Sludge pipe inspection well water level intelligent early warning method based on self-adaptive smoothing algorithm and bidirectional gating circulation unit

Abstract

The invention discloses a siltation pipe inspection well water level intelligent early warning method based on a self-adaptive smoothing algorithm and a bidirectional gating circulation unit. The method comprises the following steps that a multiphase flow control equation is established; constructing a deposition pipeline-inspection well spatio-temporal scale hydraulic response model, and configuring boundary constraint and spatio-temporal transportation state information in a fluid-solid coupling numerical simulation process; the method comprises the following steps: constructing a high-resolution deposited pipeline-inspection well numerical simulation model, and outputting a pipeline inlet and outlet flow velocity / flow sample set with an inspection well water level label; establishing an adaptive smoothing algorithm to detect missing data and heterogeneous data, and reconstructing a time-space flow velocity field and an inspection well water level of the deposited pipeline in a training period; an inspection well water level intelligent early warning model is established by taking the silted pipeline space-time flow velocity field and the inspection well water level as a training set, a verification set and a test set and utilizing a bidirectional gating circulation unit based on a dynamic conversion encoder, so that the intelligent and refined targets of inspection well water level intelligent early warning of the silted rainwater pipeline are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of municipal drainage pipe networks, and in particular to an intelligent early warning method for water level in silted manholes based on an adaptive smoothing algorithm and a bidirectional gated circulation unit. Background Art

[0002] With the continuous development of economy and the expansion of urban scale, urban rainwater pipe siltation has become one of the important factors restricting the normal operation of urban rainwater pipe network. Pipeline siltation will not only reduce the drainage capacity of the city, but also increase the load of urban rainwater pipes, greatly reducing the city's waterlogging prevention capacity.

[0003] The existing research on hydraulic characteristics of silted pipelines and manhole water level diagnosis mainly faces the following problems:

[0004] (1) There are many existing research results on the velocity distribution law of the non-full flow section of the traditional non-silted circular pipe, but there is a lack of analysis of the hydraulic response characteristics of the "pipeline-manhole" fluid-solid coupling considering the siltation blocking effect;

[0005] (2) Due to the limitations of experimental conditions, the specific changes in the transport of sediment particles and the hydraulic characteristics of the flow in the pipeline at different time and space scales cannot be accurately restored;

[0006] (3) The existing intelligent diagnosis methods for silted rainwater pipes have low diagnostic efficiency and fail to combine the fluid-solid coupling field and machine learning algorithms to build an overall intelligent detection model of "silted pipes-manholes".

[0007] Therefore, in view of the problems that the hydraulic characteristics of the silted rainwater pipe network are difficult to accurately restore and the low accuracy of traditional models, it is necessary to develop an intelligent early warning method for the water level of the manholes in the silted pipes. Summary of the invention

[0008] In order to solve the problems existing in the prior art, the present invention provides an intelligent early warning method for the water level of silted pipes and manholes based on an adaptive smoothing algorithm and a bidirectional gated circulation unit, which accurately restores the silted pipe-manhole model that shows the flow field in the pipeline, and combines artificial intelligence methods to achieve the goal of scientific, integrated and refined diagnosis of rainwater pipe networks, thereby solving the problems mentioned in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solution: an intelligent early warning method for water level of silted manholes based on an adaptive smoothing algorithm and a bidirectional gated circulation unit, comprising the following steps:

[0010] S1. Establish the continuous medium equation, pipeline hydraulic transport equation, pipeline hydraulic transport equation, particle torque equation and particle collision equation to obtain the multiphase flow control equation;

[0011] S2. Using the multiphase flow control equation in step S1, a spatiotemporal hydraulic response model of "deposition pipeline-manhole" is constructed, boundary constraints are imposed on the fluid phase and the particle phase, and the spatiotemporal transport state information of the fluid phase in the pipe is transmitted at the same time, and a fluid-solid coupling CFD-DEM numerical simulation of the deposition pipeline is performed. The spatiotemporal hydraulic response model of "deposition pipeline-manhole" is used to configure the boundary constraints and spatiotemporal transport state information in the fluid-solid coupling CFD-DEM numerical simulation process;

[0012] S3. Based on the results of the fluid-solid coupling numerical simulation of the silted pipeline, combined with Newton's second law and the continuous medium equation and particle collision equation in step S1, a high-resolution numerical simulation model of "silted pipeline-manhole" is constructed, and the high-resolution numerical simulation model of "silted pipeline-manhole" outputs a flow velocity / flow sample set of the pipeline inlet and outlet with the manhole water level label;

[0013] S4, extract features from the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, establish an adaptive smoothing algorithm to detect missing data and heterogeneous data, and reconstruct the spatiotemporal flow velocity field and manhole water level of the silted pipeline within the training period;

[0014] S5. Using the spatiotemporal velocity field of the silted pipe and the manhole water level obtained in step S4 as the training set, validation set and test set, a bidirectional gated recurrent unit TranGRU ​​based on a dynamic transformation encoder is used to establish an intelligent early warning model for the manhole water level, and the intelligent early warning results for the manhole water level are output through the model.

[0015] Preferably, in step S2, the steps specifically include the following:

[0016] Step S21: Based on the principle of constant water volume, the law of conservation of momentum and the law of conservation of energy, a hydraulic response characteristic equation of "sedimentation pipeline-manhole" is established, which is formulated as follows:

[0017]

[0018] Among them, Q in and Q out They represent the flow rates at the inlet and outlet of the pipeline at the current moment, t is the instantaneous moment, and v in and v out represents the flow rate at the inlet and outlet of the pipe, θ represents the inclination of the pipe, Δt represents an infinitesimal time interval, l s and h s are the length and thickness of the pipeline siltation (centerline position), ρ w represents the water flow density in the pipe, g represents the acceleration due to gravity, G in represents the gravity of the instantaneous flow at the pipe inlet and G in =ρ w Qin gΔt; and They represent the flow rate and velocity at the outlet of the previous pipe respectively. represents the water level of the manhole at t+Δt, Indicates the water level of the manhole at the front end of the pipeline. L p is the length of the pipeline, η is the kinetic energy conversion coefficient of the non-deposition section, r m Represents the radius of the manhole.

