A flood disaster prediction and early warning method based on DA-SSL

Abstract

The application discloses a DA-SSL-based flood disaster prediction and early warning method, which comprises the following steps: obtaining an initial sample of city pipe network diseases; inputting the initial sample of city pipe network diseases into an improved PCSWMM model; the improved PCSWMM model outputs a pipe network functional disease sample set with an in-flood result label; the initial sample of city pipe network diseases is expanded to obtain an expanded sample set with an in-flood result label; the information difference of the sample set with / without a label is measured by using cross entropy, a sample set without a label without an in-flood result label is randomly generated, and the sample set with a label and the sample set without a label are input into the improved PCSWMM model; the improved PCSWMM model is repeatedly trained to obtain a city in-flood area result under different pipe network functional disease conditions. According to the application, the drainage pipe network flow capacity and flow velocity distribution under the influence of functional diseases can be accurately fitted, and the in-flood early warning result is very accurate and reliable.

Description

Technical Field

[0001] The present invention relates to the field of flood disaster early warning, and in particular to a flood disaster prediction and early warning method based on DA-SSL. Background Art

[0002] Existing flood disaster prediction and warning methods, such as the relatively mature PCSWMM (Precipitation and Flood Management Model), can comprehensively consider surface runoff simulation and pipeline network hydrodynamic transmission processes during heavy rainstorms with different return periods when constructing urban waterlogging warning and disaster prevention decision-making systems. However, they assume that the drainage network is unobstructed, ignoring the impact of pipeline diseases on flow capacity and flow velocity distribution. However, surveys show that major cities across China have an average of 12 functional diseases per kilometer of drainage pipelines, of which scaling and siltation account for over 70%. This results in existing flood disaster prediction and warning methods being seriously inaccurate in predicting drainage network flow capacity. There is an urgent need to develop flood disaster prediction and warning methods that consider the coupled relationship between pipeline functional diseases and urban waterlogging.

[0003] Existing research on urban waterlogging early warning models mainly focuses on refined stormwater simulation and low-impact development design under heavy rains with different recurrence periods, ignoring the impact of functional defects in pipeline networks under multiphase flow and multi-field coupling on one- and two-dimensional runoff generation and runoff in urban surfaces and underground areas, making it difficult to ensure the accuracy of waterlogging early warning results.

[0004] Urban waterlogging warning is affected by multiple parameters such as pipeline slope, hydraulic radius, and wetted perimeter. Analyzing and calculating the correspondence between pipeline functional defects and waterlogged areas one by one through the control variable method is labor-intensive and difficult to implement.

[0005] In summary, existing flood disaster prediction and early warning methods have the following shortcomings:

[0006] (1) The impact of pipeline functional diseases (scaling and siltation) on flow capacity and flow velocity distribution is ignored, resulting in the serious lack of accuracy of existing flood disaster prediction and early warning methods in predicting the flow capacity of drainage pipe networks;

[0007] (2) The impact of functional defects of the pipe network on the one- and two-dimensional runoff generation and convergence of urban surface and underground watersheds under multiphase flow and multi-field coupling is ignored, making it difficult to ensure the accuracy of urban waterlogging warning results;

[0008] (3) Analyzing and calculating the corresponding relationship between pipeline functional diseases and waterlogging areas one by one through the control variable method is a huge workload and difficult to achieve.

[0009] In order to solve the above problems, the inventors of the present invention provide a flood disaster prediction and early warning method based on DA-SSL. Summary of the Invention

[0010] In order to solve the above problems, the purpose of the present invention is to provide a flood disaster prediction and early warning method based on DA-SSL, which can accurately fit the flow capacity and flow velocity distribution of the drainage network under the influence of functional diseases, and the waterlogging early warning results are very accurate and reliable.

