An urban waterlogging prevention method, device, equipment and storage medium

Abstract

Embodiments of the present application provide a city waterlogging prevention method, device, equipment and storage medium, which are applied to the technical field of rainwater system. The method comprises the following steps: inverting the liquid level and flow of the whole rainwater system pipe network, generating a ground submerged water depth map and a binary flood submerged map at a monitoring time, reasonably checking the liquid level of the whole rainwater system pipe network and the ground submerged water depth map and the binary flood submerged map, and performing fusion processing based on the hydraulic connection between one-dimensional (1D) nodes and two-dimensional (2D) ground to obtain a label, generating a new data set according to the label and forecast rainfall, mixing the historical data set and the new data set to obtain a mixed data set, updating the parameters of the city waterlogging prediction model according to the mixed data set, using the model with updated parameters to predict city waterlogging, determining the city waterlogging warning level according to the prediction result, generating a drainage scheme according to the prediction result, and adjusting and controlling the corresponding drainage facilities according to the scheme, thereby improving the city waterlogging prevention effect.

Description

Technical Field

[0001] This invention relates to the field of rainwater system technology, and in particular to a method, apparatus, equipment, and storage medium for preventing urban flooding. Background Technology

[0002] With the increase in extreme weather and rapid urban development, urban flooding caused by torrential rains has severely impacted the normal operation of cities and endangered the lives and property of residents. Therefore, urban flooding early warning technology has gradually gained attention in order to alleviate flooding disasters and reduce losses. The development of cutting-edge technologies such as big data analysis, artificial intelligence, and machine learning has brought data-driven models based on deep learning architectures into the research field of urban flooding early warning technology. Previous literature and patents have proposed simulation methods for rainfall generation and one-dimensional and two-dimensional coupled confluence processes based on large-scale datasets for urban flooding prediction, solving problems such as low computational efficiency and insufficient real-time performance of mechanistic models. However, the urban flooding prediction results of data-driven models still rely on the reliability of numerical simulation results, and their fit with observational data is poor in practical applications, resulting in unsatisfactory urban flooding prevention and control effects based on urban flooding prediction results. Summary of the Invention

[0003] Embodiments of the present invention provide a method, apparatus, equipment, and storage medium for urban flood control.

[0004] In a first aspect, embodiments of the present invention provide a method for preventing and controlling urban flooding, the method comprising:

[0005] A model for generating the liquid level and flow rate of a rainwater system is constructed. Based on this model, the global liquid level and flow rate of the rainwater system network are inverted according to the forecasted rainfall, the liquid level at the rainwater system level monitoring point, and the flow rate at the rainwater system flow monitoring point.

[0006] A flood depth compensation model is constructed, and based on the flood depth compensation model, combined with the flood depth of the ground flood depth monitoring points at the monitoring time and the output of the trained urban waterlogging prediction model, a ground flood depth map and a binarized flood inundation map at the monitoring time are generated.

[0007] A data fusion model is constructed, and based on the data fusion model, the rationality of the global liquid level and ground flooding depth map and the binary flooding map of the rainwater system network are checked. Based on the hydraulic connection between 1D nodes and 2D ground, the data is fused to obtain labels.

[0008] A new dataset is generated based on labels and forecasted rainfall. The historical dataset and the new dataset are then mixed to obtain a hybrid dataset. The parameters of the urban flooding prediction model are then updated on the hybrid dataset.

[0009] The urban flooding prediction model with updated parameters is used to predict urban flooding, and the urban flooding warning level is determined based on the prediction results.

[0010] Drainage plans are generated based on the urban flooding warning level, and corresponding drainage facilities are regulated in accordance with the drainage plans.

[0011] Among the possible implementations of the first aspect, a model for generating the liquid level and flow rate of a rainwater system is constructed, including:

[0012] A rainwater system level and flow rate generation model is constructed based on conditional variational autoencoders and similarity representations. The rainwater system level and flow rate generation model includes two conditional variational autoencoders, denoted as CVAE-1 and CVAE-2, with identical network structures. Similarity representations are defined as the encoding representations learned by the encoders in CVAE-1 and CVAE-2 being similar in the same rainfall event.

[0013] The update strategies in the rainwater system level-flow generation model include:

[0014] CVAE-1 is updated based on the forecasted rainfall, the liquid level at the rainwater system level monitoring point, and the flow rate at the rainwater system flow monitoring point to obtain the coded representation of CVAE-1;

[0015] Using the CVAE-1 encoding representation as the encoding constraint of CVAE-2, CVAE-2 is iteratively updated. Specifically, the CVAE-1 encoding representation is used as the encoding constraint of CVAE-2 and input into the decoder of CVAE-2 after loading and initializing weights to generate predicted values. These values ​​are used to complete the initial labels under the forecast rainfall and are then used as the input for the next iteration step.

[0016] Among the possible implementations of the first aspect, error compensation in the submerged water depth compensation model includes:

[0017] If the model prediction error of the inundation depth at each point in the connected inundation area is consistent, then the error term between the inundation depth of the ground inundation depth monitoring point and the output of the urban flooding prediction model will be compensated to the simulation results of each inundation point to obtain the preliminary compensated inundation depth map of the ground inundation depth monitoring point.

