Intelligent waterlogging prediction method based on time-varying slider neural network guided by probability adjustment parameters of drainage pipe siltation

Abstract

The present invention discloses a method for intelligent prediction of urban waterlogging using a time-varying slider neural network guided by a probability adjustment parameter for the sedimentation degree of a drainage pipe. The method comprises the following steps: establishing a theoretical probability structural formula for the transport of sedimentation particles in a drainage pipe; calibrating parameters in the theoretical probability structural formula for the transport of sedimentation particles in a drainage pipe; predicting the drainage pipe sedimentation probability adjustment parameter using a genetic optimization algorithm for sedimentation particles with jump feedback; introducing the algorithm into a rain and flood numerical model and combining it with an improved sample set expansion method to obtain an expanded training sample set; embedding a combined preprocessing module and a time-varying slider optimization module in each data processing unit of a bidirectional long-short-term memory neural network, and then obtaining an improved time-varying slider neural network by adding an error compensation factor; using the expanded training sample set as the input of the time-varying slider neural network to obtain an intelligent prediction of urban waterlogging using the probability adjustment parameter for the sedimentation degree of a drainage pipe.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent disaster prediction, and in particular to an intelligent waterlogging prediction method using a drainage pipe siltation probability adjustment parameter to guide a time-varying slider neural network. Background Art

[0002] Urban drainage systems are engineering facilities that treat and remove urban wastewater, including rainwater and sewage. They play an indispensable role in improving the quality of life of urban residents and preventing urban waterlogging. Sewage pipe networks are a crucial component of urban drainage systems and a vital part of urban infrastructure, responsible for collecting and transporting sewage. Solid particles carried in sewage gradually settle and accumulate as the hydraulic conditions in drainage pipes change, forming silt at the bottom of the pipes. As the thickness of the silt increases, the pipe's water-passing capacity decreases, and it can even become clogged, impacting the safe and stable operation of the drainage system.

[0003] Poor drainage caused by pipe siltation and scaling is a key factor in urban waterlogging. Existing urban waterlogging early warning systems fail to consider the impact of pipe siltation and lack intelligent, accurate urban waterlogging prediction methods.

[0004] Therefore, it is urgent to develop an intelligent waterlogging prediction method based on a time-varying slider neural network guided by probability adjustment parameters of drainage pipe siltation. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the present invention provides an intelligent prediction method for urban waterlogging using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation, establishes a theoretical probability structure for the transport of siltation particles in drainage pipes, and uses a genetic optimization algorithm for drainage pipe siltation particles with jump feedback to predict the probability adjustment parameter for drainage pipe siltation. A combined preprocessing module and a time-varying slider optimization module are embedded in each data processing unit of the bidirectional long-short memory neural network, and a two-dimensional convolutional neural network based on error compensation is added at the starting position of the forward layer and the ending position of the backward layer, thereby obtaining an intelligent prediction method for urban waterlogging using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation, thereby realizing intelligent and precise prediction of urban waterlogging taking into account the siltation of drainage pipes, and solving the problems mentioned in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligently predicting urban waterlogging using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation, specifically comprising the following steps:

[0007] Step S1: establishing a theoretical probability structure formula for the transport of sediment particles in drainage pipes based on Newton's first law and the natural distribution law of the transport of sediment particles in drainage pipes;

[0008] Step S2: combining fluid-structure coupling numerical simulation and full-scale test method to calibrate the parameters in the theoretical probability structure of the transport of sedimentation particles in the drainage pipe in step S1;

[0009] Step S3: Substituting the parameters calibrated in the theoretical probability structural formula of the transport of sedimentation particles in the drainage pipe in step S2 into step S1, constructing a genetic optimization algorithm for sedimentation particles in the drainage pipe based on jump feedback, using the theoretical probability structural formula of the transport of sedimentation particles in the drainage pipe in step S1 as the fitness of the genetic optimization algorithm for sedimentation particles in the drainage pipe, and predicting the probability adjustment parameter of the sedimentation degree of the drainage pipe;

[0010] Step S4: Introducing the drainage pipe siltation probability adjustment parameter obtained in step S3 into the rainwater numerical model, and combining it with the improved sample set expansion method, obtaining a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels;

[0011] Step S5: embedding a combined preprocessing module and a time-varying slider optimization module in each data processing unit of the bidirectional long short-term memory neural network, and simultaneously adding a two-dimensional convolutional neural network based on error compensation at the starting position of the forward layer and the ending position of the backward layer to obtain an improved time-varying slider neural network;

[0012] Step S6: Use the drainage pipe siltation probability adjustment parameter training sample set in step S4 as the input of the improved time-varying slider neural network, and use its output as the prediction result, so as to realize the intelligent prediction of urban waterlogging by guiding the time-varying slider neural network with the drainage pipe siltation probability adjustment parameter.

