An urban waterlogging accumulation prediction method based on an LSTM neural network

Abstract

The application provides a city waterlogging prediction method based on an LSTM neural network, comprising the following steps: collecting waterlogging depth of a plurality of waterlogging points in the city, height difference between a drainage outlet and a river or lake, rainfall of a waterlogging point area, and environmental data of the waterlogging points in each rainfall process; constructing a mathematical relationship between waterlogging amount and waterlogging depth of the waterlogging points, and calculating the confluence water amount of the waterlogging points in the i th sampling period; constructing a confluence water amount prediction model based on the LSTM neural network, initializing model parameters, and training the model; collecting real-time data and environmental data of a to-be-predicted point, inputting the data into the trained LSTM neural network model to predict the confluence water amount, and calculating the waterlogging depth of the to-be-predicted point through the confluence water amount and the environmental data. The application improves the prediction accuracy of the waterlogging depth by predicting the confluence water amount of the waterlogging points and calculating the drainage amount of the waterlogging points under different conditions, and provides support for the fields of city flood control and disaster reduction.

Description

Technical Field

[0001] This invention relates to the field of urban flooding prediction technology, and in particular to an urban flooding prediction method based on LSTM neural network. Background Technology

[0002] In recent years, with global warming and the impact of extreme weather events such as typhoons, many places have frequently experienced torrential rains that occur once every 50 or even 100 years. However, due to factors such as finance and construction cycles during urban development, most cities are designed with flood control capabilities based on a 20-year return period standard. In particular, some older urban areas, affected by the time of urban construction, often have flood control capabilities based on a 10-year return period standard. When extreme torrential rains occur, they can easily pose a great safety hazard to people's lives and property. Therefore, effectively predicting urban waterlogging points under extreme weather conditions and making corresponding decision-making plans has become the most important part of urban flood prevention and control.

[0003] Current urban flooding prediction methods rely on collecting data on water depth and rainfall at historical flooding points. These data are used to predict water depth for a future period. However, during extreme rainfall events, as the surface water depth changes, water that cannot drain will flow towards lower drainage outlets (such as underpasses and sunken plazas), causing significant changes in the inflow of water to these outlets. Current prediction methods are less accurate when the inflow is large. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting urban flooding based on LSTM neural networks.

[0005] The objective of this invention is achieved through the following technical solution, the specific steps of which are as follows:

[0006] 1) Data collection: At time intervals of △t, collect data on water depth at several water accumulation points in the city during each historical rainfall event, vertical height difference between the drainage outlets and drainage rivers or lakes corresponding to the water accumulation points, flow velocity data of the drainage pipes corresponding to the water accumulation points, rainfall data in the area of ​​the water accumulation points, and environmental data of the water accumulation points to obtain the collected data.

[0007] 2) Calculate the runoff volume: Construct the mathematical relationship between the water volume and water depth of the water accumulation point, and calculate the runoff volume of the water accumulation point in the i-th sampling period based on the drainage network data and rainfall data of the water accumulation point;

[0008] 3) Data splitting: Sort the collected data and calculated runoff volume of the water accumulation point during a single rainfall process according to time to obtain the original dataset. Divide the original dataset into a test set and a training set in an 8:2 ratio.

[0009] 4) Model building: Using rainfall as input and runoff volume as output, a runoff volume prediction model based on LSTM neural network is built, the model parameters are initialized, and the model is trained using training set data;

[0010] 5) Water depth prediction: Collect real-time data and environmental data of the point to be predicted, and input them into the LSTM neural network model trained in step 4) to predict the runoff volume, and calculate the water depth of the point to be predicted using the runoff volume and environmental data.

[0011] Furthermore, the water depth data mentioned in step 1) is: H = {h1, h2, ..., h j ,...,h m}, where: h j This represents the water depth data of a single water accumulation point sampled for the jth time during a single rainfall event;

[0012] The rainfall data for the waterlogged area is: P = {p1, p2, ..., p...} j ,...,p N}, where: p j This represents the rainfall data sampled for the j-th time in a single waterlogged area during a single rainfall event;

[0013] The vertical elevation difference between the drainage outlet and the drainage river or lake corresponding to the water accumulation point is: △S={△s1,△s2,...,△s j ,...,△s N}, where: △s j This refers to the vertical elevation difference between the drainage outlet and the drainage river or lake corresponding to the j-th sampling point of a single water accumulation point during a single rainfall event.

