Simulation forecasting method suitable for urban rainstorm waterlogging in area lacking pipe network data

Abstract

The invention discloses a simulation forecasting method suitable for urban rainstorm waterlogging in areas lacking pipe network data, which is characterized in that for areas without pipe network data, rainwater inspection well information is acquired by calling a Baidu map, pipeline drainage flow is generalized based on a rainwater well equivalent drainage method, and for areas with pipe network data, pipeline drainage flow is generalized by calling a Baidu map. Constructing a drainage pipe network model based on a one-dimensional Saint-View equation by adopting a finite difference method; then, constructing a hydrodynamic model, coupling a rainwater well equivalent drainage method and a drainage pipe network model with an FVCOM model, simulating an earth surface ponding evolution process, and obtaining a rainstorm waterlogging data set; and training and establishing a water depth prediction model based on the obtained data, then predicting the water depth of each prediction point after a specific time in an actual rainfall event, and giving an alarm in advance when a waterlogging early warning range is reached. According to the method, the submerged water depth of a waterlogging-prone point can be predicted in advance, waterlogging forecasting of a city lacking pipe network data can be achieved, and rainstorm disaster loss is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of urban rain and flood management, and in particular to a method for simulating and forecasting urban rainstorm waterlogging. Background Art

[0002] Climate warming and human activities have changed the natural hydrological cycle, increased the probability of extreme hydrological events, and made urban waterlogging caused by sudden rainstorms increasingly serious.

[0003] Mathematical models are an important scientific tool for studying urban flooding and waterlogging processes. With the advancement of model principles and numerical solution methods, urban stormwater simulation methods have evolved from simple hydrological or hydrodynamic approaches to coupled hydrological and hydrodynamic approaches. Among one-dimensional pipe network models, the Stormwater Runoff Management Model (SWMM), developed by the US Environmental Protection Agency, has been widely used due to its open source code and relatively independent computational components. Currently, the two most widely used hydrological and hydrodynamic coupling models are semi-distributed and fully distributed, depending on their technical strategies and methods of calculating runoff. The semi-distributed model lacks physical mechanisms, and runoff paths do not conform to actual physical processes, making it difficult to represent urban road runoff and inundation calculations in low-lying areas. The fully distributed model, on the other hand, suffers from high computational complexity and low efficiency. With the continuous development of high-performance computing technology, these computational efficiency issues have been greatly alleviated. However, even with the parallel computing of today's advanced GPU processors, the computational efficiency of large-scale fully distributed models still cannot meet the requirements for timely warnings in real-world scenarios. Therefore, at a time when artificial intelligence technology is developing rapidly, combining fully distributed hydrological and hydrodynamic models based on physical mechanisms with deep learning algorithms to conduct urban flood warning and forecasting is a technology that urgently needs to be implemented and improved.

[0004] Furthermore, constructing urban flooding models often requires underground drainage network data. However, accurate network data is often extremely difficult to obtain due to concerns about urban infrastructure safety or lack of maintenance. This lack of network data leads to low model accuracy when conducting urban flooding simulations and early warnings. Existing technologies for stormwater simulations in areas lacking network data often use empirical coefficients, assume uniform drainage, or simply ignore the effectiveness of network drainage, resulting in significant discrepancies between calculated results and actual results.

[0005] Therefore, for areas where pipeline network data is missing, it is urgent to propose a method for automatically acquiring pipeline network data, build a hydrological and hydrodynamic model based on a fully distributed coupling method with a more sound physical mechanism, and combine it with a deep learning algorithm to research and develop efficient and high-precision urban waterlogging forecasting technology. Summary of the Invention

