Urban inland inundation water depth prediction method and device and storage medium

Abstract

The invention discloses an urban inland inundation water depth prediction method and device and a storage medium, and belongs to the technical field of urban inland inundation water depth prediction.The method comprises the steps that drainage port flow in a set area around a waterlogging point and local terrain parameters and forecast precipitation of the set area around the waterlogging point in a future set time period are obtained; inputting the forecast precipitation into the constructed urban runoff simulation model to obtain drainage-free submerging parameters; the local terrain parameters, the drainage-free submerging parameters and the drainage port flow are input into a trained water depth prediction model, the maximum water depth of the waterlogging point is obtained, and the water depth prediction model is a random forest model. According to the method, the full-distributed runoff model is improved, the drainage-free submerging parameters are obtained by processing the forecast precipitation through the urban runoff simulation model, so that the maximum water depth of the waterlogging point is predicted, the maximum water depth of the waterlogging point in the urban waterlogging process can be predicted under the condition that drainage pipe network data are lacked, and the method is high in practicability. And the prediction accuracy rate reaches 86.89%.

Description

Technical Field

[0001] The present invention relates to a method, device and storage medium for predicting urban waterlogging depth, and belongs to the technical field of urban waterlogging depth prediction. Background Art

[0002] With the acceleration of the urbanization process, the problem of urban waterlogging has become one of the major challenges faced by urban sustainable development globally. Especially against the background of frequent climate change and extreme weather events, the deficiencies of urban drainage systems and the increase in impervious surface areas in cities have led to a continuous rise in the frequency and intensity of urban waterlogging disasters, seriously affecting the lives and safety of urban residents and the normal operation of cities. Using waterlogging models for urban waterlogging prediction can improve urban waterlogging early warning capabilities, reduce disaster losses, and enhance emergency response capabilities.

[0003] Currently, for the prediction of urban waterlogging depth, there are mainly methods such as the critical rainfall method and the waterlogging model method. The critical rainfall method predicts the risk of waterlogging in a region by delineating sub-watersheds or sub-regions and setting precipitation disaster thresholds. This method has a simple algorithm, but waterlogging is affected not only by the amount of rainfall but also by factors such as the current drainage state of urban impervious surfaces. It is often difficult to accurately judge the waterlogging risk at different times using a fixed value; the waterlogging model collects data such as the elevation and underground pipe networks of the city and realizes waterlogging risk early warning by simulating urban surface runoff and pipe network runoff. There are currently many mainstream models. In recent years, with the continuous development of artificial intelligence and big data technologies, urban waterlogging prediction technologies based on artificial intelligence algorithms have also begun to receive more and more research attention. Some studies use deep learning algorithms such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs) to predict urban waterlogging. Due to its clear physical meaning and the ability to conduct heterogeneous risk assessment of regions, the waterlogging model is currently adopted by most cities. However, this method has high requirements for pipe network data. It not only requires accurate basic pipe network data but also real-time pipe network state data. Many cities currently have difficulty providing accurate data, thus reducing the model accuracy. In summary, the accuracy of predicting the water depth at waterlogging points during urban waterlogging in the prior art still needs to be improved. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, device and storage medium for predicting urban waterlogging depth, so as to solve the problem of low accuracy in predicting the water depth at waterlogging points during urban waterlogging in the prior art.

[0005] To achieve the above objectives, the present invention is implemented by adopting the following technical solutions:

[0006] In the first aspect, the present invention provides a method for predicting urban waterlogging depth, including:

[0007] Obtain the local terrain parameters of the waterlogging point, the drainage outlet flow rate in the set area around the waterlogging point, and the predicted precipitation of the waterlogging point in the set future time period;

[0008] Input the predicted precipitation into the constructed urban runoff simulation model to obtain the no-drainage inundation parameters, where the no-drainage inundation parameters include the number of inundated pixel cells, the average water depth, and the number of drainage outlets in the no-drainage state;

[0009] Input the local terrain parameters, the no-drainage inundation parameters, and the drainage outlet flow rate into the trained water depth prediction model to obtain the maximum water depth of the waterlogging point, where the water depth prediction model is a random forest model;

[0010] The urban runoff simulation model is obtained by improving the pre-constructed fully distributed runoff model. The improvements include: spatially dividing the city into multiple grids of the same size, temporally dividing the urban waterlogging period to be simulated into equal time slices, and performing pixel-level water volume migration calculation and pixel-level hydrodynamic calculation in the fully distributed runoff model based on grids on each time slice.

[0011] Furthermore, the local terrain parameter is the proportion of the area higher than the average elevation.

