Central heating network pipe pressure monitoring method and system based on artificial intelligence

Abstract

The invention discloses a centralized heat supply network pipe pressure monitoring method and system based on artificial intelligence, and relates to the technical field of heat supply system monitoring, and the method comprises the steps: firstly laying optical fibers and point pressure sensors on a network pipe, and collecting acoustic strain and pressure data; constructing a digital twin grid through same-frequency alignment and Delaunay triangulation, and reconstructing a continuous pressure field; calculating the residual error between the reconstruction pressure and the measured value at the monitoring node; extracting multi-scale residual features to form a feature matrix based on a wavelet operator of a pipe network topological structure map; inputting the features into a pre-trained spectrogram wavelet neural network, and outputting a normalized abnormal score; and finally positioning a fault pipe section according to the score and performing graded warning. Physical modeling and deep learning are fused in the whole process, closed-loop monitoring from data collection to fault diagnosis is achieved, and the precision and timeliness of pipe network anomaly detection are remarkably improved. Manual detection is not needed, and the monitoring process is more flexible.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of heat supply system monitoring, in particular to a centralized heat supply network pipe pressure monitoring method and system based on artificial intelligence. BACKGROUND

[0002] The current heat supply company heat generation pipe network covers a wide area and has many nodes, and in operation, the collaborative working conditions of heat supply load and power generation equipment need to be considered. Traditional pressure regulation relies on manual regular inspection of pressure gauges and manual adjustment of valve opening, which is limited by inspection frequency and personnel experience difference, and it is difficult to respond to the dynamic changes of pipe network pressure in real time. Especially during the winter heating peak period or when the power generation load fluctuates, the complex layout of the pipe network with many branches and large differences in pipe diameters is easy to cause local pressure over-limiting or under-pressure, resulting in insufficient heat supply to remote users, heat exchanger station equipment cavitation, and even safety hazards such as pipe connection leakage.

[0003] The existing control system generally adopts a single-point pressure monitoring and independent device control mode, and lacks linkage analysis and collaborative adjustment mechanism of the whole network pressure data. The pressure sensor is sparsely deployed and the communication is lagging, and the actuator (such as a regulating valve and a circulating pump) can only be started and stopped according to the local set value, and cannot be globally optimized based on the overall pressure distribution of the pipe network. When the heat source end power generation load changes or the heat user end flow suddenly changes, the pressure imbalance problem often needs to be manually checked and repaired for several hours, during which there are multiple disadvantages such as energy invalid loss, equipment wear and tear, and rising user complaint rate. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a centralized heat supply network pipe pressure monitoring method based on artificial intelligence to solve the problems of sparse deployment of pressure sensors and communication lag in the existing point pressure monitoring technology, and to solve the problem of long identification time relying more on manual checking.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a centralized heat supply network pipe pressure monitoring method based on artificial intelligence, which comprises collecting acoustic echo, micro-strain curve and measured pressure value data to construct an acoustic strain pressure data set.

[0008] The acoustic strain pressure data set is processed for same frequency alignment, the acoustic strain pressure data set is mapped to a digital twin geometric grid constructed by using triangular subdivision, and a continuous spatial pressure field is reconstructed.

[0009] A corresponding tetrahedral unit is located for each predefined monitoring node in a continuous spatial pressure field, the barycentric coordinates of the tetrahedral unit are calculated, and the reconstructed pressure values of the tetrahedral vertices are linearly interpolated to obtain the reconstructed pressure values of the nodes, and the pressure residual values of the nodes are obtained in combination with the measured pressure values.

[0010] The pressure residual values are convoluted and projected on each scale to extract multi-scale residual signal features, and the projection results of each scale are spliced into a graph wavelet multi-scale feature matrix in the order of the nodes.

[0011] The spectral graph wavelet neural network is trained based on historical data, the original score is obtained by inputting the graph wavelet multi-scale feature matrix, and the normalized abnormal score is obtained by normalizing the original score through a Sigmoid function.

[0012] Based on the normalized abnormal score, the fault pipe section is screened and located based on topology, and risk classification and warning are performed.

[0013] As a preferred scheme of the centralized heating network pressure monitoring method based on artificial intelligence, the acoustic echo, micro-strain curve and measured pressure value data are collected to construct an acoustic strain pressure dataset, including laying a single-mode optical fiber along the outer wall of the trunk and branch pipe sections, and arranging detection units and point pressure sensors at both ends of the single-mode optical fiber and the nodes.

[0014] Periodic scanning is started to obtain acoustic echoes, micro-strain curves and measured pressure values with unified time sequence marks, and an acoustic strain pressure dataset is constructed.