[0006] In view of the above-mentioned deficiencies in the existing technology, the technical problem to be solved by the present invention is: how to provide a method for rainstorm waterlogging forecasting in cities in areas lacking pipe network data, which can reliably and effectively realize waterlogging forecasting in cities lacking pipe network data and reduce the losses caused by rainstorm disasters.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions: A method for simulating and forecasting urban rainstorm waterlogging in areas lacking pipe network data is disclosed. The method is characterized in that, for areas without pipe network data, rainwater inspection well information is obtained by calling the panoramic static API interface (data interface) of the Baidu Map Open Platform, and the pipeline drainage flow is generalized based on the rainwater well equivalent drainage method. For areas with pipe network data, a drainage pipe network model based on the one-dimensional Saint-Venant equation is constructed using the finite difference method; then, a two-dimensional hydrodynamic model based on FVCOM (finite volume ocean model) is constructed, and the rainwater well equivalent drainage method and the drainage pipe network model are coupled with FVCOM to simulate the evolution process of surface water accumulation and obtain a rainstorm waterlogging dataset; and a water depth prediction model is established based on the obtained data for training, and then the water depth after a specific time at each prediction point is predicted in actual rainfall events.

[0008] The present invention specifically comprises the following steps: S1. For areas without pipe network data, use the panoramic static API function of the Baidu Map Open Platform to batch obtain the latitude and longitude coordinates of all roads within the target area. Use a Python program to download panoramic street view images of the target area. Manually identify and locate all stormwater manholes, and generate a stormwater manhole distribution map. Use the stormwater manhole equivalent drainage method to generalize the pipe network drainage process, and calculate the drainage flow rate of each stormwater manhole using the weir flow formula. S2. For areas with available pipe network data, a finite difference method is used to construct a drainage pipe network model based on the one-dimensional Saint-Venant equation. S3. Build a two-dimensional hydrodynamic model based on the terrain data of the target area, coupling the equivalent drainage method for rainwater wells and the drainage network model with FVCOM. Calculate rainfall runoff at all grid nodes using the direct rainfall method. Use the resulting surface runoff as a boundary condition to drive a coupled model to simulate the dynamic water exchange between the network and the surface. S4. Filter rainfall events that have resulted in waterlogging from historical rainfall events recorded in the target area's hydrological and meteorological database, obtain corresponding rainfall information data, and obtain a set of rainfall information data that will result in waterlogging. The rainfall information data includes various rainfall-related data that directly affect surface water catchment, including but not limited to rainfall amount, rainfall duration, and rainfall type (to ensure that the selected rainstorm events are representative and typical, reflecting the main characteristics of waterlogging caused by rainstorms in the study area). S5. Based on the rainfall information data set obtained in S4, randomly interleave and combine the rainfall information required for simulating several single rainfall events. These are then input into the two-dimensional hydrodynamic model coupled with S3 to simulate the surface water catchment process during a single rainfall event, thereby obtaining a rainstorm waterlogging dataset. The rainstorm waterlogging dataset includes rainfall information (precipitation amount, duration, and type) corresponding to each time point during the rainfall process, as well as water depth data (including inundation depth and inundation range) for the target grid. S6. Using the rainfall information data at each time point as the input factor and the target grid water depth data after a specific time point (i.e., the forecast lead time) as the output factor, a training set and a test set are selected from the rainstorm waterlogging dataset. A long short-term memory artificial neural network (LSTM) model is used for training and testing to form a water depth prediction model. S7. In actual rainfall events, the rainfall information data at the current time point is used as the input factor and input into the water depth prediction model obtained in S6 to predict the water depth data of each prediction point (target grid) after a specific time. When the water depth data reaches the waterlogging warning range, an advance warning can be issued in advance.

[0009] Furthermore, in S1, the drainage flow of each rainwater inspection well is calculated according to the following weir flow formula:

[0010] Among them, Q i is the drainage flow of the i-th rainwater inspection well (m³ / s), C is the weir flow coefficient, L i is the circumference of the rainwater well (m), H i is the water head above the weir (m).

[0011] Furthermore, in S2, the constructed drainage network model is based on the one-dimensional Saint-Venant equation, and the Preissmann narrow gap method (a numerical method for simulating water flow movement) is introduced to simulate the hydraulic process of alternating open and full flow in a closed pipe.

[0012] Furthermore, in S3, when coupling the equivalent drainage method for rainwater wells and the drainage network model with FVCOM, the 'ADEAL' function in the FVCOM source code is used to transfer the node overflow flow value to the local grid node, and the 'NGID' function is used to transfer the infiltration flow (calculated by the equivalent drainage method for rainwater wells) or the surface runoff (calculated by the direct rainfall method) value to the inspection well node of the network.