[0018] The inundation boundary of the preliminary compensation inundation depth map is re-determined, and secondary compensation is performed on the boundary points.

[0019] The compensation results of the error terms at different ground inundation depth monitoring points at any inundation point are weighted and summed to estimate the final compensated inundation depth map.

[0020] Among the possible implementations of the first aspect, the rationality checks in generating the data fusion model include:

[0021] The correspondence between the 1D node liquid level reconstructed from the stormwater system level-flow generation model and the 2D flood depth generated at the node by the flood depth compensation model was examined, specifically including:

[0022] If the liquid level at a node is higher than the node's maximum depth, the node is considered to be overflowing, and the 2D ground label at the node's corresponding coordinates should be submerged. If the submerged water depth at the node's coordinates is 0, the node should not overflow, which is indicated by the liquid level at the node being less than the node's maximum depth.

[0023] In some possible implementations of the first aspect, the historical dataset and the new dataset are mixed to obtain a hybrid dataset, including:

[0024] Systematic sampling of historical datasets and repeated sampling of new datasets;

[0025] The system sampling results are mixed with the repeated sampling results to obtain a mixed dataset.

[0026] Among the possible implementations of the first aspect, updating the parameters of the urban flooding prediction model on a hybrid dataset includes:

[0027] A continuous learning-based model update strategy is adopted to preserve the structure and initial weights of the urban flooding prediction model, and the parameters of the urban flooding prediction model are updated on a mixed dataset.

[0028] In a second aspect, embodiments of the present invention provide an urban flood control device, the device comprising:

[0029] The first construction module is used to build a rainwater system level and flow rate generation model, and based on the rainwater system level and flow rate generation model, to invert the global level and flow rate of the rainwater system network according to the forecast rainfall and the level and flow rate of the rainwater system level monitoring point and the flow rate of the rainwater system flow monitoring point.

[0030] The second construction module is used to construct an inundation depth compensation model, and based on the inundation depth compensation model, combined with the inundation depth of the ground inundation depth monitoring points at the monitoring time and the output of the trained urban waterlogging prediction model, to generate a ground inundation depth map and a binarized flood inundation map at the monitoring time.

[0031] The third construction module is used to build a data fusion model, and based on the data fusion model, to perform a rationality check on the global liquid level and ground flooding depth map and the binarized flooding map of the rainwater system network, and to perform fusion processing based on the hydraulic connection between 1D nodes and 2D ground to obtain labels;

[0032] The parameter update module is used to generate a new dataset based on labels and forecasted rainfall, mix the historical dataset and the new dataset to obtain a mixed dataset, and update the parameters of the urban flooding prediction model on the mixed dataset.

[0033] The urban flooding early warning module is used to predict urban flooding using an updated urban flooding prediction model and to determine the urban flooding early warning level based on the prediction results.

[0034] The facility control module is used to generate drainage plans based on the urban flooding warning level and to control the corresponding drainage facilities according to the drainage plans.

[0035] Thirdly, embodiments of the present invention provide an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0036] Fourthly, embodiments of the present invention provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described above.

[0037] In embodiments of the present invention, the parameters of the urban flooding prediction model can be updated by continuously supplementing new training data, and the updated urban flooding prediction model can be used to make real-time predictions of urban flooding, so that the urban flooding prediction results of the urban flooding prediction model are more adapted to the actual observation situation. Then, based on the urban flooding prediction results, the urban flooding warning level can be determined, thereby generating a drainage plan, and the corresponding drainage facilities can be regulated according to the plan to improve the effect of urban flooding prevention and control.

[0038] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0039] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0040] Figure 1 and Figure 2 This is a flowchart of an urban flood control method provided by an embodiment of the present invention;

[0041] Figure 3This is a schematic diagram of the rainwater system network topology in the JD area according to an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the monitoring point layout in the JD area according to an embodiment of the present invention;

[0043] Figure 5 This is a structural diagram of the urban flooding prediction model RP-SN in an embodiment of the present invention;

[0044] Figure 6 This is a schematic diagram of the CVAE-1 structure in the rainwater system level and flow rate generation model in an embodiment of the present invention;

[0045] Figure 7 This is an evaluation diagram of the training effect of the rainwater system liquid level and flow rate generation model in an embodiment of the present invention;

[0046] Figure 8 This is an evaluation diagram of the training effect of the rainwater system liquid level and flow rate generation model in an embodiment of the present invention;

[0047] Figure 9 This is a box plot evaluating the update effect of the rainwater system level-flow generation model in an embodiment of the present invention;

[0048] Figure 10 This is a box plot evaluating the update effect of the rainwater system level-flow generation model in an embodiment of the present invention;

[0049] Figure 11 This is a schematic diagram of the inundation boundary treatment in the inundation depth compensation model in an embodiment of the present invention;

[0050] Figure 12 This is a comparison diagram of the compensated flooding depth and the observed value at the ground flooding depth monitoring point f1 in rainfall examples 3 and 7 of this invention;

[0051] Figure 13 This is a comparison of the compensated inundation depth map and the simulation map at certain times in the rainfall instance 7 of the updated dataset in this embodiment of the invention;

[0052] Figure 14 This is a comparison chart of 1D liquid level reconstruction values ​​and 2D flooding depth compensation values ​​and fusion values ​​at the ground corresponding to some overflow points and non-overflow points in an embodiment of the present invention.

