An intelligent prediction method for urban waterlogging considering sewer clogging

Abstract

The application discloses an urban waterlogging intelligent prediction method considering sewer pipe silting, constructs a sewer pipe silting considered underground pipe network and two-dimensional surface coupling waterlogging numerical simulation model by using underground drainage pipe network data; by using the stormwater monitoring data of the application area, the constructed underground pipe network and the numerical simulation model are subjected to model calibration and verification; a storm event data set is constructed, and storm events of different rainfall types and different return periods are designed; an urban waterlogging intelligent prediction model is fused by an improved convolutional neural network, a bidirectional long short-term memory neural network and a Bayesian optimization algorithm, a rainfall time sequence of a database is taken as input of the model, a flooded time sequence is taken as output of the model, a training set of the database is used for training and parameter optimization of the urban waterlogging intelligent prediction model; and the rainfall time sequence in the test set of the database is taken as input, and urban waterlogging intelligent prediction under a pipe silting scenario is realized by the constructed urban waterlogging intelligent prediction model.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of urban waterlogging simulation models, and relates to an intelligent urban waterlogging prediction method considering sewer pipe silting. BACKGROUND

[0002] In recent years, urban waterlogging disasters have occurred frequently in China, posing a serious threat to the daily life and safety of life and property of residents, and waterlogging risk prevention and control is facing severe challenges. For urban areas, underground drainage pipes are the most important flood control and drainage facilities. When it rains, if the pipe network drainage rate is lower than the rainwater collection rate, it will cause water accumulation on the ground, which will affect the normal life of urban residents, and even cause serious waterlogging disasters. Sewer silting will greatly reduce the flow capacity of urban drainage pipes and more easily induce urban waterlogging disasters.

[0003] However, the existing urban waterlogging models do not consider the impact of pipe silting on urban waterlogging, which will seriously reduce the simulation accuracy of the urban waterlogging model. At the same time, the current urban waterlogging model is mostly based on hydrological and hydrodynamic numerical simulation methods, and the model construction is complex and the model calibration workload is large, which is contradictory to the timeliness and accuracy of urban waterlogging prediction. How to improve the accuracy and timeliness of storm waterlogging simulation on the basis of considering pipe silting is a key problem to be solved, so it is urgent to develop an intelligent urban waterlogging prediction method considering sewer silting. SUMMARY

[0004] In view of the above problems, the purpose of the present application is to provide an intelligent urban waterlogging prediction method considering sewer silting, which improves the storm flood management model according to the pipe flow control equation under the pipe silting scenario, and couples the storm flood management model with the flow exchange equation between the one-dimensional pipe network and the two-dimensional surface under the pipe silting scenario through the flow exchange equation, to construct a coupled urban waterlogging numerical simulation model of the one-dimensional pipe network and the two-dimensional surface considering pipe silting. An intelligent urban waterlogging prediction model is constructed by fusing an improved convolutional neural network, a bidirectional long short-term memory neural network and a Bayesian optimization algorithm, to simulate the waterlogging inundation process of the storm event data set and provide training data set for the intelligent urban waterlogging prediction model of the improved convolutional neural network, bidirectional long short-term memory neural network and Bayesian optimization algorithm. Finally, the rainfall event is introduced into the trained intelligent urban waterlogging prediction model, and the waterlogging result under the pipe silting scenario is output, to realize intelligent urban waterlogging prediction considering pipe silting.

[0005] The technical scheme adopted by the present application is as follows: an intelligent urban waterlogging prediction method considering sewer silting, comprising the following steps:

[0006] Step S1, constant uniform flow in the pipeline is simulated by using Manning equation to realize numerical rating of friction loss along the way;

[0007] Step S2, the one-dimensional Saint-Venant equation and the friction loss along the way in step S1 are combined with the momentum conservation method to construct the underground pipe network water flow dynamics control equation considering pipe siltation;

[0008] Step S3, the one-dimensional Saint-Venant equation in step S2 and the hydraulic analysis method based on grid coordinates are used to construct the two-dimensional surface water flow dynamics control equation;

[0009] Step S4, the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3 are used to construct the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipe siltation;

[0010] Step S5, the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipe siltation constructed in step S4 is calibrated and verified by using the stormwater monitoring data of the application area;

[0011] Step S6, a storm event data set is constructed, the storm intensity formula of the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipe siltation in step S5 is constructed based on the recorded rainfall data of the application area, and a storm event data set of different rainfall types and different return periods is constructed;

[0012] Step S7, the storm event data set of different rainfall types and different return periods constructed in step S6 is input into the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipe siltation in step S4, the urban waterlogging inundation dynamic evolution scenario of the application area under different storm events is simulated, and a “storm-urban waterlogging” database containing rainfall time series and inundation time series under pipe siltation scenario is constructed;

[0013] Step S8, the “storm-urban waterlogging” database constructed in step S7 is normalized and divided into training set, validation set and test set;

[0014] Step S9, an urban waterlogging intelligent prediction model is constructed by fusing improved convolutional neural network, bidirectional long short-term memory neural network and Bayesian optimization, the rainfall time series of the “storm-urban waterlogging” database in step S7 is taken as the input of the prediction model, the inundation time series is taken as the output of the prediction model, and the training set and the validation set of the “storm-urban waterlogging” database after normalization in step S8 are used to train and optimize the parameters of the urban waterlogging intelligent prediction model;

[0015] Step S10, taking the rainfall time series in the test set after the "rainstorm-flooding" database normalization processing in step S8 as input, the urban flooding intelligent prediction model in step S9 is used to realize intelligent prediction of urban flooding under the pipe siltation scenario.