[0015] As a preferred scheme of the centralized heating network pressure monitoring method based on artificial intelligence, the acoustic strain pressure dataset is processed by same frequency alignment, the acoustic strain pressure dataset is mapped to a digital twin geometric grid constructed by triangular subdivision, and a continuous spatial pressure field is reconstructed, including denoising and same frequency processing of the acoustic echoes and micro-strain curves of the acoustic strain pressure dataset, constructing a digital twin geometric grid by Delaunay triangular subdivision, mapping the denoised and corrected acoustic echoes and micro-strain curves and the measured pressure values to the grid nodes to obtain discrete pressure and strain field quantities, reconstructing based on the weighted joint least squares regularization method of the discrete pressure and strain field quantities, obtaining the continuous pressure values of each node on the grid by numerical iteration, and reconstructing a continuous spatial pressure field on the digital twin geometric grid.

[0016] As a preferred scheme of the centralized heating network pipe pressure monitoring method based on artificial intelligence provided in the application, wherein: the solving of the gravity center coordinates of the tetrahedron unit and the linear interpolation of the reconstructed pressure values of the tetrahedron vertexes to obtain the reconstructed pressure values of the nodes comprises: locating the tetrahedron unit to which each monitoring node belongs through the octree index and the Delaunay adjacency query, then solving the gravity center coordinates of the tetrahedron unit, and combining the micro-strain differences of the monitoring points and the vertexes to perform weighted linear interpolation on the pressure values of the tetrahedron vertexes, to obtain the weighted reconstructed pressure values of each monitoring node, and finally subtracting the weighted reconstructed pressure values from the corresponding measured pressures point by point and collecting them in the node order to generate the pressure residual of the monitoring node.

[0017] As a preferred scheme of the centralized heating network pipe pressure monitoring method based on artificial intelligence provided in the application, wherein: the convolution projection of the pressure residual on each scale to extract the multi-scale residual signal features comprises: constructing a weighted undirected graph based on the pipe resistance inversely proportional between nodes, constructing a graph wavelet operator by adopting a Gaussian spectral wavelet function on a plurality of preset scales and through a Chebyshev polynomial approximation, and performing convolution projection on the residual vector on each scale to extract the multi-scale residual signal features, and splicing the projection results of each scale in the node order to obtain a graph wavelet multi-scale feature matrix.

[0018] As a preferred scheme of the centralized heating network pipe pressure monitoring method based on artificial intelligence provided in the application, wherein: the input of the graph wavelet multi-scale feature matrix to obtain the original score, and the normalization of the original score to obtain the normalized abnormal score through a Sigmoid function comprises: inputting the graph wavelet multi-scale feature matrix through a spectral graph wavelet neural network trained by historical data, performing two-layer graph convolution and nonlinear mapping on the input features by the spectral graph wavelet neural network during online operation, and outputting the original score, and then normalizing the original score through a Sigmoid function to obtain the normalized abnormal score.

[0019] As a preferred scheme of the centralized heating network pipe pressure monitoring method based on artificial intelligence provided in the application, wherein:

[0020] In a second aspect, the application provides a centralized heating network pipe pressure monitoring system based on artificial intelligence, comprising: a data acquisition module, a grid reconstruction module, a residual calculation module, a feature extraction module, an abnormal score module, and an alarm module.

[0021] The data acquisition module is used for acquiring acoustic echo, micro-strain curve and measured pressure value data to construct an acoustic strain pressure data set.

[0022] The grid reconstruction module is used for same-frequency alignment processing of the acoustic strain pressure data set, mapping the acoustic strain pressure data set to a digital twin geometric grid constructed by triangulation, and reconstructing a continuous spatial pressure field.

[0023] The residual calculation module is used for locating a corresponding tetrahedral unit for each pre-defined monitoring node in the continuous spatial pressure field, calculating the barycentric coordinates of the tetrahedral unit, and performing linear interpolation on the reconstructed pressure values of the tetrahedral vertices to obtain the reconstructed pressure values of the nodes, and combining the measured pressure values to obtain the pressure residuals of the nodes.

[0024] The feature extraction module is used for approximately constructing a graph wavelet operator, performing convolution projection on the residual vector at each scale to extract multi-scale residual signal features, and splicing the projection results at each scale into a graph wavelet multi-scale feature matrix according to the node order.

[0025] The anomaly scoring module is used for training a spectral graph wavelet neural network based on historical data, inputting the graph wavelet multi-scale feature matrix to obtain an original score, and normalizing the original score by a Sigmoid function to obtain a normalized anomaly score.

[0026] The alarm module is used for screening and locating a fault pipe section based on the normalized anomaly score, and performing risk classification and alarming.

[0027] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, any step of the artificial intelligence-based central heating network pressure monitoring method according to the first aspect of the present application is implemented.

[0028] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, any step of the artificial intelligence-based central heating network pressure monitoring method according to the first aspect of the present application is implemented.

