Proceedings of the 5th Management Science Informatization and Economic Innovation Development Conference, MSIEID 2023, December 8–10, 2023, Guangzhou, China

Research Article

Research on an Improved Transformer Model for Predicting Crude Oil Prices

Download97 downloads
  • @INPROCEEDINGS{10.4108/eai.8-12-2023.2344782,
        author={Heng  Guan},
        title={Research on an Improved Transformer Model for Predicting Crude Oil Prices},
        proceedings={Proceedings of the 5th Management Science Informatization and Economic Innovation Development Conference, MSIEID 2023, December 8--10, 2023, Guangzhou, China},
        publisher={EAI},
        proceedings_a={MSIEID},
        year={2024},
        month={4},
        keywords={transformer models;normal distribution functions;crude oil price prediction;},
        doi={10.4108/eai.8-12-2023.2344782}
    }
    
  • Heng Guan
    Year: 2024
    Research on an Improved Transformer Model for Predicting Crude Oil Prices
    MSIEID
    EAI
    DOI: 10.4108/eai.8-12-2023.2344782
Heng Guan1,*
  • 1: Chongqing University of Posts and Telecommunications Software Engineering Instituteline
*Contact email: 1324499282@qq.com

Abstract

To enhance the precision of crude oil price prediction, this study introduces an innovative method that integrates the attention mechanism from transformer models with normal distribution functions. The incorporation of normal distribution functions aids in capturing the inherent volatility within each segment of crude oil price data, thereby preserving the distinctive characteristics of historical data. This preservation is instrumental in achieving more accurate predictions of future crude oil prices, consequently facilitating more reasoned projections of crude oil price trends. Our investigation is centered on the daily price data of West Texas light crude oil spanning from January 1, 2001, to January 1, 2023. Subsequently, an improved transformer model was employed to train and predict the aforementioned dataset. Comparative analysis against the benchmark model reveals the superior predictive performance of the enhanced transformer model in comparison to traditional transformer models and LSTM models. Moreover, the research results have successfully withstood rigorous robustness testing, affirming the reliability of the proposed model.