Identification of Sensitive Parameters of Urban Flood Model Based on Artificial Neural Network

• Published in: Water resources management 2021-05, Vol.35 (7), p.2115-2128
• Main Authors: Wu, Zening, Ma, Bingyan, Wang, Huiliang, Hu, Caihong, Lv, Hong, Zhang, Xiangyang
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Atmospheric Sciences; Civil Engineering; Classification; Earth and Environmental Science; Earth Sciences; Environment; Environmental indicators; Floods; Geotechnical Engineering & Applied Earth Sciences; Hydrogeology; Hydrology; Hydrology/Water Resources; Identification; Identification methods; Mathematical models; Neural networks; Nodes; Optimization; Parameter identification; Parameter sensitivity; Parameters; Runoff; Sensitivity analysis; Simulation; Storms; Stormwater; Stormwater management; Urban runoff; Water management
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-021-02825-3

Sensitivity analysis of urban flood model parameters is important for urban flood simulation. Efficient and accurate acquisition of sensitive parameters is the key to real-time model calibration. In order to quickly obtain the sensitive runoff parameters of the urban flood simulation model, this study proposes an artificial neural network-based identification method for sensitive parameters. Artificial neural network (ANN) models were constructed with the binary classification and multi-classification methods, and used environmental indicators that affect the parameter sensitivity of different hydrological response units as the input, with the sensitivity parameters of the Storm water management model (SWMM) being the output. The optimization of the ANN was realized by adjusting the number of nodes in the hidden layer and the maximum number of iterations. An example application was conducted in Zhengzhou, China. The results show that the binary classification ANN quickly identified sensitive parameters, and the prediction accuracy of all parameters exceeded 96%. Convergence can be achieved when the number of nodes in the hidden layer does not exceed twice the number of input nodes, and the maximum number of iterations does not exceed 200. Rapid and accurate identification of the sensitive runoff parameters of the urban flood simulation model was achieved, which reduced the time required for parameter sensitivity analysis.