Daily Streamflow Forecasting Using Extreme Learning Machine and Optimization Algorithm. Case Study: A River in Vietnam

Abstract

Accurate prediction of streamflow plays an important role in water resource management and sustainability. Recent years have seen increased interest in data-based models, compared to the more established physics-based models, due to the accuracy of their predictions. Better results mean greater support for those who are tasked with formulating strategies and writing policy around water resource management. The objective of this study is the development of a state-of-the-art streamflow prediction method based on extreme learning machine (ELM), optimized by both hunger games search (HGS) and social spider optimization (SSO) to make accurate predictions for the Tra Khuc River in Vietnam. Rainfall and flow from 2000 to 2020 at Son Giang station on the Tra Khuc River were used to build the streamflow prediction model. The statistical indices root-mean-square error, mean absolute error, and the coefficient of determination (R²) were applied to assess the predictive ability of the proposed models. The results showed that both optimization algorithms successfully improved the ELM model to predict the streamflow for one day and six days ahead by using data from one day and three days before the day in question. Of the proposed models, the ELM-SSO model scored highest, with R²=0.891 for the one-day-ahead prediction and R²=0.701 for six days ahead. Second was ELM-HGS (R²=0.889 and R²=0.699 for one day and six days respectively), and third was ELM (R²=0.883, R²=0.696). The results demonstrate ELM to be a robust data-driven method for simulating time series regimes that is appropriate for various hydrological applications. The models proposed in this study can be generalized to predict streamflow in rivers around the world.