[0012] Furthermore, the pixel-level water volume migration calculation is carried out through the following formula:

[0013] ;

[0014] Among them, represents the proportion of the water volume flowing from grid b to the central grid c in the city. Grid b and the central grid c are adjacent. max represents taking the maximum value. Δx represents the offset of the horizontal coordinate of grid b relative to the horizontal coordinate of the central grid c, with the unit of pixel. Δy represents the offset of the vertical coordinate of grid b relative to the vertical coordinate of the central grid c. represents the horizontal water flow velocity in grid b. represents the vertical water flow velocity in grid b. represents the maximum allowable water flow velocity on the time slice;

[0015] After the pixel-level water volume migration, the expression for the remaining water volume proportion of the central grid c is:

[0016] ;

[0017] Among them, represents the remaining water volume proportion of the central grid c.

[0018] Furthermore, the expression for the interval time between two adjacent time slices after division is:

[0019] ;

[0020] Among them, Δt represents the interval time between two adjacent time slices, and C represents the grid cell resolution.

[0021] Furthermore, the pixel-level hydrodynamic calculation is performed through the following formula:

[0022] ;

[0023] Among them, represents the increment of the horizontal water flow velocity generated by grid b on the central grid in the time slice, represents the increment of the vertical water flow velocity generated by grid b on the central grid in the time slice, α is a preset positive constant, represents the water level height difference between grid b and the central grid.

[0024] Furthermore, in the constructed urban runoff simulation model, the water depth at the waterlogging point is calculated through the following formula:

[0025] ;

[0026] Among them, represents the water depth at the waterlogging point, represents the maximum water depth reduction due to drainage in the time slice, W c represents the water depth of the central grid c;

[0027] ;

[0028] Among them, represents the initial water depth of the central grid c at the th time slice, represents the initial water depth of the central grid c at the th time slice, represents the initial water depth of grid b at the th time slice, represents the th time slice, and the proportion of the water volume flowing from grid b to the central grid c in the city, represents the proportion of the remaining water volume of the central grid c after passing through the th time slice;

[0029] ;

[0030] Among them, q represents the drainage outlet flow rate.

[0031] Furthermore, the training of the water depth prediction model includes:

[0032] Obtain the precipitation within the set time before and after the historical maximum precipitation occurrence time, and then input it into the constructed urban runoff simulation model. In the urban runoff simulation model, increase the drainage outlet flow from 0 to the maximum drainage outlet flow at preset intervals, so as to obtain the maximum water depth of the waterlogging points corresponding to different drainage outlet flows within the set time before and after the historical maximum precipitation occurrence time. Put the obtained maximum water depth, drainage outlet flow, the area ratio of the area higher than the average elevation in the set area around the waterlogging point, and the no-drainage inundation parameter into a set to obtain the training data set, and train the water depth prediction model through the training data set to obtain the trained water depth prediction model.

[0033] Further, the maximum drainage outlet flow is obtained by the following method:

[0034] Gradually increase the drainage outlet flow from 0, and obtain the maximum water depth corresponding to different drainage outlet flows. If the drainage outlet flow reaches When, the maximum water depths of the first preset proportion of non-waterlogging drainage outlets are all less than the first preset value, then the At this time is the maximum drainage outlet flow.

[0035] In a second aspect, the present invention provides an urban waterlogging depth prediction device, including:

[0036] A data acquisition module, configured to: acquire the local terrain parameters of the waterlogging point, the drainage outlet flow in the set area around the waterlogging point, and the predicted precipitation of the waterlogging point in the future set time period;

[0037] A no-drainage inundation parameter solving module, configured to: input the predicted precipitation into the constructed urban runoff simulation model to obtain the no-drainage inundation parameter, and the no-drainage inundation parameter includes the number of flooded pixels, the average water depth, and the number of drainage outlets in the no-drainage state;

[0038] A waterlogging point maximum water depth prediction module, configured to: input the local terrain parameters, the no-drainage inundation parameter, and the drainage outlet flow into the trained water depth prediction model to obtain the maximum water depth of the waterlogging point, and the water depth prediction model is a random forest model;

[0039] Among them, the urban runoff simulation model is obtained by improving the pre-constructed fully distributed runoff model. The improvement includes: dividing the city into multiple grids with the same size in space, dividing the urban waterlogging time period to be simulated into equal time slices in time, and performing pixel-level water quantity migration calculation and pixel-level hydrodynamic calculation in the fully distributed runoff model based on grids on each time slice.

[0040] In a third aspect, the present invention provides a computer-readable storage medium, on which computer programs / instructions are stored. When the computer programs / instructions are executed by a processor, the steps of the urban waterlogging depth prediction method described in any one of the first aspects are implemented.

