Forecasting agricultural commodity futures with decomposition and ensemble strategy based on attentional temporal convolution network

• Published in: Nanjing Xinxi Gongcheng Daxue Xuebao Vol. 16; no. 3; pp. 311 - 320
• Main Authors: Zhang, Dabin, Huang, Junjie, Ling, Liwen, Lin, Ruibin
• Format: Journal Article
• Language: Chinese
• Published: Nanjing Nanjing University of Information Science & Technology 01.06.2024
• Subjects: Ablation; Agricultural commodities; Coders; Commodities; Decomposition; Ensemble forecasting; Forecasting; Modelling; Noise prediction; Soybeans; Time series; Transformers
• ISSN: 1674-7070
• DOI: 10.13878/j.cnki.jnuist.20230822001

To address the low prediction accuracy in agricultural commodity futures due to their nonlinear and non-smooth features resulting from various influencing factors, this paper proposes a decomposition and ensemble forecasting approach based on CEEMDAN and Transformer-Encoder-TCN. First, the Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is used to decompose the time series into multiscale Intrinsic Mode Function (IMF) and residuals, reducing the complexity of series modeling. Second, each subseries is predicted via Temporal Convolutional Network (TCN) incorporating multi-stage self-attention unit (Transformer-Encoder) , which optimizes the modeling weights of significant features. Finally, the prediction results of each subseries are linearly summed and integrated to obtain the final prediction results. The soybean futures revenue index in the agricultural commodity index of South China Futures Company is used as the research object. The model is retrained by time-series cros