[0016] Further, step S1, the Manning equation is used to simulate the constant uniform flow in the pipeline to realize the numerical rating of the friction loss along the pipeline; specifically as follows:

[0017] The expression of the friction loss along the pipeline is obtained by using the Manning equation to simulate the constant uniform flow in the pipeline as follows:

[0018] (1);

[0019] Wherein, S f is the friction loss along the pipeline, n is the Manning roughness coefficient, R is the hydraulic radius of the water section, U is the average flow velocity of the water flow in the pipeline, Q(t) is the instantaneous flow at the inlet / outlet of the pipeline at time t, and A is the area of the pipeline water section;

[0020] Further, step S2, combined with the momentum conservation method, one-dimensional Saint-Venant equation and the friction loss along the pipeline in step S1, the underground pipe network water flow dynamics control equation considering pipe siltation is constructed; specifically as follows:

[0021] Step S21, combined with the one-dimensional Saint-Venant equation, the friction loss along the pipeline in step S1 and the pipe water dynamics empirical formula, the one-dimensional control equation of the drainage pipeline water flow is constructed as follows:

[0022] (2);

[0023] Wherein, Z represents the water level height of the drainage pipeline, t represents time, represents the wave speed, is the gravitational acceleration, A is the area of the pipeline water section, Q(t) is the instantaneous flow at the inlet / outlet of the pipeline at time t, and x represents the spatial coordinate, represents the side inflow, is the local head loss on the unit length;

[0024] Step S22, using the momentum conservation method and the one-dimensional control equation of the drainage pipeline water flow constructed in step S21, the underground pipe network water flow dynamics control equation considering pipe siltation is constructed as follows:

[0025] (3);

[0026] Wherein, is the sediment-laden flow density, Q(t) is the instantaneous flow at the inlet / outlet of the pipeline at time t, represents an infinitesimal time interval, and respectively represent the frictional resistance of fluid-structure coupling in non-deposition and deposition pipe sections, represents the total length of the drainage pipe from the drainage pipe inlet a to the drainage pipe outlet b, represents the length of the deposition of the drainage pipe, is the water flow energy conversion efficiency coefficient of the non-deposition pipe section, represents the thickness of the deposition of the drainage pipe, represents the time required for water flow to move from the drainage pipe inlet a to the drainage pipe outlet b, and respectively represent the actual flow of the drainage pipe inlet a and the drainage pipe outlet b measured by the flow meter at time t, is the angle between the pipe and the horizontal line.

[0027] Further, step S3, using the one-dimensional Saint-Venant equation in step S2 and the hydraulic analysis method based on grid coordinates, a two-dimensional surface water flow dynamics control equation is constructed; the specific steps are as follows:

[0028] Step S31, a two-dimensional coordinate system of X axis and Y axis is established, and the time step of water flow in the two-dimensional coordinate system is updated is set as:

[0029] (4);

[0030] wherein, represents the fuzzy penalty coefficient, is the width of the grid unit, represents the average water depth of the grid unit;

[0031] Step S32, using the time step of water flow in the two-dimensional coordinate system in step S31 , the calculation method of the grid unit flow in the two-dimensional coordinate system is as follows:

[0032] (5);

[0033] wherein, represents the grid unit flow in the two-dimensional coordinate system, q represents the instantaneous rainfall in the grid unit, represents the ground elevation, h represents the average water depth of the grid unit, is the average elevation of the water surface in the unit width grid unit;

[0034] Step S33, using the time step of water flow in the two-dimensional coordinate system in step S31 and the calculation method of the grid unit flow in the two-dimensional coordinate system in step S32 , the average water depth of the grid unit in the two-dimensional coordinate system is updated:

[0035] (6);

[0036] wherein, represents the change of the average water depth of the grid cell at the two-dimensional coordinate i and j position within the time step , is the area of the grid cell, represents the flow along the X axis at the two-dimensional coordinate i-1 and j position, represents the flow along the X axis at the two-dimensional coordinate i and j position, represents the flow along the Y axis at the two-dimensional coordinate i and j-1 position, represents the flow along the Y axis at the two-dimensional coordinate i and j position.

[0037] Further, step S4, using the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3, an underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation is constructed; specifically as follows:

[0038] Step S41, the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3 are embedded into the rainstorm flood management model, to obtain an improved rainstorm flood management model;

[0039] Step S42, a flow exchange equation between the underground one-dimensional pipe network and the two-dimensional surface under the pipe siltation scenario is established:

[0040] (7);

[0041] (8);

[0042] (9);

[0043] wherein, is the exchange flow, is the orifice flow coefficient, is the inspection well chamber area, is the pipe internal pressure water head, is the surface flow water head, is the weir flow coefficient, is the inspection well wellhead circumference, is the ground elevation;

[0044] Step S43, the improved storm flood management model in step S41 is coupled with the flow exchange equation between the one-dimensional underground pipe network and the two-dimensional ground surface under the pipe sedimentation scenario in step S42, the coupling time step is set as the calculation time step of the improved storm flood management model, and then the underground pipe network and the two-dimensional ground surface coupling waterlogging numerical simulation model considering pipe sedimentation is obtained.

