Crude Oil Price Forecasting With Decomposition And Ensemble Deep Learning Pardigms

The development of the global economy and social stability are related to the price fluctuation of the international crude oil market.As a special commodity in the global economic market,its price is basically determined by the relationship between supply and demand.But it is also vulnerable to some irregular events such as weather,inventory levels,GDP growth,political factors and even psychological expectations.These factors have caused drastic fluctuation in the crude oil market,which is characterized by complex nonlinearity,high volatility and irregularity.To improve the prediction effect of oil price,this paper improves and innovates prediction methods in terms of the existing time series correlation.The specific research contents are shown as follows.(1)Under the strategy of "decomposition and ensemble",this paper constructs a combined prediction method of LSTM neural network based on EEMD decomposition.First,the skill of EEMD decomposes the original time series of oil price.Then it uses the LSTM neural network to predict the decomposition components.Finally,the predicted sequence is simply summed to obtain the final value.The empirical results show that compared with other benchmark models,the methods proposed in level prediction and direction prediction are more effective in predicting crude oil prices.(2)Improved oil price prediction of LSTM neural network integration model based on EEMD,this paper puts forward a combined prediction method based on EEMD decomposition,wavelet threshold denoising,the reconstruction method of fine-to-coarse,and LSTM neural network.First,the EEMD decomposes the original time series of oil prices,and wavelet threshold denoising method is used to obtain effective information of high-frequency modal components.Second,the decomposed mode components are reconstructed using the fine-to-coarse method to get reconstructed components from high frequency to low frequency.Then the LSTM neural network is used to predict the reconstructed components.Finally,individual predictions of the reconstructed sequences are fused to get the final ensemble predicted results.Compared with the other benchmark models,the proposed model improves the prediction accuracy of oil price.The two oil price forecasting models in terms of the decomposition and ensemble deep learning paradigms are proposed in this paper based on the strategy of decomposition and integration.The daily crude oil price of WTI is used for empirical analysis respectively and experimental results show both models improve the accuracy of oil price forecasting.The empirical results show that these two prediction methods are effective in predicting time series data with nonlinear and irregular characteristics.