基于STGCN的洪水预报误差实时校正方法

Real-Time Correction Method of Flood Forecasting Error Based on STCGN

中小河流具有分布广、产汇流时间短、洪水突发性强、水文资料匮乏等特点,是当前洪水防控的薄弱环节。误差实时校正是提升洪水预报精度的有效途径,针对中小河流洪水预报预见期短、预报精度不高的问题,构建基于深度学习的误差校正模型,利用时空图卷积网络寻找能反映误差序列非线性关系的映射函数,以充分挖掘水文误差序列的时序特征和局部空间特征;提出基于收敛因子和位置更新策略的改进灰狼优化算法搜索时空图卷积网络的超参数,进一步提高模型参数的适用性。实验结果证明了算法在洪水预报实时校正中的有效性和适用性,具有良好的应用前景。

The characteristics of wide distribution,short runoff time and lack of hydrological data in small and medium-sized rivers make it become the weak link of current flood prevention and control.Aiming at the problem of short forecast period and low forecast accuracy of small and medium river flood forecasting,an error correction model based on deep learning was constructed in this paper to improve the accuracy of flood forecasting.A spatio-temporal graph convolution network(STCGN)was constructed to find a mapping function that can reflect the nonlinear relationship of the error sequence,and thus to fully explore the time-series and local spatial characteristics of the hydrological error sequence.Then,a grey wolf optimization algorithm based on convergence factor and location update strategy was proposed to search for hyper-parameters of spatio-temporal graph convolutional network to further improve the applicability of model parameters.The experimental results prove the validity and applicability of the model in real-time error correction of forecasts.