[0041] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0042] The urban waterlogging depth prediction method, device and storage medium provided by the present invention improve the fully distributed runoff model to obtain an urban runoff simulation model, and obtain the undrained flood parameters by processing the predicted precipitation with the urban runoff simulation model. Then, the maximum water depth of the waterlogging point is predicted by using the undrained flood parameters, local terrain parameters and drainage outlet flow rate. It is possible to predict the maximum water depth of the waterlogging point during the urban waterlogging process in the absence of drainage pipe network data, and experiments have proved that the prediction accuracy can reach 86.89%. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flowchart of an urban waterlogging depth prediction method provided by an embodiment of the present invention;

[0044] Figure 2 is a technical roadmap of urban waterlogging simulation provided by an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of the generation of a simulated drainage outlet provided by an embodiment of the present invention;

[0046] Figure 4 is a schematic diagram of the road where the sample is located and the undrained flood area provided by an embodiment of the present invention;

[0047] Figure 5 is a schematic diagram of the drainage outlet inundation ratio of an extreme rainfall case provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The present invention will be further described below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be used to limit the protection scope of the present invention.

[0049] Embodiment 1

[0050] The present invention provides an urban waterlogging depth prediction method, including:

[0051] Obtaining local terrain parameters of the waterlogging point, the drainage outlet flow rate in a set area around the waterlogging point, and the predicted precipitation of the waterlogging point in a set future time period;

[0052] Input the predicted precipitation into the constructed urban runoff simulation model to obtain the no-drainage inundation parameters, where the no-drainage inundation parameters include the number of inundated pixels, the average water depth, and the number of drainage outlets in the no-drainage state;

[0053] Input the local terrain parameters, the no-drainage inundation parameters, and the drainage outlet flow rate into the trained water depth prediction model to obtain the maximum water depth of the waterlogging point. The water depth prediction model is a random forest model;

[0054] The urban runoff simulation model is obtained by improving the pre-constructed fully distributed runoff model. The improvements include: spatially dividing the city into multiple grids of the same size, temporally dividing the urban waterlogging period to be simulated into equal time slices, and performing pixel-level water volume migration calculations and pixel-level hydrodynamic calculations in the fully distributed runoff model based on grids for each time slice.

[0055] In the present invention, an urban runoff simulation model is obtained by improving the fully distributed runoff model. By processing the predicted precipitation through the urban runoff simulation model, the no-drainage inundation parameters are obtained. Thus, the maximum water depth of the waterlogging point can be predicted using the no-drainage inundation parameters, local terrain parameters, and drainage outlet flow rate. It is possible to predict the maximum water depth of the waterlogging point during urban waterlogging in the absence of drainage pipe network data, and through experiments (specific experimental contents are recorded in Example 2), it is proved that the prediction accuracy can reach 86.89%.

[0056] Example 2

[0057] The present invention provides a method for predicting the water depth of urban waterlogging, and its specific implementation is as Figure 2 shown.

[0058] First, construct an urban runoff simulation model. To improve the accuracy of urban waterlogging prediction, an urban runoff simulation model is constructed based on the fully distributed runoff model. First, spatially divide the city into a large number of grids of the same size, and temporally divide the urban waterlogging period to be simulated into equal time slices; on this basis, complete pixel-level water volume migration calculations and pixel-level hydrodynamic calculations for each time slice.

[0059] The water balance equation of the divided grid is:

[0060] (1);

[0061] Among them, represents the initial water depth of the central grid c at the th time slice, represents the initial water depth of the central grid c at the th time slice, represents the initial water depth of the central grid c at the Precipitation on a time slice, represents the inflow of the central grid c at the time slice, represents the infiltration of the central grid c at the time slice, represents the evaporation of the central grid c at the time slice, represents the outflow of the central grid c at the time slice, represents the drainage of the central grid c at the time slice. Among them, precipitation, infiltration, evaporation, and drainage are the water budget in the vertical direction of the grid. Precipitation is obtained based on meteorological observations, and infiltration and evaporation rates are set respectively according to the underlying surface properties, and then multiplied by respectively.

[0062] The pixel-level water migration calculation is as follows:

[0063] Inflow and outflow are the water budget in the horizontal direction of the grid, which is achieved through water migration calculation. Let the water flow velocity vector of the grid b adjacent to the central grid c be ( , ), then the proportion of water volume flowing into the central grid c is:

[0064] (2);

[0065] Among them, represents the proportion of water volume flowing from grid b into the central grid c in the city. Grid b is adjacent to the central grid c. max represents taking the maximum value. Δx represents the offset of the horizontal coordinate of grid b relative to the horizontal coordinate of the central grid c, with the unit of pixel. Δy represents the offset of the vertical coordinate of grid b relative to the vertical coordinate of the central grid c, represents the horizontal water flow velocity in grid b, represents the vertical water flow velocity in grid b, represents the maximum allowable water flow velocity on the time slice. ( , ) represents the coordinate offset of grid b relative to the coordinate of the central grid c.