[0011] Based on this, the present invention provides a flood disaster prediction and early warning method based on DA-SSL, the method comprising:

[0012] Obtaining an initial sample of functional diseases of the urban pipe network, wherein the initial sample of functional diseases of the urban pipe network includes: rainstorm data with different return periods and an existing sample set of functional diseases of the pipe network;

[0013] Input the initial sample of functional diseases of the urban pipe network into the improved PCSWMM model;

[0014] The improved PCSWMM model outputs a sample set of pipe network functional diseases with waterlogging result labels;

[0015] Expanding the sample set of the initial samples of urban pipe network diseases to obtain an expanded labeled sample set with waterlogging result labels;

[0016] Cross entropy is used to measure the information difference between labeled and unlabeled sample sets, and an unlabeled sample set without waterlogging result labels is randomly generated. The labeled sample set and the unlabeled sample set are input into the improved PCSWMM model, and the improved PCSWMM model is repeatedly trained so that the improved PCSWMM model can obtain urban waterlogging area results for different pipe network functional disease conditions.

[0017] The sample set expansion of the initial samples of urban pipe network diseases includes:

[0018] Obtaining the existing sample set of functional diseases of the pipe network and the corresponding waterlogging result label values, and assigning initial values ​​to the expanded new sample set of functional diseases of the pipe network;

[0019] Calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the functional disease sample of the pipe network, and obtaining the feature item index value with the maximum loss gradient;

[0020] Re-assign the expanded sample feature item corresponding to the index value according to the preset assignment rule;

[0021] The new sample set of functional disease of the urban pipe network with waterlogging result labels is continuously adjusted and assigned by the initial sample of functional disease of the urban pipe network to make it close to the classification boundary, thereby obtaining the expanded labeled sample set.

[0022] The step of calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the functional disease sample of the pipe network and obtaining the feature item index value with the maximum loss gradient includes:

[0023] The calculation method of the semi-supervised learning inversion loss function F is:

[0024]

[0025] Where M is the number of functional disease categories, y ic Indicates that the true category of sample i is equal to c and takes 1, otherwise it takes 0, p ic represents the predicted probability that sample i belongs to functional disease category c, and N is the number of hidden layers;

[0026] The loss gradient is obtained by semi-supervised learning inversion loss function F for any expanded functional disease sample The derivative is obtained in the direction of

[0027]

[0028] Among them, L s Indicates the waterlogging result label value corresponding to the original pipe network functional disease sample set, N s For the expanded new pipeline network functional disease sample set, For any expanded functional disease sample;

[0029] The feature item index value i with the largest loss gradient max By calculating the inversion loss function F on any expanded functional disease sample The maximum value on is solved, that is,

[0030]

[0031] Among them, n is the maximum number of rows in the pipeline network functional disease sample set, and m is the maximum number of columns in the pipeline network functional disease sample set.

[0032] The re-assigning of the expanded sample feature item corresponding to the index value according to the preset assignment rule includes:

[0033]

[0034] Among them, the left side of if For the expanded functional disease samples, if the right side This is a functional disease sample before expansion.

[0035] The method further includes: using a semi-supervised learning inversion method and a control variable method to compare and analyze the pipeline network functional disease input and waterlogging result data sets, extracting the corresponding relationship between the pipeline network functional disease and the waterlogging situation in the local area of ​​the city, and calibrating various parameters of the loss function;

[0036] The prediction results are added to the initial training set at a preset update cycle, and the error of the loss function is reduced through repeated iterations.

[0037] The semi-supervised learning inversion method and the control variable method are used to compare and analyze the pipeline network functional disease input and waterlogging result data sets, extract the corresponding relationship between the pipeline network functional disease and the waterlogging situation in local urban areas, and calibrate the various parameters of the loss function, including:

[0038] The corresponding relationship between the functional defects of the pipe network and the waterlogging situation in the local area of ​​the city is obtained by calibrating the parameters of the loss function;

[0039] The parameters of the calibration loss function are determined by continuously adjusting and reducing the comprehensive error rate of the input layer, hidden layer and output layer of the initial sample set of functional diseases of the urban pipe network.

