基于相似过程动态识别的径流预报方法OA

Based on similarity theory,this paper proposes a new runoff forecasting method that dynamically identifies similar processes.By constructing a rainfall-runoff spatiotemporal feature indicator system that in-corporates features such as rainfall magnitude,rainfall spatial distribution,and runoff processes,a sliding-window multi-timescale similarity search algorithm is employed to dynamically identify the most similar histori-cal events to the current process from a database of historical rainfall-runoff events.A dynamic weighted com-bination of their future runoff processes is then performed to generate a comprehensive similar runoff forecast.A dynamic extension method is further applied to bridge and correct the forecasted process with current mea-sured flow data,thereby yielding the runoff forecast process within the lead time.Using historical daily rainfall and runoff data from the Xiashan hydrological station in the Ganjiang River as a case study,the model perfor-mance is evaluated through a leave-one-year-out cross-validation approach.The results show that,in the ab-sence of forecasted rainfall information for the lead time,the proposed method achieves relatively high forecas-ting accuracy for 1-2 day lead times,effectively describing the overall variation trend of the runoff process and accurately capturing peak flow and the timing of peak occurrence.However,forecasting accuracy declines for a 3-day lead time.The inclusion of rainfall spatial distribution indicators significantly enhances the predictive performance.Sensitivity analysis of feature weights shows that the method is insensitive to variations in weight combinations,demonstrating good robustness.This approach can provide an effective new perspective for short-term runoff forecasting and flood early warning.