A Ensemble Learning Paradigm Using RVFL Network For Crude Oil Price Forecasting

As a major energy source,oil has a significant impact on the world's political,economic and military environment.Accurate and fast crude oil price forecasting will help to optimize the corresponding production,sales,investment and other plans to avoid potential risks and improve the profits of oil-related departments.There are a lot of influencing factors referencing to crude oil price,including not only market supply and demand balance but also various external factors such as substitutability with other resources,weather,stock levels,economic growth,political changes,psychological expectations and extreme events.Due to these various interactive factors,crude oil price forecasting has become an extremely tough task.In this context,this article focuses on accuracy and computing speed to build a crude oil price forecasting model which will improve and innovate the existing oil price forecasting model,the details are as follows:First of all,this paper constructs a new model integrating ensemble empirical mode decomposition(EEMD)and random vector functional link(RVFL)networks for crude oil price forecasting.Comparing with other single forecasting models and EEMD-based decomposition-ensemble models,it is found that the introduction of RVFL network improves the performance of the decomposition-ensemble model from the perspective of prediction accuracy and time efficiency.Then,this paper introduces three kinds of randomized algorithms,such as RVFL network,extreme learning machine(ELM)and random kitchen sinks(RKS)to the framework of decomposition-ensemble learning paradigms to build a randomized-algorithm-based decomposition-ensemble learning methodology,which is applied to energy price forecasting.Experimental results show that the introduction of randomized algorithm can effectively improve the prediction accuracy of the existing decomposition-ensemble model and save a great deal of time.Finally,this study explores diversity strategies in the RVFL network ensemble learning.The impacts of five different strategies including data diversity,sampling interval diversity,parameter diversity,ensemble number diversity and ensemble method diversity on the performance of RVFL network ensemble learning have been examined and analyzed.By using the optimal diversity strategy,a multistage nonlinear RVFL network ensemble forecasting model is proposed for crude oil price forecasting,and the corresponding performance are consistently better than that of single RVFL network model in terms of the same measurements.The RVFL network-based decomposition-ensemble learning model,randomized-algorithm-based decomposition-ensemble learning model and the multistage nonlinear RVFL network ensemble forecasting model based on the optimal diversity strategy proposed in this paper all improve the accuracy of the international crude oil price forecasting model to a certain extent and reduce the model time consuming.For decomposition-ensemble models based on traditional algorithms,introducing RVFL network and randomized algorithms can significantly improve the predictable performance of the decomposition-ensemble models.At the same time RVFL network with high predictive performance and low computation time,its single model and ensemble model can both improve the accuracy of oil price forecasting results.