[1]康得军,赖李保壹,邱福杰,等.基于机器学习理论的SWMM参数自动率定方法[J].中国给水排水,2024,40(5):122-129.
　KANGDe-jun,LAILi-bao-yi,QIUFu-jie,et al.Automatic Calibration Method of SWMM Parameters Based on Machine Learning Theory[J].China Water & Wastewater,2024,40(5):122-129.
点击复制

基于机器学习理论的SWMM参数自动率定方法

中国给水排水[ISSN:1000-4062/CN:12-1073/TU] 卷: 第40卷 期数: 2024年第5期 页码: 122-129 栏目: 出版日期: 2024-03-01
Title:
Automatic Calibration Method of SWMM Parameters Based on Machine Learning Theory
作者:
康得军1,2， 赖李保壹3， 邱福杰4， 温儒杰5， 吴端炜6， 赵颖7
（1.福州大学 土木工程学院，福建 福州 350108；2.福州大学晋江科教园，福建 晋江 362251；3.福建省龙岩市长汀县濯田镇政府，福建 龙岩 366300；4.华东勘测设计院<福建>有限公司，福建 福州 350001；5.中冶京诚工程技术有限公司，北京 100176；6.中国市政工程中南设计研究总院有限公司，湖北 武汉 430010；7.中国环境科学研究院，北京 100012）
Author(s):
KANG　De-jun1，2， LAI　Li-bao-yi3， QIU　Fu-jie4， WEN　Ru-jie5， WU　Duan-wei6， ZHAO　Ying7
（1. College of Civil Engineering， Fuzhou University， Fuzhou 350108， China； 2. Fuzhou University Jinjiang Science and Education Park， Jinjiang 362251， China； 3. Zhuotian Town Government of Changting County in Longyan City of Fujian Province， Longyan 366300， China； 4. Huadong Engineering Corporation， Fuzhou 350001， China； 5. Capital Engineering & Research Incorporation Limited， Beijing 100176， China； 6. Central & Southern China Municipal Engineering Design and Research Institute Co. Ltd.， Wuhan 430010， China； 7. Chinese Research Academy of Environmental Sciences， Beijing 100012， China）
关键词:
机器学习理论;  Metropolis-Hastings算法;  雨洪管理模型(SWMM);  MATLAB;  参数自动率定
Keywords:
machine learning theory;  Metropolis-Hastings algorithm;  storm water management model (SWMM);  MATLAB;  automatic parameter calibration
摘要:
SWMM作为一种模拟降雨情况的软件，由于其模型参数在经验范围内选取的不确定性，随着模型不断地向前运行演绎，误差会持续累积，最终影响模型的模拟结果，而传统的参数率定方法则具有效率低、准确度差等缺点。为此，利用MATLAB软件调用SWMM的水力演算程序来完成两者的数据交互与集成，结合多元逐步回归法筛选出敏感性高的参数作为率定对象，进而通过数值实验的方式以敏感性分析结果为依据使用Bayes-MH机器学习算法实现SWMM参数的自动率定。对于不同的实测降雨场景，自动率定的结果均能够顺利输出，参数匹配度达95%以上，表明了该方法具备优秀的稳定性、自适应性与全范围寻参能力。
Abstract:
As a kind of software for simulating rainfall situation, errors of SWMM will accumulate continuously with the operation of the model, and ultimately affect the simulation results of the model due to the uncertainty in the selection of model parameters within the empirical range. The traditional parameter calibration methods have the disadvantages such as low efficiency and poor accuracy. Therefore, this paper completed the data interaction and integration between MATLAB and SWMM by using MATLAB to call the hydraulic calculus program of SWMM, selected the parameters with high sensitivity as the calibration object by using multiple stepwise regression method, and eventually realized the automatic calibration of SWMM parameters by using Bayes-MH machine learning algorithm and numerical experiment based on the sensitivity analysis. In different measured rainfall scenarios, the results of automatic calibration were output smoothly, and the parameter matching degree was more than 95%, indicating that the method had excellent stability, self-adaptability and full-range parameter searching ability.
更新日期/Last Update: 2024-03-01