[0045] Further, step S5, using the rainwater-ponding monitoring data in the application area, the underground pipe network and two-dimensional ground surface coupling waterlogging numerical simulation model considering pipe sedimentation constructed in step S4 is calibrated and verified; the specific steps are as follows:

[0046] Step S51, the degree of drainage pipe sedimentation is divided into five levels, and the related configuration parameters of the surface infiltration rate under each level are measured by a full-scale test field; the five level division methods of the degree of drainage pipe sedimentation are as follows:

[0047] (10);

[0048] (11);

[0049] wherein, represents the pipe sedimentation index, represents the pipe diameter, represents the total length of the drainage pipe from the drainage pipe inlet a to the drainage pipe outlet b, represents the length of the drainage pipe sedimentation, represents the pipe sedimentation thickness, and r is the level value of the degree of pipe sedimentation;

[0050] The empirical parameters of the steady infiltration rate and the initial infiltration rate are calibrated by using the empirical formula of the infiltration curve and the soil infiltration experiment, and the Horton infiltration model is constructed;

[0051] Step S52, on the basis of the pipe sedimentation level division in step S51 and the Horton infiltration model, the permeable underlying surface infiltration rate calculation model considering the influence of pipe sedimentation is constructed as follows:

[0052] (12);

[0053] wherein, represents the surface infiltration rate of the permeable underlying surface at time t, represents the steady infiltration rate of the permeable underlying surface, is the instantaneous surface infiltration coefficient when the level value of the degree of pipe sedimentation is r, represents the initial infiltration rate of the permeable underlying surface, is the base of natural logarithm, The surface infiltration attenuation coefficient of the permeable underlying surface for the pipeline siltation grade r, The equivalent infiltration rate of the permeable underlying surface.

[0054] Further, the rainstorm event data set constructed in step S6 includes: different rainfall types of rainstorm events include uniform type, single-peak type, double-peak type, multi-peak type, and Chicago rain type with peak coefficients of 0.3, 0.5, and 0.7, and the return periods of the rainstorm are 1 year, 2 years, 5 years, 10 years, 20 years, 30 years, and 50 years, respectively.

[0055] Further, the urban waterlogging intelligent prediction model constructed in step S9 is fused with the improved convolutional neural network, the bidirectional long short-term memory neural network, and the Bayesian optimization, and specifically comprises:

[0056] Step S91, the improved convolutional neural network extracts spatial features of the input data set, which contains 、 、 and four different convolution kernels to realize multi-feature extraction of the spatial scale of waterlogging response law;

[0057] Step S92, the flow exchange control equation considering pipeline siltation is introduced into the loss function of the bidirectional long short-term memory neural network, and a theoretically guided loss function error correction is obtained, and the structural formula is as follows:

[0058] (13);

[0059] (14);

[0060] wherein, is the mean square error of the flow exchange control equation, represents the total number of drainage pipelines in the study area, 、 and respectively represent the number of pipelines in three different scenarios in the flow exchange control equation, is the loss function after the theoretically guided error correction, represents the mean square error of the original output data of the bidirectional long short-term memory neural network, 、 respectively represent the weight coefficients of the mean square errors and the mean square errors of the original output data of the flow exchange control equation;

[0061] Step S93, the Bayesian optimization is to perform hyperparameter optimization, and the optimization target equation is as follows:

[0062] (15);

[0063] wherein, denotes the hyperparameters that minimize the loss function, denotes the current value of the hyperparameters, is the loss function to be optimized in the improved convolutional neural network and bidirectional long short-term memory neural network, is the definition domain of the hyperparameters.

[0064] Further, the optimization process of the Bayesian optimization in step S9 mainly involves the following three parts:

[0065] (1) Objective function: the loss function of the improved convolutional neural network and bidirectional long short-term memory neural network in the research on the optimized hyperparameters in the validation set;

[0066] (2) Domain space: the actual value range of the hyperparameters that need to be searched in the space;

[0067] (3) Optimization algorithm: the Bayesian optimization algorithm includes a probability agent model and a collection function; wherein, the probability agent model refers to the probability distribution of the loss function dominated by the hyperparameters; the collection function adopts an upper limit function of the confidence interval.

[0068] Compared with the prior art:

[0069] (1) The present application constructs the underground pipe network water flow dynamics control equation considering pipe sedimentation, numerically simulates the pipe flow under the sedimentation scenario;

[0070] (2) The present application proposes a numerical simulation model of underground pipe network and two-dimensional surface coupling waterlogging considering pipe sedimentation, improves the prediction accuracy of the waterlogging numerical model;

[0071] (3) The underground pipe network water flow dynamics control equation considering pipe sedimentation is introduced into the loss function of the loss function of the improved convolutional mechanism, bidirectional long short-term memory neural network and Bayesian optimization algorithm (Bayes-CNN-BLSTM), a deep learning algorithm model considering pipe sedimentation is constructed, and the physical knowledge and algorithm are cooperatively driven;

[0072] (4) The present application constructs an urban waterlogging intelligent prediction model integrating the improved convolutional neural network, bidirectional long short-term memory neural network and Bayesian optimization algorithm (Bayes-CNN-BLSTM), realizes intelligent prediction of waterlogging under pipe sedimentation, and solves the problems of timeliness and accuracy of the traditional urban waterlogging numerical simulation model based on hydrology and hydrodynamics. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 show the technical roadmap of the method of the present application;

[0074] Figure 2 A rainstorm event set constructed by the method of the application is shown;