[0040] The control variable method is used to adjust the size of each parameter one by one to achieve the purpose of reducing the error rate of the output layer. i is the i-th parameter, c is a constant, and n is the total number of parameters. If the first parameter is adjusted to (A1+c,A2,...,A n ), the error rate decreases, then continue to increase the first parameter, otherwise reduce the first parameter. After the adjustment of the first parameter is completed, adjust the other parameters one by one with the constant c.

[0041] The construction process of the improved PCSWMM model is as follows:

[0042] The construction process of the improved PCSWMM model includes: engineering a one- and two-dimensional coupling model that considers functional defects of the pipeline network into a functional module, and integrating it with the existing PCSWMM model in the form of a dynamic library to obtain an improved PCSWMM model that considers pipeline network defects;

[0043] The hydrological and hydrodynamic coupling relationship model between functional diseases of the drainage network and waterlogging can be dynamically quantified using the one-dimensional clear-flow governing equation and the two-dimensional shallow-water equation, as follows:

[0044]

[0045]

[0046] Among them, Z represents the water level, A is the cross-sectional area of ​​the water flow, Q is the outlet flow of the cross section, and q Lrepresents the lateral inflow, g is the acceleration due to gravity, t and x represent the one-dimensional time and space coordinates, respectively, a represents the wave speed, S f is the frictional slope, h L is the local head loss over the unit length, h is the water depth, t is the time, x, y and z are the coordinate system, u and v are the flow velocity components in the x and y directions, respectively, is the vertical average flow velocity, p is the fluid density, b is the water bottom elevation, S ax and S ay are the x and y direction bottom slope component, respectively, S fx and S fy are the x and y direction friction component, respectively, τ zy and τ zy are the lateral stress;

[0047] A one-dimensional full-flow control equation and a two-dimensional shallow water equation are used to engineer a functional module of PCSWMM to improve the prediction accuracy of PCSWMM for waterlogging.

[0048] The present application adopts the numerical simulation and fusion technical means of "data set expansion (DA) + semi-supervised learning (SSL) + rain flood management model (PCSWMM)" to realize the DA-SSL-based flood disaster prediction and early warning method and carry out flood disaster prediction and early warning under different return periods of heavy rain. The purpose of the present application is to design and realize a DA-SSL-based flood disaster prediction and early warning method based on the coupling relationship between functional diseases of drainage pipe networks and waterlogging, DA, SSL and PCSWMM, which can accurately fit the flow capacity and flow velocity distribution of the drainage pipe network under the influence of functional diseases, accurately and efficiently and reliably predict waterlogging, and overcome many shortcomings of the current flood disaster prediction and early warning method. Based on this, the present application has the following advantages:

[0049] (1) Accurately fitting the flow capacity and flow velocity distribution of the drainage pipe network under the influence of functional diseases. The DA-SSL-based flood disaster prediction and early warning method uses the pipe confluence parameter determined by the test rate to determine the critical starting condition of pipe functional diseases and the hydrodynamic coupling relationship to improve PCSWMM, and under the joint cooperation of DLL dynamic library linkage and Arcgis software, the flow capacity and flow velocity distribution of the drainage pipe network under the influence of functional diseases are accurately fitted.

[0050] (2) Accurate waterlogging early warning results. By considering the hydrodynamic coupling relationship between pipe network functional diseases and waterlogging, the PCSWMM dynamic simulation considers the hydrodynamic coupling relationship between pipe network functional diseases and waterlogging, the urban one-two-dimensional instantaneous water flow state conversion law, accurately calculates the waterlogging indexes such as water depth, waterlogging area and waterlogging duration, and solves the problem that the influence of pipe network functional diseases on waterlogging results is not considered in the existing research.

[0051] (3) High efficiency. The flood disaster prediction and warning method based on DA-SSL uses a deep learning method that combines dataset expansion and semi-supervised learning to achieve efficient output of flood warning results.

