Short-term Thermal Coal Price Forecasting Based on EL-CLSTM Model

Coal enterprises in China have great demand for short-term thermal coal price forecasting. However, existing methods are insufficient for extracting local time-series features and screening useful feature information. To effectively extract the hidden features in thermal coal price data and improve the forecasting accuracy of thermal coal prices, we designed and implemented an EL-CLSTM model based on deep learning. First, the price series of thermal coal in Qinhuangdao Port was reconstituted into multiple components using empirical mode decomposition (EMD). The supply and demand factors, macroeconomic data, and other factors were then decomposed using EMD. The components screened by LASSO and each component of the price series of thermal coal produce feature vectors in the form of sliding windows as inputs. Then, the convolutional neural network (CNN) was used to extract the local time-series features of the high-frequency and intermediate-frequency components. These components were used as input data for the long short-term memory (LSTM) for price forecasting. Finally, the forecasting results for each thermal coal price series component were superimposed to obtain the final price forecasting value. The proposed model exhibits improved thermal coal price forecasting than ARIMA, SVR, LSTM, CNN, and LSTM with EMD and LASSO.