[0053] Figure 15 This is a box plot comparing the effects of the urban flooding prediction model RP-SN before and after parameter updates in this embodiment of the invention.

[0054] Figure 16 This is a structural diagram of an urban flood control device provided in an embodiment of the present invention;

[0055] Figure 17 This is a structural diagram of an exemplary electronic device that can implement an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0058] To address the problems in the background art, embodiments of the present invention provide a method, apparatus, equipment, and storage medium for urban flood control. The parameters of the urban flood prediction model can be continuously updated by supplementing it with new training data. The updated model is then used for real-time urban flood prediction, making the predictions more accurate and better adapted to actual observations. Based on these predictions, an urban flood warning level is determined, a drainage plan is generated, and corresponding drainage facilities are adjusted according to the plan, thereby improving the effectiveness of urban flood control.

[0059] The following detailed description, with reference to the accompanying drawings and specific embodiments, illustrates an urban flood control method, apparatus, equipment, and storage medium provided by the present invention.

[0060] Figures 1-2 A flowchart of an urban flood control method provided by an embodiment of the present invention is shown, as follows: Figures 1-2 As shown, urban flood control methods may include the following steps:

[0061] S110, construct a rainwater system level and flow rate generation model, and based on the rainwater system level and flow rate generation model, invert the global level and flow rate of the rainwater system network according to the forecast rainfall and the level and flow rate of the rainwater system level monitoring point and the flow rate of the rainwater system flow monitoring point.

[0062] S120. Construct an inundation depth compensation model, and based on the inundation depth compensation model, combine the inundation depth of the ground inundation depth monitoring points at the monitoring time with the output of the trained urban waterlogging prediction model to generate a ground inundation depth map and a binarized flood inundation map at the monitoring time.

[0063] S130. Construct a data fusion model and, based on the data fusion model, perform a rationality check on the global liquid level and ground flooding depth map and the binary flooding map of the rainwater system network. Then, perform fusion processing based on the hydraulic connection between 1D nodes and 2D ground to obtain labels.

[0064] S140: Generate a new dataset based on labels and forecasted rainfall, mix the historical dataset and the new dataset to obtain a mixed dataset, and update the parameters of the urban flooding prediction model on the mixed dataset.

[0065] S150 uses an updated urban flooding prediction model to predict urban flooding and determines the urban flooding warning level based on the prediction results.

[0066] S160 generates drainage plans based on urban flooding warning levels and regulates corresponding drainage facilities according to the drainage plans.

[0067] To facilitate further understanding, the above steps will be described in detail below with reference to specific embodiments:

[0068] Figure 3 This is a schematic diagram of the rainwater system network topology in the JD area according to an embodiment of the present invention, as shown below. Figure 3 As shown, the dots represent nodes of the rainwater system network, and the line segments between two nodes represent pipe segments. The rainwater system network in the JD area contains 340 nodes and 340 pipe segments, and there are 4 outlets among the 340 nodes.

[0069] Figure 4 This is a schematic diagram of the monitoring point layout in the JD area in an embodiment of the present invention, as shown below. Figure 4 As shown, there are three ground flooding depth monitoring points in the JD area, numbered f1, f2, and f3; two rainwater system level monitoring points, numbered n16 and n331; and one rainwater system flow monitoring point, numbered p42.

[0070] Figure 5 This is a structural diagram of the urban flooding prediction model RP-SN in an embodiment of the present invention, as shown below. Figure 5 As shown, the urban flooding prediction model RP-SN includes a rainfall-runoff simulation model RP-DL and a one-way coupled simulation model SN-DL for the runoff confluence process. The input to the urban flooding prediction model RP-SN is the rainfall time series R = [r 1·Δt ,r 2.Δt,…,r t·Δt ,…,r T ] T The topographic elevation data (DEM) includes two types of labels: one is a two-dimensional label, namely the time series inundation depth map (FD) and the time series binarized flood inundation map (FI). The label value in FI is either 0 or 1, corresponding one-to-one with FD. but otherwise, Another type is one-dimensional labels, representing the time-series DQ of liquid level and flow rate at each node of the stormwater system network. The label matrix format is shown below:

[0071]

[0072]

[0073]

[0074] Where Z represents the total number of grid cells (in meters); fd represents the flood depth (in meters); fi represents the binarized flood depth (in meters); d represents the node liquid level (in meters); and q represents the pipe flow rate (in meters). 3 / s; N represents the total number of nodes; M represents the total number of pipes.

[0075] In S110, a rainwater system level and flow rate generation model is constructed based on conditional variational autoencoders and similarity representations. This model includes two conditional variational autoencoders, denoted as CVAE-1 and CVAE-2, with identical network structures. The similarity representation is defined as the similarity of the encoded representations learned by the encoders in CVAE-1 and CVAE-2 during the same rainfall event.

[0076] The update strategies in the rainwater system level-flow generation model include:

[0077] (1) Update CVAE-1 based on the forecast rainfall and the liquid level of the rainwater system level monitoring point and the flow rate of the rainwater system flow monitoring point to obtain the coded representation of CVAE-1.