[0075] Figure 3 A schematic diagram of a fusion improved convolutional neural network, a bidirectional long short-term memory neural network and a Bayes optimization algorithm (Bayes-CNN-BLSTM) in the method of the application is shown;

[0076] Figure 4 A flooding prediction result graph of a city flooding intelligent prediction model considering pipe siltation in the method of the application in a test rainfall event 1 is shown;

[0077] Figure 5 A flooding prediction result graph of a city flooding intelligent prediction model considering pipe siltation in the method of the application in a test rainfall event 2 is shown;

[0078] Figure 6 A flooding prediction result graph of a city flooding intelligent prediction model considering pipe siltation in the method of the application in a test rainfall event 3 is shown. DETAILED DESCRIPTION

[0079] The specific embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0080] In order to make the purpose, technical scheme and advantages of the embodiments of the application more clear, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings of the embodiments of the application.

[0081] In order to achieve the above-mentioned purpose, the following technical scheme is provided: a city flooding intelligent prediction method considering drainage pipe siltation, as shown in Figure 1 Specifically, the method comprises the following steps:

[0082] Step S1: adopting the Manning equation to simulate constant uniform flow in the pipe to realize numerical rating of the along-path friction;

[0083] Step S2: combining the along-path friction in step S1, the momentum moment conservation method and the one-dimensional Saint-Venant equation to construct a groundwater pipe network water flow dynamics control equation considering pipe siltation;

[0084] Step S3: adopting the one-dimensional Saint-Venant equation in step S2 and a hydraulic analysis method based on grid coordinates to construct a two-dimensional surface water flow dynamics control equation;

[0085] Step S4: utilizing the groundwater pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3 to construct a coupled city flooding numerical simulation model considering pipe siltation of the underground pipe network and the two-dimensional surface;

[0086] Step S5, using the stormwater monitoring data of the application area, the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation constructed in step S4 are calibrated and verified;

[0087] Step S6, constructing a storm event data set, based on the recorded rainfall data of the application area, constructing the storm intensity formula of the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation in step S5, designing different rainfall types and different return period storm events;

[0088] Step S7, inputting the different rainfall types and different return period storm event data set constructed in step S6 into the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation in step S5, simulating the waterlogging inundation dynamic evolution scenario of the application area under different storm events, and constructing a "storm-waterlogging" database containing rainfall time series and inundation time series under the pipe siltation scenario;

[0089] Step S8, normalizing the "storm-waterlogging" database constructed in step S7, and dividing the training set, the validation set and the test set;

[0090] Step S9, in order to realize the intelligent prediction of the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation in step S5, an improved convolutional neural network, a bidirectional long short-term memory neural network and a Bayesian optimization algorithm are constructed, the rainfall time series of the "storm-waterlogging" database in step S7 are taken as the input of the model, the inundation time series are taken as the output of the model, and the training set and the validation set of the "storm-waterlogging" database after normalization in step S8 are used to train and optimize the parameters of the urban waterlogging intelligent prediction model;

[0091] Step S10, taking the rainfall time series in the test set of the "storm-waterlogging" database after normalization in step S8 as the input, and using the urban waterlogging intelligent prediction model in step S9 to realize the intelligent prediction of urban waterlogging under the pipe siltation scenario.

[0092] Preferably, step S1, the Manning equation is used to simulate the constant uniform flow in the pipe, and the numerical calibration of the friction along the way is realized, which is as follows:

[0093] The expression of the friction along the way S f is as follows:

[0094] (1);

[0095] Wherein, S fis the frictional resistance of the pipe, n is the Manning roughness coefficient, R is the hydraulic radius of the cross section of the pipe, U is the average flow velocity of the water flow in the pipe, Q(t) is the instantaneous flow rate of the inlet / outlet of the pipe at time t, and A is the area of the cross section of the pipe.

[0096] Preferably, step S2, in combination with the frictional resistance along the path, the method of conservation of the moment of momentum, and the one-dimensional Saint-Venant equation in step S1, constructs the underground pipe network water flow dynamics control equation considering pipe siltation, and the specific construction is as follows:

[0097] Step S21, in combination with the one-dimensional Saint-Venant equation, the frictional resistance along the path in step S1, and the pipe water dynamics empirical formula, constructs the one-dimensional control equation of the water flow in the drainage pipe as follows:

[0098] (2);

[0099] wherein Z represents the water level height of the drainage pipe, t represents time, represents the wave speed, is the gravitational acceleration, A is the cross-sectional area of the pipe, Q(t) is the instantaneous flow rate of the inlet / outlet of the pipe at time t, and x represents the spatial coordinate, represents the side inflow, is the local head loss on the unit length;

[0100] Step S22, using the method of conservation of the moment of momentum and the one-dimensional control equation of the water flow in the drainage pipe in step S21, constructs the underground pipe network water flow dynamics control equation considering pipe siltation as follows:

[0101] (3);

[0102] wherein is the sediment-laden water flow density, Q(t) is the instantaneous flow rate of the inlet / outlet of the pipe at time t, represents an infinitesimal time interval, and respectively represent the frictional resistance of the fluid-structure coupling in the non-silted and silted pipe sections, represents the total length of the drainage pipe from the inlet a to the outlet b of the drainage pipe, represents the length of the silted pipe, is the water flow kinetic energy conversion efficiency coefficient of the non-silted pipe section, represents the thickness of the silted pipe, represents the time required for the water flow to move from the inlet a to the outlet b of the drainage pipe, and respectively represent the actual flow rates of the inlet a and the outlet b of the drainage pipe measured by the flow meter at time t, is the angle between the pipe and the horizontal line.