[0066] If the central grid c flows outwards due to its own water velocity, then the proportion of the remaining water volume of the central grid c is:

[0067] (3);

[0068] Among them, Indicates the proportion of the remaining water volume in the central grid c.

[0069] According to formulas (2) and (3), the water depth of the central grid after water volume migration calculation is:

[0070] (4);

[0071] Where, Indicates the initial water depth of the central grid c at the th time slice, Indicates the initial water depth of the central grid c at the th time slice, Indicates the initial water depth of grid b at the th time slice, Indicates the th time slice, the proportion of the water volume flowing from grid b to the central grid c in the city, Indicates the proportion of the remaining water volume in the central grid c after the th time slice.

[0072] It can be seen from formulas (2) and (3) that when the water velocity in the grid is equal to , the runoff flows out completely in the horizontal direction. Therefore, also determines the size of the time slice, that is:

[0073] (5);

[0074] Where, Δt represents the interval time between two adjacent time slices, in seconds, and C represents the grid cell resolution.

[0075] The pixel-level hydrodynamic calculation is as follows:

[0076] In the runoff confluence simulation, the hydrodynamic simulation is used to set the acceleration of the water flow in the grid. The velocity vector increment ( , ) generated by the grid b adjacent to the central grid c in the time slice is expressed as:

[0077] (6);

[0078] Where, Indicates the horizontal water flow velocity increment generated by grid b on the central grid in the time slice, Indicates the vertical water flow velocity increment generated by grid b on the central grid in the time slice, α is a preset positive constant, Indicates the water level height difference between grid b and the central grid.

[0079] Due to the lack of underground pipeline network data in the selected research area in the target city (Guiyang is used as the target city in this embodiment), the simulated drainage outlets of all roads in the research area are generated by the equal-spacing method. Suppose sampling is carried out every d (unit: meter) along the road, then the generated simulated drainage outlets are as Figure 3 shown. Obviously, the smaller the spacing of the drainage outlets generated in the research area, the larger the total number of drainage outlets. According to the interval of 30 meters, the total number of simulated drainage outlets in the research area is 113,181.

[0080] Suppose the drainage flow rate of the drainage outlet is , then the expression for the maximum water depth reduced due to drainage in the time slice is:

[0081] (7);

[0082] Among them, represents the maximum water depth reduced due to drainage in the time slice.

[0083] Considering the water balance, if the water depth on the grid where the drainage outlet is located is less than the theoretical drainage volume in the time slice, the actual drainage volume is the water volume of the current grid, and the expression of the current water depth is:

[0084] (8);

[0085] Among them, represents the water depth of the waterlogging point, represents the maximum water depth reduced due to drainage in the time slice, and W c represents the water depth of the central grid c.

[0086] Based on the urban runoff simulation model, an urban waterlogging model is constructed. First, analyze the drainage capacity of drainage outlets under different drainage intervals through extreme waterlogging cases, and determine the simulated drainage outlet interval and the maximum drainage capacity of the corresponding model of the drainage outlet in combination with the actual situation of Guiyang; on this basis, increase the drainage flow rate of the drainage outlet at a certain interval to the maximum drainage flow rate , and obtain the water depth of each waterlogging point under different drainage flow rates of the drainage outlet through the urban runoff simulation model. Combine the non-drainage inundation parameters, local terrain parameters, drainage flow rate and water depth of the waterlogging point to construct a historical waterlogging point database; then, use the historical waterlogging point database as training data and use the random forest algorithm to construct a two-way prediction model of drainage capacity and water depth; based on this, use the historical rainfall process with similar time to conduct non-drainage urban runoff simulation, obtain the non-drainage inundation parameters, substitute them into the drainage flow prediction model, and estimate the drainage capacity in combination with the waterlogging disaster situation; finally, for the future rainfall process, substitute the non-drainage inundation parameters of the non-drainage urban runoff simulation into the water depth prediction model to predict the urban waterlogging water depth of the next precipitation process; the technical route of the urban waterlogging simulation corresponding to the above process is asFigure 2 as shown

[0087] Excessive precipitation and insufficient drainage capacity are the main causes of urban waterlogging. When all drainage outlets drain water at the maximum drainage flow rate of the drainage outlet, the overall drainage capacity of the city reaches the maximum. If the precipitation does not exceed the maximum drainage capacity of the city, there will be no waterlogging in the city. The following method is used to determine the maximum drainage capacity of the city:

[0088] Substitute historical extreme waterlogging cases into the urban runoff simulation model, and assume that the drainage capacity of the drainage outlets in the non-waterlogging area is unobstructed, and there are a certain number of drainage outlets whose drainage capacity reaches or is close to the maximum drainage capacity. Gradually increase the flow rate of the simulated drainage outlet from 0 m 3 / s to obtain the road waterlogging depth under different drainage volumes. If the drainage outlet flow rate reaches and the maximum simulated water depth within the simulation period of 98% (the first preset ratio) of the non-flood drainage outlets is less than 5 mm (the first preset value), then the at this time is the maximum drainage capacity of the city.

[0089] Set different drainage outlet intervals, and obtain the under each drainage interval according to the above method. Compare the model simulation results with the actual situation in Guiyang, and set the interval of the simulated drainage outlets and the corresponding maximum drainage capacity.

[0090] In this embodiment, the drainage capacity is the drainage outlet flow rate; constructing a two-way prediction model of drainage capacity and water depth is the core of the urban waterlogging model of the present invention. By constructing a prediction model from water depth to drainage outlet flow rate (i.e., the drainage flow prediction model in Figure 2 ), the current actual drainage capacity of all drainage outlets in the city can be determined through disaster cases; by constructing a prediction model from drainage capacity to water depth (i.e., the simulated water depth prediction model in Figure 2 , which is also the water depth prediction model provided by the present invention), when the urban drainage capacity changes little, the waterlogging state of the city (i.e., predicting the maximum water depth of the flooded points) can be accurately predicted according to the predicted precipitation.

[0091] Use the random forest algorithm to construct a two-way prediction model of drainage capacity and water depth. The specific method is: substitute the precipitation in the 3 hours before and after the maximum precipitation time of historical disaster cases into the urban runoff simulation model, and increase the drainage flow rate from 0 m 3 / s to at a certain interval, simulate the maximum water depth of the flooded points within the simulation period under different drainage flow rates, and generate a random forest training data set.

[0092] For each sample j in the random forest training data set, define as follows:

[0093] (9);

[0094] where S j represents the set of all data of sample j, P j represents the drainage outlet flow of sample j, W j represents the maximum water depth of sample j, D j represents the area ratio of the area within 100 m along the road of sample j that is higher than the average elevation, A j represents the number of pixels in the flooded area where sample j is located in the non-drainage state, M j represents the average water depth of the flooded area where sample j is located in the non-drainage state, M j represents the number of drainage outlets in the flooded area where sample j is located in the non-drainage state.

[0095] Figure 4 is a schematic diagram of the road where the sample is located and the non-drainage flooded area. In the non-drainage state during the current precipitation process, the maximum water depth of the city is simulated. If the water depth is greater than 5 mm, it is defined as a flooded pixel. Starting from the waterlogging point, all spatially adjacent road areas form a flooded area. The number of pixels, the average flooded water depth, the number of drainage outlets, etc. in the flooded area are counted to form a training sample. The samples of each waterlogging point simulated for all disaster cases in the region under different drainage flows are combined into a training data set.

[0096] Taking the water depth as the prediction item and the rest as input items, and substituting them into a random forest model for training, a prediction model from the drainage outlet flow to the water depth (i.e., the water depth prediction model) can be obtained; similarly, taking the drainage outlet flow as the prediction item and the rest as input items, a prediction model from the water depth to the drainage outlet flow (i.e., the drainage flow prediction model) can be obtained.

[0097] After training the water depth prediction model, when performing actual water depth prediction, the following steps are executed:

[0098] Obtain the drainage outlet flow in the set area around the waterlogging point, the local terrain parameters in the set area around the waterlogging point in the future set time period, and the predicted precipitation;

[0099] Input the predicted precipitation into the constructed urban runoff simulation model to obtain non-drainage flooding parameters;

[0100] Input the local terrain parameters, non-drainage flooding parameters, and drainage outlet flow into the trained water depth prediction model to obtain the maximum water depth of the waterlogging point.

[0101] In this embodiment, the local terrain parameter is the area ratio of the area within 100 m along the road that is higher than the average elevation, and the non-drainage flooding parameters include the number of pixels in the flooded area, the average water depth, and the number of drainage outlets in the non-drainage state.