[0029] The present application has the following beneficial effects: the present application reconstructs a continuous spatial pressure field by mapping the acoustic strain pressure data set to a digital twin geometric grid based on triangulation, realizes accurate characterization of the pressure distribution of the pipe network, provides a reference for subsequent residual analysis, and realizes the effect of high-resolution dynamic monitoring. By approximately constructing a graph wavelet operator, the residual vector is convolved and projected at multiple scales to splice a graph wavelet multi-scale feature matrix, multi-scale anomaly feature extraction is realized, multi-dimensional input is provided for anomaly scoring and pipe section positioning, and the entire detection process is more sensitive and has better robustness. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0031] Fig. 1 Flowchart of the method for monitoring pressure of central heating network based on artificial intelligence.

[0032] Fig. 2 Schematic diagram of the system for monitoring pressure of central heating network based on artificial intelligence.

[0033] Fig. 3 Flowchart of residual calculation of the method for monitoring pressure of central heating network based on artificial intelligence.

[0034] Fig. 4 Flowchart of anomaly detection of the method for monitoring pressure of central heating network based on artificial intelligence. DETAILED DESCRIPTION

[0035] In order to make the above objectives, features and advantages of the present application more apparent and comprehensible, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0036] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0037] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0038] REFERENCE Figs. 1-4 For one embodiment of the present application, the embodiment provides a method for monitoring pressure of central heating network based on artificial intelligence, comprising the following steps:

[0039] S1: Collecting acoustic echo, micro-strain curve and measured pressure value data to construct acoustic strain pressure data set.

[0040] Specifically, collecting acoustic echo, micro-strain curve and measured pressure value data to construct an acoustic strain pressure dataset includes laying single-mode optical fibers along the outer wall of the trunk and branch pipe sections, and arranging detection units and point pressure sensors at both ends of the single-mode optical fibers and nodes.

[0041] Starting a periodic scan to obtain acoustic echo, micro-strain curve and measured pressure value with unified time stamp, and constructing an acoustic strain pressure dataset.

[0042] Further, the detection unit is usually a C-OTDR (Coherent Optical Time Domain Reflectometry). The C-OTDR can provide spatially resolved vibration and strain information along the line, and can reflect dynamic changes through time series analysis. It is a key means for real-time monitoring of the entire pipeline "zero blind area".

[0043] It should be noted that the C-OTDR transmits laser pulses to each measurement point of the optical fiber in cycles, and receives scattered echo signals along the fiber length. The micro-strain curve is calculated by a difference operator. The micro-strain curve can reflect the small deformation of the pipeline under thermal expansion or external impact. At the same time, the pressure sensor on site outputs the measured pressure value with the same time sequence. After filtering, denoising and time delay correction, the acoustic vibration waveform, the micro-strain curve and the measured pressure value are uniformly stamped with PTP time stamp, and are combined into an acoustic strain pressure dataset.

[0044] S2: Same frequency alignment processing of the acoustic strain pressure dataset, mapping the acoustic strain pressure dataset to a digital twin geometric grid constructed by Delaunay triangulation, and reconstructing a continuous spatial pressure field.

[0045] Specifically, the same frequency alignment processing of the acoustic strain pressure dataset, mapping the acoustic strain pressure dataset to a digital twin geometric grid constructed by Delaunay triangulation, and reconstructing a continuous spatial pressure field includes denoising and same frequency processing of the acoustic echo and micro-strain curve of the acoustic strain pressure dataset, constructing a digital twin geometric grid by Delaunay triangulation, mapping the denoised and corrected acoustic echo and micro-strain curve and measured pressure value to the grid nodes to obtain discrete pressure and strain field quantities, reconstructing based on the weighted joint least squares regularization method of the discrete pressure and strain field quantities, obtaining the continuous pressure value of each node on the grid by numerical iteration, and reconstructing a continuous spatial pressure field on the digital twin geometric grid.

[0046] Further, the acoustic strain pressure data set is used to denoise and process the acoustic echo and micro-strain curve. The denoising is divided into two stages. The first stage uses a fourth-order Butterworth band-pass filter, and the second stage uses a Daubechies4 wavelet basis to denoise. After denoising, the same frequency processing is performed. First, the time sequence of different modal data is aligned to eliminate the clock deviation caused by the acquisition equipment and the transmission link. For each pair of acoustic echo and micro-strain curve, the cross-correlation function is calculated and the maximum correlation delay is found. Then, the wavelet denoising signal is shifted as a whole to make the two accurately coincide on the time axis. At the same time, all point pressure sensor data (the coordinates x j and the measured pressure value) are also calibrated to the same time reference by the PTP clock synchronization subsystem, and finally consistent multi-modal time sequence data is formed.

