Evaluating the performances of several artificial intelligence methods in forecasting daily streamflow time series for sustainable water resources management

Abstract

Accurate runoff forecasting plays an important role in guaranteeing the sustainable utilization and management of water resources. Artificial intelligence methods can provide new possibilities for runoff prediction when the underlying physical relationship cannot be explicitly obtained. However, few reports evaluate the performances of various artificial intelligence methods in forecasting daily streamflow time series for sustainable water resources management by far. To refill this research gap, the potentials of five artificial intelligence methods in daily streamflow series prediction are examined, including artificial neural network (ANN), adaptive neural-based fuzzy inference system (ANFIS), extreme learning machine (ELM), Gaussian process regression (GPR) and support vector machine (SVM). Four quantitative statistical indexes are chosen as the evaluation benchmarks. The results from two huge hydropower reservoirs in China show that five artificial intelligence methods can achieve satisfying forecasting results, while the SVM, GPR and ELM methods can produce better performances than ANN and ANFIS in both training and testing phases with respective to four indexes. Thus, it is of great importance to carefully choose the appropriate forecasting models based on the actual characteristics of the studied reservoir.