[0013] Preferably, in step S1, based on Newton's first law and the natural distribution law of the transport of sediment particles in the drainage pipe, a theoretical probability structure formula of the transport of sediment particles in the drainage pipe is established, and the specific steps are as follows:

[0014] Step S11: Based on Newton's first law and the principle of steady-flow fluid mechanics, the critical force equilibrium state equation of the sedimentation particles in the drainage pipe is established as follows:

[0015]

[0016] Among them, F L is the buoyancy of the water flow on the sediment particles, F N is the supporting force of the pipeline slope on the sediment particles, F f is the frictional resistance of the pipe bottom acting on the sediment particles, G is the gravity of the sediment particles, α is the slope of the pipe, τ cs is the critical shear stress on the sediment particles;

[0017] Step S12: Since the transport of sediment particles in drainage pipes obeys the natural distribution law, the probability parameter γ of the theoretical probability structure of the transport of sediment particles in drainage pipes is introduced, and the parameter obeys N(μ,δ 2 ), and then the theoretical probability structure of sedimentation particle transport in drainage pipes is constructed as follows:

[0018]

[0019] Where P represents the theoretical probability of sedimentation particle transport in the drainage pipe, N(·) represents the normal distribution, μ is the mean of the normal distribution random variable, δ represents the standard deviation of the normal distribution random variable, τ0 is the instantaneous shear stress at the bottom of the pipe on the sedimentation particles, and τ0 is calculated as follows:

[0020]

[0021] Where C is the Scheherazade coefficient, R is the hydraulic radius of the pipe, ρ fp is the average density of sediment-laden water flow, B is the instantaneous width of the water surface in the pipe, J is the hydraulic gradient of the river channel and J = sinα, u a is the average flow velocity of water depth in the direction perpendicular to the water surface.

[0022] Preferably, in step S2, the parameters in the theoretical probability structure formula of the transport of sedimentation particles in the drainage pipe include the Scheherazade coefficient, the mean of the normal distribution random variable, and the standard deviation parameter of the normal distribution random variable.

[0023] Preferably, in step S3, the parameters calibrated in the theoretical probability structure of the drainage pipe sedimentation particle transport in step S2 are substituted into step S1, and a drainage pipe sedimentation particle genetic optimization algorithm based on jump feedback is constructed. The theoretical probability structure of the drainage pipe sedimentation particle transport in step S1 is used as the fitness of the drainage pipe sedimentation particle genetic optimization algorithm, and the drainage pipe sedimentation probability adjustment parameter is predicted. The specific steps are as follows:

[0024] Step S31: Using the control variable method and full-scale / reduced-scale tests, calibrate the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameters of the normal distribution random variable in the theoretical probability structure formula of the transport of sediment particles in the drainage pipe in step S2, and substitute the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameters of the normal distribution random variable into the theoretical probability structure formula of the transport of sediment particles in the drainage pipe in step S1;

[0025] Step S32: using the theoretical probability structure of the drainage pipe sedimentation particle transport in step S1 as the fitness of the drainage pipe sedimentation particle genetic optimization algorithm, predicting the location of the drainage pipe sedimentation particle transport, and recording the predicted results of the drainage pipe sedimentation particle transport location;

[0026] Step S33: Utilize the control variable method and full-scale / reduced-scale test in step S31 to count the actual transport positions of 100 pre-marked drainage pipe sediment particles, and compare them with the predicted transport positions of drainage pipe sediment particles in step S32, calculate the error distance between the predicted transport position and the actual transport position, and use the crossover probability P of the genetic optimization algorithm to calculate the error distance between the predicted transport position and the actual transport position. c and mutation probability P m The jump feedback factor k is introduced. When the error distance is greater than 0.02m, the value of the jump feedback factor k increases by 1. When the error distance is greater than 0.02m, the value of the jump feedback factor k no longer increases. The crossover probability P of the genetic optimization algorithm for drainage pipe sedimentation particles based on jump feedback is c and mutation probability P m As shown below:

[0027]

[0028] Among them, P c3 、P c2 、P c1 、P m3 、P m2 、P m1 are constants and P c3 >P c2 >P c1 ∈(0,1),P m3 >P m2 >P m1 ∈(0,1),f min 、f avg 、f max represent the minimum, maximum and average values ​​of individual fitness of sediment particles, respectively, and f ca and f ma Represent the larger fitness value of the individuals in the population participating in crossover and mutation, A and B represent the adaptive constants of probability adjustment, k is the jump feedback factor, e s The error distance between the predicted position of sedimentation particles in the drainage pipe and the actual position of the particles;

[0029] Step S34: Use the genetic optimization algorithm for drainage pipe sediment particles based on jump feedback in step S33 to predict the transfer position of drainage pipe sediment particles. After the transfer position is determined, the average of the 10 sampling points of the sediment height of the drainage pipe centerline is used as the equivalent sediment height h. s , and obtain the drainage pipe sedimentation probability adjustment parameter D i The calculation method is as follows:

[0030]

[0031] Where i represents the number of the drainage pipe section, R represents the inner diameter of the drainage pipe, j represents the sampling point number of the siltation height of the center line of the drainage pipe section, h j is the siltation height of the jth sampling point on the center line of the drainage pipe section.

