Forex Time Series Prediction Method Based On C-LSTM

The foreign exchange market has gradually developed into the world's largest financial market.In the face of the complex nonlinear dynamic system of the foreign exchange market,traditional basic analysis and technical analysis methods have been unable to do so.Deep learning technology has a good fitting effect on complex nonlinear systems and has great potential in exchange rate forecasting.Therefore,many scholars have begun to apply deep learning techniques to analyze and predict foreign exchange time series data.However,deep learning techniques also have large computational complexity,long training time,easy over-fitting,and easy local optimal solutions.It is difficult to effectively combine different depth learning algorithms.Therefore,there are still many problems to be solved and perfected for the application of deep learning technology in the analysis of foreign exchange time series.This paper also uses Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)two deep learning algorithms to analyze and predict foreign exchange time series data,and proposes C-LSTM(CNN-LSTM)short-term forecasting method for foreign exchange time series.The three main factors(training samples,network structure,training methods)that affect the accuracy of prediction are systematically studied,and the optimal input characteristics,network structure and training methods are selected.Aiming at the problem of large data noise,a feature optimization algorithm was built based on PCA(Principal Component Analysis)to reduce the input features.Then use the Dropout and L2 regularization methods to avoid the occurrence of over-fitting problems,and further improve the prediction accuracy of the prediction method.At the same time,in order to meet the high timeliness demand of the foreign exchange market,this paper builds a parallel optimization algorithm based on GPU(Graphics Processing Unit)high-performance computing technology,which accelerates the training speed of the network model and improves the usability of the prediction method in practical application scenarios.Experimental research was conducted on nine currencies,including the euro against the US dollar(EURUSD)and the US dollar against the Canadian dollar(USDCAD).The experimental data show that compared with neural network algorithms such as BP(Back Propagation),CNN,RNN(Recurrent Neural Network),the short-term prediction method of C-LSTM forex time series constructed in this paper has the best prediction effect.It fully validates the validity and applicability of the C-LSTM foreign exchange time series short-term forecasting method in the analysis and forecast of the foreign exchange market.This paper has certain reference significance for feature selection and super-parameter tuning of deep learning algorithms such as convolution depth neural network and cyclic deep neural network.At the same time,it provides certain theoretical and practical value for the application research of deep learning technology in the foreign exchange market.