[0078] (2) Using the coded representation of CVAE-1 as the coded constraint of CVAE-2, iteratively update CVAE-2. Specifically, the coded representation of CVAE-1 is used as the coded constraint of CVAE-2 and input into the decoder of CVAE-2 after loading and initializing the weights to generate the predicted value, thereby completing the initial label under the forecast rainfall and using it as the input for the next iteration step.

[0079] In a specific example, CVAE-1 and CVAE-2 are first constructed and trained based on the designed rainfall time series. The network structures of CVAE-1 and CVAE-2 are identical.

[0080] Figure 6 This is a schematic diagram of the CVAE-1 structure in the rainwater system level-flow generation model in an embodiment of the present invention, as shown below. Figure 6 As shown, X1 and X1′ represent the input and reconstructed monitoring matrices, Condition(C) represents the rainfall, μ1 and σ1 are the mean and standard deviation of the distribution Z1 learned by encoder E-1, respectively, and z1 is the sampled value from distribution Z1 used for decoding input. The subscript of the corresponding symbols in CVAE-2 is 2.

[0081] In addition to the optimization objective of the general conditional variational autoencoder, during the training of CVAE-2, the information loss when approximating the distribution Z2 with the encoded representation Z1 was minimized, and the encoded representation Z1 learned in CVAE-1 was used as the constraint of the encoded representation in CVAE-2.

[0082] Figures 7-8 This is an evaluation diagram of the training effect of the rainwater system level and flow rate generation model in an embodiment of the present invention. It describes the comparison between the reconstructed level (flow rate) and the simulated value at some nodes (pipes) in test rainfall instances 17 and 70, reflecting the learning ability of the rainwater system level and flow rate generation model on changes in level and flow rate caused by different rainfall processes.

[0083] Table 1 shows the overall performance evaluation of the rainwater system level-flow generation model during the training phase:

[0084] Table 1: Mean NSE score of the consistency index between model reconstructed values ​​and simulated values ​​on the test set

[0085]

[0086] Table 1 shows that the rainwater system level and flow rate generation model has strong generalization ability on the test set. The model shows no obvious bias in learning level and flow rate features, and the NSE can reach above 0.9, indicating that the model can achieve high accuracy in predicting the features of each node (pipeline) in the rainwater system network.

[0087] Figures 9-10 This is a box plot evaluating the update effect of the rainwater system level-flow generation model in an embodiment of the present invention, such as... Figures 9-10 As shown, the updated model significantly improved the prediction accuracy for both liquid level and flow rate. The average MAPE for flow rate prediction was only 2.7%, while the average MAPE for liquid level prediction was slightly higher than that for flow rate prediction, at around 5%. The average deviation of the liquid level predicted by the model from the monitored value was 0.02m, and the NSE for both liquid level and flow rate prediction was higher than 0.99.

[0088] In S120, error compensation in the submerged water depth compensation model includes:

[0089] If the model prediction error of the inundation depth at each point in the connected inundation area is consistent, then the error term between the inundation depth at the ground inundation depth monitoring point and the output of the urban flooding prediction model RP-SN is compensated to the simulation results of each inundation point, thus obtaining a preliminary compensated inundation depth map of the ground inundation depth monitoring point.

[0090] The inundation boundary of the preliminary compensation inundation depth map was re-determined, and secondary compensation was performed on the boundary points.

[0091] The compensation results of the error terms at different ground inundation depth monitoring points at any inundation point are weighted and summed to estimate the final compensated inundation depth map.

[0092] In a specific example, let's first take a ground floodwater depth monitoring point p, with coordinates (a, b), and its measured floodwater depth at time step t as f. p t The model outputs the submerged water depth as Output t (a,b), the error term E between the measured value and the output value p t =f p t Output t (a,b). Assuming the model prediction error of the inundation depth at each point within the connected inundation area is consistent, the error term of the ground inundation depth monitoring point p is compensated for in the simulation results at each inundation point (x,y). t On (x,y), a preliminary compensated inundation depth map S is obtained for the ground inundation depth monitoring point p. p t (x,y)=Output t (x,y)+E p t When the error term E p t When <0, S p t Negative values ​​may appear in (x,y), and negative values ​​are set to zero.

[0093] Preliminary error compensation for ground flooding depth monitoring point p error term E p t The sign of the value determines whether the flood boundary will expand or contract, requiring a re-evaluation of the flood boundary and secondary compensation.

[0094] When E p t When the depth is less than 0, traverse the internal neighborhood (a,b) of each (x,y) at the boundary. If there exists an internal neighborhood, calculate the compensation flood depth S.p t If (a,b)>0, the flooded boundary remains unchanged; if the flooded water depth is 0 in all internal neighborhoods of the boundary point (x,y), the boundary shrinks. When E p t When the water depth at the submerged boundary is greater than 0, the water depth treatment method is divided into the following two cases based on the sign of the relative elevation Δh:

[0095] Figure 11 This is a schematic diagram of the inundation boundary treatment in the inundation depth compensation model of the present invention, as shown in the embodiment. Figure 11 As shown, the coordinates of the boundary point are (x,y), the coordinates of the internal territory point are (a,b), and the relative elevation Δh = DEM(x,y) - DEM(a,b).