[0032] Preferably, in step S4, the drainage pipe siltation probability adjustment parameter of step S3 is introduced into the rainwater numerical model, and combined with the improved sample set expansion method, a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels are obtained. The specific steps are as follows:

[0033] Step S41: constructing a rainwater flood numerical model and configuring the attribute information based on the underlying surface and runoff generation and convergence attribute information of the study area;

[0034] Step S42: using the drainage pipe siltation probability adjustment parameter of step S3, assigning the flow attribute value of each drainage pipe section in the rainwater flood numerical model of step S41, thereby obtaining a rainwater flood numerical model that takes into account the drainage pipe siltation probability adjustment parameter;

[0035] Step S43: Based on the rainfall amount and rainfall duration, the numerical model for rainwater flooding that takes into account the drainage pipe siltation probability adjustment parameter in step S42 is used to output the waterlogging result, thereby obtaining a sample set of drainage pipe siltation probability adjustment parameters with a waterlogging result label at the current moment.

[0036] Step S44: Using the drainage pipe siltation probability adjustment parameter calculation method, recalculate the drainage pipe siltation probability adjustment parameter at time period T, and input the recalculated drainage pipe siltation probability adjustment parameter into the rainwater numerical model that considers the drainage pipe siltation probability adjustment parameter in step S42, to obtain a drainage pipe siltation probability adjustment parameter sample set with a waterlogging result label at the next moment, and add this sample set to step S43 to form the original training sample set;

[0037] Step S45: Since the original training sample set in step S44 is related to the spatiotemporal characteristics, a hybrid cross neural network consisting of a two-dimensional convolutional neural network and a three-dimensional convolutional neural network is added in front of the generator of the generative adversarial network. The two-dimensional convolutional neural network contains spatial feature extraction convolution kernels of five different scales, namely 1×1, 1×3, 1×9, 3×3, and 9×9, thereby extracting the spatial features of the waterlogging result label. The three-dimensional convolutional neural network only contains two different scales, namely 2×3×3 and 2×9×9, thereby extracting the features of the spatial structure composed of the waterlogging result, rainfall duration, and rainfall intensity. At the same time, the output result of the hybrid cross neural network is XORed with the noise and the attribute parameters of the generative adversarial network. The operation result is input into the generator to obtain an improved generative adversarial network. The improved generative adversarial network is used to expand the original training sample set in step S44 to obtain a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels, i.e., the expanded training sample set.

[0038] Preferably, in step S5, a combined preprocessing module and a time-varying slider optimization module are embedded in each data processing unit of the bidirectional long short-term memory neural network, and a two-dimensional convolutional neural network based on error compensation is added at the starting position of the forward layer and the end position of the backward layer to obtain an improved time-varying slider neural network. The specific steps are as follows:

[0039] Step S51: Introduce a 34-layer deep residual network and a one-dimensional capsule network to form a combined preprocessing module for the expanded training sample set. The combined preprocessing module is placed at the forefront of the data processing unit in the bidirectional long short-term memory neural network to achieve data preprocessing and dimensionality reduction of the training sample set.

[0040] Step S52: Using the sliding time window to construct a time-varying slider optimization module, which is placed in front of the combined preprocessing module in step S51. The time-varying slider optimization module dynamically adjusts the length of the sliding time window based on the error percentage between the prediction result output by the bidirectional long short-term memory neural network and the actual result. The initial value of the length of the sliding time window is 10. When the error percentage between the prediction result and the actual result is greater than 10%, the length of the sliding time window is reduced by 1, and then the bidirectional long short-term memory neural network is used again to output the prediction result until the error percentage between the prediction result and the actual result is less than 10%.

[0041] Step S53: using the combined pre-processing module in step S51, the time-varying slider optimization module in step S52, and the bidirectional long short-term memory neural network to establish a time-varying slider neural network;

[0042] Step S54: In step S53, a two-dimensional convolutional neural network group is added to the starting position of the forward layer and the ending position of the backward layer of the time-varying slider neural network. The two-dimensional convolutional neural network group is constructed in parallel by three two-dimensional convolution kernels of 1×1, 3×3, and 9×9. The error percentage between the prediction result and the actual result in step S52 is compared, and an error compensation factor is added to the loss function of the two-dimensional convolutional neural network group to dynamically optimize the waterlogging prediction results of the two-dimensional convolutional neural network group and the time-varying slider neural network. The two-dimensional convolutional neural network group, the error compensation factor, and the slider neural network together constitute an improved time-varying slider neural network.

[0043] Preferably, in step S6, the expanded training sample set is used as the input of the improved time-varying slider neural network in step S5, and the support vector machine is used to classify the waterlogging results, and its output is used as the intelligent prediction result of waterlogging in the study area. The output indicators of the intelligent prediction result of waterlogging in the study area include flooding depth and flooding area, thereby realizing the intelligent prediction of waterlogging by guiding the time-varying slider neural network with the probability adjustment parameters of drainage pipe siltation.

[0044] The beneficial effects of the present invention are:

[0045] 1) Based on Newton's first law and the natural distribution pattern of sedimentation particle transport in drainage pipes, this paper establishes a theoretical probability structural formula for the transport of sedimentation particles in drainage pipes. This theoretical probability structural formula is introduced into the calculation process of the drainage pipe sedimentation probability adjustment parameter, which can effectively improve the calculation accuracy of the drainage pipe sedimentation probability adjustment parameter.

