| Gold futures is an important trading product in the futures market,which plays an important role in the futures market.Its transaction characteristics prove that it is not a common time series data,but a financial and economic time series data.The trading characteristics of gold futures price also make it become a common and indispensable financial tool in modern society investment and financing,so all financial tools are in line with the market rules,and people who pay attention to the market trading rules are very concerned about the price fluctuations.Because of its trading characteristics,the time series of gold futures price becomes a complex,accidental and turbulent data.These three characteristics make it a complex nonlinear data.The complexity of gold futures price data determines that the traditional model can not fully describe the law,and also makes it difficult to fully describe the law.The results show that there are some errors in the traditional model.However,neural network has strong learning ability in building complex nonlinear relationship model.After inputting the initial value and relationship of the data,we can infer the unknown relationship between the unknown data through self-learning,so that the model has the ability to improve and predict the unknown data.Therefore,this paper proposes a new method to predict unknown data,in which the neural network method is used for in-depth learning,and the gold futures price is regarded as a time series change.Then,through the establishment of E-BiLSTM network model and E-CBiLNN network model,the price and development trend of gold futures are predictedBecause the data of gold futures is complex,the prediction of neural network model can achieve better results,but the original model produced some errors with data information in the prediction,and it was discarded.Although no big impact to model the trend,but in order to guarantee the precision of the model,this paper improved the BiLSTM model,joined the error correction unit in the hidden layer,named EC unit,through EC units produced in the process of fitting the model error correction,improve the accuracy of the model which is named E-BiLSTM in price forecasting.After improving the model,the running accuracy of the model has been improved,and the running time has not been greatly increased.In order to make the prediction more accurate,the CNN-LSTM model is improved on the basis of E-BiLSTM model,and it is named E-CBiLNN model.After the establishment of the two models presented in this paper,the two models are analyzed by numerical simulation.The comparison results show that the two models presented in this paper have better practicability.This paper uses the historical data of three kinds of gold futures prices(GC,XAU,AU)of different units of measurement in three exchanges(2009-2019)to predict the closing price in a certain period.In terms of data division,according to the short-term and medium-term rules of the market,the obtained data are manually cut,in which the short-term and medium-term data are the same data volume,and there are five groups of data at different times;the long-term data are five groups,and the data volume is different,and the data volume gradually increases.After the full training and parameter setting of the model,the historical price data of the first three days and the common technical indicators of the quantitative futures market are the training characteristics of all the data.At this time,the fourth day closing price is predicted according to the training characteristics,and the model data from the same source are integrated into one,and the vertical comparison of the model is also compared with the reference model.The model has achieved good prediction results.This paper proves the effectiveness of deep learning neural network model in gold futures price forecasting from an empirical point of view.The predicted price curve trend can provide some scientific and effective references for market traders,and has corresponding practical value. |
PDF Full Text Request