[0013] The specific coupling process may include the following steps: Step 1: Store the rainwater well equivalent drainage method and drainage network model in the FVCOM source code folder FVCOM_source as a Fortran 95 programming language module (*.F format). Add the module names of the rainwater well equivalent drainage method and drainage network model to the original makefile file in the folder. Step 2: Call the rainwater well equivalent drainage method and drainage network model module in the 'BCOND_GCN' subroutine (SUBROUTINE) of the FVCOM model source code file 'bcond_gcn.F'; Step 3: In the 'BCOND_GCN' subroutine (SUBROUTINE) of the FVCOM model source code file 'bcond_gcn.F', use the 'ADEAL' function to read the node overflow flow value calculated by the drainage network model; Step 4: In the 'BCOND_GCN' subroutine (SUBROUTINE) of the FVCOM model source code file 'bcond_gcn.F', use the 'NGID' function to assign the infiltration flow (calculated by the equivalent drainage method of the rainwater well) or surface runoff (calculated by the direct rainfall method) value to the corresponding inspection well node.

[0014] Furthermore, in S3, the runoff simulation using the direct rainfall method is calculated using the following formula:

[0015] Among them, Q m is the flow rate of the mth calculation unit (m³ / s), I is the rainfall intensity (mm / min), f t is the infiltration rate at time t calculated by the Horton model (mm / h), A m is the unit area (m²).

[0016] Furthermore, the terrain data of the target area in S3 are generated based on the DEM data obtained from the geospatial data cloud platform (https: / / www.gscloud.cn / home).

[0017] Furthermore, the terrain data of the target area in S3 is constructed using a high-precision digital elevation model (DEM).

[0018] This can more accurately reflect the topographic and geomorphic characteristics of the study area. The terrain data format is usually *.asc, which contains the elevation information of each grid point.

[0019] In S3, the direct rainfall method is a common approach for calculating surface runoff. Based on rainfall data and the surface infiltration capacity, the runoff at each grid point is calculated as one of the model inputs. Specifically, the groundwater input file format specified in the FVCOM model manual is prepared, and the groundwater module is enabled in the model run management file (_run.nml).

[0020] Furthermore, in the rainstorm waterlogging dataset simulated in S5, the rainfall information data corresponding to each time point includes rainfall amount, rainfall duration and rainfall type; the target grid water depth data includes the flooded water depth and flooded range of the target grid.

[0021] Furthermore, the target grid in S6 refers to the grid position in the model area corresponding to the coordinates of each flood-prone location within the prediction range city.

[0022] Compared with the prior art, the present invention has at least the following beneficial effects: 1. Innovative data acquisition method: Utilizing the panoramic static API function of the Baidu Maps open platform to obtain the distribution of stormwater inspection wells, this method addresses the issue of missing critical data in areas lacking pipe network data. Compared to traditional manual surveys, this method is over 10 times more efficient and 80% less costly, providing a viable solution for flood simulation in areas lacking pipe network data.

[0023] 2. In terms of mechanism model construction, by proposing a two-dimensional hydrodynamic model bidirectional coupling method and combining it with the direct rainfall method to calculate the entire hydrodynamic process, the shortcomings of the existing unidirectional coupling and semi-distributed hydrodynamic models with insufficient physical mechanisms are compensated.

[0024] 3. Based on the fusion of physical mechanisms and data-driven approaches, a mechanistic model is constructed to ensure the physical rationality of the simulation process. An LSTM network is used to capture complex nonlinear relationships, forming a comprehensive modeling framework combining physical mechanisms with temporal features. Tests have shown that this hybrid model achieves 25% higher prediction accuracy than a purely data-driven model.

[0025] 4. In terms of prediction efficiency, through model coupling and parallel computing optimization, the prediction time is shortened from hours in traditional mechanism models to minutes, meeting the needs of real-time prediction.

[0026] 5. Wide applicability: The present invention is not only applicable to mountainous cities, but can also be applied to different types of urban areas such as plain cities and coastal cities, providing a universal technical solution to solve the problem of urban waterlogging in my country. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic flow chart of the specific steps of the present invention in a specific implementation manner. DETAILED DESCRIPTION

[0028] The present invention will be further described in detail below with reference to specific embodiments.