[0103] Preferably, step S3, using the one-dimensional Saint-Venant equation in step S2 and the hydraulic analysis method based on grid coordinates, the two-dimensional surface water flow dynamics control equation is constructed, and the hydraulic analysis method based on grid coordinates is as follows:

[0104] Step S31, a two-dimensional coordinate system of X axis and Y axis is established, and the time step of water flow in the two-dimensional coordinate system is updated Set to:

[0105] (4);

[0106] Wherein, The fuzzy penalty coefficient is represented, The grid cell width is, The average water depth of the grid cell is represented;

[0107] Step S32, using the time step of water flow in the two-dimensional coordinate system in step S31 , the calculation method of the grid cell flow Q in the two-dimensional coordinate system is as follows:

[0108] (5);

[0109] Wherein, The grid cell flow in the two-dimensional coordinate system is represented, q represents the instantaneous rainfall in the grid cell, The ground elevation is represented, h represents the average water depth of the grid cell, The average elevation of the water surface in the unit width grid cell is;

[0110] Step S33, using the time step of water flow in the two-dimensional coordinate system in step S31 And the calculation method of the grid cell flow Q in step S32, the average water depth of the grid cell in the two-dimensional coordinate system is updated:

[0111] (6);

[0112] Wherein, The change of the average water depth of the grid cell at the two-dimensional coordinates i and j position within the time step Is represented, The grid cell area is, The flow along the X axis of the two-dimensional coordinates i-1 and j position is represented, The flow along the X axis of the two-dimensional coordinates i and j position is represented, The flow along the Y axis of the two-dimensional coordinates i and j-1 position is represented, The flow along the Y axis of the two-dimensional coordinates i and j position is represented.

[0113] Preferably, step S4, using the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3, an underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation is constructed, specifically as follows:

[0114] Step S41, the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3 are embedded into the rainstorm flood management model to obtain an improved rainstorm flood management model.

[0115] Step S42, a flow exchange equation between the underground one-dimensional pipe network and the two-dimensional surface under the pipe siltation scenario is established:

[0116] (7);

[0117] (8);

[0118] (9);

[0119] Wherein, is the exchange flow, is the weir flow coefficient, is the orifice flow coefficient, is the circumference of the inspection well, is the inspection well chamber area, is the surface flow water head, is the pipe internal pressure water head, is the ground elevation;

[0120] Step S43, the improved rainstorm flood management model in step S41 is coupled with the flow exchange equation between the underground one-dimensional pipe network and the two-dimensional surface under the pipe siltation scenario in step S42, and the coupling time step is set to the calculation time step of the improved rainstorm flood management model, thereby obtaining an underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation.

[0121] Preferably, step S5, using the rainstorm-ponding monitoring data of the application area, the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation constructed in step S4 is calibrated and verified, specifically as follows:

[0122] Step S51, the degree of siltation of the drainage pipe is divided into five levels, and the surface infiltration rate related configuration parameters under each level are determined by full-scale test field. The five level division method of the degree of siltation of the drainage pipe is shown as follows:

[0123] (10);

[0124] (11);

[0125] wherein, represents the pipe siltation index, represents the pipe diameter, represents the total length of the drainage pipe from the drainage pipe inlet a to the drainage pipe outlet b, represents the length of the drainage pipe siltation, represents the pipe siltation thickness, r is the grade value of the pipe siltation degree;

[0126] The empirical formula of the infiltration curve and the empirical parameters of the soil infiltration rate calibrated by the soil infiltration experiment are used to further construct the Horton infiltration model.

[0127] Step S52, on the basis of the pipe siltation grade division and the Horton infiltration model in step S51, a calculation model of the infiltration rate of the permeable underlying surface considering the influence of the pipe siltation is constructed as follows:

[0128] (12);

[0129] wherein, represents the surface infiltration rate of the permeable underlying surface at time t, represents the stable surface infiltration rate of the permeable underlying surface, is the instantaneous surface infiltration coefficient when the grade value of the pipe siltation degree is r, represents the initial surface infiltration rate of the permeable underlying surface, is the base number of the natural logarithm, is the surface infiltration attenuation coefficient of the permeable underlying surface when the pipe siltation grade is r, represents the equivalent surface infiltration rate of the permeable underlying surface.

[0130] Preferably, step S6, a rainstorm event data set is constructed, a rainstorm intensity formula of the step S5 considering the pipe siltation of the underground pipe network and the two-dimensional surface coupled waterlogging numerical simulation model is constructed based on the recorded rainfall data of the application area, different rainfall types and different return periods of rainstorm events are designed, and the rainstorm event data set includes:

[0131] The different rainfall types of the rainstorm event include uniform type, single-peak type, double-peak type, multi-peak type, and Chicago rain type with a peak coefficient of 0.3, 0.5 and 0.7, and the return periods of the rainstorm are 1 year, 2 years, 5 years, 10 years, 20 years, 30 years and 50 years.

[0132] In particular, the urban waterlogging intelligent prediction model in step S9 is specifically composed of three parts of an improved convolutional neural network (CNN), a bidirectional long short-term memory neural network (BLSTM) and a Bayesian optimization (Bayes) submodule.