[0052] (4) High reliability. This paper uses the numerical simulation and fusion technology of "dataset augmentation (DA) + semi-supervised learning (SSL) + stormwater management model (PCSWMM)" to implement a flood disaster prediction and early warning method based on DA-SSL. The PCSWMM, which considers the coupling relationship between functional defects of the pipeline network and the hydrological and hydrodynamic characteristics of waterlogging, obtains an accurate deep learning input sample set. Combined with the data set augmentation method, it can fully guarantee the reliability of the output results of the flood disaster prediction and early warning method based on DA-SSL. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 Schematic diagram of a flood disaster prediction and early warning method based on DA-SSL provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] Figure 1 FIG. 1 is a schematic diagram of a flood disaster prediction and early warning method based on DA-SSL provided by an embodiment of the present invention, the method comprising:

[0057] S101. Obtaining an initial sample of urban pipe network disease defects, wherein the initial sample of urban pipe network disease defects includes: rainstorm data with different return periods and an existing sample set of pipe network functional disease defects;

[0058] The existing sample set of functional diseases of the pipeline network includes the siltation situation (whether there is silt material), siltation degree, siltation length, scaling situation (whether there is scaling body), scaling degree, and scaling length of pipelines in different sections.

[0059] S102, inputting the initial sample of urban pipe network diseases into the improved PCSWMM model;

[0060] Improvement process of the improved PCSWMM model: The one- and two-dimensional coupling model considering functional diseases of the pipeline network is engineered into a functional module, and integrated with the existing PCSWMM model in the form of a dynamic library to obtain the improved PCSWMM model considering pipeline network diseases.

[0061] The construction process of the improved PCSWMM model is as follows:

[0062] Improvement process of the improved PCSWMM model: The one- and two-dimensional coupling model considering functional diseases of the pipeline network is engineered into a functional module, and integrated with the existing PCSWMM model in the form of a dynamic library to obtain the improved PCSWMM model considering pipeline network diseases.

[0063] The hydrological and hydrodynamic coupling relationship model between functional diseases of the drainage network and waterlogging can be dynamically quantified using the one-dimensional clear-flow governing equation and the two-dimensional shallow-water equation, as follows:

[0064]

[0065]

[0066] Among them, Z represents the water level, A is the cross-sectional area of ​​the water flow, Q is the outlet flow of the cross section, and q L represents the lateral inflow, g is the gravitational acceleration, t and x represent the one-dimensional time and space coordinates, a represents the wave velocity, S f is the friction ratio, h L is the local head loss over the unit length, h is the water depth, t is the time, x, y and z are the coordinate systems, u and v are the velocity components in the x and y directions respectively, is the vertical average velocity, ρ is the fluid density, b is the bottom elevation, S ax and S ay are the bottom slope components in the x and y directions, S fx and S fy are the friction components in the x and y directions, τ zy and τ zy Both are lateral stresses.

[0067] The one-dimensional clearwater flow governing equation and the two-dimensional shallow water equation were engineered into a functional module of PCSWMM to improve the accuracy of PCSWMM's prediction of waterlogging.

[0068] S103, the improved PCSWMM model outputs a pipe network functional disease sample set with waterlogging result labels;

[0069] S104: Expand the sample set of the initial samples of urban pipe network diseases to obtain an expanded labeled sample set with waterlogging result labels;

[0070] Since the initial samples of urban pipe network diseases are limited, the initial samples are expanded by random sample set expansion method and input into the improved PCSWMM to obtain a labeled sample set with waterlogging result labels;

[0071] Here, only the existing sample set of network functional diseases in the initial sample of urban pipe network diseases is input into the improved PCSWMM model.

[0072] The sample set expansion of the initial samples of urban pipe network diseases includes:

[0073] Obtaining the existing sample set of functional diseases of the pipe network and the corresponding waterlogging result label values, and assigning initial values ​​to the expanded new sample set of functional diseases of the pipe network;

[0074] Calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the functional disease sample of the pipe network, and obtaining the feature item index value with the maximum loss gradient;

[0075] Re-assign the expanded sample feature item corresponding to the index value according to the preset assignment rule;

[0076] The new sample set of functional disease of the pipeline network with waterlogging result labels is continuously adjusted and assigned by the initial sample set to make it close to the classification boundary, thereby obtaining the expanded labeled sample set.