[0047] In the micro-strain curve after synchronization correction, the curve value of each sampling point reflects the strain accumulation of the pipeline position. In order to depict the instantaneous dynamic change, the local deformation rate ε j is used, and the difference in micro-strain between adjacent two time points is taken as the deformation rate of the node, which quantifies the instantaneous change rate of the small deformation of the pipeline.

[0048] Based on the BIM pipe network three-dimensional grid, a digital twin geometric grid is constructed by Delaunay triangulation. The coordinates x j , the measured pressure value P j and the deformation rate ε j of each sampling point are used to calculate the spatial distance between the monitoring point node and all node coordinates in the digital twin geometric grid, M is the total number of grid nodes, and k is the grid node. The grid node with the smallest distance is found. Then the measured pressure value of the sampling point with the smallest distance to each digital twin geometric grid node is constructed as the pressure value set of the grid node, and the same method is used to construct the monitoring point deformation rate set of the grid node. When the same grid node corresponds to multiple sampling points, the pressure values of all sampling points are averaged to obtain the grid pressure field vector b of the grid node, which is represented as:

[0049] b=[b1,…,b M ] T ;

[0050] Where T is the transpose.

[0051] Correspondingly, the strain values of all sampling points are also averaged to obtain the micro-strain field vector f of the grid node, which is represented as:

[0052] f=[f1,…,f M ] T ;

[0053] Subsequently, the joint least square objective function J(P) with micro-strain and pressure deviation weighting is constructed, expressed as:

[0054]

[0055] where P(x k ) is the value of the continuous pressure field at the grid node coordinate x k , b k is the grid pressure field vector of the kth grid node, w b (k) is the fitting item weight, and the weight function is defined as w b (k) = 1 + γ|f k |, γ is the strain weight coefficient, γ > 0, and δ is the pressure deviation weight coefficient, δ ≤ 0.5, is the pressure gradient vector at x k , the node pressure gradient is calculated by the central difference method, the regularization coefficient λ is determined by the L-curve method to balance between fitting accuracy and gradient consistency, f k is the micro-strain field vector of the kth node. The derivative of J(P) with respect to the pressure vector p = [P(x1), …, P(x M )] T is taken and set to zero, and the weighted normal equation is obtained, expressed as:

[0056] (A T W b A+λD T W f D)p=A T W b b+λD T W f f;

[0057] wherein is the measurement point mapping matrix, representing the correspondence between the sampling points and the grid nodes, is a real matrix of L columns and L rows, is the gradient difference matrix, used to approximately calculate the node gradient, and the three-axis three-direction difference of the pressure of each node is constructed into three rows of equations, totaling L rows, and the diagonal weight matrix W b = diag(w b (1), …, w b (L)) and W f = diag(w f (1), …, w f (L)) correspond to the weighting of the fitting item and the gradient item, respectively. Finally, the conjugate gradient iteration method is used for solving, and the iteration is stopped when the residual norm is lower than the value of the order of magnitude of the standard deviation of the field pressure and micro-strain measurement noise, and the reconstructed pressure value of the solution vector is output with the grid node coordinates {x k} to complete the generation of the reconstructed pressure field.

[0058] S3: Locating the corresponding tetrahedral unit for each pre-defined monitoring node in the continuous spatial pressure field, calculating the barycentric coordinates of the tetrahedral unit and performing linear interpolation on the reconstructed pressure values of the tetrahedral vertices to obtain the reconstructed pressure values of the nodes, and combining the measured pressure values to obtain the pressure residual of the nodes.

[0059] Specifically, the acoustic strain pressure dataset is processed for same frequency alignment, the acoustic strain pressure dataset is mapped to the digital twin geometric grid constructed by Delaunay triangulation, and the continuous spatial pressure field is reconstructed, including denoising and same frequency processing of the acoustic echo and micro-strain curve of the acoustic strain pressure dataset, constructing the digital twin geometric grid by Delaunay triangulation, and mapping the denoised and corrected acoustic echo and micro-strain curve and the measured pressure value to the grid node to obtain the discrete pressure and strain field.

[0060] Based on the weighted joint least squares regularization method of the discrete pressure and strain field, the continuous pressure value of each node on the grid is obtained by numerical iteration, and the continuous spatial pressure field is reconstructed on the digital twin geometric grid.

[0061] Further, based on the right-hand Cartesian coordinate system (X axis points to the east, Y axis points to the north, and Z axis is the elevation), the reconstructed pressure field and the measured pressure value are input to calculate the pressure residual of the node.

[0062] First, in all grid node coordinates {x k} and all monitoring point node coordinates {x j}, the octree acceleration index is combined with the Delaunay adjacency query to locate each monitoring point node x j and the tetrahedral unit to which it belongs.