[0029] Specific implementation method: A method for simulating and forecasting urban rainstorm waterlogging in areas lacking pipe network data, characterized in that, for areas without pipe network data, rainwater inspection well information is obtained by calling the panoramic static API interface of Baidu Map Open Platform (https: / / lbsyun.baidu.com / faq / api?title=viewstatic), and the pipeline drainage flow is generalized based on the rainwater well equivalent drainage method; for areas with pipe network data, a finite difference method is used to construct a drainage pipe network model based on the one-dimensional Saint-Venant equation; then a two-dimensional hydrodynamic model based on FVCOM (finite volume ocean model) is constructed, and the rainwater well equivalent drainage method and the drainage pipe network model are coupled with FVCOM to simulate the evolution process of surface water accumulation and obtain a rainstorm waterlogging data set; and a water depth prediction model is established based on the obtained data training, and then the water depth after a specific time at each prediction point is predicted in an actual rainfall event.

[0030] See also Figure 1 , the present invention specifically includes the following steps: S1. For areas without pipe network data, use the panoramic static API function of the Baidu Map Open Platform to batch obtain the latitude and longitude coordinates of all roads within the target area. Use a Python program to download panoramic street view images of the target area. Manually identify and locate all stormwater manholes, and generate a stormwater manhole distribution map. Use the stormwater manhole equivalent drainage method to generalize the pipe network drainage process, and calculate the drainage flow rate of each stormwater manhole using the weir flow formula. S2. For areas with available pipe network data, a finite difference method is used to construct a drainage pipe network model based on the one-dimensional Saint-Venant equation. S3. Build a two-dimensional hydrodynamic model based on the terrain data of the target area, coupling the equivalent drainage method for rainwater wells and the drainage network model with FVCOM. Calculate rainfall runoff at all grid nodes using the direct rainfall method. Use the resulting surface runoff as a boundary condition to drive a coupled model to simulate the dynamic water exchange between the network and the surface. S4. Filter rainfall events that have resulted in waterlogging from historical rainfall events recorded in the target area's hydrological and meteorological database, obtain corresponding rainfall information data, and obtain a set of rainfall information data that will result in waterlogging. The rainfall information data includes various rainfall-related data that directly affect surface water catchment, including but not limited to rainfall amount, rainfall duration, and rainfall type (to ensure that the selected rainstorm events are representative and typical, reflecting the main characteristics of waterlogging caused by rainstorms in the study area). S5. Based on the rainfall information data set obtained in S4, randomly interleave and combine the rainfall information required for simulating several single rainfall events. These are then input into the two-dimensional hydrodynamic model coupled with S3 to simulate the surface water catchment process during a single rainfall event, thereby obtaining a rainstorm waterlogging dataset. The rainstorm waterlogging dataset includes rainfall information (precipitation amount, duration, and type) corresponding to each time point during the rainfall process, as well as water depth data (including inundation depth and inundation range) for the target grid. S6. Using the rainfall information data at each time point as the input factor and the target grid water depth data after a specific time point (i.e., the forecast lead time) as the output factor, a training set and a test set are selected from the rainstorm waterlogging dataset. A long short-term memory artificial neural network (LSTM) model is used for training and testing to form a water depth prediction model. S7. In actual rainfall events, the rainfall information data at the current time point is used as the input factor and input into the water depth prediction model obtained in S6 to predict the water depth data of each prediction point (target grid) after a specific time. When the water depth data reaches the waterlogging warning range, an advance warning can be issued in advance.