[0014] The flow velocity V = {v1, v2, ..., v} in the drain pipe corresponding to the water accumulation point j ,...,v N}, where: v j This represents the flow velocity data of the drainage pipe corresponding to the j-th sampling point of a single water accumulation point during a single rainfall event;

[0015] The environmental data of the water accumulation point includes the geographical type data of the water accumulation point and the drainage pipe network data of the water accumulation point. The geographical type data of the water accumulation point includes two types: urban overpass underpasses and urban low-lying areas. The urban low-lying areas include urban sunken plazas or flat roads.

[0016] The drainage network data includes the design flow velocity ν of the drainage pipe at the water accumulation point and the flow constant b of the drainage pipe.

[0017] Furthermore, the specific steps for calculating the runoff volume in step 2) are as follows:

[0018] 2-1) Construct the water accumulation volume Q at the water accumulation point z With water depth h z Mathematical relationship:

[0019] Q z =f(h) z )

[0020] In the formula, f(h) z Q represents the water volume at the water accumulation point. z With water depth h z Mathematical relational functions;

[0021] Constructing a mathematical function f1(h) for underpass tunnels at urban interchanges z A mathematical function f2(h) is constructed for low-lying urban areas. z ):

[0022]

[0023] f2(h z ) = h z .S j

[0024] In the formula, ax 2 Let S be the cross-sectional curve of an underpass tunnel at an urban interchange, w be the width of the underpass tunnel, and S be the width of the tunnel. j The area of ​​sunken plazas or flooded roads in the city;

[0025] 2-2) Calculate the amount of water flowing into the water accumulation point Q during the i-th sampling period. hi :

[0026]

[0027] In the formula, Q zi Let ν be the water volume at the i-th sampling point. i denoted as , where is the flow velocity of the drain pipe at the water accumulation point during the i-th sampling; b is the flow rate constant, which is related to the cross-sectional area of ​​the drain pipe at the water accumulation point; and n is the number of samplings.

[0028] Furthermore, the specific steps for constructing the model in step 4) are as follows:

[0029] 3-1) with p t-r△t ... p t-2△t p t-△t Rainfall p at time t-r△t ... p t-2△t p t-△t As input, the runoff volume Q at time t. hiFor the output, a runoff volume prediction model based on LSTM neural network is constructed, where: r is the data length of the moving sliding window of LSTM neural network;

[0030] 3-2) The LSTM neural network optimizer uses the Adam function, and the LSTM neural network uses the root mean square error (RMSE), mean absolute error (MAE), and absolute percentage error (MAPE) as loss functions. The parameters of the LSTM neural network are initialized as follows: initial learning rate l r Number of hidden layer units, number of iterations (epochs), minimum training batch size, and error threshold;

[0031] 3-3) Train the LSTM neural network using the training set data, and detect whether the prediction loss LOSS of the LSTM neural network is less than the error threshold or whether the number of iterations is greater than or equal to the maximum number of iterations. If yes, output the model parameters and output the trained runoff volume prediction model. If no, continue training.

[0032] Furthermore, the specific steps for predicting water depth in step 5) are as follows:

[0033] 4-1) Let the time to be predicted be t. d Collect the points t to be predicted d -r△t、...、t d -2△t、t d Rainfall data at time -Δt Input it into the LSTM neural network model trained in step 3), for t d The amount of water flowing at any given moment Make predictions;

[0034] 4-2) Collect t d -Δt is the vertical elevation difference between the point to be predicted at time t and the drainage outlet or drainage channel or lake. And the water depth at the point to be predicted and the vertical height of the corresponding drainage outlet at the point to be predicted.

[0035] like and Then t d The increase in the flow rate at the predicted point at any given time for:

[0036]

[0037] like and or Then t d The increase in the flow rate at the predicted point at any given time for:

[0038]

[0039]

[0040] In the formula, ν is the design flow velocity of the drainage pipe at the point to be predicted, and η is... td Let η1, η2, ..., η be the flow velocity influencing factors. g All are preset flow velocity influence factor thresholds, △S1, △S2, ..., △S g All are preset height difference thresholds

[0041] 4-3) Collect geographic type data for the points to be predicted, and determine the increase in runoff based on the geographic type data. Substitute f1(h) z ) or f2(h z Calculate the increase in water level Δh td , then t d The water depth at the point to be predicted at any given time for:

[0042]

[0043] In the formula, For t d -Δt is the depth of water accumulation at the point to be predicted.