[0063] If the query is not hit or the unit geometry is degraded, switch to the nearest neighbor mapping to ensure that each x j has a mapping unit. Secondly, the barycentric coordinate matrix is constructed, which is represented as:

[0064]

[0065] Where V j is the barycentric coordinate coefficient matrix, l j is the target coordinate vector, and represent the X, Y, Z coordinate components of the first vertex of the tetrahedral unit of the network node x j , x j , y j,z j is the monitoring point node x j X, Y, and Z coordinates in the same three-dimensional Cartesian coordinate system.

[0066] Solve M by singular value decomposition j λ j =l j Get the barycentric coordinates, barycentric coordinate vector λ j =[λ1,…,λ4] T Then, for each vertex of the tetrahedron element, a weighting coefficient w based on the difference in local deformation rate between the monitoring point and the vertex is used. n,j , expressed as:

[0067]

[0068] Among them, ε j and are the deformation rate of the monitoring point and the local deformation rate of the vertex, β>0 is the microstrain weight coefficient, λ n (x j ) represents the weight from the monitoring point to each vertex of the tetrahedron unit. n,j Reconstructed pressure values ​​for tetrahedron vertices Perform normalized interpolation, expressed as:

[0069]

[0070] Among them, P recon (x j ,t) is x j The weighted reconstructed pressure is realized to achieve adaptive amplification response to the spatial gradient change of the pressure field.

[0071] Finally, the difference between the weighted reconstructed pressure and the measured pressure is calculated point by point, which is expressed as

[0072] r j (t) = P(x j ,t)-P recon (x j ,t);

[0073] Among them, r j (t) is the pressure residual of the jth monitoring node at time t, P(x j ,t) is the measured pressure value of the monitoring point at time t.

[0074] All residuals are combined into a monitoring node pressure residual vector according to the monitoring node number.

[0075] S4: The wavelet operator is constructed approximately, the multi-scale residual signal features are extracted by convolution projection of the pressure residual at each scale, and the projection results of each scale are spliced into a graph wavelet multi-scale feature matrix in node order.

[0076] Specifically, the center of gravity coordinates of the tetrahedral unit is calculated, and the reconstructed pressure value of the tetrahedral vertex is linearly interpolated to obtain the reconstructed pressure value of the node, including locating the tetrahedral unit to which each monitoring node belongs through octree index and Delaunay adjacency query.

[0077] The center of gravity coordinates of the tetrahedral unit is calculated, and the pressure value of the tetrahedral vertex is weighted linearly interpolated in combination with the micro-strain difference between the monitoring point and the vertex to obtain the weighted reconstructed pressure value of each monitoring node.

[0078] The weighted reconstructed pressure value is subtracted from the corresponding measured pressure point by point, and the pressure residual of the monitoring node is generated in node order.

[0079] Further, first, an undirected weighted graph G(V, E) is constructed based on the inverse relationship of the pipe segment resistance coefficient, the node set V is the monitoring node, the edge weight w ij = 1 / R ij , wherein R ij is the pipe segment resistance between the monitoring nodes i and j, and i represents a monitoring node different from j. Then, the degree matrix D and the weight matrix W are calculated, and the normalized graph Laplacian operator is constructed, denoted as:

[0080]

[0081] wherein I is a unit matrix, W = [w ij ] is a weight matrix, and D is a degree matrix. I is a JxJ diagonal matrix, all elements on the diagonal are 1, and non-diagonal elements are 0, and J is the total number of monitoring nodes. W is a JxJ symmetric matrix. D is a JxJ diagonal matrix, and the diagonal elements in the row are equal to the sum of all elements of the corresponding row of W.

[0082] Then, the spectral domain scaling maps the spectral range to [-1, 1], denoted as:

[0083]

[0084] wherein λ max is the maximum eigenvalue of L norm estimated by the power method. Then, the adaptive spectral wavelet kernel function is defined as:

[0085] g s (λ) = (1 + η‖r(t)‖2)e -sλ ;

[0086] where s is the wavelet scale, ‖r(t)‖2 is the 2-norm of the residual vector, η > 0 is the global residual gain coefficient, and g is fitted by the least square method s (λ) to obtain the expansion coefficient α s,K , K is the Chebyshev polynomial approximation expansion to the order.

[0087] The Chebyshev polynomial approximation expansion to the order K (4 is taken in this embodiment) is adopted to construct the wavelet operator, which is expressed as:

[0088]

[0089] where T n is the Chebyshev polynomial basis. Then, the multi-scale projection of the residual vector is expressed as:

[0090] r s (t) = Ψ s r(t);

[0091] The local and global abnormal responses at each scale are obtained, respectively.

[0092] Finally, the projection results of all scales are horizontally spliced in the node order to form a graph wavelet multi-scale feature matrix R(t) with the dimension of node number x scale number, which is expressed as:

[0093]

[0094] S5: Training the graph wavelet neural network based on historical data, inputting the graph wavelet multi-scale feature matrix to obtain the original score, and normalizing the original score to obtain the normalized abnormal score through the Sigmoid function.