[0031] During implementation, S1 specifically includes the following processes: (1) Extracting the coordinate points of all roads in the area lacking pipe network data in the coordinate picking interface provided by the Baidu Map open platform. After entering the coordinate picking system interface from Baidu Map, simply click on the corresponding location to obtain the corresponding latitude and longitude coordinates. Each time the coordinates are obtained, they must be exported and used as input information when downloading the street view panorama; (2) Take all the coordinates extracted in step (1) as input, run the Python program based on the panoramic static API interface, and batch download street panoramic photos of the specified coordinate location. This step requires applying for the creation of a panoramic static image API application on the Baidu Map Open Platform in advance, completing developer certification and submitting information such as the service key (AK) as required. After a successful application, there is a 3-month trial period. To use the panoramic static image API function, you only need to set parameters such as image size, latitude and longitude coordinates, send an HTTP request to access the Baidu Map panoramic static image service, and you can display the panoramic image as an image in the browser; (3) The location distribution of the inspection wells is determined by screening the photos downloaded in step (2). In the Python program, the downloaded street view panoramic photos are set to latitude and longitude coordinates and direction names. Through manual screening, it is determined whether there are rainwater inspection wells at the location, and the coordinates of the rainwater inspection wells are recorded and organized.

[0032] In this way, the distribution information of rainwater inspection wells in areas lacking pipe network data is obtained through the Baidu Map Open Platform, which provides useful information for improving the accuracy of rainstorm waterlogging simulation in areas lacking pipe network data. At the same time, it also avoids the shortcomings of high cost and low efficiency of manual field surveys in large areas lacking pipe network data.

[0033] During implementation, in S1, the drainage flow of each rainwater inspection well is calculated according to the following weir flow formula:

[0034] Among them, Q i is the drainage flow of the i-th rainwater inspection well (m³ / s), C is the weir flow coefficient, L i is the circumference of the rainwater well (m), H i is the water head above the weir (m). The specific coefficient can be adjusted during implementation based on the designed discharge value and actual waterlogging. This is a conventional technology and will not be described in detail here.

[0035] The stormwater well equivalent drainage method employed requires generalizing the network drainage to the location of the stormwater manhole. This is achieved by deducting the equivalent infiltration volume from the grid node corresponding to the stormwater manhole. The correspondence between stormwater manholes and grid nodes follows the principle of proximity. Generalizing the drainage network through stormwater wells more closely resembles the actual network drainage process: rainwater, after generating runoff, flows through roads, into stormwater wells, and then into pipes, ultimately being transported downstream or to other water storage facilities. The stormwater well equivalent drainage method takes into account the impact of actual surface topography on rainwater generation and convergence, accurately reflecting the flow of rainwater into the network and improving the accuracy of stormwater and flood model simulations.

[0036] During implementation, in S2, the constructed drainage network model is based on the one-dimensional Saint-Venant equation, and the Preissmann narrow gap method (a numerical method for simulating water flow movement) is introduced to simulate the hydraulic process of alternating open and full flow in a closed pipe.

[0037] During implementation, in S3, when coupling the stormwater well equivalent drainage method and the drainage network model with FVCOM, the 'ADEAL' function in the FVCOM source code is used to transfer the node overflow flow value to the local grid node, and the 'NGID' function is used to transfer the infiltration flow (calculated by the stormwater well equivalent drainage method) or surface runoff (calculated by the direct rainfall method) value to the pipe network inspection well node.

[0038] During implementation, the coupling steps specifically include: Step 1: Store the rainwater well equivalent drainage method and drainage network model in the FVCOM source code folder FVCOM_source as a Fortran 95 programming language module (*.F format). Add the module names of the rainwater well equivalent drainage method and drainage network model to the original makefile file in the folder. Step 2: Call the rainwater well equivalent drainage method and drainage network model module in the 'BCOND_GCN' subroutine (SUBROUTINE) of the FVCOM model source code file 'bcond_gcn.F'; Step 3: In the 'BCOND_GCN' subroutine (SUBROUTINE) of the FVCOM model source code file 'bcond_gcn.F', use the 'ADEAL' function to read the node overflow flow value calculated by the drainage network model; Step 4: In the 'BCOND_GCN' subroutine (SUBROUTINE) of the FVCOM model source code file 'bcond_gcn.F', use the 'NGID' function to assign the infiltration flow (calculated by the equivalent drainage method of the rainwater well) or surface runoff (calculated by the direct rainfall method) value to the corresponding inspection well node.

[0039] During implementation, in S3, the following formula is used to calculate rainfall runoff using the direct rainfall method:

[0040] Among them, Q m is the flow rate of the mth calculation unit (m³ / s), I is the rainfall intensity (mm / min), f t is the infiltration rate at time t calculated by the Horton model (mm / h), A m is the unit area (m²).