[0046] 2) This paper constructs a genetic optimization algorithm for drainage pipe siltation particles based on jump feedback. The theoretical probability structure of drainage pipe siltation particle transport is used as the fitness of the genetic optimization algorithm for drainage pipe siltation particles. The algorithm predicts the probability adjustment parameter of drainage pipe siltation. This parameter is used to configure a numerical model for rainfall and flooding, significantly improving the accuracy of the model's waterlogging predictions.

[0047] 3) The present invention embeds a combined preprocessing module and a time-varying slider optimization module in each data processing unit of the bidirectional long-short-term memory neural network, and at the same time adds a two-dimensional convolutional neural network based on error compensation at the starting position of the forward layer and the ending position of the backward layer to obtain an improved time-varying slider neural network. The drainage pipe siltation probability adjustment parameter training sample set is used as the time-varying slider neural network input, and then the drainage pipe siltation probability adjustment parameter guides the time-varying slider neural network to obtain an intelligent waterlogging prediction method, thereby achieving the goals of intelligent and precise urban waterlogging prediction and improving the accuracy of intelligent waterlogging prediction when siltation exists in drainage pipes. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the steps of the method of the present invention;

[0049] Figure 2 Schematic diagram of the process of establishing the theoretical probability structure formula for the transport of sedimentation particles in drainage pipes according to the method of the present invention;

[0050] Figure 3 Schematic diagram of the prediction process of the drainage pipe siltation probability adjustment parameter according to the method of the present invention;

[0051] Figure 4 A flow chart of the generation and expansion of a training sample set for the method of the present invention;

[0052] Figure 5 This is a flow chart for constructing the improved time-varying slider neural network according to the method of the present invention. DETAILED DESCRIPTION

[0053] 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] See also Figure 1-Figure 5 Taking the full-scale test of underground drainage pipe siltation at the full-scale test site of the dam project as an example, the present invention provides a technical solution: an intelligent prediction method for urban waterlogging using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation, such as Figure 1 As shown, the specific steps include:

[0055] Step S1: Based on Newton's first law and the natural distribution law of the transport of sediment particles in the drainage pipe, a theoretical probability structure formula of the transport of sediment particles in the drainage pipe is established. The process of establishing the theoretical probability structure formula of the transport of sediment particles in the drainage pipe is as follows: Figure 2 The specific steps are as follows:

[0056] Step S11: Based on Newton's first law and the principle of steady-flow fluid mechanics, the critical force equilibrium state equation of the sedimentation particles in the drainage pipe is established as follows:

[0057]

[0058] Among them, F L is the buoyancy of the water flow on the sediment particles, F N is the supporting force of the pipeline slope on the sediment particles, F f is the frictional resistance of the pipe bottom acting on the sediment particles, G is the gravity of the sediment particles, α is the slope of the pipe, τ csis the critical shear stress on the sediment particles;

[0059] Step S12: Since the transport of sediment particles in drainage pipes obeys the natural distribution law, the probability parameter γ of the theoretical probability structure of the transport of sediment particles in drainage pipes is introduced, and the parameter obeys N(μ,δ 2 ), and then the theoretical probability structure of sedimentation particle transport in drainage pipes is constructed as follows:

[0060]

[0061] Where P represents the theoretical probability of sedimentation particle transport in the drainage pipe, N(·) represents the normal distribution, μ is the mean of the normal distribution random variable, δ represents the standard deviation of the normal distribution random variable, τ0 is the instantaneous shear stress at the bottom of the pipe on the sedimentation particles, and τ0 is calculated as follows:

[0062]

[0063] Where C is the Scheherazade coefficient, R is the hydraulic radius of the pipe, ρ fp is the average density of sediment-laden water flow, B is the instantaneous width of the water surface in the pipe, J is the hydraulic gradient of the river channel and J = sinα, u a is the average flow velocity of water depth in the direction perpendicular to the water surface.

[0064] Step S2: Combine fluid-solid coupling numerical simulation and full-scale test method to realize parameter calibration in the theoretical probability structural formula of sedimentation particle transport in the drainage pipe in step S1; the parameters in the theoretical probability structural formula of sedimentation particle transport in the drainage pipe include the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameter of the normal distribution random variable.

[0065] Step S3: Substitute the parameters calibrated in the theoretical probability structure of the drainage pipe sedimentation particle transport in step S2 into step S1, construct a drainage pipe sedimentation particle genetic optimization algorithm based on jump feedback, use the theoretical probability structure of the drainage pipe sedimentation particle transport in step S1 as the fitness of the drainage pipe sedimentation particle genetic optimization algorithm, and predict the drainage pipe sedimentation probability adjustment parameter. The prediction process of the drainage pipe sedimentation probability adjustment parameter is as follows: Figure 3 The specific steps are as follows:

[0066] Step S31: Using the control variable method and full-scale / reduced-scale tests, calibrate the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameters of the normal distribution random variable in the theoretical probability structure formula of the transport of sediment particles in the drainage pipe in step S2, and substitute the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameters of the normal distribution random variable into the theoretical probability structure formula of the transport of sediment particles in the drainage pipe in step S1;

