Shapelet-based International Oil Price Forecasting

Crude oil is the cornerstone of energy security,and its price fluctuation affects the trend of the world pattern and economic development.Crude oil and its products play an irreplaceable role in many industries,thus profoundly affecting the daily life of the people.Therefore,the accurate prediction of crude oil price can provide an effective reference for the strategic planning of the state and enterprises,and provide a basis for the formulation of relevant policies.In this context,this paper proposes three effective international oil price forecasting models based on the idea of "shapelet","similar pattern matching","deep learning","decomposition and ensemble" and "data triat driven modeling".The main contents are shown as follows:(1)International oil price forecasting based on shapelet and similar pattern matchingIn order to effectively improve the continuous multi-step prediction accuracy of international crude oil price,an oil price forecasting model based on shapelet and similar pattern matching is proposed in this study,which outputs the continuous multi-step ahead prediction results at one time and effectively alleviates the error accumulation problem caused by multi-step point-by-point prediction.This model involves three steps: test of memory characteristics,shapelet search and scale transformation prediction.Specifically,the long memory of the oil price series is tested firstly,and then the similar historical patterns of the current oil price pattern are determined by the shapelet search and clustering algorithm.Finally,the scale transformation is used for prediction,and the average method and genetic algorithm are used to integrate the results to obtain more robust prediction results.The WTI crude oil futures price data is selected to test the effectiveness of the proposed model,and it is found that the proposed model is superior to many mainstream prediction models in both horizontal and directional perspectives,which can improve the accuracy of multi-step oil price prediction to a certain extent.(2)International oil price forecasting based on shapelet and deep learningAiming at the problem that the length of shapelet is set subjectively and the result of shapelet selection is accidental,this study proposes an oil price prediction model combining shapelet method with deep learning.This model mainly includes four steps: memory testing,shapelet matching,shapelet weighted training and extrapolation prediction.Firstly,the memory characteristics of the international oil price data are tested,and the length of the rolling window is determined according to the test results,and then three different distance measures are selected to find shapelet patterns similar to the current oil price pattern in the historical data,and the robust shapelets are determined by the intersection method.Finally,the training data are weighted by the results of shapelet matching,and the weighted data are input into the deep learning model to train the prediction model with weight factors for extrapolation prediction.In the empirical study,the Brent crude oil futures price data is used to test the prediction capability of the proposed model,and the empirical results found that the proposed model is significantly better than most of the benchmark models in multi-step prediction tasks,and can be used as an effective tool for multi-step oil price prediction,and the proposed "shapelet weighted training" technique can enhance the training performance of the deep learning model.(3)International oil price forecasting based on shapelet and decomposition-ensemble methodIn order to solve the problem that the strong volatility of the series leads to large errors between the similar patterns found by the shapelet method and the current pattern at individual points,thus affecting the subsequent prediction accuracy,this study proposes an oil price forecasting model based on shapelet and decomposition-ensemble method.The model mainly includes three steps.First,the fluctuation characteristics of oil price series are described by using fractal analysis,and the reconstruction methods of different frequency components obtained by decomposing the original series are determined accordingly.Then,in the prediction stage,the original series is decomposed and reconstructed into a smooth trend term and a high-frequency fluctuation term by using the decomposition-reconstruction framework,so that the shapelet method only acts on the smooth trend term to maximize its role,and the machine learning model is used to predict the fluctuation term.Finally,the prediction result is output by ensemble strategy.Empirical analysis of WTI international crude oil futures price data shows that the proposed model is superior to most of the classical benchmark models in single-step and multi-step forecasting tasks,and can improve the prediction accuracy to a certain extent.The three oil price forecasting models proposed in this paper are all based on the shapelet method,and have achieved good empirical prediction results,which shows that it is reasonable and reliable to jointly build prediction models based on the memory and fractal characteristics of international crude oil price and the use of shapelets,combined with decomposition-ensemble method,deep learning and other technologies.Meanwhile,the effectiveness of "shapelet method" and "data trait driven modeling" is proved again.