[0041] The direct rainfall method employed for runoff simulation is a grid-based runoff simulation method. Areas lacking pipe network data are divided into structural units based on a topographic grid. Each calculation unit contains information such as elevation, roughness coefficient, permeability coefficient, percentage of impervious area, and depression depth. Surface runoff is generated independently for each unit. Each calculation unit is divided into permeable and impervious areas based on land use type or surface cover type, and runoff is calculated separately. For impervious areas, runoff is equal to rainfall intensity; for pervious areas, runoff is equal to rainfall intensity minus infiltration losses.

[0042] During implementation, infiltration models such as the Horton model and the Green-Ampt model can be used to estimate the infiltration rate of the permeable zone.

[0043] In this way, the obtained surface runoff flow is used as the boundary condition to drive the coupling model to simulate the dynamic water exchange process between the pipeline network and the surface. This not only takes into account the surface flow process of runoff before entering the pipeline system, but also makes up for the shortcomings of time-consuming and inefficient calculation of rainfall runoff in software such as InfoWorksICM.

[0044] During implementation, the terrain data of the target area in S3 were generated based on the DEM data obtained from the Geospatial Data Cloud Platform (https: / / www.gscloud.cn / home).

[0045] During implementation, the terrain data of the target area in S3 is constructed using a high-precision digital elevation model (DEM).

[0046] This can more accurately reflect the topographic and geomorphic characteristics of the study area. The terrain data format is usually *.dat, which contains the elevation information of each grid point.

[0047] In S3, equivalent boundary conditions for stormwater wells are key to simulating drainage processes in areas lacking pipe network data. Based on the distribution of stormwater manholes, the equivalent drainage method for stormwater wells is used to generalize the drainage network into equivalent boundary conditions for stormwater wells. The equivalent boundary condition file, typically in the *.nc format, contains information such as the location and drainage capacity of each stormwater well. Prepare the file according to the groundwater input format specified in the FVCOM model manual, and enable the groundwater module in the model run management file (_run.nml).

[0048] In S3, the direct rainfall method is a common method for calculating surface runoff. Based on rainfall data and the surface infiltration capacity, the runoff at each grid point is calculated as one of the model inputs.

[0049] In S3, the coupling model is compiled in the Linux system terminal to generate an executable file for simulation, and the hydrological and hydraulic element information such as water depth and flow velocity at each grid point is obtained.

[0050] During implementation, in the rainstorm waterlogging dataset simulated in S5, the rainfall information data corresponding to each time point includes rainfall amount, rainfall duration and rainfall type; the target grid water depth data includes the flooded water depth and flooded range of the target grid.

[0051] During implementation, the target grid in S4 refers to the grid position within the model area corresponding to the coordinates of each flood-prone location within the city within the prediction range.

[0052] At the time of implementation, the model information in this application is: software version: FVCOM (https: / / www.fvcom.org / ). Time step: 0.05 s, based on the CFL stability condition. Boundary conditions: Equivalent drainage from stormwater wells and surface runoff are calculated using the direct rainfall method. Manning's coefficient and infiltration parameters are adjusted based on actual conditions.

Claims

1. A method for simulating and forecasting urban rainstorm waterlogging in areas lacking pipe network data, characterized in that: For areas without pipe network data, the panoramic static API interface of the Baidu Map Open Platform is called to obtain rainwater inspection well information, and the pipeline drainage flow is generalized based on the equivalent drainage method of the rainwater well. For areas with pipe network data, the finite difference method is used to construct a drainage pipe network model based on the one-dimensional Saint-Venant equation; then a two-dimensional hydrodynamic model based on FVCOM is constructed, and the equivalent drainage method of the rainwater well and the drainage pipe network model are coupled with FVCOM to simulate the evolution process of surface water accumulation and obtain a rainstorm waterlogging data set; and a water depth prediction model is established based on the obtained data training, and then the water depth at each prediction point after a specific time is predicted in actual rainfall events.

2. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data according to claim 1, characterized in that: The specific steps include: S1. For areas without pipe network data, use the panoramic static API function of the Baidu Map Open Platform to batch obtain the latitude and longitude coordinates of all roads within the target area. Use a Python program to download all street view panoramic images within the target area. Manually identify and locate all stormwater manholes, and generate a stormwater manhole distribution map. Use the stormwater manhole equivalent drainage method to generalize the pipe network drainage process, and calculate the drainage flow rate of each stormwater manhole using the weir flow formula. S2. For areas with available pipe network data, a finite difference method is used to construct a drainage pipe network model based on the one-dimensional Saint-Venant equation. S3. Construct a two-dimensional hydrodynamic model based on the terrain data of the target area, and couple the equivalent drainage method of the rainwater well and the drainage network model with FVCOM. The direct rainfall method is used to calculate rainfall runoff at all grid nodes, and the resulting surface runoff flow is used as a boundary condition to drive the coupling model to simulate the dynamic water exchange process between the pipe network and the surface. S4. Filtering rainfall events that have caused waterlogging from historical rainfall events recorded in the target area's hydrological and meteorological database, obtaining corresponding rainfall information data, and obtaining a set of rainfall information data that may have caused waterlogging. The rainfall information data includes various rainfall-related data that directly affect surface water collection, including but not limited to rainfall amount, rainfall duration, and rainfall type. S5. Based on the rainfall information data set obtained in S4, randomly interweave and combine the rainfall information required for simulating several single rainfall events. These are then input into the coupled two-dimensional hydrodynamic model in S3 to simulate the surface water catchment process during a single rainfall event, thereby obtaining a rainstorm waterlogging dataset. The rainstorm waterlogging dataset includes rainfall information data corresponding to each time point during the rainfall process, as well as water depth data for the target grid. S6. Using the rainfall information data at each time point as input factors and the water depth data of the target grid after a specific time point as output factors, the rainstorm waterlogging dataset is divided into training and test sets, and a long-short-term memory artificial neural network model is used for training and testing to form a water depth prediction model; S7. In actual rainfall events, the rainfall information data at the current time point is used as an input factor and input into the water depth prediction model obtained in S6 to predict the water depth data for each prediction point after a specific time. When the water depth data reaches the waterlogging warning range, a pre-alarm can be issued in advance.

3. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data as claimed in claim 2, characterized in that: In S1, the drainage flow of each rainwater inspection well is calculated according to the following weir flow formula: Among them, Q i is the drainage flow of the i-th rainwater inspection well (m 3 / s), C is the weir flow coefficient, L i is the circumference of the rainwater well (m), H i is the water head above the weir (m).

4. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data according to claim 2, characterized in that: In S2, the constructed drainage network model is based on the one-dimensional Saint-Venant equation, and the Preissmann narrow slot method is introduced to simulate the hydraulic process of alternating open and full flow in a closed pipe.

5. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data as claimed in claim 2, characterized in that: In S3, when coupling the equivalent drainage method for rainwater wells and the drainage network model with FVCOM, the 'ADEAL' function in the FVCOM source code is used to transfer the node overflow flow value to the local grid node, and the 'NGID' function is used to transfer the infiltration flow or surface runoff value to the pipe network inspection well node.

6. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data as claimed in claim 2, characterized in that: In S3, the following formula is used to calculate runoff flow using the direct rainfall method: Among them, Q m is the flow rate of the mth calculation unit (m 3 / s), I is the rainfall intensity (mm / min), f t is the infiltration rate at time t calculated by Horton model (mm / h), A m is the unit area (m 2 ).

7. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data as claimed in claim 2, characterized in that: The terrain data of the target area in S3 is generated based on the DEM data obtained from the geospatial data cloud platform.

8. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data as claimed in claim 2, characterized in that: The terrain data of the target area in S3 is constructed using a high-precision digital elevation model (DEM).

9. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data as claimed in claim 2, characterized in that: In the simulated rainstorm waterlogging dataset in S5, the rainfall information data corresponding to each time point includes rainfall amount, rainfall duration, and rainfall type; the target grid water depth data includes the flooded water depth and flooded range of the target grid.

10. The method for predicting urban rainstorm waterlogging in areas lacking pipe network data according to claim 2, characterized in that: The target grid in S6 refers to the grid position in the model area corresponding to the coordinates of each flood-prone location within the prediction range city.

Images