[0067] Step S32: using the theoretical probability structure of the drainage pipe sedimentation particle transport in step S1 as the fitness of the drainage pipe sedimentation particle genetic optimization algorithm, predicting the location of the drainage pipe sedimentation particle transport, and recording the predicted results of the drainage pipe sedimentation particle transport location;

[0068] Step S33: Utilize the control variable method and full-scale / reduced-scale test in step S31 to count the actual transport positions of 100 pre-marked drainage pipe sediment particles, and compare them with the predicted transport positions of drainage pipe sediment particles in step S32, calculate the error distance between the predicted transport position and the actual transport position, and use the crossover probability P of the genetic optimization algorithm to calculate the error distance between the predicted transport position and the actual transport position. c and mutation probability P m The jump feedback factor k is introduced. When the error distance is greater than 0.02m, the value of the jump feedback factor k increases by 1. When the error distance is greater than 0.02m, the value of the jump feedback factor k no longer increases. The crossover probability P of the genetic optimization algorithm for drainage pipe sedimentation particles based on jump feedback is c and mutation probability P m As shown below:

[0069]

[0070] Among them, P c3 、P c2 、P c1 、P m3 、P m2 、P m1 are constants and P c3 >P c2 >P c1 ∈(0,1),P m3 >P m2 >P m1 ∈(0,1),f min 、f avg 、f max represent the minimum, maximum and average values ​​of individual fitness of sediment particles, respectively, and f ca and f ma Represent the larger fitness value of the individuals in the population participating in crossover and mutation, A and B represent the adaptive constants of probability adjustment, k is the jump feedback factor, e s The error distance between the predicted position of sedimentation particles in the drainage pipe and the actual position of the particles;

[0071] Step S34: Use the genetic optimization algorithm for drainage pipe sediment particles based on jump feedback in step S33 to predict the transfer position of drainage pipe sediment particles. After the transfer position is determined, the average of the 10 sampling points of the sediment height of the center line of the drainage pipe section is used as the equivalent sediment height. Get the drainage pipe sedimentation probability adjustment parameter D i The calculation method is as follows:

[0072]

[0073] Where i represents the number of the drainage pipe section, R represents the inner diameter of the drainage pipe, j represents the sampling point number of the siltation height of the center line of the drainage pipe section, h j is the siltation height of the jth sampling point on the center line of the drainage pipe section.

[0074] Step S4: Introduce the drainage pipe siltation probability adjustment parameter in step S3 into the rainwater numerical model, and combine it with the improved sample set expansion method to obtain a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels. The generation and expansion process of the training sample set is as follows: Figure 4 The specific steps are as follows:

[0075] Step S41: constructing a rainwater flood numerical model and configuring the attribute information based on the underlying surface and runoff generation and convergence attribute information of the study area;

[0076] Step S42: using the drainage pipe siltation probability adjustment parameter of step S3, assigning the flow attribute value of each drainage pipe section in the rainwater flood numerical model of step S41, thereby obtaining a rainwater flood numerical model that takes into account the drainage pipe siltation probability adjustment parameter;

[0077] Step S43: Based on the rainfall amount and rainfall duration, the numerical model for rainwater flooding that takes into account the drainage pipe siltation probability adjustment parameter in step S42 is used to output the waterlogging result, thereby obtaining a sample set of drainage pipe siltation probability adjustment parameters with a waterlogging result label at the current moment.

[0078] Step S44: Using the drainage pipe siltation probability adjustment parameter calculation method, recalculate the drainage pipe siltation probability adjustment parameter at time period T, and input the recalculated drainage pipe siltation probability adjustment parameter into the rainwater numerical model that considers the drainage pipe siltation probability adjustment parameter in step S42, to obtain a drainage pipe siltation probability adjustment parameter sample set with a waterlogging result label at the next moment, and add this sample set to step S43 to form the original training sample set;

[0079] Step S45: Since the original training sample set in step S44 is related to the spatiotemporal characteristics, a hybrid cross neural network consisting of a two-dimensional convolutional neural network and a three-dimensional convolutional neural network is added in front of the generator of the generative adversarial network. The two-dimensional convolutional neural network contains spatial feature extraction convolution kernels of five different scales, namely 1×1, 1×3, 1×9, 3×3, and 9×9, thereby extracting the spatial features of the waterlogging result label. The three-dimensional convolutional neural network only contains two different scales, namely 2×3×3 and 2×9×9, thereby extracting the features of the spatial structure composed of the waterlogging result, rainfall duration, and rainfall intensity. At the same time, the output result of the hybrid cross neural network is XORed with the noise and the attribute parameters of the generative adversarial network. The operation result is input into the generator to obtain an improved generative adversarial network. The improved generative adversarial network is used to expand the original training sample set in step S44 to obtain a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels, i.e., the expanded training sample set.