[0133] The improved convolutional neural network (CNN) is responsible for extracting the spatial features of the input data set, and the sub-module contains 、 、 and Four different convolution kernels realize the multi-feature extraction of the spatial scale of the waterlogging response law.

[0134] The loss function of the bidirectional long short-term memory neural network (BLSTM) introduces the flow exchange control equation considering pipe siltation, and obtains the theoretically guided loss function error correction, and its structural formula is as follows:

[0135] (13);

[0136] (14);

[0137] Among them, is the mean square error of the flow exchange control equation, represents the total number of drainage pipes in the study area, 、 and respectively represent the number of pipes in three different scenarios in the flow exchange control equation, is the loss function after the theoretical guided error correction, represents the mean square error of the BLSTM original output data, 、 respectively represent the weight coefficients of and .

[0138] The purpose of the Bayesian optimization (Bayes) is to optimize the hyperparameters of CNN and BLSTM, and the optimization objective equation is as follows:

[0139] (15);

[0140] Among them, represents the hyperparameters that make the loss function take the minimum value, represents the current value of the hyperparameters, is the loss function to be optimized in CNN and BLSTM, is the definition domain of the hyperparameters.

[0141] Preferably, the optimization process of the Bayesian optimization in step S9 mainly involves the following three parts:

[0142] (1) Objective function: The loss function of the CNN and BLSTM to be optimized in this study on the validation set.

[0143] (2) Domain space: the actual value range of the hyperparameters that need to be searched in the space.

[0144] (3) Optimization algorithm: the Bayesian optimization algorithm includes a probability surrogate model and an acquisition function. The probability surrogate model refers to a probability distribution of a loss function dominated by hyperparameters; and the acquisition function adopts an upper limit function of a confidence interval.

[0145] Embodiment: Taking the waterlogging test field of Huizhou University as an example, a kind of intelligent prediction method for urban waterlogging considering pipe siltation is further illustrated:

[0146] Step one, the Manning equation is used to simulate the constant uniform flow in the pipe to realize the numerical calibration of the friction loss along the way;

[0147] Step two, the one-dimensional Saint-Venant equation considering pipe siltation is constructed by combining the friction loss along the way in step one, the momentum conservation method and the one-dimensional Saint-Venant equation;

[0148] Step three, the two-dimensional surface water flow dynamics control equation is constructed by using the one-dimensional Saint-Venant equation in step two and the hydraulic analysis method based on grid coordinates;

[0149] Step four, the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation is constructed by using the underground pipe network water flow dynamics control equation in step two and the two-dimensional surface water flow dynamics control equation in step three, combined with the basic data of the waterlogging test field of Huizhou University;

[0150] Step five, the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation constructed in step four is calibrated and verified by using the stormwater monitoring data of the waterlogging test field of Huizhou University;

[0151] Step six, the storm event data set is constructed, the storm intensity formula of the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation in step five is constructed based on the recorded rainfall data of the application area, and different rainfall types and different return period storm events are designed, and the specific storm events are shown in Figure 2 ;

[0152] Step seven, the storm event data set of different rainfall types and different return period storm events constructed in step six is input into the underground pipe network and two-dimensional surface coupled waterlogging numerical simulation model considering pipe siltation in step five, the waterlogging inundation dynamic evolution scenario of the application area under different storm events is simulated, and the "storm-waterlogging" database containing the rainfall time series and the inundation time series under the pipe siltation scenario is constructed;

[0153] Step eight, the "storm-waterlogging" database constructed in step seven is normalized and divided into training set, validation set and test set;

[0154] Step nine, in order to realize the intelligent prediction of the underground pipe network and two-dimensional ground coupling flood numerical simulation model considering pipe siltation in step five, an urban flood intelligent prediction model is constructed by combining the improved convolutional neural network, bidirectional long short-term memory neural network and Bayesian optimization algorithm. The rainfall time series in the "rainstorm-flood" database in step seven is used as the input of the model, and the flood time series is used as the output of the model. The training set and the validation set of the "rainstorm-flood" database after normalization processing in step eight are used to train and optimize the parameters of the urban flood intelligent prediction model. The schematic diagram of the urban flood intelligent prediction model is shown in Figure 3 .

[0155] Step ten, the rainfall time series in the test set of the "rainstorm-flood" database after normalization processing in step eight is used as the input, and the urban flood intelligent prediction model constructed in step nine is used to realize the intelligent prediction of urban flood under the condition of pipe siltation. Three test rainfall events R1, R2 and R3 are selected as the verification of the model. The specific prediction results are shown in Figure 4 、 Figure 5 、 Figure 6 .

[0156] Table 1 Precision evaluation results of the model in the example under three test rainfall events

[0157]