[0044] Because of the adoption of the above technical solution, the present invention has the following advantages:

[0045] 1. This application improves the accuracy of water depth prediction by predicting the runoff volume of water accumulation points and calculating the drainage volume of water accumulation points under different conditions, providing support for urban flood control, disaster reduction and other fields.

[0046] 2. This application predicts the runoff volume of water accumulation points per unit time, effectively solving the technical problem of large variations in water depth due to changes in rainfall, which cause changes in the runoff area of ​​water accumulation points.

[0047] 3. This application calculates the relative elevation difference between the drainage outlet of the water accumulation point and the drainage river or lake, and introduces the flow velocity influence factor. It fully considers the impact of the drainage outlet of the water accumulation point on the flow velocity of the pipeline network when the drainage outlet of the water accumulation point is flooded by the river, thereby improving the prediction accuracy of the pipeline network drainage volume.

[0048] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0049] The accompanying drawings of this invention are described below.

[0050] Figure 1 This is a flowchart of the urban flooding prediction method based on LSTM neural network of the present invention. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0052] Example 1:

[0053] like Figure 1 The method for predicting urban flooding based on LSTM neural networks, as shown, includes the following steps:

[0054] 1) Data collection: At time intervals of △t, collect data on water depth at several water accumulation points in the city during each historical rainfall event, vertical height difference between the drainage outlets and drainage rivers or lakes corresponding to the water accumulation points, flow velocity data of the drainage pipes corresponding to the water accumulation points, rainfall data in the area of ​​the water accumulation points, and environmental data of the water accumulation points to obtain the collected data.

[0055] The water depth data is: H = {h1, h2, ..., h j ,...,h m}, where: h j This represents the water depth data of a single water accumulation point sampled for the jth time during a single rainfall event;

[0056] The rainfall data for the waterlogged area is: P = {p1, p2, ..., p...} j ,...,p N}, where: p j This represents the rainfall data sampled for the j-th time in a single waterlogged area during a single rainfall event;

[0057] The vertical elevation difference between the drainage outlet and the drainage river or lake corresponding to the water accumulation point is: △S={△s1,△s2,...,△s j ,...,△s N}, where: △s j This refers to the vertical elevation difference between the drainage outlet and the drainage river or lake corresponding to the j-th sampling point of a single water accumulation point during a single rainfall event.

[0058] The flow velocity V = {v1, v2, ..., v} in the drain pipe corresponding to the water accumulation point j ,...,v N}, where: v j This represents the flow velocity data of the drainage pipe corresponding to the j-th sampling point of a single water accumulation point during a single rainfall event;

[0059] The environmental data of the water accumulation point includes the geographical type data of the water accumulation point and the drainage pipe network data of the water accumulation point. The geographical type data of the water accumulation point includes two types: urban overpass underpasses and urban low-lying areas. The urban low-lying areas include urban sunken plazas or flat roads.

[0060] The drainage network data includes the design flow velocity ν of the drainage pipe at the water accumulation point, the flow constant b of the drainage pipe, and the vertical height difference △S′ between the water accumulation point and the corresponding drainage outlet.

[0061] In this invention example, the sampling time interval is set to 1 hour, and the drainage outlets of the water accumulation points are all connected to natural rivers or large lakes.

[0062] 2) Calculate the runoff volume: Construct a mathematical relationship between the runoff volume and depth of the water accumulation point, and calculate the runoff volume of the water accumulation point in the i-th sampling period based on the drainage network data and rainfall data; the specific steps are as follows:

[0063] 2-1) Construct the water accumulation volume Q at the water accumulation point z With water depth h z Mathematical relationship:

[0064] Q z =f(h) z )

[0065] In the formula, f(h) z Q represents the water volume at the water accumulation point. z With water depth h z Mathematical relational functions;

[0066] Constructing a mathematical function f1(h) for underpass tunnels at urban interchanges z A mathematical function f2(h) is constructed for low-lying urban areas. z ):

[0067]

[0068] f2(h z ) = h z .S j

[0069] In the formula, ax 2 Let S be the cross-sectional curve of an underpass tunnel at an urban interchange, w be the width of the underpass tunnel, and S be the width of the tunnel. j The area of ​​sunken plazas or flooded roads in the city;

[0070] 2-2) Calculate the amount of water flowing into the water accumulation point Q during the i-th sampling period. hi :

[0071]

[0072] In the formula, Q zi Let ν be the water volume at the i-th sampling point. i denoted as , where is the flow velocity of the drain pipe at the water accumulation point during the i-th sampling; b is the flow rate constant, which is related to the cross-sectional area of ​​the drain pipe at the water accumulation point; and n is the number of samplings.