[0077] The step of calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the pipe network functional disease sample and obtaining the feature item index value with the maximum loss gradient includes:

[0078] The calculation method of the semi-supervised learning inversion loss function F is:

[0079]

[0080] Where M is the number of functional disease categories, y ic Indicates that the true category of sample i is equal to c and takes 1, otherwise it takes 0, p ic represents the predicted probability that sample i belongs to functional disease category c, and N is the number of hidden layers;

[0081] The loss gradient is obtained by semi-supervised learning inversion loss function F for any expanded functional disease sample The derivative is obtained in the direction of

[0082]

[0083] Among them, L s Indicates the waterlogging result label value corresponding to the original pipe network functional disease sample set, N s For the expanded new pipeline network functional disease sample set, For any expanded functional disease sample;

[0084] The index value i of the feature item with the largest loss gradient max By calculating the inversion loss function F on any expanded functional disease sample The maximum value on is solved to obtain , that is .

[0085]

[0086] Among them, n is the maximum number of rows in the pipeline network functional disease sample set, and m is the maximum number of columns in the pipeline network functional disease sample set.

[0087] The re-assigning of the expanded sample feature item corresponding to the index value according to the preset assignment rule includes:

[0088] The principle of expanding the sample set of the initial urban pipe network disease samples is as follows: FGSM calculates the gradient of the semi-supervised learning inversion loss function with respect to the input data set, finds the feature item with the largest gradient value, and then changes the value of this feature item to move the input in the direction of increasing the model loss, thereby generating new sample points that are closer to the decision boundary of the model. The specific process is as follows:

[0089] 1) The original pipe network functional disease sample set and the corresponding waterlogging result label values ​​are I S , L s ;

[0090]

[0091] There are n·m-dimensional feature items, where m represents the number of flood warning areas and n represents the number of functional disease types in drainage pipes. Each feature item takes a value set of {0, 0.25, 0.5, 0.75, 1}. A value of 0 indicates that the drainage pipe does not have that functional disease; a value between 0.25 and 1 indicates the severity of that functional disease. Initial values ​​are assigned to the expanded new pipe network functional disease sample set Ns.

[0092]

[0093] 2) Calculate the loss gradient of the semi-supervised learning inversion loss function F for the original waterlogging result label and find the feature item index value i with the largest gradient value max ,Right now.

[0094]

[0095] 3) In order to make the new sample value with waterlogging label obtained by data expansion closer to the actual situation, the index value i max The corresponding new sample feature items are re-adjusted and assigned, and the assignment rules are:

[0096]

[0097] 4) New label L with waterlogging results s The functional disease sample set N of the pipeline network s From the initial sample set I s The assignment is continuously adjusted to make it close to the classification boundary, thereby obtaining an expanded set of labeled samples.

[0098] S105. Use cross entropy to measure the information difference between labeled and unlabeled sample sets, randomly generate unlabeled sample sets without waterlogging result labels, input the labeled sample sets and the unlabeled sample sets into the improved PCSWMM model, repeatedly train the improved PCSWMM model, and obtain urban waterlogging regional results for different pipe network functional disease conditions.

[0099] The method further includes: using a semi-supervised learning inversion method and a control variable method to compare and analyze the pipeline network functional disease input and waterlogging result data sets, extracting the corresponding relationship between the pipeline network functional disease and the waterlogging situation in the local area of ​​the city, and calibrating various parameters of the loss function;

[0100] The prediction results are added to the initial training set at a preset update cycle, and the error of the loss function is reduced through repeated iterations.

[0101] The corresponding relationship between the functional defects of the pipe network and the waterlogging situation in the local area of ​​the city is obtained by calibrating the parameters of the loss function;

[0102] The parameters of the calibration loss function are determined by continuously adjusting and reducing the comprehensive error rate of the input layer, hidden layer and output layer of the initial sample set of functional diseases of the urban pipe network.

