Research On Crude Oil Price Forecasting Model Based On Meta-learning

Energy is an crucial strategy for the development of a country's modernization.And,crude oil,a main energy source and international commodity,has direct or indirect connection with domestic industries.However,the crude oil price is affected by not only supply and demand but also various factors and the change and trend of it are greatly uncertain.To abate the uncertainty,accurate crude oil price information has always attracting much attention.For crude oil price forecasting methods,there are two main categories: direct forecasting method and hybrid forecasting method.Generally,these methods consider only one forecasting model in their forecasting stages and that limits their forecasting results.Besides,tradition model selection methods are based on the expertise or the trial-and-error method which respectively cause limitation of adaptation and performance and incensement of consumption.Therefore,this paper introduces meta-learning to the crude oil price forecasting problem and proposes a meta-learning decision-making model(MDM)to adaptively decide using of forecasting models according to the characteristics of forecasting tasks.The proposed model has been implemented to optimize the two kinds of methods.Firstly,MDM is studied and applied to the direct forecasting method to verify that the model can capture relationships between forecasting tasks and performances of forecasting models and provides the use of models with accurate decision support for crude oil price and related price forecasting tasks.Then,the optimized MDM is integrated into the hybrid forecasting method with “decomposition-forecasting-ensemble” paradigm.A hybrid forecasting model of crude oil price based on decision support from meta-learning is proposed.This novel model aims to achieve an overall improvement of prediction by optimizing decisions of prediction tasks in the prediction state.In this research,the daily,weekly and monthly crude oil price series are considered as experimental data to identify the decision-making ability of MDM and its generalization to different forecasting tasks.According to the experiment results,the proposed model has more accurate and efficient decision-making effects than the traditional model selection method.And,the novel hybrid forecasting model is effectively enhanced comparing to the original hybrid forecasting method.The performances are significantly improved in terms of the prediction accuracy and direction of movement.In addition,MDM is more superior than the traditional model selection methods attributing to the ability of more reliable decision and lower consumption.This more accurate and reliable price forecasting method can be conducive to the government,enterprises and individual investment units to better formulate plans related to crude oil to prevent the uncertain influences of the price fluctuations.This improvement owns important practical value and significance for the crude oil-related industries.