[0096] (1) When Δh>0, if the flood depth S after compensation in any internal neighborhood of the boundary p t (a,b) is less than the relative elevation Δh, the boundary point is not submerged, S p t (x,y)=0; if there exists an internal neighborhood point S p t If (a,b)>Δh, then the boundary (x,y) is submerged. Assume the newly generated submerged point has the same water surface elevation as its internal points, and the submerged depth is S. p t (x,y)=S p t (a,b)―Δh, update the flooded boundary.

[0097] (2) When Δh < 0, if the basic assumption of uniform water height is followed (S p t (x,y)=S p t (a,b)―Δh), then the compensated submerged water depth S p t (x,y) is prone to maxima. Therefore, since the inundation depth near the boundary is generally very small, it can be approximated that the inundation depth at the boundary is the same as the inundation depth within its interior, i.e., S p t (x,y)≈S p t (a,b).

[0098] The flooded boundary is processed according to the above rules for secondary compensation until the compensation result tends to stabilize and no new boundary points are generated.

[0099] It should be noted that due to the limited number of ground flooding depth monitoring points and the highly susceptible nature of connectivity within the flooded area to changes caused by rainfall, it is impractical to deploy ground flooding depth monitoring points in every connected region. Therefore, this study uses the error terms of several deployed ground flooding depth monitoring points to compensate for any flooded point, and then performs a weighted sum to approximate the final compensated flooding depth map.

[0100] The calculation process for the weighted sum is as follows:

[0101] Connect any inundation point (x, y) to the set of ground inundation depth monitoring points M. F ={1,2,…N f} Various ground inundation depth monitoring points p(x p ,y p The Manhattan distance of ) is denoted as {L1(x,y),L2(x,y),…,L P (x,y)}:

[0102] L p (x,y)=|x―x p |+|y―y p | (4)

[0103] Because the closer the coordinate point (x, y) is to the ground inundation depth monitoring point p, the greater the influence of the error term of that monitoring point on the compensated inundation depth value. Therefore, for Manhattan distance... Perform reciprocal consistency processing, where L p ―1 (x,y) represents the error compensation result S of the ground floodwater depth monitoring point p relative to the coordinate point (x,y). p t The final compensation result S at the inundation point (x,y) t The weight ω of (x,y) p (x, y). The error compensation results for each surface inundation depth monitoring point are weighted according to the calculated weights, and the weighted sum S of the information from each surface inundation depth monitoring point is calculated. t (x, y) represents the submerged water depth at coordinate (x, y) at time t.

[0104]

[0105] For example, here we construct an inundation depth compensation model using the error terms of ground inundation depth monitoring points f2 and f3, and verify the reliability of the error compensation method by comparing the compensation value of the model at ground inundation depth monitoring point f1 with the observed value.

[0106] Figure 12This is a comparison chart of the compensated inundation depth at monitoring point f1 and the observed value in rainfall examples 3 and 7 of this invention. Figure 13 This is a comparison between the compensated inundation depth map and the simulation map at certain times in the rainfall instance 7 of the updated dataset in this embodiment of the invention.

[0107] In S130, the rationality of the generated 1D and 2D labels is first checked and compared with the label check results in the historical dataset, i.e., the simulated dataset. The correspondence between the 1D node liquid level reconstructed by the stormwater system level-flow generation model and the 2D flood depth generated by the flood depth compensation model at that node is checked as follows: If the node liquid level is higher than the difference between the ground elevation and the inner bottom elevation of the node (i.e., the maximum depth of the node), the node is considered to be overflowing, and the ground 2D label at the corresponding coordinate of the node should be flooded; similarly, if the ground flood depth at the node coordinate is 0, then the node should not overflow, which is manifested as the node liquid level being less than the maximum depth of the node. Here, the 1D To 2D and 2D To 1D indices are selected to describe the proportion of correctly classified overflow points and non-overflow points. The calculation formulas for each index are as follows:

[0108] (1) 1D to 2D

[0109]

[0110]

[0111] (2) 2D to 1D

[0112]

[0113]

[0114] The results of the label rationality check are shown in Table 2:

[0115] Table 2: Accuracy of 1D and 2D Label Matching

[0116]

[0117] In Table 2, the label pairs represent the 1D and 2D data sources for calculating the matching degree; Sim1D and Sim2D indicate that the node liquid level and surface flooding depth data come from the simulation results of the mechanistic model; Rec1D refers to the reconstructed liquid level data after the update of the stormwater system liquid level and flow generation model, and Err2D is the flooding depth data obtained by error compensation of the surface flooding depth monitoring points. The matching accuracy of the simulation data label pairs is considered to be the allowable error in the hydraulic simulation process and is used to compare the rationality of the generated labels. The third row and third column of the table show that the average proportion of overflow nodes correctly classified in the simulated flooding depth map is 0.8465; the fifth row and fourth column show that, on both temporal and spatial scales, the average proportion of nodes classified as non-flooded in the 2D compensation map as having no overflow in the 1D reconstruction is 0.8306. As can be seen from the accuracy scores in the third and fourth columns of the table above, when judging the non-flooded state, the spatiotemporal correspondence between the corrected 1D and 2D labels is close to the state during simulation, with an accuracy of over 80%; however, when judging the overflow or flooded state, the matching accuracy between the corrected labels is lower.