[0019] Step S22: Based on the hydraulic response characteristic equation of the “deposition pipeline-manhole” in step S21, a lateral vibration differential equation of the hydraulic transport of the “deposition pipeline-manhole” under the excitation of the basic acceleration is constructed, which is formulated as follows:

[0020]

[0021] Where E is the elastic modulus, I is the moment of inertia of the pipeline interface, x is the horizontal coordinate variable parallel to the horizontal plane, w(x,t) and w r (x, t) represent the absolute displacement and relative base displacement of the pipe flow, respectively, w g (t) represents the basic displacement of the pipe flow and satisfies w(x,t) = w r (x,t)+w g (t), p in is the fluid pressure at the centerline of the pipeline inlet, S f is the cross-sectional area of ​​the fluid, m p is the mass of the pipe per unit length, m f is the fluid mass per unit length, c is the structural damping coefficient, r p is the radius of the stormwater pipe.

[0022] Step S23: Using the hydraulic response characteristic equation of the "silted pipe-manhole" in step S21 and the lateral vibration differential equation of the hydraulic transport of the "silted pipe-manhole" under the excitation of the basic acceleration in step S22, the siltation length, siltation thickness, pipeline water flow velocity and flow coupling relationship are dynamically simulated, and then the "silted pipe-manhole" time-space scale hydraulic response model is established.

[0023] Step S24: Construct a "silted pipe-manhole" fluid-solid coupling (CFD-DEM) numerical simulation model, and use the "silted pipe-manhole" spatiotemporal scale hydraulic response model in step S23 to configure the boundary constraints and spatiotemporal transport state information in the "silted pipe-manhole" fluid-solid coupling (CFD-DEM) numerical simulation model.

[0024] Preferably, in step S3, the specific steps include:

[0025] Step S31: Determine the hydraulic parameters of the rainwater pipe using Newton's second law and the continuous medium equation and particle collision equation in step S1.

[0026] Step S32: Using the hydraulic parameters of the rainwater pipe obtained in step S31 as input items for the flow-solid coupling of the rainwater pipe, a high-resolution "silted pipe-manhole" fluid-solid coupling (CFD-DEM) numerical simulation model is constructed.

[0027] Step S33: Utilize the high-resolution “silted pipe-manhole” fluid-structure coupling (CFD-DEM) numerical simulation model in step S32 to output a pipe inlet and outlet velocity / flow rate sample set with manhole water level labels.

[0028] Preferably, in step S4, the following steps are specifically included:

[0029] Step S41: using a multi-layer autoencoder stack to learn the deep features of the data in the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, adding sparsity constraints, mining the useful structure when there are more hidden neurons, and extracting the feature information of the data in the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label;

[0030] Step S42: Based on the characteristic information of the pipeline inlet and outlet velocity / flow sample set data with the manhole water level label extracted in step S41, a three-dimensional nonlinear spatiotemporal interpolation method is used to estimate the spatiotemporal velocity field and manhole water level of the silted pipeline, and a continuous function V of the local average velocity at time t and position x is constructed. free (t,x) and V cong (t,x). Based on the discrete convolution of the kernel containing all observed data points, two smooth velocity fields in the free flow direction and the congestion flow direction are given respectively. The formula is as follows:

[0031]

[0032] Among them, V free (t,x) and V cong (t,x) is the continuous interpolation speed function based on the existing i-th data point, t i and x i are the timestamp and position corresponding to the i-th data point, c free represents the propagation velocity in the free stream direction, c cong Indicates the congestion propagation speed in the congestion flow direction.

[0033] The kernel function φ is given by the following symmetric exponential form:

[0034]

[0035] Where exp is an exponential function, and τ and δ are specified constants. The exponential function smoothes temporal fluctuations on scales smaller than τ and smoothes spatial fluctuations on scales smaller than δ.

[0036] Step S43: V free (t,x) and V cong (t, x) superposition, the adaptive smoothing method (ASM) is used to calculate the velocity V(t, x) of the point to be estimated, which is expressed in the following formula:

[0037] V(t,x)=ω(t,x)V cong (t,x)+[1-ω(t,x)]V free (t,x)

[0038] Among them, ω(t,x) is the weighted function of congestion and free-flow traffic states, which is represented by the following Sigmoid nonlinear function:

[0039]

[0040] Among them, tanh represents the hyperbolic tangent function, V c represents the critical threshold between free flow and congested flow, and ΔV is the bandwidth for the free flow state to transition to the congested flow state.

[0041] Step S44: The reconstructed spatiotemporal velocity field of the silted pipe within the training period obtained in step S4 is used as the data source k and input into the extended generalized Treiber-Helbing filter to detect data loss, noise reduction and data fusion problems of heterogeneous data caused by detector failure in the data source k. The spatiotemporal velocity field z(t,x) of the silted pipe output by the extended generalized Treiber-Helbing filter is formulated as follows:

[0042]

[0043] Among them, α (k) (t,x) represents a specific data source from data source k at the spatiotemporal location (t,x) and can be interpreted as a dynamic indicator of the reliability of the data source. i (k) (t, x) is a weight function corresponding to data source k and measurement data point i, and the weight can be determined based on a priori estimation of the measurement accuracy of data source k.

[0044] Weight function φ i (k) (t,x) reflects that data from different data sources will lead to different spatiotemporal differences between congestion and free flow conditions. i (k) (t,x) can be expressed as:

[0045]

[0046] in, represents the spatiotemporal coordinates of the spatiotemporal data point i of the kth data source, ω (k) (t,x) represents the weighted function of the congestion and free-flow traffic status of the k-th data source, φ represents the generalized weight function, and represent the timestamp and location of the spatiotemporal data point i of the kth data source, c cong Represents a weight constant.

[0047] Preferably, in step S5, the following steps are specifically included:

[0048] Step S51: Integrate the bidirectional gated recurrent unit (BiGRU) into the dynamic transformation encoder (Transformer), and use the self-attention mechanism to enhance the feature search capability of the bidirectional gated recurrent unit, thereby constructing a bidirectional gated recurrent unit (TranGRU) based on the dynamic transformation encoder and the self-attention mechanism.

