A Novel Methanol Futures Price Prediction Method Based on Multicycle CNN-GRU and Attention Mechanism

Making accurate prediction of methanol price is great important for intelligent coking plant production scheduling, and it is essentially a time-series prediction problem. Efficient extraction of temporal and spatial features of methanol price data is conducive to improve the prediction accuracy of the future methanol price, and there are cyclical influences in methanol price prediction. Considering the advantages of convolutional neural networks (CNN) and gated recurrent unit (GRU) in extracting temporal and spatial features, a novel methanol price prediction method based on multicycle CNN-GRU and attention mechanism is proposed, abbreviated MCGAT. In the model construction, a parallel hybrid network architecture is designed. First, CNN is used for extracting static spatial features of methanol price data, and GRU is employed to mine its dynamic temporal features; then, the attention mechanism is utilized to fuse the static spatial and dynamic temporal features attained by CNN and GRU; finally, the methanol price prediction model-based multicycle CNN-GRU is constructed to probe into the cyclical influences. Experimental results on two real methanol price datasets in the past six years manifest the proposed MCGAT outperforms other state-of-the-art methods, and it provides a useful prediction tool of future methanol price for intelligent coking plant.