[0118] It should be noted that if the reconstruction result of the rainwater system level-flow generation model has higher reliability than the ground floodwater depth monitoring point error-compensated floodwater depth map, then the unreasonable labels in the above inspection process will be reprocessed. Specifically, as shown below:

[0119] (1) Select liquid level point: Traverse the reconstructed 1D node liquid level in the rainwater system liquid level and flow generation model. If the node overflows and there is no water accumulation at the point in the compensation diagram, add the node to the node sampling space of the flood depth compensation process.

[0120] (2) Calculate the correction coefficient of the node from the simulation data, correct the overflow part in the node reconstruction result, and obtain the flooding depth value at the ground coordinates of the node.

[0121] Taking node k as an example, the average ratio of the simulated liquid level value when the stormwater system node overflows (after deducting the maximum depth of the node) to the simulated inundation depth value at the corresponding ground coordinates at that moment is used as the correction coefficient α for the node. (k) :

[0122]

[0123] in, The maximum value represents the simulated liquid level at node k at time step t in the training sample e, in meters (m). k The maximum depth of node k (ground elevation - inner bottom elevation) is represented in meters. The simulated flooding depth at point (i,j) on the ground corresponding to node k is represented in meters.

[0124] Then, the liquid level at time step k of the reconstructed time step t is denoted as like And the submerged water depth compensated at the corresponding coordinates at that moment. Then the corrected submersion depth at liquid level k is...

[0125] (3) Repeat the 2D submerged water depth map error compensation process and correct the compensation map.

[0126] Figure 14 This is a comparison chart of 1D liquid level reconstruction values ​​and 2D flooding depth compensation values ​​and fusion values ​​at the ground corresponding to some overflow points and non-overflow points in an embodiment of the present invention, as shown in the figure. Figure 14 As shown, during the fusion of 1D and 2D labels, the reprocessing of the compensated flooding depth map makes the change in flooding depth at the overflow point match the time when the liquid level at that point reaches the maximum depth, ultimately achieving the goal of label matching. Non-overflow points may also be flooded due to the overflow from upstream nodes. In this case, the trend of the fused value and the compensation value of the flooding depth at the node are basically the same, indicating that label fusion does not significantly affect the error compensation result.

[0127] In step S140, a new dataset is generated based on labels and forecast rainfall. Systematic sampling is performed on the historical dataset (i.e., the simulated dataset), and repeated sampling is performed on the new dataset. The systematic sampling results and the repeated sampling results are then combined to obtain a mixed dataset. The ratio of new to old data samples in the mixed dataset is approximately 1:1, which avoids catastrophic forgetting of knowledge learned by the old model while considering the model's performance on the old dataset. Then, a continuous learning-based dynamic model update strategy is employed, preserving the structure and initial weights of the urban flooding prediction model RP-SN, and updating the parameters of RP-SN on the mixed dataset.

[0128] Figure 15 This is a box plot comparing the performance of the urban flooding prediction model RP-SN before and after parameter updates in this embodiment of the invention, such as... Figure 15 As shown, Y represents the corrected state, and N represents the uncorrected state. The boxes from left to right represent the scores of the model before and after correction on the consistency between the predicted and monitored inundation depths at the ground inundation depth monitoring points, under the indices ACC, FNR, RMSE, 2D-CC, and NSE. Here, ACC represents accuracy, FNR represents false negative rate, RMSE represents root mean square error, 2D-CC represents spatial Pearson correlation coefficient, and NSE represents Nash efficiency coefficient.

[0129] The accuracy of the revised model in flood inundation assessment has been significantly improved, while the false negative rate has been reduced to below 1%. In the inundation depth prediction task, the model's prediction accuracy for spatial inundation depth maps (2D-CC) reaches 0.9, and its prediction accuracy for temporal trend changes at each inundation point (NSE) is also close to 0.9. Overall, compared with the previous model, the average deviation (RMSE) of the revised model's prediction results at both temporal and spatial scales has decreased from 0.08m to 0.03m.

[0130] In S150, the updated urban flooding prediction model is used to predict the inundation depth map to complete the urban flooding prediction, and the urban flooding warning level is determined based on the urban flooding prediction results.

[0131] In S160, a drainage plan is generated based on the urban flooding warning level, and the corresponding drainage facilities are regulated according to the plan. For example, the opening and closing of the gates at the outlets and their degree of opening are adjusted.

[0132] According to embodiments of the present invention, at least the following technical effects are achieved:

[0133] This invention continuously updates the parameters of an urban flooding prediction model by supplementing it with new training data. The updated model is then used for real-time urban flooding prediction, making the predictions more adaptable to actual observations. Based on these predictions, an urban flooding warning level is determined, a drainage plan is generated, and corresponding drainage facilities are regulated to improve urban flooding prevention. Furthermore, this invention constructs a global inversion model based on stormwater system network observation data. The structure and update strategy of the stormwater system level and flow rate generation model offer a feasible approach to addressing the problem of missing inputs or labels in neural networks, providing data support for the applicability of subsequent data-driven models in handling label completion issues. Finally, this invention employs continuous training and mixed datasets as update strategies, ensuring update efficiency while avoiding catastrophic forgetting.