[0095] Specifically, inputting the graph wavelet multi-scale feature matrix to obtain the original score, and normalizing the original score to obtain the normalized abnormal score through the Sigmoid function includes inputting the graph wavelet multi-scale feature matrix through the graph wavelet neural network trained based on historical data, the graph wavelet neural network performing two-layer graph convolution and nonlinear mapping on the input feature and outputting the original score when running online, and then normalizing the original score through the Sigmoid function to obtain the normalized abnormal score.

[0096] Further, the historical operation and monitoring logs of the central heating network management are called to extract the manual inspection reports and alarm events of each monitoring node, including the alarm time and fault type, and these events are accurately aligned with the time series data collected by the sensor through PTP clock or timestamp. For each monitoring node j, the label y j is set to fault (i.e., y j = 1) in the time window corresponding to the alarm event, and the label is set to normal (i.e., y j = 0) for the node that does not trigger any alarm in the time window.

[0097] It should be noted that the existing spectral wavelet neural network (GW-GNN) model is used for real-time anomaly score calculation. The GW-GNN model uses historical operation data and fault labels for supervised training in the offline stage. The total number of training set samples N s Take 3000 (range 1000~5000), and divide it into 2400 training samples and 600 validation samples in the ratio of 8:2. Each sample is composed of the graph wavelet multi-scale feature matrix and the corresponding node fault label vector y=[y1,…,y J ] T Composition, label y j =1 indicates a fault at node j, y j =0 means normal. The training adopts the binary cross entropy loss function, which is expressed as:

[0098]

[0099] Among them, q j is the raw score output by the GW-GNN at node j. The optimization algorithm used was Adam, with a learning rate of 0.0005 (range 0.0001–0.001), 100 epochs (range 50–200), and a batch size of 32. After training, save the GW-GNN model parameters.

[0100] After offline training of the GW-GNN model, the spectral graph wavelet neural network (GW-GNN) was evaluated using an independent validation set. First, the true fault labels of the nodes in the validation set were matched to the raw scores output by the model. The true positive rate (recall) and true negative rate (specificity) were calculated at different score thresholds, and receiver operating characteristic (ROC) curves were plotted. Next, a high-risk threshold was selected as the score point where the specificity reached 95%. At this point, the GW-GNN model achieved a recall rate of 90% or higher while maintaining a false alarm rate of only 5%, effectively identifying the majority of faulty nodes. Next, the point on the same ROC curve where the recall and specificity curves intersected or minimized their difference was found. This score was used as the medium-risk threshold to achieve a balanced distinction between moderately abnormal nodes.

[0101] During online reasoning, R(t) is input into GW-GNN, and after two layers of spectral wavelet convolution and nonlinear activation operations, the original scores of each node {u j}. Subsequently, the original score is normalized by the Sigmoid function to obtain the normalized anomaly score o j , expressed as

[0102]

[0103] The scoring range is limited to 0 to 1.

[0104] To facilitate the hierarchical response of operation and maintenance, define the high-risk threshold T high = 0.8 and the medium-risk threshold T mid = 0.5, which is determined according to the ROC curve drawn by GW-GNN on the validation set, under the premise of ensuring the specificity of node failure of 95% and the recall rate of 90%. Normalize the abnormal score o j If greater than or equal to T high , it is determined that the node j is high-risk. If a j is between T mid and T high , it is determined to be medium-risk. If a j is lower than T mid , it is determined to be low-risk.

[0105] S6: Screening based on the normalized abnormal score and locating the fault pipe section through topology to perform risk classification alarm.

[0106] Specifically, screening based on the normalized abnormal score and locating the fault pipe section through topology to perform risk classification alarm includes screening high-risk nodes with the node abnormal score vector, performing node positioning combined with pipe network topology, classifying each section according to network type and scoring high, medium and low risk, and generating a hierarchical alarm work order.

[0107] Further, first, according to the preset high-risk threshold T high , high-risk nodes with abnormal scores greater than or equal to the high-risk threshold are screened from all monitored nodes to obtain a preliminary fault node set, and then based on the pipe network topology relationship, depth-first search is performed on these nodes to extract a preliminary fault node subset connected to each other, each preliminary fault node subset corresponds to a candidate range of a fault pipe section.

[0108] Next, the maximum normalized abnormal score in each fault pipe section subset is taken as the abnormal score of the pipe section, and the risk classification is performed combined with the network type identifier (main network, backwater trunk or branch network) of each node:

[0109] If there is a main network node in the fault pipe section subset and the abnormal score reaches the high-risk threshold, it is determined to be high-risk of the main network.