[0080] Step S5: embed a combined preprocessing module and a time-varying slider optimization module in each data processing unit of the bidirectional long short-term memory neural network, and at the same time add a two-dimensional convolutional neural network based on error compensation at the starting position of the forward layer and the end position of the backward layer to obtain an improved time-varying slider neural network. The construction process of the improved time-varying slider neural network is as follows: Figure 5 The specific steps are as follows:

[0081] Step S51: Introduce a 34-layer deep residual network and a one-dimensional capsule network to form a combined preprocessing module for the expanded training sample set. The combined preprocessing module is placed at the forefront of the data processing unit in the bidirectional long short-term memory neural network to achieve data preprocessing and dimensionality reduction of the training sample set.

[0082] Step S52: Using the sliding time window to construct a time-varying slider optimization module, which is placed in front of the combined preprocessing module in step S51. The time-varying slider optimization module dynamically adjusts the length of the sliding time window based on the error percentage between the prediction result output by the bidirectional long short-term memory neural network and the actual result. The initial value of the length of the sliding time window is 10. When the error percentage between the prediction result and the actual result is greater than 10%, the length of the sliding time window is reduced by 1, and then the bidirectional long short-term memory neural network is used again to output the prediction result until the error percentage between the prediction result and the actual result is less than 10%.

[0083] Step S53: using the combined pre-processing module in step S51, the time-varying slider optimization module in step S52, and the bidirectional long short-term memory neural network to establish a time-varying slider neural network;

[0084] Step S54: In step S53, a two-dimensional convolutional neural network group is added to the starting position of the forward layer and the ending position of the backward layer of the time-varying slider neural network. The two-dimensional convolutional neural network group is constructed in parallel by three two-dimensional convolution kernels of 1×1, 3×3, and 9×9. The error percentage between the prediction result and the actual result in step S52 is compared, and an error compensation factor is added to the loss function of the two-dimensional convolutional neural network group to dynamically optimize the waterlogging prediction results of the two-dimensional convolutional neural network group and the time-varying slider neural network. The two-dimensional convolutional neural network group, the error compensation factor, and the slider neural network together constitute an improved time-varying slider neural network.

[0085] Step S6: Use the drainage pipe siltation probability adjustment parameter training sample set in step S4 as the input of the improved time-varying slider neural network, and use its output as the prediction result, so as to realize the intelligent prediction of urban waterlogging by guiding the time-varying slider neural network with the drainage pipe siltation probability adjustment parameter.

[0086] The expanded training sample set is used as the input of the improved time-varying slider neural network in step S5, and the support vector machine is used to classify the waterlogging results. The output is used as the intelligent prediction result of waterlogging in the study area. The output indicators of the intelligent prediction result of waterlogging in the study area include flooding depth and flooding area, thereby realizing the intelligent prediction of waterlogging by guiding the time-varying slider neural network with the probability adjustment parameter of drainage pipe siltation.

[0087] The waterlogging results of a full-scale test site of a drainage pipe dam project were obtained by numerical simulation of 300 sets of rainfall with different return periods and rainfall durations. The results were compared with the waterlogging prediction output of the intelligent waterlogging prediction method for drainage pipe siltation probability adjustment parameter guided by a time-varying slider neural network proposed in the present invention. The performance indicators are shown in Table 1.

[0088] Table 1 Statistical results of performance indicators of the patent algorithm in the embodiment

[0089]

[0090] As shown in Table 1, the root mean square error, mean absolute percentage error, and goodness of fit of the intelligent waterlogging prediction method using the drainage pipe siltation probability adjustment parameter-guided time-varying slider neural network proposed in the present invention are 0.417, 0.051, and 0.938, respectively. This shows that the intelligent waterlogging prediction method has the advantages of high accuracy and good reliability, which is conducive to achieving the goal of high-precision and intelligent waterlogging intelligent prediction.

[0091] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligent waterlogging prediction based on a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation, characterized in that: The specific steps include: Step S1: Based on Newton's first law and the natural distribution law of the transport of sediment particles in drainage pipes, a theoretical probability structure formula for the transport of sediment particles in drainage pipes is established. The specific steps are as follows: Step S11: Based on Newton's first law and the principle of steady-flow fluid mechanics, the critical force equilibrium state equation of the sedimentation particles in the drainage pipe is established as follows: Among them, F L is the buoyancy of the water flow on the sediment particles, F N is the supporting force of the pipeline slope on the sediment particles, F f is the frictional resistance of the pipe bottom acting on the sediment particles, G is the gravity of the sediment particles, α is the slope of the pipe, τ cs is the critical shear stress on the sediment particles; Step S12: Since the transport of sediment particles in drainage pipes obeys the natural distribution law, the probability parameter γ of the theoretical probability structure of the transport of sediment particles in drainage pipes is introduced, and the parameter obeys N(μ,δ 2 ), and then the theoretical probability structure of sedimentation particle transport in drainage pipes is constructed as follows: Where P represents the theoretical probability of sedimentation particle transport in the drainage pipe, N(·) represents the normal distribution, μ is the mean of the normal distribution random variable, δ represents the standard deviation of the normal distribution random variable, τ0 is the instantaneous shear stress at the bottom of the pipe on the sedimentation particles, and τ0 is calculated as follows: Where C is the Scheherazade coefficient, R is the hydraulic radius of the pipe, ρ fp is the average density of sediment-laden water flow, B is the instantaneous width of the water surface in the pipe, J is the hydraulic gradient of the river channel and J = sinα, u a is the average velocity of water depth in the direction perpendicular to the water surface; Step S2: combining fluid-structure coupling numerical simulation and full-scale test method to calibrate the parameters in the theoretical probability structure of the transport of sedimentation particles in the drainage pipe in step S1; Step S3: Substituting the parameters calibrated in the theoretical probability structural formula of the transport of sedimentation particles in the drainage pipe in step S2 into step S1, constructing a genetic optimization algorithm for sedimentation particles in the drainage pipe based on jump feedback, using the theoretical probability structural formula of the transport of sedimentation particles in the drainage pipe in step S1 as the fitness of the genetic optimization algorithm for sedimentation particles in the drainage pipe, and predicting the probability adjustment parameter of the sedimentation degree of the drainage pipe; Step S4: Introducing the drainage pipe siltation probability adjustment parameter obtained in step S3 into the rainwater numerical model, and combining it with the improved sample set expansion method, obtaining a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels; Step S5: embedding a combined preprocessing module and a time-varying slider optimization module in each data processing unit of the bidirectional long short-term memory neural network, and simultaneously adding a two-dimensional convolutional neural network based on error compensation at the starting position of the forward layer and the ending position of the backward layer to obtain an improved time-varying slider neural network; Step S6: Use the drainage pipe siltation probability adjustment parameter training sample set in step S4 as the input of the improved time-varying slider neural network, and use its output as the prediction result, so as to realize the intelligent prediction of urban waterlogging by guiding the time-varying slider neural network with the drainage pipe siltation probability adjustment parameter.

