机理和数据混合驱动的城市排水管网建模方法研究
Hybrid modelling study for urban drainage systems simulation based on data-driven and mechanism-driven methods
黄振宇 1王一茗 1张大臻 1董欣2
作者信息
- 1. 清华大学环境学院,北京 100084
- 2. 清华大学环境学院,北京 100084;环境模拟与污染控制国家重点联合实验室,北京 100084
- 折叠
摘要
针对城市排水管网的高精度仿真,提出一种机理与数据混合驱动的城市排水管网水力建模方法.该方法利用一维圣维南方程组和人工神经网络分别刻画排水系统中的水量传输过程、机理方程模拟的误差.为评估模型性能,以某市排水系统某排水片区为模拟对象建模,分别利用实际监测和数学试验进行混合驱动模型测试,并与纯机理驱动模型进行对比.结果表明混合驱动模型在实际降雨下流量、液位模拟准确性比纯机理模型提升26.5％;在数学试验场景下流量、液位准确性分别提升59.0％和6.1％,稳定性提升54.5％,2.6％;而模拟耗时仅增加了 1.1％.研究提出的混合驱动仿真模型在一定程度上提升了模拟的准确性和稳定性,可更好地支撑城市排水系统的运行管理.
Abstract
This paper proposes a high-precision simulation method for urban drainage pipe net-work based on a hybrid mechanism-driven and data-driven approach.The method utilizes the one-dimensional Saint Venant equations and artificial neural networks to respectively depict the water transmission process in the drainage system and the errors in the mechanism equation simulation.To evaluate the model performance,a certain drainage sub-area of a city's drainage system is mod-elled as the simulation object.Hybrid-driven model testing is conducted through both actual real system monitoring and mathematical experiments,and compared with a pure mechanism-driven model.The results show that under actual rainfall,the accuracy of flow and water level simulation of the hybrid-driven model is 26.5％higher than that of the pure mechanism model;under mathe-matical experimental scenarios,the accuracy of flow and water level simulation is improved by 59.0％and 6.1％respectively,stability is improved by 54.5％and 2.6％,while the simulation time only increases by 1.1％.The proposed hybrid-driven simulation model has improved the accu-racy and stability of the simulation to a certain extent,which can better support the operation and management of urban drainage systems.
关键词
排水管网/混合驱动建模/人工神经网络Key words
Urban drainage networks/Hybrid-driven modelling/Artificial neural network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据