[0049] Step S52: Using the spatiotemporal velocity field of the silted pipeline and the manhole water level obtained in step S4 as the training set, validation set and test set, extract the local and global features of the pipeline inlet and outlet velocity / flow time series and the manhole water level height, collaboratively learn the local velocity information, siltation degree and manhole water level of the pipeline, and calibrate the hyperparameters in the bidirectional gated recurrent unit based on the dynamic transformation encoder.

[0050] Step S53: configure the hyperparameters calibrated in step S52 into a bidirectional gated recurrent unit based on a dynamic conversion encoder, thereby obtaining an intelligent warning model for manhole water level, and outputting the intelligent warning result for manhole water level through the model.

[0051] Preferably, in step S51, the bidirectional gated recurrent unit (BiGRU) is integrated into the dynamic transformation encoder (Transformer), and the self-attention mechanism is used to enhance the feature search capability of the bidirectional gated recurrent unit, thereby constructing a bidirectional gated recurrent unit (TranGRU) based on the dynamic transformation encoder and the self-attention mechanism, and the specific steps are:

[0052] Step S511: a multi-layer dynamic transformer is introduced, wherein the first layer of the dynamic transformer receives the pipeline inlet and outlet flow velocity / flow time series as input.

[0053] Step S512: A bidirectional gated recurrent unit (BiGRU) and a self-attention mechanism are embedded in the dynamic transformation encoder (Transformer) of step S511. The first layer of the dynamic transformation encoder (Transformer) in step S511 is responsible for receiving input information and performing normalization operations, and then inputting them into the bidirectional gated recurrent unit neural network and the self-attention neural network respectively. The bidirectional gated recurrent unit (BiGRU) then encodes the local features of the flow velocity / flow time series at the pipeline inlet and outlet and the manhole water level height, and the self-attention mechanism encodes the global features of the flow velocity / flow time series at the pipeline inlet and outlet and the manhole water level height.

[0054] Step S513: Add a preprocessing module at the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512, add a Softmax function gating mechanism at the output position, adaptively adjust the ratio of the acquired local features and global features in the encoding state, and output the intelligent warning result of the manhole water level.

[0055] Step S514: Take the flow velocity / flow time series at the inlet and outlet of the pipeline under different siltation intensities and the water level height of the manhole as the training sample set, and obtain the hyperparameters of the bidirectional gated recurrent unit (BiGRU) and the self-attention mechanism embedded in the dynamic transformation encoder (Transformer) in step S512 through training. Substitute the hyperparameters into the bidirectional gated recurrent unit (BiGRU) and the self-attention mechanism embedded in the dynamic transformation encoder (Transformer) to obtain the bidirectional gated recurrent unit (TranGRU) based on the dynamic transformation encoder and the self-attention mechanism, and use the Softmax function gating mechanism in step S513 as the final output unit.

[0056] Preferably, in step S513, a preprocessing module is added to the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512, a Softmax function gating mechanism is added to the output position, and the intelligent warning result of the manhole water level is output. The specific steps are:

[0057] Step S5131: For the pipeline inlet and outlet flow velocity / flow time series X, the preprocessing module added to the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512 receives the input and performs preprocessing operations to obtain a hidden state layer Z of the same length i The specific calculation process is as follows:

[0058] Z i =SA(W Q LN(X i ),W K LN(X i ),W V LN(X i))+X i

[0059] Among them, i is the time series number, Z i represents the hidden state of the current layer after self-attention network encoding, X i represents the time series data of the flow velocity / flow rate at the inlet and outlet of the pipeline numbered i. LN(·) represents the normalization operation, and SA(·) represents the multi-head self-attention neural network sublayer. W Q , W K and W V Represents the state input parameters of the self-attention query, key, and value respectively.

[0060] Z i Input into the feedforward neural network sublayer of the preprocessing module to obtain the hidden state H i , the calculation method is as follows:

[0061] H i =FFN(LN(Z i ))+Z i

[0062] Where FFN(·) represents the feed-forward neural network sublayer. Residual connections are applied to the feed-forward neural network sublayer to alleviate the gradient problem.

[0063] Step S5132: Z output by the preprocessing module in step S5131 i Input to the bidirectional gated recurrent unit and self-attention mechanism, the specific processing process is as follows:

[0064]

[0065] in, Represents the output state of the self-attention mechanism. Represents the output state of a bidirectional gated recurrent unit.

[0066] Step S5133: Using the and It takes as input, and outputs the intelligent warning result of manhole water level through the Softmax function gating mechanism.

[0067] The beneficial effects of the present invention are as follows: the present invention integrates the fluid-solid coupling method, the adaptive smoothing method and the bidirectional gated cyclic unit model, uses the fluid-solid coupling method to obtain the index with flow velocity labels at the inlet and outlet of the silted pipe, and uses the adaptive smoothing method to reconstruct the missing values ​​of the input flow velocity / flow sample set with manhole water level labels output by numerical simulation on the spatiotemporal scale. Finally, the reconstructed pipeline flow field and the bidirectional gated unit based on the dynamic conversion encoder are integrated to establish an intelligent early warning model for manhole water level, and realize intelligent early warning of manhole water level in silted rainwater pipes. Compared with the prior art, the present invention closely combines the high-resolution fluid-solid coupling model with the neural network, so that the manhole water level of the silted pipe can be warned more accurately and timely, providing technical support for water level alarm diagnosis in urban flood scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of the intelligent early warning method for water level of silted pipe manholes based on the adaptive smoothing algorithm and the bidirectional gated circulation unit of the present invention;

[0069] Figure 2 A flow chart for outputting a pipeline inlet and outlet velocity / flow sample set with manhole water level labels;

[0070] Figure 3 A flow chart for establishing an intelligent early warning model for manhole water level;

[0071] Figure 4 Schematic diagram of the process of constructing a bidirectional gated recurrent unit TranGRU ​​based on a dynamic transformation encoder and self-attention mechanism. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0073] See also Figure 1-Figure 4 The present invention provides a technical solution: an intelligent early warning method for water level of silted manholes based on an adaptive smoothing algorithm and a bidirectional gated circulation unit, such as Figure 1 As shown, the following steps are included:

[0074] S1. Establish the continuous medium equation, pipeline hydraulic transport equation, pipeline hydraulic transport equation, particle torque equation and particle collision equation, and on this basis, obtain the multiphase flow control equation;

[0075] S2. Using the multiphase flow control equation in step S1, a spatiotemporal hydraulic response model of "deposition pipeline-manhole" is constructed, boundary constraints are imposed on the fluid phase and the particle phase, and the spatiotemporal transport state information of the fluid phase in the pipe is transmitted at the same time, and a fluid-solid coupling CFD-DEM numerical simulation of the deposition pipeline is performed. The spatiotemporal hydraulic response model of "deposition pipeline-manhole" is used to configure the boundary constraints and spatiotemporal transport state information in the fluid-solid coupling CFD-DEM numerical simulation process;

[0076] S3. Based on the results of the fluid-solid coupling numerical simulation of the silted pipeline, combined with Newton's second law and the continuous medium equation and particle collision equation in step S1, a high-resolution numerical simulation model of "silted pipeline-manhole" is constructed, and the high-resolution numerical simulation model of "silted pipeline-manhole" outputs a flow velocity / flow sample set of the pipeline inlet and outlet with the manhole water level label;

[0077] S4, extract features from the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, establish an adaptive smoothing algorithm to detect missing data and heterogeneous data, and reconstruct the spatiotemporal flow velocity field and manhole water level of the silted pipeline within the training period;

[0078] S5. Using the spatiotemporal velocity field of the silted pipe and the manhole water level obtained in step S4 as the training set, validation set and test set, a bidirectional gated recurrent unit TranGRU ​​based on a dynamic transformation encoder is used to establish an intelligent early warning model for the manhole water level, and the intelligent early warning results for the manhole water level are output through the model.

[0079] Furthermore, in step S2, the specific steps include:

[0080] Step S21: Based on the principle of constant water volume, the law of conservation of momentum and the law of conservation of energy, a hydraulic response characteristic equation of "sedimentation pipeline-manhole" is established, which is formulated as follows:

[0081]

[0082] Among them, Q in and Q out They represent the flow rates at the inlet and outlet of the pipeline at the current moment, t is the instantaneous moment, and v in and v out represents the flow rate at the inlet and outlet of the pipe, θ represents the inclination of the pipe, Δt represents an infinitesimal time interval, l s and h s are the length and thickness of the pipeline siltation (centerline position), ρ w represents the water flow density in the pipe, g represents the acceleration due to gravity, G in represents the gravity of the instantaneous flow at the pipe inlet and G in =ρ w Qin gΔt; and They represent the flow rate and velocity at the outlet of the previous pipe respectively. represents the water level of the manhole at t+Δt, Indicates the water level of the manhole at the front end of the pipeline. L p is the length of the pipeline, η is the kinetic energy conversion coefficient of the non-deposition section, r m Represents the radius of the manhole.

[0083] Step S22: Based on the hydraulic response characteristic equation of the “deposition pipeline-manhole” in step S21, a lateral vibration differential equation of the hydraulic transport of the “deposition pipeline-manhole” under the excitation of the basic acceleration is constructed, which is formulated as follows:

[0084]

[0085] Where E is the elastic modulus, I is the moment of inertia of the pipeline interface, x is the horizontal coordinate variable parallel to the horizontal plane, w(x,t) and w r (x, t) represent the absolute displacement and relative base displacement of the pipe flow, respectively, w g (t) represents the basic displacement of the pipe flow and satisfies w(x,t) = w r (x,t)+w g (t), p in is the fluid pressure at the centerline of the pipeline inlet, S f is the cross-sectional area of ​​the fluid, m p is the mass of the pipe per unit length, m f is the fluid mass per unit length, c is the structural damping coefficient, r p is the radius of the stormwater pipe.

[0086] Step S23: Using the hydraulic response characteristic equation of the "silted pipe-manhole" in step S21 and the lateral vibration differential equation of the hydraulic transport of the "silted pipe-manhole" under the excitation of the basic acceleration in step S22, the siltation length, siltation thickness, pipeline water flow velocity and flow coupling relationship are dynamically simulated, and then the "silted pipe-manhole" time-space scale hydraulic response model is established.

[0087] Step S24: Construct a "silted pipe-manhole" fluid-solid coupling (CFD-DEM) numerical simulation model, and use the "silted pipe-manhole" spatiotemporal scale hydraulic response model in step S23 to configure the boundary constraints and spatiotemporal transport state information in the "silted pipe-manhole" fluid-solid coupling (CFD-DEM) numerical simulation model.

[0088] Further, in step S3, if Figure 2As shown in the figure, outputting the flow velocity / flow sample set of the pipe inlet and outlet with the manhole water level label includes the following steps:

[0089] Step S31: Determine the hydraulic parameters of the rainwater pipe using Newton's second law and the continuous medium equation and particle collision equation in step S1.

[0090] Step S32: Using the hydraulic parameters of the rainwater pipe obtained in step S31 as input items for the flow-solid coupling of the rainwater pipe, a high-resolution "silted pipe-manhole" fluid-solid coupling (CFD-DEM) numerical simulation model is constructed.

[0091] Step S33: Utilize the high-resolution “silted pipe-manhole” fluid-structure coupling (CFD-DEM) numerical simulation model in step S32 to output a pipe inlet and outlet velocity / flow rate sample set with manhole water level labels.

[0092] Furthermore, in step S4, feature extraction is performed on the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, and an adaptive smoothing algorithm is established to detect missing data and heterogeneous data, and the spatiotemporal flow velocity field and manhole water level of the silted pipeline within the training period are reconstructed, which specifically includes the following steps:

[0093] Step S41: using a multi-layer autoencoder stack to learn the deep features of the data in the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, adding sparsity constraints, mining the useful structure when there are more hidden neurons, and extracting the feature information of the data in the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label;

[0094] Step S42: Based on the characteristic information of the pipeline inlet and outlet velocity / flow sample set data with the manhole water level label extracted in step S41, a three-dimensional nonlinear spatiotemporal interpolation method is used to estimate the spatiotemporal velocity field and manhole water level of the silted pipeline, and a continuous function V of the local average velocity at time t and position x is constructed. free (t,x) and V cong (t,x). Based on the discrete convolution of the kernel containing all observed data points, two smooth velocity fields in the free flow direction and the congestion flow direction are given respectively. The formula is as follows:

[0095]

[0096] Among them, V free (t,x) and V cong (t,x) is the continuous interpolation speed function based on the existing i-th data point, t i and x i are the timestamp and position corresponding to the i-th data point, c free represents the propagation speed in the free stream direction, c congIndicates the congestion propagation speed in the congestion flow direction.

