Design And Implementation Of ARIMA-LSTM Based Forex Trend Forecasting System

With the popularity of foreign exchange trading and the development of neural network technology,the acquisition of foreign exchange data is becoming more and more convenient,which makes the research on the trend of foreign exchange more and more concerned.Compared with other developed countries,the development of foreign exchange in China starts relatively late,and the material of research results on the timing of foreign exchange trading is very less.Therefore,how to judge the trend of foreign exchange from the complicated market rules has become the research focus of many traders and researchers.This paper combines linear and non-linear strategies to forecast the trend of foreign exchange based on foreign exchange data.Aiming at the problem of forecasting the trend of foreign exchange,this paper takes the euro against the dollar(EURUSD)as the research object.By analyzing the influencing factors of the EURUSD trend and obtaining the historical data of foreign exchange trends,we construct a time series of influencing factors to train in the model and use the saved model to predict the future closing price trend of foreign exchange.The main research work is as follows:(1)The improvement of classical long-term and short-term memory networks is realized.Considering the nonlinearity of foreign exchange historical data,this paper chooses LSTM(Long Short-Term Memory,long-term and short-term memory network)to analyze the data,and improves it based on LSTM,introduces 1-arctan function into the output gate,changes the value range of the output gate from(0,1)to(0.21,1),to make the output value of the output gate in a more obvious range,which can preserve the data features as much as possible,and thus improve the prediction performance of the neural network.(2)A model is established to predict the trend of foreign exchange in the future.In this paper,an ARIMA-LSTM combination model is designed to predict the trend of foreign exchange.First of all,the model obtains the linear characteristics of the foreign exchange historical data through ARIMA and then uses the improved LSTM model to obtain the nonlinear characteristics of the foreign exchange historical data.Finally,combined with the current opening price,the information fusion is carried out through the CRITIC in the objective weighting method to obtain the forecasting results of the combined model.To verify the performance of the ARIMA-LSTM model,a comparative experiment was carried out with six models such as RNN,ARIMA,and LSTM.The experimental results show that the MAPE,RMSE,and MAE of the ARIMA-LSTM model are the smallest.(3)The foreign exchange trend forecasting system is completed the design and implementation.The system is developed by the powerful Java language.It designs and implements five functional modules including user management module,personal information management module,data collection module,foreign exchange trend analysis module,and foreign exchange trend prediction module.The graphical display interface containing foreign exchange trend information is realized through Echarts technology,which is convenient for users to understand the foreign exchange trend intuitively.