The CEEMDAN-MPE-TCN Model For The Price Prediction Of Crude Oil

As an important energy resource,oil is gradually becoming one of the important factors affecting economic development,social stability,and national security.With the development of global economic integration,the change of oil price will attract extensive attention from all walks of life.The price of futures is closely related to the price of oil.Therefore,improving the prediction ability of the futures price has great significant to the study of oil spot market.In recent years,with the rapid development of the computer industry,the neural network model has been applied to various fields.Using a convolutional neural network to predict time series is also becoming a hotspot in academic circles.In this paper,the empirical mode decomposition algorithm is introduced into the modeling of crude oil futures price prediction and combined with the TCN model to extract the long-term and short-term relationship among the time-series data.To improve the accuracy and reduce the modeling complexity of TCN,the model used an algorithm to denoise the original data based on the multi-scale permutation entropy theory before training the neural network.The forecasting method CEEMDAN-MPE-TCN adopts the idea of "decomposition-reconstruction-prediction-integration",and is used to predict futures price.According to the idea of modeling,there are four steps to make it work.First,using CEEMDAN algorithm to decompose the original data into several independent IMFs.Second,using MPE and mean test to denoise these IMFs and reconstruct them to three sequences which carried the special information of original data.These sequences are called high-frequency,low-frequency,and trend terms.Then,every reconstructed sequence is put into TCN to train the model.And every model finally has a predicted value.Finally,these predicted values are added to be the final predictor result of the crude-oil-futures price.In the research,this paper selects a variety of error evaluation indicators to evaluate the combinatorial model.The CEEMDAN-MPE-TCN model proposed in this paper has been proved that is better than any other models.It has the highest prediction accuracy.The consequence is attributed to that these reconstruct sequences have effectively extracted important information from the original data.