基于数值模拟与BP神经网络的城市调蓄池调度快速预报方法  

Rapid Forecasting Method for Urban Storage Pond Scheduling Based on Numerical Simulation and BP Neural Network

作　　者：潘鑫鑫 侯精明 陈光照 李东来 周聂 梁鑫 吕佳豪 乔贤玲 呼媛[1] 高徐军 Pan Xinxin;Hou Jingming;Chen Guangzhao;Li Donglai;Zhou Nie;Liang Xin;Lyu Jiahao;Qiao Xianling;Hu Yuan;Gao Xujun(State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China,Xi'an University of Technology,Xi'an 710048,China;State Key Laboratory of Water Resources Engineering and Management,Wuhan University,Wuhan 430072,China;China Power Construction Group,Co.LTD,Northwest Engineering Corporation Limited,Xi'an 710048,China)

机构地区：[1]西安理工大学省部共建西北旱区生态水利国家重点实验室,西安710048 [2]武汉大学水资源与水电工程科学国家重点实验室,武汉430072 [3]中国电建集团西北勘测设计研究院有限公司,西安710065

出　　处：《水动力学研究与进展（A辑）》2023年第3期409-420,共12页Chinese Journal of Hydrodynamics

基　　金：国家自然科学基金项目(52009104、52079106);中德合作交流项目(M-0427)。

摘　　要：调蓄池是当前削减暴雨引起城市内涝问题的主要工程措施,然而调蓄池何时开启一般通过人工观测判定,难以及时有效地发挥调蓄池的调蓄作用。如何通过数值模拟和AI技术对调蓄池控制节点及相应的内涝点进行精准模拟和预测是亟待解决的问题,该文将一维SWMM模型和具有高精度的二维水动力模型进行耦合,再结合具有高效率的BP神经网络,提出了一种快速预报城市调蓄池调度的方法。该方法以耦合模型模拟出的结果作为数据驱动,构建调蓄池控制节点及内涝点的BP神经网络快速预报模型。结果表明:快速预报模型可在24 s内快速预测出调蓄池的控制节点水位及内涝点的积水深度,且预报模型的纳什效率系数NSE不低于0.90,误差较小。可以满足日常防汛应急的需要,有效降低生命财产损失。Currently,storage ponds are the main engineering measure to reduce urban waterlogging inundation caused by rainstorm.However,when to open storage ponds is usually determined through manual observation,which makes it difficult to play the role of storage ponds in a timely and effective manner.How to accurately simulate and predict the control nodes of storage pond and the corresponding points of waterlogging through numerical simulation and AI technology is an urgent problem that needs to be solved.This paper couples the one-dimensional SWMM model with the high-precision two-dimensional hydrodynamic model,and combines it with the highly efficient BP neural network to propose a fast-forecasting method for urban storage pond scheduling.This method uses the simulated results of the coupled model as the data driver to construct a BP neural network fast forecasting model for key control nodes and inundation points of storage ponds.The results show that the fast-forecasting model can quickly predict the water level at the key control nodes of the storage pond and the water depth at the inundation point within 24 s,and the NSE of the forecasting model is not less than 0.90.It can meet the needs of daily flood control emergency and effectively reduce the loss of life and property.

关 键 词：数值模拟 BP神经网络 快速预报 水文水动力模型 调蓄池 

分 类 号：TV139.2[水利工程—水力学及河流动力学]