Proposing two novel hybrid intelligence models for forecasting copper price based on extreme learning machine and meta-heuristic algorithms

作     者：Zhang, Hong Hoang Nguyen Bui, Xuan-Nam Pradhan, Biswajeet Ngoc-Luan Mai Diep-Anh Vu 

作者机构：Changsha Univ Dept Econ & Management Changsha Peoples R China Hanoi Univ Min & Geol Min Fac Dept Surface Min Duc Thang Ward 18 Vien St Hanoi 100000 Vietnam Hanoi Univ Min & Geol Ctr Min Electromech Res Duc Thang Ward 18 Vien St Hanoi 100000 Vietnam Univ Technol Ctr Adv Modelling & Geospatial Informat Syst CAMG Sydney NSW 2007 Australia Sejong Univ Dept Energy & Mineral Resources Engn 209 Neungdong Ro Seoul 05006 South Korea Visagio Australia 6-189 St Georges Terrace Perth WA Australia Hanoi Univ Min & Geol Fac Econ & Business Adm Basic Econ Dept Duc Thang Ward 18 Vien St Hanoi 100000 Vietnam 

出 版 物：《RESOURCES POLICY》 (资源政策)

年 卷 期：2021年第73卷

页      面：1页

核心收录：

学科分类：0830[工学-环境科学与工程（可授工学、理学、农学学位）] 0819[工学-矿业工程] 08[工学] 

基　　金：Center for Mining and Electro-Mechanical Research, Hanoi University of Mining and Geology (HUMG), Hanoi, Vietnam Innovations for Sustainable and Responsible Mining (ISRM) research group of HUMG 

主　　题：Copper price Forecasting price Natural resources Extreme learning machine Optimization algorithms Hybrid models 

摘      要：The focus of this study aims at developing two novel hybrid intelligence models for forecasting copper prices in the future with high accuracy based on the extreme learning machine (ELM) and two meta-heuristic algorithms (i.e., particle swarm optimization (PSO) and genetic algorithm (GA)), named as PSO-ELM and GA-ELM models. Accordingly, the time series datasets of the copper price for thirty years were collected based on the influencing parameters, such as crude oil, iron ore, gold, silver, and natural gas prices. Furthermore, the exchange rate of the four largest countries in copper-producing, including Chile (USD/CLP), China (USD/CNY), Peru (USD/PEN), and Australia (USD/AUD), were also considered to evaluate the copper prices. The GA and PSO algorithms then optimized the weights and biases of the ELM model to reduce the error of the ELM model for forecasting copper price. The traditional ELM model (without optimization), and artificial neural networks (ANN) were also developed as the comparative models for resulting in convincing experimental results in the proposed PSO-ELM and GA-ELM models. The results indicated that the proposed hybrid PSO-ELM and GA-ELM models could forecast copper price with higher accuracy and reliability over the traditional ELM and ANN models. Of those, the PSOELM yielded the most dominant accuracy with a root-mean-squared error (RMSE) of 304.943, mean absolute error (MAE) of 241.946, mean absolute percentage error (MAPE) of 0.037, and mean absolute scaled error (MASE) of 0.933. The t-test and Wilcoxon test also demonstrated the statistical significance of the proposed models and the best 95% confident interval of the PSO-ELM model with the range of $177.046 to $67.054 with pvalue = 2.589e-05. Whereas, the GA-ELM model provided the forecasted copper price higher $137.233 than the actual copper price, and the 95% confidence interval is from $189.672 to $84.793 with p-value = 1.027e-06.