基于GBDT的实时洪水预报误差校正方法

洪水预报实时校正是提高预报精度的有效途径。通过研究实时洪水预报误差系列构建方法,引入GBDT方法建立误差校正模型,并采用粒子群算法优选模型参数,选用洪峰段洪量相对误差、洪峰流量相对误差、确定性系数等指标评估实时校正效果。对淮河流域王家坝站点的实例应用结果表明,无论是率定期还是验证期,基于GBDT的实时预报误差校正方法精度均优于经典AR方法和KNN方法,各项指标精度均有不同程度提升,可有效提高实时洪水预报效果,且稳定性较高。

Real-time Correction Method of Flood Forecasting Error Based on GBDT

Real-time correction is an important way to improve the accuracy of flood forecasting. By constructing real-time flood forecasting error series, gradient boosting decision tree (GBDT) was used to establish error correction model of flood forecasting, and the model parameters were optimized by particle swarm optimization algorithm. The effect of real-time correction was evaluated in terms of flood volume relative error, flood peak discharge relative error and certainty coefficient. Taking Wangjiaba site in the Huaihe basin as an example, the results show that the method based GBDT is better than the classical AR and KNN method, and can improve the accuracy of each criterion in varying degrees and the accuracy of flood forecasting. The stability of this method is also higher.