[0097] The kernel function φ is given by the following symmetric exponential form:

[0098]

[0099] Where exp is an exponential function, and τ and δ are specified constants. The exponential function smoothes temporal fluctuations on scales smaller than τ and smoothes spatial fluctuations on scales smaller than δ.

[0100] Step S43: V free (t,x) and V cong (t, x) superposition, the adaptive smoothing method (ASM) is used to calculate the velocity V(t, x) of the point to be estimated, which is expressed in the following formula:

[0101] V(t,x)=ω(t,x)V cong (t,x)+[1-ω(t,x)]V free (t,x)

[0102] Among them, ω(t,x) is the weighted function of congestion and free-flow traffic states, which is represented by the following Sigmoid nonlinear function:

[0103]

[0104] Among them, tanh represents the hyperbolic tangent function, V c represents the critical threshold between free flow and congested flow, and ΔV is the bandwidth for the free flow state to transition to the congested flow state.

[0105] Step S44: The reconstructed spatiotemporal velocity field of the silted pipe within the training period obtained in step S4 is used as the data source k and input into the extended generalized Treiber-Helbing filter to detect data loss, noise reduction and data fusion problems of heterogeneous data caused by detector failure in the data source k. The spatiotemporal velocity field z(t,x) of the silted pipe output by the extended generalized Treiber-Helbing filter is formulated as follows:

[0106]

[0107] Among them, α (k) (t,x) represents a specific data source from data source k at the spatiotemporal location (t,x) and can be interpreted as a dynamic indicator of the reliability of the data source. i (k) (t, x) is a weight function corresponding to data source k and measurement data point i, and the weight can be determined based on a priori estimation of the measurement accuracy of data source k.

[0108] Weight function φ i (k) (t,x) reflects that data from different data sources will lead to different spatiotemporal differences between congestion and free flow conditions. i (k) (t,x) can be expressed as:

[0109]

[0110] in, represents the spatiotemporal coordinates of the spatiotemporal data point i of the kth data source, ω (k) (t,x) represents the weighted function of the congestion and free-flow traffic status of the k-th data source, φ represents the generalized weight function, and represent the timestamp and location of the spatiotemporal data point i of the kth data source, c cong Represents a weight constant.

[0111] Further, such as Figure 3 As shown, in step S5, the spatiotemporal velocity field of the silted pipe and the manhole water level obtained in step S4 are used as the training set, the validation set and the test set, and a bidirectional gated recurrent unit (TranGRU) based on a dynamic transformation encoder is used to establish an intelligent warning model for the manhole water level, and the intelligent warning result of the manhole water level is output through the model, which specifically includes the following steps:

[0112] Step S51: Integrate the bidirectional gated recurrent unit (BiGRU) into the dynamic transformation encoder (Transformer), and use the self-attention mechanism to enhance the feature search capability of the bidirectional gated recurrent unit, thereby constructing a bidirectional gated recurrent unit (TranGRU) based on the dynamic transformation encoder and the self-attention mechanism.

[0113] Step S52: Using the spatiotemporal velocity field of the silted pipeline and the manhole water level obtained in step S4 as the training set, validation set and test set, extract the local and global features of the pipeline inlet and outlet velocity / flow time series and the manhole water level height, collaboratively learn the local velocity information, siltation degree and manhole water level of the pipeline, and calibrate the hyperparameters in the bidirectional gated recurrent unit based on the dynamic transformation encoder.

[0114] Step S53: configure the hyperparameters calibrated in step S52 into a bidirectional gated recurrent unit based on a dynamic conversion encoder, thereby obtaining an intelligent warning model for manhole water level, and outputting the intelligent warning result for manhole water level through the model.

[0115] Further, such as Figure 4As shown, in step S51, the bidirectional gated recurrent unit (BiGRU) is integrated into the dynamic transformation encoder (Transformer), and the self-attention mechanism is used to enhance the feature search capability of the bidirectional gated recurrent unit, thereby constructing a bidirectional gated recurrent unit (TranGRU) based on the dynamic transformation encoder and the self-attention mechanism. The specific steps are:

[0116] Step S511: a multi-layer dynamic transformer is introduced, wherein the first layer of the dynamic transformer receives the pipeline inlet and outlet flow velocity / flow time series as input.

[0117] Step S512: A bidirectional gated recurrent unit (BiGRU) and a self-attention mechanism are embedded in the dynamic transformation encoder (Transformer) of step S511. The first layer of the dynamic transformation encoder (Transformer) in step S511 is responsible for receiving input information and performing normalization operations, and then inputting them into the bidirectional gated recurrent unit neural network and the self-attention neural network respectively. The bidirectional gated recurrent unit (BiGRU) then encodes the local features of the flow velocity / flow time series at the pipeline inlet and outlet and the manhole water level height, and the self-attention mechanism encodes the global features of the flow velocity / flow time series at the pipeline inlet and outlet and the manhole water level height.

[0118] Step S513: Add a preprocessing module at the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512, add a Softmax function gating mechanism at the output position, adaptively adjust the ratio of the acquired local features and global features in the encoding state, and output the intelligent warning result of the manhole water level.

[0119] Step S514: Take the flow velocity / flow time series at the inlet and outlet of the pipeline under different siltation intensities and the water level height of the manhole as the training sample set, and obtain the hyperparameters of the bidirectional gated recurrent unit (BiGRU) and the self-attention mechanism embedded in the dynamic transformation encoder (Transformer) in step S512 through training. Substitute the hyperparameters into the bidirectional gated recurrent unit (BiGRU) and the self-attention mechanism embedded in the dynamic transformation encoder (Transformer) to obtain the bidirectional gated recurrent unit (TranGRU) based on the dynamic transformation encoder and the self-attention mechanism, and use the Softmax function gating mechanism in step S513 as the final output unit.