[0102] In this embodiment, the waterlogging on June 18, 2023 is selected as an extreme case. On the evening of June 18, 2023, Guiyang City encountered heavy rainfall. During this precipitation process, the cumulative daily rainfall at the national meteorological station in Guiyang on June 18 was 130.8 mm, reaching the heavy rainstorm level. According to the monitoring, the precipitation occurred from 14:00 on the 18th to 8:00 on the 19th, and the precipitation reached the maximum at 0:00 on June 19, 2023. The maximum hourly precipitation at the Guiyang Station was 99.8 mm. This rainfall caused relatively serious waterlogging on 30 roads in Yunyan District, Nanming District, Huaxi District, etc.

[0103] According to the "Code for Design of Outdoor Drainage (2016 Edition) of GB50014-2006", the spacing of drainage outlets should be preferably 25 m to 50 m, and the interval of drainage facilities summarized from the experience of Guiyang municipal management is about 20 m to 30 m. In the case where the accurate distribution of drainage settings cannot be obtained, the interval of drainage outlets in the study area is set to be distributed at an interval of 30 m along the road.

[0104] Substitute the precipitation from 21:00 on the 18th to 3:00 on the 19th into the urban runoff simulation model, and set the drainage outlet flow rate to be 0 to 0.6 m³ / s respectively. Simulate the water depth at each drainage outlet position during the simulation period, and count the proportion of drainage outlets with a water depth greater than 5 mm. The results are as Figure 5 shown. As Figure 5 can be seen, under the determined drainage outlet interval, as the drainage outlet flow rate increases, the proportion of drainage outlets with a water depth greater than 5 mm gradually decreases. According to the definition of the maximum drainage capacity mentioned above, the maximum drainage outlet flow rate is 0.17 m³ / s.

[0105] Collect 10 waterlogging cases from 2023 to 2024, input the precipitation 3 hours before and after the maximum precipitation into the urban runoff simulation model, and increase the drainage outlet flow rate from 0 m³ / s to 0.17 m³ / s at an interval of 0.003. Simulate the water depth of each waterlogging point in each disaster case, and combine them into a training database. The training database has a total of 919 records. Using this data as the driving force, construct a random forest model for waterlogging points.

[0106] Take the non-drainage inundation parameter, local terrain parameter, and water depth as independent variables, and the drainage outlet flow rate as the dependent variable, and substitute them into the random forest model for learning to construct a drainage flow prediction model; similarly, take the non-drainage inundation parameter, local terrain parameter, and drainage outlet flow rate as independent variables, and the water depth as the dependent variable, and substitute them into the random forest model for learning to construct a water depth prediction model.

[0107] To evaluate the model accuracy, 80% of the samples in the database were used as training samples, and the remaining 20% of the samples were used as test samples. For the drainage flow prediction, the absolute error evaluation method was used for accuracy evaluation. If the absolute error between the predicted flow and the sample flow was less than 0.01 m³ / s, the prediction was correct, and its prediction accuracy was statistically calculated; for the water depth prediction, the relative error evaluation method was used for accuracy evaluation. If the relative error between the predicted water depth and the sample water depth was less than 20%, the prediction was correct.

[0108] The evaluation formula for the relative error is:

[0109] (10);

[0110] where RE represents the relative error, P represents the predicted water depth, and O represents the sample water depth.

[0111] The average errors of training and testing and the prediction accuracy rates are statistically calculated as shown in Table 1:

[0112] Table 1 - Statistical table of average error and accuracy rate of prediction

[0113]

[0114] As can be seen from Table 1, for the test samples, the average error of the drainage flow prediction was 0.0027 m³ / s, and the prediction accuracy rate reached 93.44%; in terms of the water depth prediction, the average error of the test samples was 0.022 m, and the prediction accuracy rate was 86.89%. This indicates that the two-way prediction model has high accuracy and can be effectively applied to the prediction of urban waterlogging.

[0115] Assuming that the drainage capacity of the drainage outlet remains stable within a certain period of time, the drainage capacity of the waterlogging point was obtained through historical event analysis, and the water depth of the waterlogging point on June 29, 2024 was simulated with the same drainage flow and compared with the actual observed water depth. Since the waterlogging points actually collected were reported manually according to the disaster locations, it was difficult to ensure the same disaster situation points in different event cases. Therefore, in contrast to the waterlogging points on June 29, 2024, 16 waterlogging points on June 18, 2023 and April 18, 2024 were selected. The drainage flow of the drainage outlet near the waterlogging point was inversely calculated using the observed water depth, and the inversely calculated drainage flow of the drainage outlet was substituted into the water depth prediction model to predict the water depth situation at the corresponding positions on June 29, 2024. The results are shown in Table 2:

[0116] Table 2 - Table of waterlogging simulation results in the study area on June 29, 2024

[0117]

