Combined Modelling for Iron Ore Demand Forecasting with Intelligent Optimization Algorithms

The stable supply of iron ore resources is not only related to energy security, but also to a country's sustainable development. The accurate forecast of iron ore demand is of great significance to the industrialization development of a country and even the world. Researchers have not yet reached a consensus about the methods of forecasting iron ore demand. Combining different algorithms and making full use of the advantages of each algorithm is an effective way to develop a prediction model with high accuracy, reliability and generalization performance. The traditional statistical and econometric techniques of the Holt-Winters (HW) non-seasonal exponential smoothing model and autoregressive integrated moving average (ARIMA) model can capture linear processes in data time series. The machine learning methods of support vector machine (SVM) and extreme learning machine (ELM) have the ability to obtain nonlinear features from data of iron ore demand. The advantages of the HW, ARIMA, SVM, and ELM methods are combined in various degrees by intelligent optimization algorithms, including the genetic algorithm (GA), particle swarm optimization (PSO) algorithm and simulated annealing (SA) algorithm. Then the combined forecast models are constructed. The contrastive results clearly show that how a high forecasting accuracy and an excellent robustness could be achieved by the particle swarm optimization algorithm combined model, it is more suitable for predicting data pertaining to the iron ore demand.