Twin extreme learning machine model and cooperation search algorithm for multi-step-ahead point and interval runoff prediction

Abstract

Accurate runoff predictions provide crucial technical supporting information for water resource decision-makers, offering insights into future runoff changes. This study investigates the effectiveness of twin extreme learning machine (TELM) and cooperation search algorithm (CSA) in multi-step-ahead point and interval runoff prediction. Then, three multi-step-ahead forecasting strategies are considered to develop various models: recursive, direct, and direct-recursive. The results show that the developed model consistently delivers superior accuracy and reliability in predicting runoff, while CSA outperforms other evolutionary methods in determining model parameters. However, no single forecasting strategy consistently outshines others across all scenarios, with the recursive strategy showing a slight edge in performance. Besides, the interval runoff predictions confirm the effectiveness of TELM in yielding high-quality prediction intervals across various experiments by incorporating upper and lower boundary estimation and boundary functions. For station A with a 98% confidence level, the proposed method achieves prediction interval coverage probability, prediction interval normalized average width, and coverage width criterion of 0.9904, 0.1138, and 0.6828, respectively, indicating overall high interval prediction quality. Thus, a novel artificial intelligence model is developed for multi-step-ahead point and interval runoff prediction.