2. The method for intelligent waterlogging prediction using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation according to claim 1 is characterized by: In step S2, the parameters in the theoretical probability structure of the transport of sediment particles in the drainage pipe include the Scheherazade coefficient, the mean of the normal distribution random variable, and the standard deviation parameter of the normal distribution random variable.

3. The method for intelligent waterlogging prediction using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation according to claim 1 is characterized by: In step S3, the parameters calibrated in the theoretical probability structure of the drainage pipe sedimentation particle transport in step S2 are substituted into step S1, and a drainage pipe sedimentation particle genetic optimization algorithm based on jump feedback is constructed. The theoretical probability structure of the drainage pipe sedimentation particle transport in step S1 is used as the fitness of the drainage pipe sedimentation particle genetic optimization algorithm, and the drainage pipe sedimentation probability adjustment parameter is predicted. The specific steps are as follows: Step S31: Using the control variable method and full-scale / reduced-scale tests, calibrate the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameters of the normal distribution random variable in the theoretical probability structure formula of the transport of sediment particles in the drainage pipe in step S2, and substitute the Xie Cai coefficient, the mean of the normal distribution random variable, and the standard deviation parameters of the normal distribution random variable into the theoretical probability structure formula of the transport of sediment particles in the drainage pipe in step S1; Step S32: using the theoretical probability structure of the drainage pipe sedimentation particle transport in step S1 as the fitness of the drainage pipe sedimentation particle genetic optimization algorithm, predicting the location of the drainage pipe sedimentation particle transport, and recording the predicted results of the drainage pipe sedimentation particle transport location; Step S33: Utilize the control variable method and full-scale / reduced-scale test in step S31 to count the actual transport positions of 100 pre-marked drainage pipe sediment particles, and compare them with the predicted transport positions of drainage pipe sediment particles in step S32, calculate the error distance between the predicted transport position and the actual transport position, and use the crossover probability P of the genetic optimization algorithm to calculate the error distance between the predicted transport position and the actual transport position. c and mutation probability P m The jump feedback factor k is introduced. When the error distance is greater than 0.02m, the value of the jump feedback factor k increases by 1. When the error distance is greater than 0.02m, the value of the jump feedback factor k no longer increases. The crossover probability P of the genetic optimization algorithm for drainage pipe sedimentation particles based on jump feedback is c and mutation probability P m As shown below: Among them, P c3 、P c2 、P c1 、P m3 、P m2 、P m1 are constants and P c3 >P c2 >P c1 ∈(0,1),P m3 >P m2 >P m1 ∈(0,1),f min 、f avg 、f max represent the minimum, maximum and average values ​​of individual fitness of sediment particles, respectively, and f ca and f ma Represent the larger fitness value of the individuals in the population participating in crossover and mutation, A and B represent the adaptive constants of probability adjustment, k is the jump feedback factor, e s The error distance between the predicted position of sedimentation particles in the drainage pipe and the actual position of the particles; Step S34: Use the genetic optimization algorithm for drainage pipe sediment particles based on jump feedback in step S33 to predict the transfer position of drainage pipe sediment particles. After the transfer position is determined, the average of the 10 sampling points of the sediment height of the center line of the drainage pipe section is used as the equivalent sediment height. Get the probability adjustment parameter D of the drainage pipe siltation degree i The calculation method is as follows: Where i represents the number of the drainage pipe section, R represents the inner diameter of the drainage pipe, j represents the sampling point number of the siltation height of the center line of the drainage pipe section, h j is the siltation height of the jth sampling point on the center line of the drainage pipe section.

