Research On Natural Gas Futures Price Prediction Based On LSTM Model

As a clean,efficient and low-carbon energy source,natural gas is one of the world’s most important strategic energy sources.With the introduction of carbon emission reduction and carbon neutrality,the importance of energy commodity futures to the industry continues to grow,and natural gas price forecasting has become a key concern for investors and industrial clients.A reasonable and effective forecasting analysis of natural gas futures prices can play an anticipatory and pioneering role in futures themselves,providing more accurate price guidance to relevant participants and improving hedging returns.Given that China is a major importer of natural gas,effective forecasting of international natural gas futures prices can provide a reference for the Chinese natural gas market and provide more comprehensive forecasting information for decision makers such as the government and industry associations,which can help decision makers to make better judgements.This paper selects normative and representative daily price data for natural gas futures(trading symbol NG c1)on the Chicago Mercantile Exchange from April 3,1990 to August31,2022 for the study.Input features affecting natural gas futures prices are screened using e Xtreme Gradient Boosting(XGBoost)and Principal Component Analysis(PCA).Long Short-Term Memory(LSTM)has a long-term memory function and is suitable for processing time series,while the Attention Mechanism places particular emphasis on certain time steps which contain the most important discriminatory information.The combination of the two allows for longer time series to be processed,improving the generalization ability of the forecasting model and its efficiency in processing the data.Given the characteristics of strong volatility and complexity of futures price data,the LSTM model incorporating attention mechanism(LSTM-Attention)is selected for empirical analysis,and it is compared with single-feature LSTM,Gated Recurrent Unit(GRU),and multi-feature LSTM model.For the evaluation metrics,R~2,MSE,MAE and MAPE on the test set were chosen to assess the differences between the predicted results and the true labels.The experimental results showed that the R~2,MSE,MAE and MAPE values of the LSTM-Attention model were better than those of the single-feature LSTM,GRU and multi-feature LSTM model,which improved the accuracy and applicability of the model prediction.This paper makes natural gas futures price forecasts based on the analysis of natural gas impact factors and puts forward relevant policy recommendations.The results of the study provide a certain reference role for natural gas futures price forecasting and impact factor research,and provide new ideas for the study of natural gas and other energy commodities.