM compared with other time series models,and continues to improve the model.Through the analysis and comparison with other algorithm,this paper points out the defect of Adam algorithm in the processing of learning rate,puts forward and applies the improved Adam algorithm to the prediction model of BiLSTM,and proves the rationality and feasibility of the model prediction through the comparative experiment and the related technical index experiment.