超前洪水预估精度不高一直困扰防洪决策，为解决这一技术难题，提出一种基于暴雨洪水知识的相似性分 析方法，进行洪水预报预测。该方法从历史典型暴雨洪水知识中提取多要素特征指标，基于欧氏距离进行当前与 历史暴雨洪水的特征指标相似性判别，根据判别出的最相似洪水，经“峰-量”联合修正消除非一致性后，实时外推 预估未来洪水过程，构成一套完整的“多要素特征指标提取-历史暴雨洪水相似性判别-实时洪水修正外推预估” 技术。在沂河蒙阴站的应用结果表明，基于“降雨-径流”关系，对判别出的最相似洪水进行修正，显著提高了外推 预估洪水精度，洪峰流量相对误差δQm 的范围降至 10% 附近，峰现时间绝对误差ΔT的范围降至±2h 以内，径流深 相对误差δR的范围降至±20% 以内，且随着时间推移，精度水平不断提高。该套技术方法能够挖掘隐含在历史暴 雨洪水数据中的相似性，超前预估当前洪水变化过程，为洪水预报提供一种新的技术参考。

Flood forecasting and prediction are integral components of non-structural flood management measures. Methods for flood forecasting and prediction can generally be classified into two categories: process-driven approaches (hydrological models) and data-driven approaches. Traditionally, the focus has been on process-driven approaches, but with the accumulation of hydrological data and advancements in big data analytics, data-driven approaches have gained increasing attention. In particular, the application of artificial intelligence technology in the water industry has led to the emergence of hydrological data mining-based forecasting and prediction methods as a research hotspot. Conducting hydrological knowledge mining and prediction based on the principle of similarity has become an important research direction, offering a new technical means to uncover hidden patterns within rainfall, floods, and watershed surface information. This approach also promotes the automation and intelligence of water resources data processing, assisting in improving the accuracy of flood forecasting and prediction, thereby facilitating the modernization and precision of the water industry.In theory, the longer the series of hydrological data, the more torrential rain-induced flood knowledge can be extracted. However, hydrological data series in a changing environment often exhibit inconsistencies, which affect the accuracy of flood forecasting and prediction based on torrential rain-induced flood knowledge. Currently, research on historical similar torrential rain-induced flood knowledge considering inconsistencies in guiding real-time flood forecasting is relatively limited. In this context, a methodology based on the knowledge of torrential rain-induced floods for real-time flood forecasting and prediction is proposed. The proposed method focuses on historical records of typical torrential rain-induced floods and extracts rainfall feature indicators, such as the position of the rainstorm center, antecedent precipitation, total average rainfall, and rainfall processes. Multiple feature indicators are simultaneously assessed for their similarity using criteria such as Euclidean distance. By inferring historical typical floods based on similarity knowledge and incorporating the "rainfall-peak flow" or "rainfall-runoff" relationship before and after the change, a combined "peak-flow" correction approach is applied to ensure consistency. Real-time rolling extrapolation is then performed to estimate future flood processes, forming a comprehensive "multi-feature indicator extraction-historical torrential rain-induced flood similarity determination-real-time flood correction and extrapolation" technique.The application results at the Mengyin Station on the Yi River demonstrate the effectiveness of the proposed methodology. For any given torrential rain-induced flood event, the most similar historical flood event can be accurately identified through multiple feature indicators, ensuring the theoretical correctness of the technique. By considering the most identified similar flood event and applying suitable corrections to ensure consistency, the extrapolation and prediction of future flood processes significantly improve the accuracy of real-time flood forecasting compared to the direct application of similar flood processes.In summary, the suggested methodology, grounded in torrential rain-induced flood knowledge, introduces an effective avenue for real-time flood forecasting and prediction. By extracting multiple feature indicators, evaluating their similarity, and incorporating correction and extrapolation steps, it enables accurate identification of similar historical flood events and enhances the precision of real-time flood forecasting. This study contributes to the progression of flood management and establishes the groundwork for further research aimed at enhancing flood forecasting accuracy and propelling the modernization of the water industry.