[0120] Further, in step S513, a preprocessing module is added to the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512, and a Softmax function gating mechanism is added to the output position to output the intelligent warning result of the manhole water level. The specific steps are as follows:

[0121] Step S5131: For the pipeline inlet and outlet flow velocity / flow time series X, the preprocessing module added to the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512 receives the input and performs preprocessing operations to obtain a hidden state layer Z of the same length i The specific calculation process is as follows:

[0122] Z i =SA(W Q LN(X i ),W K LN(X i ),W V LN(X i ))+X i

[0123] Among them, i is the time series number, Z i represents the hidden state of the current layer after self-attention network encoding, X i represents the time series data of the flow velocity / flow rate at the inlet and outlet of the pipeline numbered i. LN(·) represents the normalization operation, and SA(·) represents the multi-head self-attention neural network sublayer. W Q , W K and W V Represents the state input parameters of the self-attention query, key, and value respectively.

[0124] Z i Input into the feedforward neural network sublayer of the preprocessing module to obtain the hidden state H i , the calculation method is as follows:

[0125] H i =FFN(LN(Z i ))+Z i

[0126] Where FFN(·) represents the feed-forward neural network sublayer. Residual connections are applied to the feed-forward neural network sublayer to alleviate the gradient problem.

[0127] Step S5132: Z output by the preprocessing module in step S5131 i Input to the bidirectional gated recurrent unit and self-attention mechanism, the specific processing process is as follows:

[0128]

[0129] in, Represents the output state of the self-attention mechanism. Represents the output state of a bidirectional gated recurrent unit.

[0130] Step S5133: Using the and It takes as input, and outputs the intelligent warning result of manhole water level through the Softmax function gating mechanism.

[0131] Test verification

[0132] Taking the "silted pipe-manhole" test field of Huizhou University as an example, the intelligent early warning method of the water level of the silted pipe manhole based on the adaptive smoothing algorithm and the bidirectional gated circulation unit of the present invention is further explained:

[0133] Step 1: Establish the continuous medium equation, pipeline hydraulic transport equation, pipeline hydraulic transport equation, particle torque equation and particle contact equation, and on this basis, obtain the control equation of multiphase flow.

[0134] Step 2: Using the multiphase flow control equations in step 1, construct a spatiotemporal hydraulic response model of the "silted pipe-manhole", impose boundary constraints on the fluid phase and particle phase, and simultaneously transmit the spatiotemporal transport state information of the fluid phase in the pipe to perform fluid-solid coupling (CFD-DEM) numerical simulation of the silted pipe. Use the spatiotemporal hydraulic response model of the "silted pipe-manhole" to configure the boundary constraints and spatiotemporal transport state information in the fluid-solid coupling (CFD-DEM) numerical simulation process.

[0135] Step 3: Based on the numerical simulation results of fluid-solid coupling of the silted pipe in step 2, combined with Newton's second law and the continuous medium equation and particle collision equation in step 1, a high-resolution numerical simulation model of "silted pipe-manhole" is constructed. On this basis, the relationship between the flow characteristic index of the silted pipe and the siltation state and the manhole water level is analyzed, and the high-resolution numerical simulation model of "silted pipe-manhole" outputs the flow velocity / flow sample set of the pipe inlet and outlet with the manhole water level label.

[0136] Step 4: Extract features from the pipeline inlet and outlet velocity / flow sample set with manhole water level labels in step 3, establish an adaptive smoothing algorithm to detect missing data and heterogeneous data, and reconstruct the spatiotemporal velocity field and manhole water level of the silted pipeline during the training period.

[0137] Step 5: Using the spatiotemporal velocity field of the silted pipe and the manhole water level obtained in step 4 as the training set, validation set and test set, an intelligent early warning model for manhole water level is established using a bidirectional gated recurrent unit (TranGRU) based on a dynamic transformation encoder.

[0138] Step 6: Taking the test sets of three different pipe siltation conditions collected at the "Silted Pipe-Manhole" test site of Huizhou University as an example, the performance of the manhole water level intelligent early warning model in step 5 is statistically shown in Table 1.

[0139] Table 1 Evaluation results of the manhole water level warning accuracy of the model under three different pipeline siltation conditions

[0140]

[0141] Compared with the existing technology, the present invention closely combines a high-resolution fluid-solid coupling model with a neural network, so that it can provide more accurate and timely early warning of the manhole water level in the silted pipe, and provide technical support for water level alarm diagnosis in urban flood scenarios.

[0142] 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 protection scope of the present invention.

Claims

1. An intelligent early warning method for silted manhole water level based on adaptive smoothing algorithm and bidirectional gated circulation unit, It is characterized in that The steps include: S1. Establish the continuous medium equation, pipeline hydraulic transport equation, pipeline hydraulic transport equation, particle torque equation and particle collision equation to obtain the multiphase flow control equation; S2. Using the multiphase flow control equations in step S1, a spatiotemporal hydraulic response model of "deposition pipeline-manhole" is constructed, boundary constraints are imposed on the fluid phase and the particle phase, and the spatiotemporal transport state information of the fluid phase in the pipe is transmitted at the same time, and the fluid-solid coupling CFD-DEM numerical simulation of the deposition pipeline is performed. The spatiotemporal hydraulic response model of "deposition pipeline-manhole" is used to configure the boundary constraints and spatiotemporal transport state information in the fluid-solid coupling CFD-DEM numerical simulation process; S3. Based on the results of the fluid-solid coupling numerical simulation of the silted pipeline, combined with Newton's second law and the continuous medium equation and particle collision equation in step S1, a high-resolution numerical simulation model of "silted pipeline-manhole" is constructed, and the high-resolution numerical simulation model of "silted pipeline-manhole" outputs a sample set of flow velocity / flow at the inlet and outlet of the pipeline with the manhole water level label; S4, extract features from the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, establish an adaptive smoothing algorithm to detect missing data and heterogeneous data, and reconstruct the spatiotemporal flow velocity field and manhole water level of the silted pipeline within the training period; S5. Using the spatiotemporal velocity field of the silted pipe and the manhole water level obtained in step S4 as the training set, validation set and test set, a bidirectional gated recurrent unit TranGRU ​​based on a dynamic transformation encoder is used to establish an intelligent early warning model for the manhole water level, and the intelligent early warning results for the manhole water level are output through the model.