[0134] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0135] The above is an introduction to the method embodiments. The following describes the solution of the present invention further through device embodiments.

[0136] Figure 16This is a structural diagram of an urban flood control device provided in an embodiment of the present invention, as shown below. Figure 16 As shown, urban flood control devices may include:

[0137] The first construction module 1610 is used to construct a rainwater system level and flow rate generation model, and based on the rainwater system level and flow rate generation model, according to the forecast rainfall and the level and flow rate of the rainwater system level monitoring point and the flow rate of the rainwater system flow monitoring point, the global level and flow rate of the rainwater system pipe network are inverted.

[0138] The second construction module 1620 is used to construct an inundation depth compensation model, and based on the inundation depth compensation model, combined with the inundation depth of the ground inundation depth monitoring points at the monitoring time and the output of the trained urban waterlogging prediction model, to generate a ground inundation depth map and a binarized flood inundation map at the monitoring time.

[0139] The third construction module 1630 is used to construct a generated data fusion model. Based on the generated data fusion model, it performs a rationality check on the global liquid level and ground flooding depth map and the binary flooding map of the rainwater system network. It also performs fusion processing based on the hydraulic connection between 1D nodes and 2D ground to obtain labels.

[0140] The parameter update module 1640 is used to generate a new dataset based on labels and forecast rainfall, mix the historical dataset and the new dataset to obtain a mixed dataset, and update the parameters of the urban flooding prediction model on the mixed dataset.

[0141] The urban flooding early warning module 1650 is used to predict urban flooding using an updated urban flooding prediction model and to determine the urban flooding early warning level based on the prediction results.

[0142] The facility control module 1560 is used to generate drainage plans based on the urban flooding warning level and to control the corresponding drainage facilities according to the drainage plans.

[0143] Understandable, Figure 16 Each module / unit in the urban flood control device shown has the ability to achieve Figure 1-2 The functions of each step in the urban flood control method shown are explained, and their corresponding technical effects are achieved. For the sake of brevity, these will not be elaborated here.

[0144] Figure 17 This is a structural diagram of an exemplary electronic device that can implement embodiments of the present invention, as provided in the embodiments of the present invention. Figure 17As shown, the electronic device may include a computing unit 1701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1702 or a computer program loaded from a storage unit 1708 into a random access memory (RAM) 1703. The RAM 1703 may also store various programs and data required for the operation of the electronic device. The computing unit 1701, ROM 1702, and RAM 1703 are interconnected via a bus 1704. An input / output (I / O) interface 1705 is also connected to the bus 1704.

[0145] Multiple components in the electronic device are connected to the I / O interface 1705, including: input units 1706, such as a keyboard, mouse, etc.; output units 1707, such as various types of displays, speakers, etc.; storage units 1708, such as disks, optical disks, etc.; and communication units 1709, such as network interface cards, modems, wireless transceivers, etc. The communication unit 1709 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0146] The computing unit 1701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1701 performs the various methods and processes described above. For example, in some embodiments, the methods may be implemented as a computer program product, including a computer program tangibly contained in a computer-readable medium, such as storage unit 1708. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM 1702 and / or communication unit 1709. When the computer program is loaded into RAM 1703 and executed by the computing unit 1701, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 1701 may be configured to perform methods by any other suitable means (e.g., by means of firmware).

[0147] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0148] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for preventing urban flooding, characterized in that, The method includes: A rainwater system level and flow rate generation model is constructed, and based on the rainwater system level and flow rate generation model, the global level and flow rate of the rainwater system network are inverted according to the forecast rainfall and the level and flow rate of the rainwater system level monitoring point and the flow rate of the rainwater system flow monitoring point. A flood depth compensation model is constructed, and based on the flood depth compensation model, combined with the flood depth of the ground flood depth monitoring points at the monitoring time and the output of the trained urban waterlogging prediction model, a ground flood depth map and a binarized flood inundation map at the monitoring time are generated. A data fusion model is constructed, and based on the data fusion model, the rationality of the global liquid level and ground flooding depth map and the binary flooding map of the rainwater system network are checked. The data is then fused based on the hydraulic connection between 1D nodes and 2D ground to obtain labels. A new dataset is generated based on labels and forecasted rainfall. The historical dataset and the new dataset are then mixed to obtain a hybrid dataset. The parameters of the urban flooding prediction model are then updated on the hybrid dataset. The urban flooding prediction model with updated parameters is used to predict urban flooding, and the urban flooding warning level is determined based on the prediction results. A drainage plan is generated based on the urban flooding warning level, and the corresponding drainage facilities are regulated according to the drainage plan. The construction of the rainwater system level-flow generation model includes: A rainwater system level and flow rate generation model is constructed based on conditional variational autoencoders and similarity representations. The rainwater system level and flow rate generation model includes two conditional variational autoencoders, denoted as CVAE-1 and CVAE-2, with identical network structures. The similarity representation is defined as the similarity of the encoded representations learned by the encoders in CVAE-1 and CVAE-2 in the same rainfall event. The update strategy in the rainwater system level-flow generation model includes: CVAE-1 is updated based on the forecasted rainfall, the liquid level at the rainwater system level monitoring point, and the flow rate at the rainwater system flow monitoring point to obtain the coded representation of CVAE-1; Using the CVAE-1 encoding representation as the encoding constraint of CVAE-2, CVAE-2 is iteratively updated. Specifically, this includes: using the CVAE-1 encoding representation as the encoding constraint of CVAE-2 as input to the decoder in CVAE-2 after loading the initialized weights, generating predicted values, thereby completing the initial labels under the forecast rainfall, and using them as input for the next iteration step. Error compensation in the inundation depth compensation model includes: If the model prediction error of the inundation depth at each point in the connected inundation area is consistent, then the error term between the inundation depth at the ground inundation depth monitoring point and the output of the urban flooding prediction model is compensated to the simulation results of each inundation point to obtain a preliminary compensated inundation depth map of the ground inundation depth monitoring point. The inundation boundary of the preliminary compensation inundation depth map is re-determined, and secondary compensation is performed on the boundary points. The compensation results of the error terms at different ground inundation depth monitoring points at any inundation point are weighted and summed to estimate the final compensated inundation depth map.