[0158] Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

Claims

1. An intelligent prediction method of urban waterlogging considering sewer clogging, characterized in that: The method comprises the following steps: Step S1, using Manning equation to simulate the constant uniform flow in the pipeline, to realize the numerical calibration of the frictional resistance along the pipeline; Step S2, combining the momentum conservation method, one-dimensional Saint-Venant equation and the frictional resistance along the pipeline in step S1, to build the underground pipe network water flow dynamics control equation considering pipeline siltation; Step S3, using the one-dimensional Saint-Venant equation in step S2 and the hydraulic analysis method based on grid coordinates, to build the two-dimensional surface water flow dynamics control equation; Step S4, using the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3, to build the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipeline siltation; Step S5, using the stormwater monitoring data of the application area, to calibrate and verify the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipeline siltation built in step S4; Step S6, building the storm event data set, building the storm intensity formula of the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipeline siltation in step S5 based on the recorded rainfall data of the application area, and building the storm event data set of different rainfall types and different return periods; Step S7, inputting the storm event data set of different rainfall types and different return periods built in step S6 into the coupled urban waterlogging numerical simulation model of underground pipe network and two-dimensional surface considering pipeline siltation in step S4, simulating the urban waterlogging dynamic evolution scenario of the application area under different storm events, and building the "storm-waterlogging" database containing the rainfall time sequence and the waterlogging time sequence under the pipeline siltation scenario; Step S8, normalizing the "storm-waterlogging" database built in step S7, and dividing the training set, the verification set and the test set; Step S9, building the urban waterlogging intelligent prediction model fusing the improved convolutional neural network, the bidirectional long short-term memory neural network and the Bayesian optimization, taking the rainfall time sequence of the "storm-waterlogging" database in step S7 as the input of the prediction model and the waterlogging time sequence as the output of the prediction model, and training and optimizing the parameters of the urban waterlogging intelligent prediction model by using the training set and the verification set of the "storm-waterlogging" database after normalization in step S8; Step S10, taking the rainfall time sequence in the test set of the "storm-waterlogging" database after normalization in step S8 as the input, and realizing the urban waterlogging intelligent prediction under the pipeline siltation scenario by using the urban waterlogging intelligent prediction model in step S9; Step S1, using Manning equation to simulate the constant uniform flow in the pipeline, to realize the numerical calibration of the frictional resistance along the pipeline; specifically as follows: Using Manning equation to simulate the constant uniform flow in the pipeline, the expression of the frictional resistance along the pipeline is as follows: （1）； where S f is the frictional resistance along the path, n is the Manning roughness coefficient, R is the hydraulic radius of the water section, U is the average flow rate of the water flow inside the pipe, Q(t) is the instantaneous flow rate at the inlet / outlet of the pipe at time t, and A is the area of the water section of the pipe. Step S2, combining the momentum conservation method, one-dimensional Saint-Venant equation and the frictional resistance along the pipeline in step S1, to build the underground pipe network water flow dynamics control equation considering pipeline siltation; specifically as follows: Step S21, combining one-dimensional Saint-Venant equation, the frictional resistance along the pipeline in step S1 and the pipeline water dynamics empirical formula, to build the one-dimensional control equation of the drainage pipeline water flow as follows: （2）； where Z represents the water level height of the sewer, t represents time, represents the wave speed, is the gravitational acceleration, A is the cross-sectional area of the pipe, Q(t) is the instantaneous flow rate at the inlet / outlet of the pipe at time t, and x represents the spatial coordinate, represents the lateral inflow, is the local head loss over a unit length; Step S22, using the method of momentum conservation and the one-dimensional control equation of the sewer flow constructed in step S21, the underground pipe network water flow dynamics control equation considering pipe siltation is constructed as follows: （3）； wherein, is the density of the sediment-laden flow, Q(t) is the instantaneous flow rate at time t at the inlet / outlet of the pipe, denotes an infinitesimal time interval, and denote the frictional resistance due to fluid-structure interaction in the non-deposited and deposited pipe sections, respectively, denotes the total length of the sewer pipe from the inlet a to the outlet b of the sewer pipe, denotes the length of the deposited section of the sewer pipe, is the water flow kinetic energy conversion efficiency coefficient in the non-deposited pipe section, denotes the thickness of the deposited section of the sewer pipe, denotes the time required for the water flow to move from the inlet a to the outlet b of the sewer pipe, and denote the actual flow rates at the inlet a and the outlet b of the sewer pipe at time t, respectively, measured by the flow meters, is the angle of the pipe with the horizontal line; Step S3, using the one-dimensional Saint-Venant equation in step S2 and the hydraulic analysis method based on grid coordinates, the two-dimensional surface water flow dynamics control equation is constructed; the specific is as follows: Step S31, establish X-axis and Y-axis two-dimensional coordinate system, update the time step of water flow in the two-dimensional coordinate system Set to: （4）； wherein, represents a fuzzy penalty coefficient, is the width of the grid cell, represents the average water depth of the grid cell; Step S32, using the time step of the water flow in the two-dimensional coordinate system in step S31 The calculation method of the grid cell flow in the two-dimensional coordinate system is as follows: （5）； wherein, represents the grid cell flow in a two-dimensional coordinate system, q represents the instantaneous rainfall in the grid cell, represents the ground elevation, h represents the average water depth in the grid cell, is the average elevation of the water surface in the unit width grid cell; Step S33, update the average water depth of the grid cell in the two-dimensional coordinate system by using the calculation method of the time step of the water flow in the two-dimensional coordinate system in step S31 and the grid cell flow in the two-dimensional coordinate system in step S32 ​ （6）； wherein, represents the change in the average water depth of the grid cell at the two-dimensional coordinate i and j location within the time step represents the change in the average water depth of the grid cell at the two-dimensional coordinate i and j location within the time step is the area of the grid cell, represents the flow along the X-axis at the two-dimensional coordinate i-1 and j location, represents the flow along the X-axis at the two-dimensional coordinate i and j location, represents the flow along the Y-axis at the two-dimensional coordinate i and j-1 location, represents the flow along the Y-axis at the two-dimensional coordinate i and j location.