4. The method for intelligent waterlogging prediction using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation according to claim 1 is characterized by: In step S4, the drainage pipe siltation probability adjustment parameter from step S3 is introduced into the rainwater numerical model. Combined with the improved sample set expansion method, a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels are obtained. The specific steps are as follows: Step S41: constructing a rainwater flood numerical model and configuring the attribute information based on the underlying surface and runoff generation and convergence attribute information of the study area; Step S42: using the drainage pipe siltation probability adjustment parameter of step S3, assigning the flow attribute value of each drainage pipe section in the rainwater flood numerical model of step S41, thereby obtaining a rainwater flood numerical model that takes into account the drainage pipe siltation probability adjustment parameter; Step S43: Based on the rainfall amount and rainfall duration, the numerical model for rainwater flooding that takes into account the drainage pipe siltation probability adjustment parameter in step S42 is used to output the waterlogging result, thereby obtaining a sample set of drainage pipe siltation probability adjustment parameters with a waterlogging result label at the current moment. Step S44: Using the drainage pipe siltation probability adjustment parameter calculation method, recalculate the drainage pipe siltation probability adjustment parameter at time period T, and input the recalculated drainage pipe siltation probability adjustment parameter into the rainwater numerical model that considers the drainage pipe siltation probability adjustment parameter in step S42, to obtain a drainage pipe siltation probability adjustment parameter sample set with a waterlogging result label at the next moment, and add this sample set to step S43 to form the original training sample set; Step S45: Since the original training sample set in step S44 is related to the spatiotemporal characteristics, a hybrid cross neural network consisting of a two-dimensional convolutional neural network and a three-dimensional convolutional neural network is added in front of the generator of the generative adversarial network. The two-dimensional convolutional neural network contains spatial feature extraction convolution kernels of five different scales, namely 1×1, 1×3, 1×9, 3×3, and 9×9, thereby extracting the spatial features of the waterlogging result label. The three-dimensional convolutional neural network only contains two different scales, namely 2×3×3 and 2×9×9, thereby extracting the features of the spatial structure composed of the waterlogging result, rainfall duration, and rainfall intensity. At the same time, the output result of the hybrid cross neural network is XORed with the noise and the attribute parameters of the generative adversarial network. The operation result is input into the generator to obtain an improved generative adversarial network. The improved generative adversarial network is used to expand the original training sample set in step S44 to obtain a large number of drainage pipe siltation probability adjustment parameter training sample sets with waterlogging prediction result labels, i.e., the expanded training sample set.

5. The method for intelligent waterlogging prediction using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation according to claim 1 is characterized by: In step S5, a combined preprocessing module and a time-varying slider optimization module are embedded in each data processing unit of the bidirectional long short-term memory neural network. At the same time, a two-dimensional convolutional neural network based on error compensation is added at the starting position of the forward layer and the end position of the backward layer to obtain an improved time-varying slider neural network. The specific steps are as follows: Step S51: Introduce a 34-layer deep residual network and a one-dimensional capsule network to form a combined preprocessing module for the expanded training sample set. The combined preprocessing module is placed at the forefront of the data processing unit in the bidirectional long short-term memory neural network to achieve data preprocessing and dimensionality reduction of the training sample set. Step S52: Using the sliding time window to construct a time-varying slider optimization module, which is placed in front of the combined preprocessing module in step S51. The time-varying slider optimization module dynamically adjusts the length of the sliding time window based on the error percentage between the prediction result output by the bidirectional long short-term memory neural network and the actual result. The initial value of the length of the sliding time window is 10. When the error percentage between the prediction result and the actual result is greater than 10%, the length of the sliding time window is reduced by 1, and then the bidirectional long short-term memory neural network is used again to output the prediction result until the error percentage between the prediction result and the actual result is less than 10%. Step S53: using the combined pre-processing module in step S51, the time-varying slider optimization module in step S52, and the bidirectional long short-term memory neural network to establish a time-varying slider neural network; Step S54: In step S53, a two-dimensional convolutional neural network group is added to the starting position of the forward layer and the ending position of the backward layer of the time-varying slider neural network. The two-dimensional convolutional neural network group is constructed in parallel by three two-dimensional convolution kernels of 1×1, 3×3, and 9×9. The error percentage between the prediction result and the actual result in step S52 is compared, and an error compensation factor is added to the loss function of the two-dimensional convolutional neural network group to dynamically optimize the waterlogging prediction results of the two-dimensional convolutional neural network group and the time-varying slider neural network. The two-dimensional convolutional neural network group, the error compensation factor, and the slider neural network together constitute an improved time-varying slider neural network.

6. The method for intelligent waterlogging prediction using a time-varying slider neural network guided by a probability adjustment parameter for drainage pipe siltation according to claim 1 is characterized by: In step S6, the expanded training sample set is used as the input of the improved time-varying slider neural network in step S5, and the support vector machine is used to classify the waterlogging results. The output is used as the intelligent prediction result of waterlogging in the study area. The output indicators of the intelligent prediction result of waterlogging in the study area include flooding depth and flooding area, thereby realizing the intelligent prediction of waterlogging by guiding the time-varying slider neural network based on the probability adjustment parameters of the drainage pipe siltation degree.

Images