Coal Price Forecast Based On Decomposition Ensemble Mode

Accurate prediction of coal prices is important for making coal policies and preventing coal market risks.In this thesis,the decomposition integration model(VADM)and the decomposition integration model(VANQM)for deterministic coal price forecasting are proposed respectively to gradually improve the problems in coal price forecasting.Specifically,the VADM model firstly improves the variational modal decomposition(VMD)method using arithmetic optimization algorithm(AOA)to obtain AOA-VMD,and decomposes coal price series using AOA-VMD.Secondly,the decomposed subsequence is predicted using deep time convolutional network(Deep TCN),and the prediction results of each subsequence are summed to obtain the prediction results of coal price.Finally,the mean impact value algorithm(MIV)is used to analyze the importance of the drivers.The first and third steps of VANQM are similar to the VADM model,and its second step is to improve the neural basis extended analysis of interpretable time series forecasts(N-BEATS)model using quantile regression to obtain interval forecasts of coal prices for the decomposed subseries.Both models are finally validated for their respective superiority by cross-comparison between models.The empirical study shows that the decomposed sub-series of AOA-VMD are smoother and more linear compared with the original series;compared with the benchmark model,MAPE,MASE and SMAPE of the VADM model all show different degrees of decreases;the forecasting errors of coal prices are significantly reduced after considering the overnight Shanghai Interbank Offered Rate and June Treasury Yield,two newly proposed interest rate factors;the Compared with the benchmark model,the PICP values of the VANQM model are greater than their corresponding confidence levels at 70%,80% and 90% confidence levels;the A-share power industry index,hydropower industry index,coal mining industry index and coal industry index are the four most significant factors affecting coal prices.The innovation of this thesis is that a new driving factor,interest rate,is considered in predicting coal price,and the impact of interest rate on coal price prediction is verified by comparing the MIV algorithm with the prediction effect;using a newly proposed meta-heuristic optimization algorithm,the arithmetic optimization algorithm(AOA),the combination of parameters for the variational modal decomposition is determined that avoids the problem of weakening the decomposition effect due to the direct adoption of the empirical judgment method;using AOA-VMD significantly improves the forecasting performance of Deep TCN and N-BEATS models;using quantile regression to optimize the N-BEATS model to achieve interval prediction of coal prices,which makes up for the deficiency of kernel density estimation.