[0110] If all nodes are backwater trunks and the score is between the medium-risk and high-risk thresholds, it is determined to be medium-risk of backwater. The rest of the pipe sections are all determined to be low-risk of branch. Finally, corresponding alarm work orders are automatically generated according to different risk levels and sent to dispatching and maintenance terminals through the operation and maintenance platform, and the location of the fault pipe section is highlighted on the GIS map to guide the on-site personnel to respond quickly and complete accurate disposal.

[0111] The embodiment also provides a centralized heating network pipe pressure monitoring system based on artificial intelligence, comprising:

[0112] a data acquisition module, a grid reconstruction module, a residual calculation module, a feature extraction module, an anomaly scoring module, and an alarm module.

[0113] The data acquisition module is configured to acquire acoustic echo, micro-strain curve and measured pressure value data to construct an acoustic strain pressure dataset.

[0114] The grid reconstruction module is configured to perform same-frequency alignment processing on the acoustic strain pressure dataset, map the acoustic strain pressure dataset to a digital twin geometric grid constructed by using triangular partitioning, and reconstruct a continuous spatial pressure field.

[0115] The residual calculation module is configured to locate a corresponding tetrahedral unit for each predefined monitoring node in the continuous spatial pressure field, calculate the barycentric coordinates of the tetrahedral unit, perform linear interpolation on the reconstructed pressure values of the vertices of the tetrahedral unit to obtain a reconstructed pressure value of the node, and combine the measured pressure value to obtain a pressure residual of the node.

[0116] The feature extraction module is configured to approximately construct a graph wavelet operator, perform convolution projection on the residual vector at each scale to extract a multi-scale residual signal feature, and splice the projection results of each scale into a graph wavelet multi-scale feature matrix according to the node order.

[0117] The anomaly scoring module is configured to train a spectral graph wavelet neural network based on historical data, input the graph wavelet multi-scale feature matrix to obtain an original score, and normalize the original score by using a Sigmoid function to obtain a normalized anomaly score.

[0118] The alarm module is configured to filter and locate a fault pipe section based on the normalized anomaly score, and perform risk classification and alarm.

[0119] The embodiment also provides a computer device suitable for the centralized heating network pipe pressure monitoring method based on artificial intelligence, comprising a memory and a processor.

[0120] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0121] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for monitoring pressure of a central heating network based on artificial intelligence as described in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0122] In summary, the present application achieves the following effects: by processing the acoustic strain pressure data set in the same frequency, the data is mapped to the digital twin geometric grid based on triangulation to reconstruct the continuous space pressure field, which achieves accurate description of the pressure distribution of the pipe network and provides a benchmark for subsequent residual analysis, and achieves the effect of high-resolution dynamic monitoring. By approximately constructing the graph wavelet operator, the residual vector is convolved and projected on multiple scales and spliced into a graph wavelet multi-scale feature matrix, which achieves multi-scale anomaly feature extraction and provides multi-dimensional input for anomaly scoring and pipe segment positioning, making the entire detection process more sensitive and more robust.

[0123] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for monitoring central heating network pressure based on artificial intelligence, characterized by: include, Acoustic echo, microstrain curve and measured pressure value data are collected to construct an acoustic strain pressure data set; The acoustic strain and pressure data sets are aligned with the same frequency and mapped to the digital twin geometric grid constructed using triangulation to reconstruct a continuous spatial pressure field. In the continuous spatial pressure field, the corresponding tetrahedron element is located for each predefined monitoring node, the centroid coordinates of the tetrahedron element are solved, and the reconstructed pressure values ​​of the tetrahedron vertices are linearly interpolated to obtain the reconstructed pressure value of the node. The pressure residual of the node is obtained by combining the measured pressure value. Approximately construct a graph wavelet operator, perform convolution projection on the pressure residual at each scale to extract multi-scale residual signal features, and splice the projection results of each scale into a graph wavelet multi-scale feature matrix in node order; Based on historical data, the spectral wavelet neural network is trained, and the multi-scale feature matrix of the wavelet is input to obtain the raw score. The raw score is normalized by the Sigmoid function to obtain the normalized anomaly score. Based on the normalized anomaly score screening and topological location of the faulty pipe section, risk-graded alarms are issued.

2. The method for monitoring central heating network pressure based on artificial intelligence according to claim 1, characterized in that: The acquisition of acoustic echoes, microstrain curves, and measured pressure value data to construct an acoustic strain and pressure data set includes laying single-mode optical fibers along the outer walls of the pipeline network trunk and branch pipe sections, and arranging detection units and point pressure sensors at both ends and nodes of the single-mode optical fibers; Start periodic scanning to obtain acoustic echoes, microstrain curves and measured pressure values ​​with uniform time series markings, and construct an acoustic strain and pressure data set.