2. The method according to claim 1, characterized in that, The rationality check in the generated data fusion model includes: The correspondence between the 1D node liquid level reconstructed by the rainwater system liquid level and flow rate generation model and the 2D flood depth generated at the node by the flood depth compensation model is examined, specifically including: If the liquid level at a node is higher than the node's maximum depth, the node is considered to be overflowing, and the 2D ground label at the node's corresponding coordinates should be submerged. If the submerged water depth at the node's coordinates is 0, the node should not overflow, which is indicated by the liquid level at the node being less than the node's maximum depth.

3. The method according to claim 1, characterized in that, The process of mixing historical and new datasets to obtain a hybrid dataset includes: The historical dataset is systematically sampled, and the new dataset is repeatedly sampled; The system sampling results are mixed with the repeated sampling results to obtain the mixed dataset.

4. The method according to claim 1, characterized in that, The step of updating the parameters of the urban flooding prediction model on the hybrid dataset includes: A continuous learning-based model update strategy is adopted to preserve the structure and initial weights of the urban flooding prediction model, and the parameters of the urban flooding prediction model are updated on the hybrid dataset.

5. An urban flood control device, characterized in that, The device includes: The first construction module is used to construct a rainwater system level and flow rate generation model, and based on the rainwater system level and flow rate generation model, according to the forecast rainfall and the level and flow rate of the rainwater system level monitoring point and the flow rate of the rainwater system flow monitoring point, to invert the global level and flow rate of the rainwater system pipe network. The second construction module is used to construct an inundation depth compensation model, and based on the inundation depth compensation model, combined with the inundation depth of the ground inundation depth monitoring points at the monitoring time and the output of the trained urban waterlogging prediction model, to generate a ground inundation depth map and a binarized flood inundation map at the monitoring time. The third construction module is used to construct a data fusion model, and based on the data fusion model, to perform a rationality check on the global liquid level and ground flooding depth map and the binarized flooding map of the rainwater system network, and to perform fusion processing based on the hydraulic connection between 1D nodes and 2D ground to obtain labels; The parameter update module is used to generate a new dataset based on labels and forecasted rainfall, and to mix the historical dataset and the new dataset to obtain a mixed dataset, on which the parameters of the urban flooding prediction model are updated; The urban flooding early warning module is used to predict urban flooding using an updated urban flooding prediction model and to determine the urban flooding early warning level based on the prediction results. The facility control module is used to generate a drainage plan based on the urban flooding warning level and to control the corresponding drainage facilities according to the drainage plan. The construction of the rainwater system level-flow generation model includes: A rainwater system level and flow rate generation model is constructed based on conditional variational autoencoders and similarity representations. The rainwater system level and flow rate generation model includes two conditional variational autoencoders, denoted as CVAE-1 and CVAE-2, with identical network structures. The similarity representation is defined as the similarity of the encoded representations learned by the encoders in CVAE-1 and CVAE-2 in the same rainfall event. The update strategy in the rainwater system level-flow generation model includes: CVAE-1 is updated based on the forecasted rainfall, the liquid level at the rainwater system level monitoring point, and the flow rate at the rainwater system flow monitoring point to obtain the coded representation of CVAE-1; Using the CVAE-1 encoding representation as the encoding constraint of CVAE-2, CVAE-2 is iteratively updated. Specifically, this includes: using the CVAE-1 encoding representation as the encoding constraint of CVAE-2 as input to the decoder in CVAE-2 after loading the initialized weights, generating predicted values, thereby completing the initial labels under the forecast rainfall, and using them as input for the next iteration step. Error compensation in the inundation depth compensation model includes: If the model prediction error of the inundation depth at each point in the connected inundation area is consistent, then the error term between the inundation depth at the ground inundation depth monitoring point and the output of the urban flooding prediction model is compensated to the simulation results of each inundation point to obtain a preliminary compensated inundation depth map of the ground inundation depth monitoring point. The inundation boundary of the preliminary compensation inundation depth map is re-determined, and secondary compensation is performed on the boundary points. The compensation results of the error terms at different ground inundation depth monitoring points at any inundation point are weighted and summed to estimate the final compensated inundation depth map.

6. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-4.

7. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-4.

Images