2. The intelligent prediction method of urban waterlogging considering sewer clogging according to claim 1, wherein, Step S4, using the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3, the underground pipe network and two-dimensional surface coupling waterlogging numerical simulation model considering pipe siltation is constructed; the specific is as follows: Step S41, the underground pipe network water flow dynamics control equation in step S2 and the two-dimensional surface water flow dynamics control equation in step S3 are embedded into the stormwater management model to obtain an improved stormwater management model; Step S42, the flow exchange equation between the underground one-dimensional pipe network and the two-dimensional surface under the pipe siltation scenario is established: （7）； （8）； （9）； wherein, is the exchange flow, is the orifice flow coefficient, is the inspection well chamber area, is the pressure head in the pipe, is the surface overland flow head, is the weir flow coefficient, is the inspection well wellhead perimeter, is the ground elevation; Step S43, the improved stormwater management model in step S41 and the flow exchange equation between the underground one-dimensional pipe network and the two-dimensional surface under the pipe siltation scenario in step S42 are coupled, and the coupling time step is set as the calculation time step of the improved stormwater management model, and then the underground pipe network and two-dimensional surface coupling waterlogging numerical simulation model considering pipe siltation is obtained.

3. The intelligent prediction method of urban waterlogging considering sewer clogging according to claim 1, wherein, Step S5, the model calibration and verification of the underground pipe network and two-dimensional surface coupling waterlogging numerical simulation model considering pipe siltation constructed in step S4 are carried out by using the stormwater monitoring data of the application area; the specific is as follows: Step S51, the degree of sewer siltation is divided into five levels, and the related configuration parameters of the surface infiltration rate under each level are determined by using the full-scale test field; the five level division method of the degree of sewer siltation is shown as follows: （10）； （11）； wherein, represents the pipe clogging index, represents the pipe diameter, represents the total length of the drainage pipe from the drainage pipe inlet a to the drainage pipe outlet b, represents the length of the drainage pipe clogging, represents the pipe clogging thickness, r is the grade value of the pipe clogging degree; The empirical parameters of stable infiltration rate and initial infiltration rate are calibrated by using the empirical formula of infiltration curve and soil infiltration experiment, and the Horton infiltration model is constructed; Step S52, on the basis of the pipe siltation level division in step S51 and the Horton infiltration model, the permeable underlying surface infiltration rate calculation model considering the influence of pipe siltation is constructed as follows: （12）； wherein, K(t) represents the subsurface infiltration rate of the pervious underlying surface at time t, Ks represents the steady subsurface infiltration rate of the pervious underlying surface, Ks(r) represents the instantaneous subsurface infiltration coefficient of the pervious underlying surface for a pipe clogging degree level r, K0 represents the initial subsurface infiltration rate of the pervious underlying surface, e represents the base of the natural logarithm, Kd(r) represents the subsurface infiltration decay coefficient of the pervious underlying surface for a pipe clogging degree level r, Ks,eff represents the equivalent subsurface infiltration rate of the pervious underlying surface.

4. The intelligent prediction method of urban waterlogging considering sewer clogging according to claim 1, wherein, The storm event data set constructed in step S6 includes: different rainfall types of storm events including uniform type, single-peak type, double-peak type, multi-peak type and Chicago rain type with peak coefficient of 0.3, 0.5 and 0.7, and the return period of storm is 1 year, 2 years, 5 years, 10 years, 20 years, 30 years and 50 years.

5. The intelligent prediction method of urban waterlogging considering sewer clogging according to claim 1, wherein, The urban waterlogging intelligent prediction model fused with the improved convolutional neural network, bidirectional long short-term memory neural network and Bayesian optimization is constructed in step S9, which is specifically: Step S91, the improved convolutional neural network extracts the spatial features of the input data set, containing 、 、 And Four different convolution kernels, realize the multi-feature extraction of the spatial scale of waterlogging response law; Step S92, the flow exchange control equation considering pipe siltation is introduced into the loss function of bidirectional long short-term memory neural network, and the theoretically guided loss function error correction is obtained, and its structure formula is as follows: （13）； （14）； wherein, is the mean square error of the flow exchange control equation, represents the total number of sewer pipes within the study area, , and represent the number of pipes in three different scenarios of the flow exchange control equation, respectively, is the loss function after the theoretical guidance error correction, represents the mean square error of the original output data of the bidirectional long short-term memory neural network, , represent the weight coefficients of the mean square error of the flow exchange control equation and the mean square error of the original output data , respectively; Step S93, the Bayesian optimization is a super parameter optimization, and its optimization target equation is as follows: （15）； wherein, denotes the hyperparameters that minimize the loss function, denotes the current value of the hyperparameters, is the loss function to be optimized in the improved convolutional neural network and bidirectional long short-term memory neural network, is the definition domain of the hyperparameters.

6. The intelligent prediction method of urban waterlogging considering sewer clogging according to claim 5, wherein, The optimization process of Bayesian optimization in step S9 mainly involves the following three parts: (1) Objective function: the loss function of the improved convolutional neural network and bidirectional long short-term memory neural network in the research on the optimized hyperparameters in the validation set; (2) Domain space: the actual range of values that can be taken by the hyperparameters that need to be searched in space; (3) Optimization algorithm: Bayesian optimization algorithm includes probability agent model and acquisition function; among them, the probability agent model refers to the probability distribution of the loss function dominated by hyperparameters; the upper limit function of the confidence interval is used as the acquisition function.

Images