3. The method for monitoring central heating network pressure based on artificial intelligence according to claim 2, characterized in that: The said co-frequency alignment processing of the acoustic strain and pressure data set, mapping the acoustic strain and pressure data set to a digital twin geometric grid constructed by triangulation, and reconstructing a continuous spatial pressure field includes obtaining the acoustic strain and pressure data set, denoising and co-frequency processing the acoustic echo and microstrain curve, constructing the digital twin geometric grid by Delaunay triangulation, and mapping the denoised and corrected acoustic echo and microstrain curve and the measured pressure value to the grid nodes to obtain discrete pressure and strain field quantities; Reconstruction is performed based on the weighted joint least squares regularization method of discrete pressure and strain field quantities. The continuous pressure value of each node on the grid is obtained through numerical iterative solution, and the continuous spatial pressure field is reconstructed on the digital twin geometric grid.

4. The method for monitoring central heating network pressure based on artificial intelligence according to claim 3, characterized in that: The said calculating the centroid coordinates of the tetrahedron unit and performing linear interpolation on the reconstructed pressure values ​​of the tetrahedron vertices to obtain the reconstructed pressure value of the node includes locating the tetrahedron unit to which each monitoring node belongs through octree indexing and Delaunay adjacency query; The centroid coordinates of the tetrahedron unit are solved and the pressure values ​​of the tetrahedron vertices are weighted linearly interpolated based on the microstrain difference between the monitoring point and the vertex to obtain the weighted reconstructed pressure value of each monitoring node. The weighted reconstructed pressure value is subtracted from the corresponding measured pressure point by point and summarized in node order to generate the pressure residual of the monitoring node.

5. The method for monitoring central heating network pressure based on artificial intelligence according to claim 4, characterized in that: The method of performing convolution projection on the pressure residual at each scale to extract multi-scale residual signal features, and splicing the projection results of each scale into a graph wavelet multi-scale feature matrix in node order includes: constructing a weighted undirected graph based on the inverse ratio of pipe section resistance between nodes, using a Gaussian spectral wavelet function at multiple preset scales and constructing a graph wavelet operator through Chebyshev polynomial approximation, performing convolution projection on the residual vector at each scale to extract multi-scale residual signal features, and splicing the projection results of each scale in node order to obtain a graph wavelet multi-scale feature matrix.

6. The method for monitoring central heating network pressure based on artificial intelligence according to claim 5, characterized in that: The input graph wavelet multi-scale feature matrix is ​​used to obtain a raw score, and the raw score is normalized by a Sigmoid function to obtain a normalized anomaly score. The method includes inputting the graph wavelet multi-scale feature matrix into a spectral graph wavelet neural network trained with historical data, performing two-layer graph convolution and nonlinear mapping on the input features during online operation and outputting a raw score, which is then normalized by a Sigmoid function to obtain a normalized anomaly score.

7. The method for monitoring central heating network pressure based on artificial intelligence according to claim 6, characterized in that: The method of screening faulty pipe sections based on normalized anomaly scores and locating faulty pipe sections through topology to generate risk-graded alarms includes screening high-risk nodes using node anomaly score vectors, performing node location based on the pipe network topology, and grading each section into high, medium, and low risk categories according to network management type and score, thereby generating graded alarm work orders.

8. An artificial intelligence-based centralized heating network pressure monitoring system, based on the artificial intelligence-based centralized heating network pressure monitoring method according to any one of claims 1 to 7, characterized in that: Including data acquisition module, grid reconstruction module, residual calculation module, feature extraction module, anomaly scoring module, and alarm module; The data acquisition module is used to collect acoustic echoes, microstrain curves and measured pressure value data to construct an acoustic strain pressure data set; The grid reconstruction module is used to align the acoustic strain and pressure data sets with the same frequency, map the acoustic strain and pressure data sets to the digital twin geometric grid constructed by triangulation, and reconstruct a continuous spatial pressure field; The residual calculation module is used to locate the corresponding tetrahedral unit for each predefined monitoring node in the continuous spatial pressure field, solve the centroid coordinates of the tetrahedral unit and perform linear interpolation on the reconstructed pressure values ​​of the tetrahedral vertices to obtain the reconstructed pressure value of the node, and combine the measured pressure value to obtain the pressure residual of the node; The feature extraction module is used to approximately construct a graph wavelet operator, perform convolution projection on the residual vector at each scale to extract multi-scale residual signal features, and splice the projection results of each scale into a graph wavelet multi-scale feature matrix in node order; The anomaly scoring module is used to train a spectral wavelet neural network based on historical data, input a spectral wavelet multi-scale feature matrix to obtain a raw score, and normalize the raw score using a Sigmoid function to obtain a normalized anomaly score; The alarm module is used to screen the faulty pipe section based on the normalized anomaly score and locate the faulty pipe section through the topology to issue risk graded alarms.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based centralized heating network pressure monitoring method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based centralized heating network pressure monitoring method according to any one of claims 1 to 7 are implemented.

Images