[0118] If the predicted water depth is less than 0.05m and no waterlogging is observed at the point on June 29, 2024, the prediction is correct; if the predicted water depth is greater than 0.05m and the relative error from the observed water level on June 29, 2024 is less than 20%, the prediction is also correct. According to such a standard, the waterlogging status of 10 points in Table 2 was correctly predicted (the rows marked with √ in the last column of the table), and the prediction accuracy was 62.5%. Among them, there were 13 historical waterlogging points where no waterlogging was observed on June 29, 2024. For 8 of them, the predicted water depth simulated by the model was less than 0.05m; for 3 identical or nearby waterlogging points (the last 3 rows in Table 2), the predicted water depth RE of Renmin Avenue Longquan Building and Fountain Roundabout was ≤20%, and the prediction was correct; the water depth error at the underpass of Beijing Road Intersection reached 0.145m, and RE was 68.40%. Referring to the water depth on June 18, 2023, which was 1.2m, this point was an extreme water accumulation area, so the error caused by its entry into the model was relatively large.

[0119] Example 3

[0120] The present invention provides a device for predicting urban waterlogging water depth, including:

[0121] A data acquisition module, configured to: acquire local terrain parameters of the waterlogging point, the drainage flow at the drainage outlets in a set area around the waterlogging point, and the predicted precipitation at the waterlogging point in a set future time period;

[0122] A non-drainage inundation parameter solving module, configured to: input the predicted precipitation into the pre-constructed urban runoff simulation model to obtain non-drainage inundation parameters, where the non-drainage inundation parameters include the number of inundated pixels, the average water depth, and the number of drainage outlets in the non-drainage state;

[0123] A maximum water depth prediction module for the waterlogging point, configured to: input the local terrain parameters, the non-drainage inundation parameters, and the drainage flow into the trained water depth prediction model to obtain the maximum water depth of the waterlogging point, where the water depth prediction model is a random forest model;

[0124] Among them, the urban runoff simulation model is obtained by improving a pre-constructed fully distributed runoff model. The improvement includes: spatially dividing the city into multiple grids of the same size, temporally dividing the urban waterlogging period to be simulated into equal time slices, and performing pixel-level water volume migration calculation and pixel-level hydrodynamic calculation in the fully distributed runoff model based on grids on each time slice.

[0125] Example 4

[0126] The present invention provides a computer-readable storage medium, on which computer programs / instructions are stored. When the computer programs / instructions are executed by a processor, the steps of the urban waterlogging water depth prediction method provided in Example 1 are implemented:

[0127] Obtain the local terrain parameters of the waterlogging point, the drainage outlet flow rate in a set area around the waterlogging point, and the predicted precipitation of the waterlogging point in a set future time period;

[0128] Input the predicted precipitation into the constructed urban runoff simulation model to obtain the no-drainage inundation parameters, which include the number of inundated pixels, the average water depth, and the number of drainage outlets in the no-drainage state;

[0129] Input the local terrain parameters, the no-drainage inundation parameters, and the drainage outlet flow rate into the trained water depth prediction model to obtain the maximum water depth of the waterlogging point. The water depth prediction model is a random forest model;

[0130] The urban runoff simulation model is obtained by improving a pre-constructed fully distributed runoff model. The improvements include: spatially dividing the city into multiple grids of the same size, temporally dividing the urban waterlogging period to be simulated into equal time slices, and performing pixel-level water volume migration calculation and pixel-level hydrodynamic calculation in the fully distributed runoff model based on grids on each time slice.

[0131] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0133] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one process or multiple processes and / or blocksFigure 1 The functions specified in one or more boxes.

[0134] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the steps of the functions specified in one or more processes and / or boxes Figure 1 One process or more processes and / or boxes Figure 1 The steps of the functions specified in one or more boxes.

[0135] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting urban waterlogging depth, characterized in that: include: Obtain local terrain parameters of the waterlogging point, the drainage outlet flow in the set area around the waterlogging point, and the predicted precipitation of the waterlogging point in the future set period; The predicted precipitation is input into the constructed urban runoff simulation model to obtain the no-drainage flooding parameters, which include the number of pixels in the flooded area, the average water depth and the number of drainage outlets under the no-drainage state; Inputting local terrain parameters, no-drainage inundation parameters and drainage outlet flow into a trained water depth prediction model to obtain the maximum water depth of the waterlogged point, wherein the water depth prediction model is a random forest model; The urban runoff simulation model is obtained by improving a pre-built fully distributed runoff model, and the improvement includes: dividing the city into multiple grids of the same size in space, dividing the urban waterlogging period to be simulated into equal time slices in time, and performing pixel-level water migration calculation and pixel-level hydrodynamic calculation in the grid-based fully distributed runoff model on each time slice.

