Study On The Optimization Of Coal Product Structure Based On Steam Coal Price Prediction Using Deep Learning

Nowadays,in the face of increasingly fierce market competition,coal enterprises need to adjust the structure of coal preparation products timely through scientifically and effectively predicting the coal market dynamics and targeting at high-value and marketable products.This has become an urgent measure for coal enterprises to circumvent financial risks and promote economic efficiency.Focusing on the above issues,this thesis mainly carries out the following research:Firstly,the feature construction and selection for the steam coal price prediction models are analyzed.Daily data of Qinhuangdao Port's Q5500 K,Q5000K,Q4500 K steam coal prices and 47 kinds of influential factors during nearly 10 years from mid-2010 to the end of 2020 have been collected and analyzed.Through trend,correlation and causality analysis,their relationship was qualitatively and quantitatively mined.Five influencing factors such as international coal prices are finally selected to explain the fluctuation of coal price,and can be used as the preferred feature input for subsequent forecasting models.Then,LSTM/GRU fusion deep learning models based on empirical mode decomposition(EMD)and Attention mechanism are constructed to realize the accurate prediction of steam coal price.The model prediction results show that the mean values of mean square error,mean absolute error,determination coefficient and accuracy of the proposed fusion model in three steam coal price prediction tasks are 9.939,7.655,0.928 and 0.820,respectively.Compared with two conventional machine learning models of RF and GBDT,the average improvement rates of the four evaluation metrics are 30.00%,25.75%,9.44% and 72.33%,respectively.Compared with two ordinary deep learning models of LSTM and GRU,the average improvement rates of the four evaluation metrics are 24.50%,25.32%,6.10% and 16.73%,respectively.In terms of model stability,the average variance of prediction error of the proposed model is 32.475,which is lower than other baseline models.Comprehensive performance metrics including regression,accuracy,improvement rate and stability have been verified that the fusion model proposed in this thesis combines the advantages of EMD and Attention mechanism,and can achieve accurate prediction of steam coal price.Finally,an objective function for maximizing the sales revenue of coal preparation plant is constructed.The annual sales revenue of a steam coal preparation plant in Shanxi Province under different coal preparation schemes is simulated for real and predicted steam coal market prices.The case study results show that the application of coal price forecast to adjust the product structure of coal preparation plants can increase annual sales revenue of 6.2-7.8 million yuan compared with the fixed coal preparation product structure.This can provide a theoretical basis for coal enterprises to maximize economic benefits from the perspective of operation and management.There are 33 figures,14 tables and 93 references in this paper.