2. According to claim 1, the intelligent early warning method for water level of silted manhole based on adaptive smoothing algorithm and bidirectional gated circulation unit, Features: In step S2, the specific steps include: Step S21: Based on the principle of constant water volume, the law of conservation of momentum and the law of conservation of energy, the hydraulic response characteristic equation of "sedimentation pipeline-manhole" is established, which is formulated as follows: Among them, Q in and Q out They represent the flow rates at the inlet and outlet of the pipeline at the current moment, t is the instantaneous moment, and v in and v out represents the flow rate at the inlet and outlet of the pipe, θ represents the inclination of the pipe, Δt represents an infinitesimal time interval, l s and h s are the length and thickness of the pipeline siltation, ρ w represents the water flow density in the pipe, g represents the acceleration due to gravity, G in represents the gravity of the instantaneous flow at the pipe inlet and G in =ρ w Q in gΔt; and They represent the flow rate and velocity at the outlet of the previous pipe respectively. represents the water level of the manhole at t+Δt, Indicates the water level of the manhole at the front end of the pipeline, L p is the length of the pipeline, η is the kinetic energy conversion coefficient of the non-deposition section, r m Represents the radius of the manhole; Step S22: Based on the hydraulic response characteristic equation of the "deposition pipeline-manhole" in step S21, a lateral vibration differential equation of the hydraulic transport of the "deposition pipeline-manhole" under the excitation of the basic acceleration is constructed, which is formulated as follows: Where E is the elastic modulus, I is the moment of inertia of the pipeline interface, x is the horizontal coordinate variable parallel to the horizontal plane, w(x,t) and w r (x, t) represent the absolute displacement and relative base displacement of the pipe flow, respectively, w g (t) represents the basic displacement of the pipe flow and satisfies w(x,t) = w r (x,t)+w g (t), p in is the fluid pressure at the centerline of the pipeline inlet, S f is the cross-sectional area of ​​the fluid, m p is the mass of the pipe per unit length, m f is the fluid mass per unit length, c is the structural damping coefficient, r p is the radius of the stormwater pipe; Step S23: using the hydraulic response characteristic equation of the "deposition pipeline-manhole" in step S21 and the lateral vibration differential equation of the hydraulic transport of the "deposition pipeline-manhole" under the excitation of the basic acceleration in step S22, dynamically simulate the coupling relationship between the siltation length, siltation thickness, pipeline water flow velocity and flow rate, and then establish the "deposition pipeline-manhole" time-space scale hydraulic response model; Step S24: construct a "silted pipe-manhole" fluid-solid coupling CFD-DEM numerical simulation model, and use the "silted pipe-manhole" spatiotemporal scale hydraulic response model in step S23 to configure the boundary constraints and spatiotemporal transport state information in the "silted pipe-manhole" fluid-solid coupling CFD-DEM numerical simulation model.

3. According to claim 1, the intelligent early warning method for water level of silted manhole based on adaptive smoothing algorithm and bidirectional gated circulation unit, Features: In step S3, the specific steps include: Step S31: using Newton's second law and the continuous medium equation and particle collision equation in step S1, determine the hydraulic parameters of the rainwater pipe; Step S32: Using the hydraulic parameters of the rainwater pipe obtained in step S31 as input items of the flow-solid coupling of the rainwater pipe, a high-resolution "silted pipe-manhole" flow-solid coupling CFD-DEM numerical simulation model is constructed; Step S33: Use the high-resolution "silted pipe-manhole" fluid-solid coupling CFD-DEM numerical simulation model in step S32 to output the pipe inlet and outlet velocity / flow sample set with the manhole water level label.

4. According to claim 1, the intelligent early warning method for water level of silted manhole based on adaptive smoothing algorithm and bidirectional gated circulation unit, Features: In step S4, the specific steps include: Step S41: using a multi-layer autoencoder stack to learn the deep features of the data in the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label, adding sparsity constraints, mining the useful structure when there are more hidden neurons, and extracting the feature information of the data in the flow velocity / flow sample set at the inlet and outlet of the pipeline with the manhole water level label; Step S42: Based on the characteristic information of the pipeline inlet and outlet velocity / flow sample set data with the manhole water level label extracted in step S41, a three-dimensional nonlinear spatiotemporal interpolation method is used to estimate the spatiotemporal velocity field and manhole water level of the silted pipeline, and a continuous function V of the local average velocity at time t and position x is constructed. free (t,x) and V cong (t, x); Based on the discrete convolution of the kernel containing all observed data points, two smooth velocity fields in the free flow direction and the congestion flow direction are given respectively, which are formulated as follows: Among them, V free (t,x) and V cong (t,x) is the continuous interpolation speed function based on the existing i-th data point, t i and x i are the timestamp and position corresponding to the i-th data point, c free represents the propagation velocity in the free stream direction, c cong represents the congestion propagation speed in the congestion flow direction; The kernel function φ is given by the following symmetric exponential form: Where exp is an exponential function, τ and δ are specified constants; the exponential function smoothes temporal fluctuations on a scale smaller than τ, and smoothes spatial fluctuations on a scale smaller than δ; Step S43: V free (t,x) and V cong (t, x) superposition, the adaptive smoothing method ASM is used to calculate the velocity V(t, x) of the point to be estimated, and the formula is as follows: V(t,x)=ω(t,x)V cong (t,x)+[1-ω(t,x)]V free (t,x) Among them, ω(t,x) is the weighted function of congestion and free-flow traffic states, which is represented by the following Sigmoid nonlinear function: Among them, tanh represents the hyperbolic tangent function, V c represents the critical threshold between free flow and congested flow, ΔV is the bandwidth for the transition from free flow state to congested flow state; Step S44: The obtained reconstructed spatiotemporal velocity field of the silted pipe within the training period is used as the data source k and input into the extended generalized Treiber-Helbing filter to detect the data missing, noise reduction and data fusion problems of heterogeneous data caused by the detector failure in the data source k. The spatiotemporal velocity field z(t,x) of the silted pipe output by the extended generalized Treiber-Helbing filter is formulated as follows: Among them, α (k) (t,x) represents a specific data source from data source k at the spatiotemporal location (t,x), is the weight function corresponding to the data source k and the measured data point i; Weight function This reflects that data from different data sources will result in different spatial and temporal differences between congestion and free flow conditions. It is expressed as: in, represents the spatiotemporal coordinates of the spatiotemporal data point i of the kth data source, ω (k) (t,x) represents the weighted function of the congestion and free-flow traffic status of the k-th data source, φ represents the generalized weight function, and represent the timestamp and location of the spatiotemporal data point i of the kth data source, c cong Represents a weight constant.