2. The urban waterlogging depth prediction method according to claim 1, characterized in that: The local terrain parameter is the proportion of area above the average elevation.

3. The urban waterlogging depth prediction method according to claim 1, characterized in that: The pixel-level water migration calculation is performed using the following formula: ； in, It represents the proportion of water flowing from grid b to the central grid c in the city. Grid b and the central grid c are adjacent. max represents the maximum value. Δx represents the offset of the horizontal coordinate of grid b relative to the horizontal coordinate of the central grid c, in pixels. Δy represents the offset of the vertical coordinate of grid b relative to the vertical coordinate of the central grid c. represents the water velocity in the horizontal direction in grid b, represents the vertical water velocity in grid b, Indicates the maximum water flow velocity allowed on a time slice; After pixel-level water migration, the expression of the remaining water ratio of the central grid c is: ； in, Indicates the remaining water ratio of the central grid c.

4. The urban waterlogging depth prediction method according to claim 3, characterized in that: The expression of the interval time between two adjacent time slices after division is: ； Among them, Δt represents the duration of the time slice, and C represents the grid pixel resolution.

5. The urban waterlogging depth prediction method according to claim 3, characterized in that: The pixel-level hydrodynamic calculation is performed using the following formula: ； in, It represents the horizontal water velocity increment caused by grid b on the time slice to the central grid. It represents the vertical flow velocity increment of the grid b on the time slice to the central grid. α is a preset positive constant. Represents the water level difference between grid b and the central grid.

6. The urban waterlogging depth prediction method according to claim 4, characterized in that: In the constructed urban runoff simulation model, the water depth of the flood point is calculated by the following formula: ； in, Indicates the depth of water in the flood point. Indicates the maximum water depth reduced by drainage on a time slice, W c represents the water depth of the center grid c; ； in, Indicates that the center grid c is The initial water depth at a time slice is Indicates that the center grid c is The initial water depth at a time slice is Indicates that grid b is The initial water depth at a time slice is Indicates The proportion of water flowing from grid b to the central grid c in the city in the time slice is: Indicates that after The remaining water content ratio of the central grid c after the time slice; ； Where q represents the outlet flow rate.

7. The urban waterlogging depth prediction method according to claim 1, characterized in that: The training of the water depth prediction model includes: The precipitation within a set time before and after the time of the maximum historical precipitation is obtained, and then input into the constructed urban runoff simulation model, and the outlet flow is increased from 0 to the maximum outlet flow at a preset interval in the urban runoff simulation model, so as to obtain the maximum water depth of the waterlogged point corresponding to different outlet flows within a set time before and after the time of the maximum historical precipitation. The obtained maximum water depth, outlet flow, area ratio above the average elevation of the set area around the waterlogged point, and no drainage inundation parameters are put into a set to obtain a training data set, and the water depth prediction model is trained by the training data set to obtain a trained water depth prediction model.

8. The urban waterlogging depth prediction method according to claim 7, characterized in that: The maximum drain outlet flow rate is obtained by the following method: The outfall flow rate is gradually increased from 0, and the maximum water depth corresponding to different outfall flow rates is obtained. If the outfall flow rate reaches When the maximum water depth of the non-flood point drainage outlets with the first preset ratio is less than the first preset value, then is the maximum outfall flow rate.

9. A device for predicting urban waterlogging depth, characterized in that: include: The data acquisition module is configured to: acquire local terrain parameters of the waterlogging point, the flow rate of the drainage outlet in the set area around the waterlogging point, and the predicted precipitation of the waterlogging point in the future set period; The no-drainage flooding parameter solving module is configured to: input the forecast precipitation into the constructed urban runoff simulation model to obtain the no-drainage flooding parameters, wherein the no-drainage flooding parameters include the number of pixels in the flooded area under the no-drainage state, the average water depth and the number of drainage outlets; The maximum water depth prediction module of the waterlogged point is configured to: input the local terrain parameters, the non-drainage inundation parameters and the drainage outlet flow into the trained water depth prediction model to obtain the maximum water depth of the waterlogged point, wherein the water depth prediction model is a random forest model; Among them, the urban runoff simulation model is obtained by improving the pre-built fully distributed runoff model, and the improvement includes: dividing the city into multiple grids of the same size in space, dividing the urban flooding period to be simulated into equal time slices in time, and performing pixel-level water migration calculation and pixel-level hydrodynamic calculation in the grid-based fully distributed runoff model on each time slice.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the urban waterlogging depth prediction method according to any one of claims 1 to 8 are implemented.

Images