CN-116090625-B - Quick prediction method for coastal city flood based on LightGBM and hydrologic hydrodynamic model

CN116090625BCN 116090625 BCN116090625 BCN 116090625BCN-116090625-B

Abstract

The invention discloses a rapid prediction method of coastal city flood based on LightGBM and a hydrographic model, which comprises the steps of firstly, obtaining sample data, designing various rainfall-tide level combined scenes, constructing a flood simulation model of a target area based on the hydrographic model of PCSWMM, simulating the maximum water depth of each selected flood point under different rainfall-tide level combined scenes by using the flood simulation model, extracting characteristic variables related to rainfall and characteristic variables related to tide level, and constructing and training a rapid prediction model of city flood based on LightGBM, and thirdly, predicting the maximum water accumulation depth of the flood points in city flood by using the trained city flood prediction model. The method has excellent performance and calculation efficiency, can realize accurate and rapid prediction of urban flood, and can also realize efficient and accurate prediction of urban areas with large space range or high space resolution.

Inventors

XU HONGSHI
XU KUI
HAN ZHENTAO
WU ZENING
WANG HUILIANG
Wang tianye

Assignees

郑州大学

Dates

Publication Date: 20260505
Application Date: 20230101

Claims (8)

1. The rapid prediction method for coastal city flood based on LightGBM and hydrologic hydrodynamic models is characterized by comprising the following steps: 1. The method comprises the steps of obtaining sample data, namely designing various rainfall-tide level combination scenes, constructing a flood simulation model of a target area based on a PCSWMM hydrologic hydrodynamic model, simulating the maximum water depth of each selected flood point under different rainfall-tide level combination scenes by using the flood simulation model, extracting characteristic variables related to rainfall and characteristic variables related to tide level, and taking the characteristic variables, the position of the flood point and the maximum water depth as sample data; 2. Constructing and training a city flood prediction model, wherein the constructing and training of the city flood prediction model comprises constructing the city flood prediction model based on LightGBM and training the constructed city flood prediction model by adopting sample data; 3. Predicting the maximum ponding depth of a submerged point in urban flooding by adopting a trained urban flooding prediction model; The PCSWMM-based hydrokinetic model constructs a flood simulation model of a target area, comprising: Firstly, collecting data of a digital elevation model, a drainage pipe network, a one-dimensional inspection well, a sub-catchment area and an obstacle, then, constructing a one-dimensional drainage model of the inspection well and a pipeline to obtain two-dimensional nodes and a two-dimensional network, then, constructing a two-dimensional earth surface inundation model according to the two-dimensional nodes and the two-dimensional grids, and finally, performing coupling of the two-dimensional model by using a bottom orifice connection mode to obtain a inundation simulation model.
2. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, wherein the method is characterized by: The characteristic variables related to rainfall comprise one or more of accumulated rainfall, rainfall recurrence period, rainfall peak value, maximum 2h rainfall, maximum 3h rainfall and accumulated rainfall before peak value, and the characteristic variables related to the tide level comprise one or more of maximum tide level, tide level recurrence period, average tide level and maximum 5h average tide level.
3. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, wherein the method is characterized by: the design of multiple rainfall-tidal level combination scenarios includes: The method comprises the steps of fitting frequency distribution of rainfall and tide level by using a generalized extremum function, estimating parameters of the generalized extremum function, including position parameters, scale parameters and shape parameters, obtaining more than one design rainfall and tide level peak value of a reproduction period from 5 years to 100 years by calculating the generalized extremum function, scaling typical rainfall and tide level processes by using a same-magnification ratio method to obtain design rainfall and tide level processes of different reproduction periods, and combining rainfall and tide level data to obtain multiple design rainfall-tide level combined scenes serving as boundary conditions of a flood simulation model.
4. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, further comprising: And selecting rainfall and tide level sequences actually measured in the target area and determining the constructed flood simulation model by submerged water depth.
5. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, wherein the method is characterized by: the city flood prediction model constructed by adopting sample data training comprises the following steps: selecting a plurality of flood points from a flood-prone area, extracting a plurality of characteristic variables related to rainfall and a plurality of characteristic variables related to tide level, inputting the positions of the flood points and the extracted characteristic variables into a flood simulation model for simulation, simulating the maximum water depths of the flood points under different rainfall-tide level combined scenes, taking the different rainfall-tide level combined scenes and the flood points as the input of a LightGBM model, taking the maximum water depths of the flood points as the output of a LightGBM model, and training a LightGBM model.
6. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, wherein the method is characterized by: the method also comprises the following steps of optimizing the super parameters of the LightGBM model by adopting a K-fold cross validation and grid search method, and specifically comprises the following steps: Defining a super-parameter range and interval of LightGBM models, forming a three-dimensional super-parameter grid by the super-parameter range and interval, traversing the super-parameter grids by utilizing a grid search method, performing K-fold cross validation on each super-parameter grid, and selecting the super-parameter with the best generalization capability; The K-fold cross validation comprises the steps of dividing sample data into K subsets randomly and uniformly, using K-1 subsets for training a model each time a model is trained and tested, using the rest subsets for testing, and averaging K test results after repeating K times to obtain a model evaluation value.
7. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, wherein the method is characterized by: the super parameters of the LightGBM model include learning rate, estimator number, and leaf node number.
8. The rapid prediction method for coastal city flooding based on LightGBM and hydrokinetic models as claimed in claim 1, wherein the method is characterized by: The method also comprises the step of analyzing LightGBM the contribution degree of the input characteristic variable of the model to urban flood prediction of the research area by using the keni index.