5. According to claim 1, the intelligent early warning method for water level of silted manhole based on adaptive smoothing algorithm and bidirectional gated circulation unit, Features: In step S5, the specific steps include: Step S51: Integrate the bidirectional gated recurrent unit BiGRU into the dynamic transformation encoder Transformer, and use the self-attention mechanism to enhance the feature search ability of the bidirectional gated recurrent unit, thereby constructing a bidirectional gated recurrent unit TranGRU ​​based on the dynamic transformation encoder and the self-attention mechanism; Step S52: Using the spatiotemporal velocity field of the silted pipeline and the manhole water level obtained in step S4 as the training set, validation set and test set, extract the local and global features of the flow velocity / flow time series at the pipeline inlet and outlet and the manhole water level height, collaboratively learn the local flow velocity information, siltation degree and manhole water level of the pipeline, and calibrate the hyperparameters in the bidirectional gated recurrent unit based on the dynamic conversion encoder; Step S53: configure the hyperparameters calibrated in step S52 into a bidirectional gated recurrent unit based on a dynamic conversion encoder, thereby obtaining an intelligent warning model for manhole water level, and outputting the intelligent warning result for manhole water level through the model.

6. According to claim 5, the intelligent early warning method for water level of silted manhole based on adaptive smoothing algorithm and bidirectional gated circulation unit, Features: In step S51, the bidirectional gated recurrent unit BiGRU is integrated into the dynamic transformation encoder Transformer, and the self-attention mechanism is used to enhance the feature search capability of the bidirectional gated recurrent unit, thereby constructing a bidirectional gated recurrent unit TranGRU ​​based on the dynamic transformation encoder and the self-attention mechanism. The specific steps include the following: Step S511: introducing a multi-layer dynamic transformation encoder Transformer, wherein the first layer of the dynamic transformation encoder Transformer receives the flow velocity / flow rate time series of the pipeline inlet and outlet as input; Step S512: embedding a bidirectional gated recurrent unit BiGRU and a self-attention mechanism in the dynamic transformation encoder Transformer of step S511. The first layer of the dynamic transformation encoder Transformer in step S511 is responsible for receiving input information and performing a normalization operation, and then inputting the information into the bidirectional gated recurrent unit neural network and the self-attention neural network respectively. Then, the bidirectional gated recurrent unit BiGRU encodes the local features of the flow velocity / flow rate time series at the inlet and outlet of the pipeline and the water level height of the manhole, and the self-attention mechanism encodes the global features of the flow velocity / flow rate time series at the inlet and outlet of the pipeline and the water level height of the manhole; Step S513: Add a preprocessing module at the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512, add a Softmax function gating mechanism at the output position, adaptively adjust the ratio of the acquired local features and global features in the encoding state, and output the intelligent warning result of the manhole water level; Step S514: Take the flow velocity / flow time series at the inlet and outlet of the pipeline under different siltation intensities and the water level height of the manhole as the training sample set, and obtain the hyperparameters of the bidirectional gated recurrent unit BiGRU and the self-attention mechanism embedded in the dynamic transformation encoder Transformer in step S512 through training. Substitute the hyperparameters into the bidirectional gated recurrent unit BiGRU and the self-attention mechanism embedded in the dynamic transformation encoder Transformer to obtain the bidirectional gated recurrent unit TranGRU ​​based on the dynamic transformation encoder and the self-attention mechanism, and use the Softmax function gating mechanism in step S513 as the final output unit.

7. According to claim 6, the intelligent early warning method for water level of silted manholes based on adaptive smoothing algorithm and bidirectional gated circulation unit, Features: In step S513, a preprocessing module is added to the input position of the bidirectional gated recurrent unit and the self-attention mechanism, a Softmax function gating mechanism is added to the output position, and the intelligent warning result of the manhole water level is output. The specific steps include the following: Step S5131: For the pipeline inlet and outlet flow velocity / flow time series X, the preprocessing module added to the input position of the bidirectional gated recurrent unit and the self-attention mechanism in step S512 receives the input and performs preprocessing operations to obtain a hidden state layer Z of the same length i ; The specific calculation process is as follows: Z i =SA(W Q LN(X i ),W K LN(X i ),W V LN(X i ))+X i Among them, i is the time series number, Z i represents the hidden state of the current layer after self-attention network encoding, X i represents the flow velocity / flow time series data of the inlet and outlet of the pipeline numbered i, LN(·) represents the normalization operation, SA(·) represents the multi-head self-attention neural network sublayer; W Q , W K and W V The state input parameters representing the query, key, and value of self-attention respectively; Z i Input into the feedforward neural network sublayer of the preprocessing module to obtain the hidden state H i , the calculation method is as follows: H i =FFN(LN(Z i ))+Z i Among them, FFN(·) represents the feedforward neural network sublayer, and the residual connection is applied to the feedforward neural network sublayer to alleviate the gradient problem; Step S5132: Z output by the preprocessing module in step S5131 i Input to the bidirectional gated recurrent unit and self-attention mechanism, the specific processing process is as follows: in, represents the output state of the self-attention mechanism, Represents the output state of the bidirectional gated recurrent unit; Step S5133: Using the and It takes the data as input and outputs the intelligent warning result of manhole water level through the Softmax function gating mechanism.

Images