[0073] In this invention, by statistically analyzing urban waterlogging points, it is determined that during rainfall, when the rainfall exceeds the maximum designed drainage capacity of the drainage pipes, rainwater will flow along roads or green belts towards urban overpasses or low-lying urban areas. The cross-section of urban overpasses is generalized to satisfy a parabolic shape, which is achieved by constructing the parabolic equation ax for urban overpasses. 2 The depth of the water accumulation is calculated. Since the low-lying areas of the city are all regular squares, the depth of the water accumulation is calculated by measuring the area of ​​the sunken plazas or waterlogged roads in the city.

[0074] 3) Data splitting: Sort the collected data and calculated runoff volume of the water accumulation point during a single rainfall process according to time to obtain the original dataset. Divide the original dataset into a test set and a training set in an 8:2 ratio.

[0075] 4) Model Construction: Using rainfall as input and runoff volume as output, construct a runoff volume prediction model based on an LSTM neural network, initialize the model parameters, and train the model using training set data. The specific steps are as follows:

[0076] 3-1) with p t-r△t ... p t-2△t p t-△t Rainfall p at time t-r△t ... p t-2△t p t-△t As input, the runoff volume Q at time t. hi To produce the output, a runoff volume prediction model based on an LSTM neural network is constructed, where r is the data length of the moving sliding window of the LSTM neural network.

[0077] In this invention, when the rainfall exceeds the maximum drainage capacity designed for the drainage pipes, rainwater will flow along roads or green belts to urban overpasses, underpasses, or low-lying urban areas. As the rainfall changes, the amount of water flowing into the water accumulation points will change in real time. The water depth is related not only to the amount of water flowing into the water accumulation points but also to the drainage speed of the water accumulation points. Therefore, the prediction accuracy of the rainfall-water accumulation prediction model is higher, and the value of r is 5.

[0078] 3-2) The LSTM neural network optimizer uses the Adam function, and the LSTM neural network uses the root mean square error (RMSE), mean absolute error (MAE), and absolute percentage error (MAPE) as loss functions. The parameters of the LSTM neural network are initialized as follows: initial learning rate l r Number of hidden layer units, number of iterations (epochs), minimum training batch size, and error threshold;

[0079] In this invention example, the LSTM neural network includes an input gate, a forget gate, and an output gate. The forget gate controls the memory unit, determining how much of the previous state is retained in the current state. This forgetting is achieved through a sigmoid layer. The input gate determines how much information is added to the unit. The information from the sigmoid layer and the tanh layer are used together to update the unit state. The output gate determines which part of the current unit state is output, again achieved through the sigmoid and tanh layers.

[0080] In this embodiment of the invention, the LSTM neural network has 3 input layer neurons and 1 output layer neuron. The model optimizer uses the Adam function to train the parameters of the LSTM neural network, with an initial learning rate of l. r The value is 0.005, the number of hidden layer units is 50, the number of iterations (epochs) is 200, and the minimum training batch size is 10.

[0081] 3-3) Train the LSTM neural network using the training set data, and detect whether the prediction loss LOSS of the LSTM neural network is less than the error threshold or whether the number of iterations is greater than or equal to the maximum number of iterations. If yes, output the model parameters and output the trained runoff volume prediction model. If no, continue training.

[0082] In this invention, the root mean square error (RMSE) reflects the degree of deviation between the flood forecast and the measured value; a smaller value indicates a smaller deviation. The mean absolute error (MAE) reflects the degree of error between the forecast and the measured value; a smaller value indicates that the forecast and the measured value are closer. The absolute percentage error (MAPE) reflects the percentage of the error between the forecast and the measured value relative to the measured value; a MAPE of 0% indicates a perfect model, while a MAPE greater than 100% indicates a poor model.

