A system and method for detecting leakage of urban sewage pipe network

Abstract

The present invention relates to the technical field of urban sewage pipe network detection, and discloses a system and method for detecting leakage in urban sewage pipe networks. The system includes a data acquisition module that uses multi-source sensors to collect real-time operating data such as pipe network pressure and flow; a feature extraction module that extracts spatiotemporal features based on a multi-scale convolutional neural network to generate a pipe network state feature matrix; an anomaly detection module that inputs the feature matrix into a pre-trained model to mark potential leakage areas; an optimization analysis module that constructs a multi-constraint dynamic optimization model and optimizes pipe network pressure parameters using an adaptive particle swarm algorithm; and a hierarchical execution module that generates a global control sequence, dynamically matches local pressure parameters, and adjusts valve opening and pump station power through a decision layer, a regional coordination layer, and an execution layer. The system and method offer precise detection and reasonable control optimization, effectively reducing leakage risks, improving pipe network operation and management, and reducing water resource waste and environmental pollution.

Description

Technical Field

[0001] The present invention relates to the technical field of urban sewage pipe network detection, and in particular to a system and method for detecting leakage in urban sewage pipe networks. Background Art

[0002] In urban sewage pipe networks, leakage has always been a major challenge plaguing the management and maintenance of urban infrastructure. With the rapid advancement of urbanization, urban sewage pipe networks have continued to expand in scale and become increasingly complex. Factors such as aging pipes, construction defects, and external loads have led to frequent leakage.

[0003] Traditional manual inspection methods are inefficient and inaccurate, making it difficult to cover large-scale pipe network systems and easily affected by the experience and subjective factors of the inspectors. For example, manual inspection can only determine whether there is leakage by simple methods such as observing whether there is sewage seepage on the ground and listening to whether there are abnormal sounds in the pipes. It is difficult to detect leakage points in pipes that are deeply buried and highly concealed in time. Some existing detection technologies, such as acoustic detection and pressure detection, although they have improved the accuracy of detection to a certain extent, also have problems such as limited detection range and poor adaptability to complex pipe networks. These technologies can often only detect local areas of the pipeline, and it is difficult to fully grasp the leakage situation of the entire pipeline network, and cannot meet the needs of efficient and precise management of modern urban sewage pipeline networks. Therefore, it is urgent to develop an efficient, accurate and adaptable urban sewage pipeline network leakage detection system and method. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for detecting leakage in urban sewage pipe networks to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a leakage detection system for urban sewage pipe networks, the system comprising:

[0006] Data acquisition module: used to collect real-time operation data of the pipe network through multi-source sensors, including pressure, flow, acoustic wave signals and water quality parameters;

[0007] Feature extraction module: extracts spatiotemporal features of the real-time operation data based on a multi-scale convolutional neural network to generate a pipeline network state feature matrix;

[0008] Anomaly detection module: inputs the pipeline network state feature matrix into the pre-trained spatiotemporal anomaly detection model to generate potential leakage area markers;

[0009] Optimization analysis module: constructs a multi-constraint dynamic optimization model based on the leakage area markers. The model takes pressure balance adjustment and leakage location accuracy as optimization goals and uses an adaptive particle swarm algorithm to globally optimize the pipe network pressure parameters.

[0010] Hierarchical execution module: Based on the dynamic optimization model, the optimal control strategy is output and the distributed response of instructions is realized through the hierarchical execution architecture. The architecture includes a decision-making layer, a regional coordination layer and an execution layer. The decision-making layer generates a global pressure control sequence, the regional coordination layer uses a sliding window optimization algorithm to dynamically match local pressure parameters, and the execution layer realizes the adjustment of valve opening and pump station power based on a fuzzy proportional integral control algorithm.

[0011] Preferably, the collecting of real-time pipe network operation data by multi-source sensors includes:

[0012] Multi-source sensors include pressure transmitters, electromagnetic flow meters, acoustic wave sensors, and dissolved oxygen sensors;

[0013] The pressure transmitter data and electromagnetic flowmeter data are aligned in time and space to construct a pressure-flow distribution map of the pipe network; the acoustic sensor data is decomposed by wavelet packets to generate a high-frequency vibration feature sequence;

[0014] Constructing a dual-branch feature fusion network, wherein the first branch uses a three-dimensional convolutional network to extract the spatial correlation features of the pressure-flow distribution graph, and the second branch uses a temporal convolutional network to extract the time-varying features of the high-frequency vibration feature sequence;

[0015] The spatial correlation features and time-varying features are fused through a cross-modal gating mechanism to generate a joint feature tensor. The joint feature tensor is dynamically modeled based on a bidirectional long short-term memory network to output a pipeline network state feature matrix containing pressure fluctuations, vibration anomalies, and water quality mutations.

[0016] Preferably, the spatiotemporal anomaly detection model adopts a multi-level graph neural network structure and generates leakage area markers based on a dynamic graph attention mechanism; the structure includes:

[0017] Construct a pipe network-environment interaction diagram. The nodes in the diagram include pipeline nodes, pump station nodes, valve nodes, and environmental monitoring nodes. The node attributes include pressure value, flow value, and sound wave energy.