[0103] The control variable method is used to adjust the size of each parameter one by one to achieve the purpose of reducing the error rate of the output layer. i is the i-th parameter, c is a constant, and n is the total number of parameters. If the first parameter is adjusted to (A1+c, A2, ..., A n ), the error rate decreases, then continue to increase the first parameter, otherwise reduce the first parameter. After the first parameter is adjusted, adjust the other parameters one by one with the constant c.

[0104] The present invention adopts a flood prediction and early warning system of "data set expansion (DA) + semi-supervised learning (SSL) + rainwater management model (PCSWMM)", which fully considers the impact of the coupling relationship between the functional diseases of the underground drainage network and the hydrology and hydrodynamics of waterlogging on the flood results, and combines the sample set expansion based on FGSM and the semi-supervised learning inversion technology to improve the output efficiency and accuracy of the flood prediction and early warning results. Therefore, the present invention has the advantages of high prediction efficiency, high accuracy and high reliability. The new output regional flood results of the flood disaster prediction and early warning method based on DA-SSL and the input samples of the functional diseases of the pipeline network together constitute a labeled sample set, and the semi-supervised learning inversion is error corrected with a period T. Therefore, the present invention has the advantage that the prediction and early warning accuracy gradually increases with the increase of the number of iterations.

[0105] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.

Claims

1. A flood disaster prediction and early warning method based on DA-SSL, characterized in that: include: Obtaining an initial sample of functional diseases of the urban pipe network, wherein the initial sample of functional diseases of the urban pipe network includes: rainstorm data with different return periods and an existing sample set of functional diseases of the pipe network; Inputting the initial sample of functional defects of the urban pipe network into an improved PCSWMM model, the construction process of the improved PCSWMM model comprising: engineering a one- and two-dimensional coupled model that considers functional defects of the pipe network and can be dynamically quantified by a one-dimensional full flow control equation and a two-dimensional shallow water equation into a functional module, and integrating it with the existing PCSWMM model in the form of a dynamic library to obtain an improved PCSWMM model that considers pipe network defects; The improved PCSWMM model outputs a sample set of pipe network functional diseases with waterlogging result labels; Expanding the sample set of the initial samples of urban pipe network diseases, wherein expanding the sample set of the initial samples of urban pipe network diseases includes: Obtaining the existing sample set of functional diseases of the pipe network and the corresponding waterlogging result label values, and assigning initial values ​​to the expanded new sample set of functional diseases of the pipe network; Calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the functional disease sample of the pipe network, and obtaining the feature item index value with the maximum loss gradient; Re-assign the expanded sample feature item corresponding to the index value according to the preset assignment rule; The new pipe network functional disease sample set with waterlogging result labels is continuously adjusted and assigned by the initial sample of urban pipe network functional disease to make it close to the classification boundary, thereby obtaining an expanded labeled sample set; Obtain the expanded labeled sample set with waterlogging result labels; Cross entropy is used to measure the information difference between labeled and unlabeled sample sets, and an unlabeled sample set without waterlogging result labels is randomly generated. The labeled sample set and the unlabeled sample set are input into the improved PCSWMM model, and the improved PCSWMM model is repeatedly trained so that the improved PCSWMM model can obtain urban waterlogging area results for different pipe network functional disease conditions.

2. The flood disaster prediction and early warning method based on DA-SSL according to claim 1, characterized in that: Expanding the sample set of the initial sample of urban pipe network diseases includes: Obtaining the existing sample set of functional diseases of the pipe network and the corresponding waterlogging result label values, and assigning initial values ​​to the expanded new sample set of functional diseases of the pipe network; Calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the functional disease sample of the pipe network, and obtaining the feature item index value with the maximum loss gradient; Re-assign the expanded sample feature item corresponding to the index value according to the preset assignment rule; The new sample set of functional disease of the urban pipe network with waterlogging result labels is continuously adjusted and assigned by the initial sample of functional disease of the urban pipe network to make it close to the classification boundary, thereby obtaining the expanded labeled sample set.