[0083] 5) Water depth prediction: Collect real-time data and environmental data of the point to be predicted, and input them into the LSTM neural network model trained in step 4) to predict the runoff volume, and calculate the water depth of the point to be predicted using the runoff volume and environmental data.

[0084] 4-1) Let the time to be predicted be t. d Collect the points t to be predicted d-r△t、...、t d -2△t、t d Rainfall data at time -Δt Input it into the LSTM neural network model trained in step 3), for t d The amount of water flowing at any given moment Make predictions;

[0085] In this embodiment of the invention, r is the data length of the moving sliding window of the LSTM neural network, and the value of r is 5.

[0086] 4-2) Collect t d -Δt is the vertical elevation difference between the point to be predicted at time t and the drainage outlet or drainage channel or lake. And the water depth at the point to be predicted and the vertical height of the corresponding drainage outlet at the point to be predicted.

[0087] like and Then t d The increase in the flow rate at the predicted point at any given time for:

[0088]

[0089] like and or Then t d The increase in the flow rate at the predicted point at any given time for:

[0090]

[0091]

[0092] In the formula, ν is the design flow velocity of the drainage pipe at the point to be predicted. Let η1, η2, ..., η be the flow velocity influencing factors. g All are preset flow velocity influence factor thresholds, △S1, △S2, ..., △S g All are preset height difference thresholds.

[0093] In this invention, the water level of drainage channels or ponds changes with rainfall. The elevation difference between the drainage outlet corresponding to the predicted point and the water level of the drainage channel or pond has a significant impact on the drainage speed of the drainage pipe. Therefore, by collecting real-time data on the vertical elevation difference between the drainage outlet corresponding to the predicted point and the drainage channel or lake, the drainage speed of the water accumulation point can be calculated. and When the water level in the drainage channel or lake is higher than the vertical height of the water at the accumulation point, the water cannot be drained, meaning all the flowing water accumulates at the accumulation point (the drainage network is equipped with anti-convective devices to prevent backflow of water from the drainage channel or lake); when... and or That is, the water level in the drainage channel or lake is lower than the vertical height of the water accumulation point. The velocity of the drainage pipe is related to the vertical height difference between the drainage outlet corresponding to the point to be predicted and the drainage channel or lake. In the example of this invention, η1, η2, ..., η g Statistical calculations were performed using historical data on the vertical elevation difference between the drainage outlet and the drainage river or lake corresponding to the water accumulation point, and the flow velocity data of the drainage pipe corresponding to the water accumulation point:

[0094] In this embodiment of the invention, g is 5, η1, η2, ..., η g The values ​​are 1, 0.8, 0.65, 0.5, and 0.4 respectively; △S g When it is 0mm, hour, This means that at this time, the drainage outlet corresponding to the water accumulation point is not submerged by the river or lake, and the flow velocity of the drainage pipe corresponding to the water accumulation point is equal to the design flow velocity. When the drainage outlet corresponding to the water accumulation point is not submerged by the river or lake, the flow velocity slows down as the submersion depth increases.

[0095] 4-3) Collect geographic type data for the points to be predicted, and determine the increase in runoff based on the geographic type data. Substitute f1(h) z ) or f2(h z Calculate the increase in water level Δh td , then t d The water depth at the point to be predicted at any given time for:

[0096]

[0097] In the formula, For t d -Δt is the depth of water accumulation at the point to be predicted.

