A novel decomposition integration model for power coal price forecasting

Accurate prediction of steam coal prices is important for stabilizing the coal trading market and formulating coal use strategies scientifically. In this paper, a new decomposition integration model (VADM) is proposed to predict coal prices by combining the variational modal decomposition (VMD), arithmetic optimization algorithm (AOA), deep temporal convolutional network (DeepTCN), and mean impact value algorithm (MIV). Firstly, the AOA optimization algorithm is used to improve the VMD, AOA-VMD was obtained. It is used to decompose the steam coal price series. Then, the decomposed subsequences are predicted for the prediction of steam coal prices by using DeepTCN. Finally, the MIV algorithm is applied to analyze the impact of different factors on the price of steam coal. It is found that: the steam coal price sub-series decomposed by AOA-VMD are smoother and more linear compared with the original series; the errors in forecasting steam coal prices are significantly reduced after considering newly proposed factors, interest rates, such as the overnight Shanghai interbank offered rate and the six-month treasury bond yield; the MAPE, MASE and SMAPE of the VADM model all show different degrees of decline compared with benchmark models. The forecasting effect of VADM model is better than the benchmark model in terms of stability and accuracy, and can be used for short-term forecasting of coal prices.

Automated summary

A new decomposition integration model (VADM) is proposed in this paper to predict coal prices by combining variational modal decomposition (VMD), arithmetic optimization algorithm (AOA), deep temporal convolutional network (DeepTCN), and mean impact value algorithm (MIV). The AOA optimization algorithm is used to improve VMD, and the smoothed and linear decomposed subsequences are predicted by DeepTCN. The MIV algorithm is then applied to analyze the impact of different factors. Experimental results show that the VADM model outperforms the benchmark models in terms of stability and accuracy, with significantly reduced forecasting errors.