3. The flood disaster prediction and early warning method based on DA-SSL according to claim 2, characterized in that: Calculating the loss gradient of the semi-supervised learning inversion loss function for the waterlogging result label value corresponding to the pipe network functional disease sample and obtaining the feature item index value with the maximum loss gradient includes: The calculation method of the semi-supervised learning inversion loss function F is: Where M is the number of functional disease categories, y ic Indicates that the true category of sample i is equal to c and takes 1, otherwise it takes 0, p ic represents the predicted probability that sample i belongs to functional disease category c, and N is the number of hidden layers; The loss gradient is obtained by semi-supervised learning inversion loss function F for any expanded functional disease sample The derivative is obtained in the direction of Among them, L s Indicates the waterlogging result label value corresponding to the original pipe network functional disease sample set, N s For the expanded new pipeline network functional disease sample set, For any expanded functional disease sample; The index value i of the feature item with the largest loss gradient max By calculating the inversion loss function F on any expanded functional disease sample The maximum value on is solved, that is, Among them, n is the maximum number of rows in the pipeline network functional disease sample set, and m is the maximum number of columns in the pipeline network functional disease sample set.

4. The flood disaster prediction and early warning method based on DA-SSL according to claim 2, characterized in that: The re-assigning of the expanded sample feature item corresponding to the index value according to the preset assignment rule includes: Among them, the left side of if For the expanded functional disease samples, if the right side This is a functional disease sample before expansion.

5. The flood disaster prediction and early warning method based on DA-SSL according to claim 1, characterized in that: The method also includes: using a semi-supervised learning inversion method and a control variable method to compare and analyze the pipeline network functional disease input and urban waterlogging result data sets, extracting the corresponding relationship between the pipeline network functional disease and the urban local area waterlogging situation, and calibrating various parameters of the loss function; adding the prediction results to the initial training set at a preset update cycle, and repeatedly iterating to reduce the error of the loss function.

6. The flood disaster prediction and early warning method based on DA-SSL according to claim 3, characterized in that: The semi-supervised learning inversion method and the control variable method are used to compare and analyze the pipeline network functional disease input and waterlogging result data sets, extract the corresponding relationship between the pipeline network functional disease and the waterlogging situation in local urban areas, and calibrate the various parameters of the loss function, including: The corresponding relationship between the functional defects of the pipe network and the waterlogging situation in the local area of ​​the city is obtained by calibrating the parameters of the loss function; The parameters of the calibration loss function are determined by continuously adjusting and reducing the comprehensive error rate of the input layer, hidden layer and output layer of the initial sample set of functional diseases of the urban pipe network. The control variable method is used to adjust the size of each parameter one by one to achieve the purpose of reducing the error rate of the output layer. i is the i-th parameter, c is a constant, and n is the total number of parameters. If the first parameter is adjusted to (A1+c,A2,...,A n ), the error rate decreases, then continue to increase the first parameter, otherwise reduce the first parameter. After the adjustment of the first parameter is completed, adjust the other parameters one by one with the constant c.

7. The flood disaster prediction and early warning method based on DA-SSL according to claim 1, characterized in that: The method further includes The hydrological and hydrodynamic coupling relationship model between functional diseases of the drainage network and waterlogging can be dynamically quantified using the one-dimensional clear-flow governing equation and the two-dimensional shallow-water equation, as follows: Among them, Z represents the water level, A is the cross-sectional area of ​​the water flow, Q is the outlet flow of the cross section, and q L represents the lateral inflow, g is the gravitational acceleration, t and x represent the one-dimensional time and space coordinates, a represents the wave velocity, S f is the friction ratio, h L is the local head loss over the unit length, h is the water depth, t is the time, x, y and z are the coordinate systems, u and v are the velocity components in the x and y directions respectively, is the vertical average velocity, ρ is the fluid density, b is the bottom elevation, S ax and S ay are the bottom slope components in the x and y directions, S fx and S fy are the friction components in the x and y directions, τ zy and τ zy All are lateral stresses; The one-dimensional clear-flow governing equation and the two-dimensional shallow water equation were engineered into a functional module of PCSWMM to improve the accuracy of PCSWMM's prediction of waterlogging.

Images