Abstract: Due to the nonlinear and stochastic characteristics of runoff, a single-point prediction model is inadequate in accurately capturing and describing the runoff evolution process.To address this issue, we introduced an intelligent interval prediction method that effectively quantifies the range of runoff fluctuations.Firstly, the complete ensemble empirical mode decomposition with adaptive noise was utilized to partition the nonlinear runoff sequence into multiple subseries, and the sample entropy method was employed to reconstruct the modified subseries.Secondly, employing the twin support vector regression as a foundation, interval prediction models were constructed for the more complex subseries, while point prediction models were established for the relatively simpler ones.Simultaneously, the whale optimization algorithm was employed to seek satisfactory combinations of model parameters.Finally, the prediction results from each submodel were aggregated to derive the ultimate prediction interval.Application results demonstrated that the proposed method exhibits remarkable stability and dependability, surpassing comparative models in various scenarios and prediction periods, including both point prediction and interval prediction.When the forecast period was 3 d, the anticipated interval derived for the Tangnaihai hydrological station in the Yellow River Basin had good reliability and clarity, with a Prediction Interval Coverage Probability(PICP)value of 98.30% and a Prediction Interval Normalized Average Width(PINAW)value of 0.079 2.Average increases in reliability and clarity were 9.47% and 32.66%,respectively.The results can provide effective models for intelligent runoff interval prediction.