Research On Forecasting Method For Thermal Coal Price Under Policy Intervention

As the main energy in China,coal plays an important role in China’s energy security and economic development.In recent years,with the impact of domestic energy structure optimization and emergencies,the mismatch of supply and demand in domestic thermal coal market and the sharp fluctuation of coal price occur frequently.In order to keep the coal market stability,government regulates and controls the coal market via polices.As results,the price fluctuation of thermal coal in China is the outcome of both market spontaneous regulation and policy intervention.In this situation,accurate prediction of the coal price and its trend will help producers and operators of coal upstream and downstream to avoid the risks caused by coal price fluctuations,and provide decision-making basis for energy policy makers as well.Due to the complexity of the policy’s impact on the coal market and the text nature of the policy itself,most existing machine learning and deep learning models on coal price forecast ignore the significant effect on price by polices and focus on the prediction model design from numerical variables from market supply and demand,macroeconomic and so on,which cannot accurately model the trend and the pattern of price fluctuation.Therefore,in this thesis,both the policies and market factors are concerned for great improvement of price prediction.Based on the analysis on the effect of policies on thermal coal price,a quantitative model is designed for the policies,from which a deep learning model is developed to forecast the thermal coal spot price.The main contributions of this thesis include two aspects.(1)A novel quantitative method for energy policy is proposed,which takes both the subjective evaluation and objective influence into account and can support real-time prediction of thermal coal price.Firstly,the coal price formation mechanism and the transport mechanism from policy to coal price fluctuation are analyzed,the relevant concepts are defined,and then the main transport channels from the policy to the coal price are determined.After collecting regulatory policies,data mining is performed via statistics and text analysis,and the rules on text characteristics,change trends,content themes and many others are qualitatively and quantitatively distilled,and then a policy quantification method is proposed,whose effectiveness is verified by experiments on the real coal dataset.(2)A short-term thermal coal spot price forecasting model is developed by thoroughly considering energy regulation policy and market supply and demand.Firstly,the effect analysis on price fluctuation shows that much difference exist between policy and market factors,which calls for different modeling strategies to design the mapping from factors to price.Through correlation analysis,index set of price influence factors is constructed,and a two-stage deep learning model is designed,which separates the effects by market and policy factors.Comparative experiments are implemented on the real data set of the thermal coal price,and the experimental results verify that the proposed model outperforms baseline ones on all concerned performance metrics,including mean absolute error and mean absolute percentage error.The study on the intelligent computing method by fusing domain knowledge and data driven deep learning can greatly improve both the efficiency of model design and prediction performance thermal coal price.