GECA Proposed Ensemble–KNN Method for Improved Monthly Runoff Forecasting

Medium- to long-term runoff forecasting on monthly timescales is an important aspect of formulating long-term water resource dispatch plans, making it of great significance to water resource management. Current research on such forecasting mostly focuses on attempts to determine the correlation between various factors and the runoff process using mathematical methods. Linear or nonlinear methods are used to establish direct or indirect conversion equations that express the relationship between high-correlation factors and runoff. However, the hydrologic cycle is a large and complex system that has chaotic characteristics. Because the system cannot be accurately depicted using existing mathematical methods, models based on these methods have limited forecasting abilities. This study developed and tested an Ensemble–KNN forecasting method based on historical samples, thereby partially avoiding uncertainties caused by modeling inaccuracies. Precipitation disturbances were used to generate the precipitation dataset that enabled the model to produce ensemble forecasts. This approach somewhat reduced the impact of uncertainties inherent in precipitation forecasts. The Ensemble–KNN forecasting method was then used to generate medium- to long-term inflow forecasts for the Danjiangkou Reservoir in China, which is the water source for the South-to-North Water Diversion Middle Route Project. The results proved the validity and reliability of the proposed modeling method.