Point And Interval Forecasting For Metal Prices Based On Variational Mode Decomposition And An Optimized Outlier-Robust Extreme Learning Machine

Metal prices forecasting has practical implications for a variety of participants, including producers, consumers, governments and investors. However, metal prices forecasting is a complex and challenging issue widely discussed among researchers, due to the complexity and significant fluctuations observed in metal market. Most of present metal price forecasting models only emphasized deterministic prediction, usually ignored the uncertainty analysis and prediction, and eventually provided limited reference information. Under this background, in this research a novel hybrid system, consisting of distribution functions estimation, point and interval prediction, was proposed. More specifically, five distribution functions based on optimization algorithms were estimated to mine and analyze the traits of metal prices. With respect to the point prediction, an innovative hybrid forecasting models using variational mode decomposition and an optimized outlier-robust extreme learning machine by an optimization algorithm was established for metal prices prediction. Finally, based on the results of the distribution functions and point forecasting, interval forecasting was designed to provide predictive range, confidence level and the other uncertain information. Three metal prices series were taken as illustrated example to test the effective of the presented system and numerical results showed that the developed hybrid system can not only obtain higher prediction accuracy than that of the comparison models and also offer more valuable suggestions for enterprise administrators and investors in financial market.