CN-121998203-A - Deep learning flood evolution prediction method coupled with flood prediction and flood simulation

CN121998203ACN 121998203 ACN121998203 ACN 121998203ACN-121998203-A

Abstract

The invention discloses a deep learning flood prediction method for coupling flood prediction and inundation simulation, which comprises the steps of 1, obtaining underlying data and historical flood data of a research area, determining an accuracy evaluation index, 2, setting an input period and a prediction period, dividing the historical flood data into a deep learning flood prediction training set and a deep learning flood prediction verification set, constructing a deep learning flood prediction model, 3, driving a two-dimensional hydrodynamic model to generate a historical inundation process data set, 4, taking the historical flood data as input and the historical inundation process data set as output, training the deep learning inundation simulation model, 5, connecting the deep learning flood prediction model and the deep learning inundation simulation model in series to form a rapid deep learning flood inundation prediction framework, 6, evaluating the inundation prediction accuracy, and S7, improving analysis efficiency and prolonging effective prediction period. The invention strives for precious rescue time window for urban flood control, personnel evacuation and emergency material scheduling.

Inventors

CHEN XUANCHI
LI CHAOQUN
ZHANG YONGYONG
YAN DENGMING
LIANG GUOJIE
LIAN XIE
ZHANG DONGQING
SUN MINGKUN
Lei Kaixuan

Assignees

黄河勘测规划设计研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260401

Claims (6)

1. The deep learning flood evolution prediction method coupling flood prediction and inundation simulation is characterized by comprising the following steps of: Step S1, acquiring underlying surface data and historical flood data of a research area, and determining an accuracy evaluation index; step S2, setting an input period and a foreseeing period, dividing the historical flood data into a deep learning flood forecast training set and a deep learning flood forecast verification set, constructing a deep learning flood forecast model by using a gating circulation unit model, and verifying flood forecast accuracy; S3, driving a two-dimensional hydrodynamic model to generate a historical submerged process data set by using the underlying surface data of the research area and the historical flood data; s4, taking historical flood data as input, taking a historical flooding process data set as output, dividing a deep learning flooding simulation training set and a deep learning flooding simulation verification set, and training a deep learning flooding simulation model; S5, connecting the deep learning flood forecast model and the deep learning flooding simulation model in series to form a deep learning flood flooding rapid prediction framework; S6, selecting sampling points, and evaluating the precision of submerged prediction based on the deep learning submerged simulation verification set and the precision evaluation index; And S7, counting the operation time consumption of the deep learning flood inundation rapid prediction framework, comparing the operation time consumption with the two-dimensional hydrodynamic model, and improving the analysis efficiency and prolonging the effective prediction period.
2. The method for deep learning flood evolutionary prediction coupling flood forecast with inundation simulation according to claim 1, wherein in step S1, the determining the precision evaluation index comprises the following steps: s1.1, collecting a digital elevation model of a research area and upstream boundary conditions of an inflow river corresponding to a plurality of flood events; S1.2, preprocessing the digital elevation model by using a geographic information technology to generate standardized digital elevation model topographic data; s1.3, the precision evaluation indexes selected and used comprise root mean square error and Nash efficiency coefficient, and are used for evaluating the precision of flood forecast and flooding prediction, wherein the specific calculation formulas are as follows: Wherein RMSE is root mean square error, NSE is nash efficiency coefficient, yi is predicted value, ŷ i is observed value, ȳ is average value of observed values, and N is sample size.
3. The method for deep learning flood evolution prediction by coupling flood prediction and inundation simulation according to claim 1 or 2, wherein in step S2, the verification of flood prediction accuracy is specifically: S2.1, setting the input time period as a hours and the foreseeing time period as b hours according to the research purpose and the requirement, wherein the input of the deep learning flood prediction model is an a-hour flow monitoring value, and the output of the deep learning flood prediction model is a b-hour flow prediction value; s2.2, dividing historical flood data into the deep learning flood forecast training set and the deep learning flood forecast verification set according to a set proportion; s2.3, constructing the deep learning flood forecast model by using the gating circulating unit model, wherein a calculation formula is as follows: Wherein, the And The outputs of the update gate and the reset gate respectively, Is a candidate hidden state for use in the method, Is the hidden state of the current time step; representing a Sigmoid function, limiting the value between 0 and 1; Is a weight matrix; is the hidden state of the last time step; Is the input of the current time step; and S2.4, calculating the accuracy of the flood forecast result by using the deep learning flood forecast verification set so as to evaluate the forecast performance of the deep learning flood forecast model under different forecast periods.
4. The method for predicting deep learning flood by coupling flood prediction and inundation simulation according to claim 1 or 2, wherein in step S4, the training deep learning inundation simulation model is constructed by connecting long-time and short-time memory network modules in series on a convolutional neural network model, and the specific calculation formula of the modules is as follows: Wherein, the Representing a forget gate, an input gate and an output gate respectively; Is a cell state; Is in a hidden state; And Respectively a weight matrix and a bias term; the function is activated for Sigmoid, Is a hyperbolic tangent activation function.
5. The method for deep learning flood evolution prediction according to claim 1 or 2, wherein in step S5, the fast prediction framework for deep learning flood flooding is: And the deep learning flood prediction model reads flood monitoring data to conduct flood prediction, and a prediction result is provided for the deep learning flooding simulation model to conduct rapid prediction of flood flooding.
6. The method for deep learning flood evolution prediction according to claim 1 or 2, wherein in step S6, the estimated flooding prediction accuracy is: And selecting a plurality of key control points in the research area, extracting a predicted water depth sequence and a real water depth sequence of the points in the whole flood process, and calculating root mean square error and Nash efficiency coefficient indexes.