[0098] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0099] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting urban flooding based on LSTM neural networks, characterized in that, The specific steps are as follows: 1) Data collection: At time intervals, data on water depth at several waterlogged points in the city during each historical rainfall event, vertical elevation difference between the drainage outlets and drainage rivers or lakes corresponding to the waterlogged points, flow velocity in the drainage pipes corresponding to the waterlogged points, rainfall in the area of ​​the waterlogged points, and environmental data of the waterlogged points are collected to obtain the collected data. 2) Calculate the runoff volume: Construct a mathematical relationship between the water volume and depth at each water accumulation point, and calculate the runoff volume based on the drainage network data and rainfall data at the water accumulation point. The water volume collected at the water accumulation point within each sampling period; 3) Data splitting: Sort the collected data and calculated runoff volume of the water accumulation point during a single rainfall process according to time to obtain the original dataset. Divide the original dataset into a test set and a training set in an 8:2 ratio. 4) Model building: Using rainfall as input and runoff volume as output, a runoff volume prediction model based on LSTM neural network is built, the model parameters are initialized, and the model is trained using training set data; 5) Water depth prediction: Collect real-time data and environmental data of the point to be predicted, and input them into the LSTM neural network model trained in step 4) to predict the runoff volume, and calculate the water depth of the point to be predicted based on the runoff volume and environmental data. The specific steps for calculating the runoff volume in step 2) are as follows: 2-1) Construct the water accumulation volume at the water accumulation point With water depth Mathematical relationship: In the formula, Water volume at the water accumulation point With water depth Mathematical relational functions; Mathematical functions for constructing underpasses in urban interchanges Mathematical functions are constructed for low-lying urban areas. : In the formula, The cross-sectional curve of an underpass tunnel at an urban interchange. The width of the underpass tunnel at the city interchange. The area of ​​sunken plazas or flooded roads in the city; 2-2) Calculate the first The amount of water flowing into the water accumulation point within each sampling period : In the formula, For the first The amount of water collected at each sampling point For the first The flow rate of the drainage pipe at the secondary sampling point was measured. This is the flow constant, which is related to the cross-sectional area of ​​the drain pipe at the water accumulation point; Number of samples; The specific steps for predicting water depth in step 5) are as follows: 4-1) Let the time to be predicted be... Collect points to be predicted Rainfall data at any time Input it into the LSTM neural network model trained in step 3), and... The amount of water flowing at any given moment Make predictions; 4-2) Collection The vertical elevation difference between the drainage outlet and the drainage channel or lake at the point to be predicted at any given time. And the water depth at the point to be predicted and the vertical height of the corresponding drainage outlet at the point to be predicted. ; like and ,but The increase in the flow rate at the predicted point at any given time for: like and ,or ,but The increase in the flow rate at the predicted point at any given time for: In the formula, The design flow velocity of the drainage pipe at the point to be predicted. For flow velocity influencing factors, All are preset threshold values ​​for flow velocity influence factors. All are preset height difference thresholds; 4-3) Collect geographic type data for the points to be predicted, and determine the increase in runoff volume based on the geographic type data. Substitute or Calculate the increase in water level ,but The water depth at the point to be predicted at any given time for: In the formula, for The water depth at the point to be predicted at any given time.

2. The urban flooding prediction method based on LSTM neural network as described in claim 1, characterized in that, The water depth data mentioned in step 1) is as follows: ,in: For a single water accumulation point during a single rainfall event The water depth data from the second sampling; The rainfall data for the waterlogged area is as follows: ,in: For a single waterlogged area during a single rainfall event Rainfall data from the second sampling; The vertical elevation difference between the drainage outlet and the drainage river or lake corresponding to the water accumulation point is as follows: ,in: For a single water accumulation point during a single rainfall event The vertical elevation difference data between the drainage outlet and the drainage river or lake corresponding to the water accumulation point was sampled. The flow velocity of the drain pipe corresponding to the water accumulation point ,in: For a single water accumulation point during a single rainfall event The flow velocity data of the drainage pipe corresponding to the water accumulation point was sampled. The environmental data of the water accumulation point includes the geographical type data of the water accumulation point and the drainage pipe network data of the water accumulation point. The geographical type data of the water accumulation point includes two types: urban overpass underpasses and urban low-lying areas. The urban low-lying areas include urban sunken plazas or flat roads. The drainage network data includes the design flow velocity of the drainage pipes at water accumulation points. Flow constant of drain pipe .

3. The urban flooding prediction method based on LSTM neural network as described in claim 2, characterized in that, The specific steps for building the model in step 4) are as follows: 3-1) with Rainfall at any time As input, with The amount of water flowing at any given moment For the output, a runoff volume prediction model based on an LSTM neural network is constructed, where: The data length of the sliding window in the LSTM neural network; 3-2) The LSTM neural network optimizer uses the Adam function, and the LSTM neural network uses the root mean square error (RMSE), mean absolute error (MAE), and absolute percentage error (MAPE) as loss functions. The parameters of the LSTM neural network are initialized as follows: initial learning rate. Number of hidden layer units, number of iterations (epochs), minimum training batch size, and error threshold; 3-3) Train the LSTM neural network using the training set data, and detect whether the prediction loss LOSS of the LSTM neural network is less than the error threshold or whether the number of iterations is greater than or equal to the maximum number of iterations. If yes, output the model parameters and output the trained runoff volume prediction model. If no, continue training.

Images