Research On Exchange Rate Forecast Based On LSTM-ELM Hybrid Model

With the increasing proportion of RMB in international settlements,the fluctuation of USD/RMB exchange rate has always been the focus of investors’ attention.The fluctuation of exchange rate is very important to a country’s economic and trade.Due to the non-stationary and nonlinear structural nature of exchange rate time series,it is a challenging task to accurately predict exchange rate time series.Therefore,developing a more effective exchange rate forecast model is very important for the government.,banks,enterprises and individuals and other economic or management entity systems have an important impact.The data in this article comes from the S&P Capital IQ and Wind databases,and selects the daily average price data,as well as the exchange rate daily transaction data and stock data,a total of 8 indicators.Firstly,simple descriptive statistics are carried out on the daily data of USD/CNY,and then the prediction values of the exchange rate data by the LSTM network and the ELM model are obtained respectively,and the two prediction results are multiplied by the optimized ocean predator algorithm.The respective weights are added together to obtain the final result of the combination method,that is,the LSTM-ELM weighted combination exchange rate prediction model is constructed,and its prediction value is the final prediction result.This article is divided into univariate forecast and multivariate forecast.The univariate forecast uses the daily average price data to predict the USD/CNY exchange rate on the next trading day,and the multivariate forecast uses 8 indicators including the exchange rate daily transaction data and stock data to predict the USD/RMB exchange rate for the next day.Closing price.And selected two different distribution methods of training set and test set,namely 9:1 and 8:2,to explore the impact of different training set and test set distribution on exchange rate prediction.The LSTM-ELM weighted combination exchange rate prediction model proposed in this paper is compared with SVM,random forest,ELM,LSTM,LSTM-ELM average combination model.The experimental results show that the LSTM-ELM weighted combination exchange rate prediction model has the smallest MSE and MAPE in terms of univariate data and multivariate data prediction,and is closest to 1,that is,the model proposed in this paper has higher prediction accuracy and better fitting effect.Therefore,LSTM-ELM weighted combination prediction model can effectively predict exchange rate.In addition,if the model proposed in this paper is used for exchange rate-related prediction,and the distribution of training set and test set of 8:2 is used,the prediction effect will be better.