Description

Deep learning flood evolution prediction method coupled with flood prediction and flood simulation Technical Field The invention belongs to the field of natural disaster early warning, and particularly relates to a deep learning flood evolution prediction method for coupling flood prediction and inundation simulation. Background Flood is one of the most destructive natural disasters worldwide, forms serious threat to personal safety, economy and water resource management, and has great significance for quick and effective forecast of the evolution process. Traditional two-dimensional hydrodynamic models such as LISFLOOD-FP can be used for simulating the time-space evolution of flood range and water depth, but mainly depend on flood flow real-time monitoring data, and have low calculation efficiency, so that the high-resolution rapid forecasting requirement is difficult to meet. With the development of deep learning, on one hand, a deep learning model such as a Convolutional Neural Network (CNN) can establish a mapping relation between input and output by analyzing meteorological, hydrologic and historical flood data to realize rapid submerged simulation, which is a main means of rapid prediction research of current flood submerged, but mainly focuses on real-time simulation of a flood submerged process rather than pre-prediction with effective prediction period. On the other hand, a deep learning model such as a cyclic neural network (RNN) like a gate-controlled cyclic unit (GRU) can capture time evolution information from data, and provide possibility for flood forecasting. However, prior studies have not seen any information about combining the two to achieve an effective forecast of the future flooding process. Disclosure of Invention The invention aims to provide a deep learning flood evolution prediction method for coupling flood prediction and inundation simulation, which can improve the prediction timeliness and the calculation efficiency of future flood inundation prediction. In order to achieve the above purpose, the invention adopts the following technical scheme: The invention relates to a deep learning flood evolution prediction method for coupling flood prediction and inundation simulation, which comprises the following steps: Step S1, acquiring underlying surface data and historical flood data of a research area, and determining an accuracy evaluation index; Step S2, setting an input period and a foreseeing period, dividing the historical flood data into a deep learning flood forecast training set and a deep learning flood forecast verification set, constructing a deep learning flood forecast model (M1) by using a gate control circulation unit (GRU) model, and verifying flood forecast accuracy; S3, driving a two-dimensional hydrodynamic model (LISFLOOD-FP) by using the underlying surface data of the research area and the historical flood data to generate a historical submerged process data set; S4, taking historical flood data as input, taking a historical flooding process data set as output, dividing a deep learning flooding simulation training set and a deep learning flooding simulation verification set, and training a deep learning flooding simulation model (M2); s5, connecting the deep learning flood forecast model (M1) and the deep learning flooding simulation model (M2) in series to form a deep learning flood flooding rapid prediction framework; S6, selecting sampling points, and evaluating the precision of submerged prediction based on the deep learning submerged simulation verification set and the precision evaluation index; And S7, counting the operation time consumption of the deep learning flood inundation rapid prediction framework, and comparing with a traditional two-dimensional hydrodynamic model (LISFLOOD-FP), and improving the analysis efficiency and the effective prediction period. Further, in step S1, the acquiring the underlying data and the historical flood data of the research area, and determining the precision evaluation index specifically includes: S1.1, acquiring a Digital Elevation Model (DEM) of a research area and upstream boundary conditions of an inflow river corresponding to a multi-field flood event; S1.2, preprocessing the Digital Elevation Model (DEM) by using a geographic information technology, wherein the preprocessing comprises the steps of eliminating landform noise points, correcting interference of bridges, trees and artificial/natural structures on elevation data of a research area, and generating standardized Digital Elevation Model (DEM) landform data; S1.3, the precision evaluation indexes selected include, but are not limited to, root Mean Square Error (RMSE) and Nash efficiency coefficient (NSE), and are used for evaluating the precision of flood forecast and flooding forecast, wherein specific calculation formulas are as follows: Where yi is the predicted value, ŷ i is the observed value, ȳ is the mean of the observed values, and N is the sample