Description

Quick prediction method for coastal city flood based on LightGBM and hydrologic hydrodynamic model Technical Field The application belongs to the technical field of urban flood early warning, and particularly relates to a coastal urban flood rapid prediction method based on LightGBM and a hydrologic hydrodynamic model. Background At present, urban flood prediction models can be divided into physical models and data driving models, wherein the physical models mainly refer to hydrologic hydrodynamic models, the technical development of the model is relatively mature, and the calculation accuracy is high. With the progress of numerical simulation technology in recent years, a hydrokinetic model has been widely applied in the field of urban flood simulation. However, the complexity and long calculation time of the hydrographic hydrodynamic model limit the development of the model, especially in urban areas with large space range or high resolution requirements, the calculation time of the hydrographic hydrodynamic model will be longer, and the real-time flood prediction requirement is difficult to meet. In addition, the hydrologic hydrodynamic model has high requirements on flood monitoring data, a large number of parameters need to be calibrated to accurately describe the flood process, and the popularization and application of the flood monitoring data are affected by the serious lack of the flood monitoring data. In addition to physical models, data driven models are also widely used in the field of urban flood prediction. The data-driven model recognizes a pattern between input and output by learning a large amount of data, and directly gives model output according to the model input, thereby enabling fast prediction of flooding. The data driven model requires a large amount of flood data to construct an effective flood prediction model. However, in many areas, the use of data driven models is limited due to the lack of measured data. One possible solution is to combine a physical model with a data-driven model, and utilize the physical model to simulate a large amount of flood data to train the data-driven model, so as to replace the physical model to realize rapid prediction of the flood. For example, kabir [1] in combination with LISFLOOD-FP and Convolutional Neural Network (CNN) is used for rapid prediction of river flood water depth. Berkhahn [2] utilizes HYSTEM-EXTRAN D simulation to generate a flood map database, and constructs a city flood rapid prediction model based on an integrated neural network. Lowe [3] proposes a method combining MIKE and convolutional neural networks for urban river flood depth prediction. However, the gradient promotion tool in the above study uses a pre-ordering based algorithm and a step-by-step growth tree strategy, and needs to represent category features through single-hot coding, which increases memory consumption and reduces training speed. The following references are referred to herein: [1] Kabir S, Patidar S, Xia X, et al. A deep convolutional neural network model for rapid prediction of fluvial flood inundation [J]. Journal of Hydrology, 2020, 590, 125481. [2] Berkhahn S, Fuchs L, Neuweiler I. An ensemble neural network model for real-time prediction of urban floods [J]. Journal of Hydrology, 2019, 575: 743-754. [3] Löwe R, Böhm J, Jensen D G, et al. U-FLOOD – Topographic deep learning for predicting urban pluvial flood water depth [J]. Journal of Hydrology, 2021, 603, 126898. Disclosure of Invention In view of this, the present application proposes a rapid urban flood prediction method based on LightGBM and hydrokinetic models to further reduce memory consumption and increase training speed. The technical scheme of the application is realized as follows: the application provides a rapid prediction method for coastal city flood based on LightGBM and a hydrokinetic model, which comprises the following steps: 1. The method comprises the steps of obtaining sample data, namely designing various rainfall-tide level combination scenes, constructing a flood simulation model of a target area based on a PCSWMM hydrologic hydrodynamic model, simulating the maximum water depth of each selected flood point under different rainfall-tide level combination scenes by using the flood simulation model, extracting characteristic variables related to rainfall and characteristic variables related to tide level, and taking the characteristic variables, the position of the flood point and the maximum water depth as sample data; 2. Constructing and training a city flood prediction model, wherein the constructing and training of the city flood prediction model comprises constructing the city flood prediction model based on LightGBM and training the constructed city flood prediction model by adopting sample data; 3. and predicting the maximum ponding depth of the submerged point in the urban flood by adopting a trained urban flood prediction model. In some embodiments, the number of chara