International Crude Oil Futures Price Forecasting And Application Research Based On BiLSTM-Attention Combined Model

Crude oil is one of the main indicators of the global economy and plays an important role in economic growth.Fluctuations in crude oil prices can affect all aspects of people’s lives and even hinder the development of national economies,and crude oil futures prices are the most direct indicator of changes in crude oil prices.Therefore,analysing the volatility mechanism of crude oil futures prices and accurately and reliably forecasting crude oil futures prices is a challenging and very meaningful issue.Firstly,this study analyses the development status of the crude oil futures market from both international and domestic levels in chronological order,analyses the mechanism of supply and demand,economic and financial and geopolitical factors on the fluctuation of crude oil futures prices from a theoretical perspective,and composes21 relevant indicators affecting crude oil futures prices from them,and uses grey correlation analysis to filter out 15 correlation indicators with a correlation of 0.8 or above The key indicators are used as input indicators for the construction of the international crude oil futures price forecasting model.Secondly,two neural network models,LSTM and GRU,were selected as the base models in this study,and the bi-directional neural network structure and attention mechanism were fused to construct six types of prediction in two categories,Bi LSTM,Bi GRU,Bi LSTM-Attention,Bi GRU-Attention,LSTM-Attention and GRU-Attention models,and the performance of the models was evaluated and analysed in terms of MAPE,RMSE,MAE and R~2.The results of the comparison experiments show that the bidirectional structure helps to improve the forecasting performance of the neural network models,which can better predict indicators such as crude oil futures prices that are influenced by prior and current period data;the attention mechanism improves the model fit while significantly improving the forecasting accuracy of the neural network models;the optimal model for WTI crude oil futures price forecasting is Bi LSTM-Attention,whose MAPE,RMSE,MAE and R~2values of 1.842%,1.360,0.942 and 0.989.In addition,to test the robustness of the optimal forecasting model,this study uses Bi LSTM-Attention to conduct forecasting experiments with the same training set and different test sets to reflect the degree of crude oil futures price oscillation in different periods through the different test sets,so as to illustrate the forecasting performance of the model in various periods.The results show that the Bi LSTM-Attention model is less influenced by the amplitude of crude oil futures prices and the model has strong robustness.Finally,to further validate the applicability of the optimal model,this study forecasts Brent crude oil futures prices based on the Bi LSTM-Attention model.The results show that the optimal model proposed in this study can be applied to the forecasting of various international crude oil futures prices,and the forecasting performance is excellent.