[0018] A dual-path attention mechanism is used. The first path calculates the interaction weights between pipeline nodes and adjacent nodes through a spatial graph aggregation layer, and the second path performs feature screening on historical operation data through a temporal graph pooling layer.

[0019] Node attributes are iteratively updated based on a multi-head graph attention module, and each attention head fuses node status with environmental monitoring parameters. The training process is stabilized through skip connections and layer normalization mechanisms, and the final output is a leakage area label that contains the network topology constraints.

[0020] Preferably, the adaptive particle swarm algorithm integrates inertia weight adjustment and constraint repair strategy, including:

[0021] The pressure optimization problem is modeled as a multi-constrained nonlinear programming problem, where the decision variables include discrete valve state variables and continuous pump station power variables.

[0022] Initialize the particle swarm and calculate the Pareto front candidate solutions. Use the inertia weight adaptive mechanism to adjust the particle search range according to the real-time pipe network load.

[0023] In the optimization phase, the global problem is divided into multiple subgroup optimization problems based on the target decomposition technique; in the constraint repair phase, relaxation factors are introduced to balance the conflicting constraints between subgroups;

[0024] A parallel evolutionary strategy is used to iteratively optimize the sub-swarms, updating the local optimal solution in each iteration and synchronizing the optimal solution through global information sharing.

[0025] Preferably, the regional coordination layer uses a sliding window optimization algorithm to dynamically match local pressure parameters, including:

[0026] A time-varying pipe network pressure propagation model is constructed, and the fluid dynamics equation is discretized into a state transition equation, which includes a pressure decay gradient, a flow mutation term, and a valve response delay term;

[0027] The window optimization objective function is designed, which includes the pressure fluctuation suppression term, the energy consumption minimization term, and the leakage positioning error correction term.

[0028] Preferably, the execution layer realizes the adjustment of valve opening and pump station power based on fuzzy proportional integral control algorithm, including:

[0029] A multi-level fuzzy inference system is designed to map pressure error and flow error into fuzzy rule triggering degrees; an adaptive proportional-integral parameter adjustment mechanism is constructed to dynamically adjust the slope of the fuzzy membership function according to the accumulated error.

[0030] Preferably, the three-dimensional convolutional network adopts a multi-scale dilated convolution structure to accelerate feature extraction, including:

[0031] The pipeline network pressure-flow distribution map is divided into multi-level grids, and each grid cell stores the statistics of pressure gradient and flow rate change rate;

[0032] In the encoding stage, dilated convolution kernels are used to expand the spatial perception range, and in the decoding stage, transposed convolution layers are used to restore local details;

[0033] The spatial attention module is introduced to assign regional weights to the feature map.

[0034] Preferably, the spatial graph aggregation layer adopts a relative distance coding mechanism, including:

[0035] Define the relative distance vector between the pipeline node and the adjacent nodes, including the hydraulic gradient, pipeline material attenuation coefficient and connection direction;

[0036] The relative distance vector is converted into the offset parameter of the graph convolution kernel through a nonlinear mapping layer;

[0037] The offset parameter is added to the standard graph convolution operation.

[0038] Preferably, the inertia weight adaptation mechanism is implemented based on an online feedback strategy, including:

[0039] Collect the particle fitness values and constraint violation degrees during the historical optimization process as training data;

[0040] Construct the mapping relationship between the extreme learning machine network fitting weight coefficient and the pipe network status;

[0041] The network parameters are updated through online recursive least squares method, and the particle swarm search step size is adjusted in real time;

[0042] When a sudden change in pipe network pressure is detected, a rapid weight reset operation is triggered.

[0043] Preferably, the present invention further includes a method for detecting leakage in a town sewage pipe network, which is applied to the above-mentioned town sewage pipe network leakage detection system, and the method comprises the following steps:

[0044] Step 1: Using a data acquisition module to collect real-time operation data of the pipe network using multi-source sensors, the real-time operation data includes pressure, flow, acoustic wave signals and water quality parameters;

[0045] Step 2: Using the feature extraction module, the spatiotemporal features of the real-time operation data collected in step 1 are extracted based on a multi-scale convolutional neural network to generate a pipeline network status feature matrix;

[0046] Step 3: Input the pipeline network status feature matrix generated in step 2 into the pre-trained spatiotemporal anomaly detection model, and generate potential leakage area markers through the anomaly detection module;

[0047] Step 4: The optimization analysis module constructs a multi-constraint dynamic optimization model based on the leakage area markers obtained in step 3. The model uses pressure balance adjustment and leakage location accuracy as optimization objectives and uses an adaptive particle swarm algorithm to globally optimize the pipe network pressure parameters;

[0048] Step 5: Based on the optimal control strategy output by the dynamic optimization model in step 4, the distributed response of the instructions is realized through the hierarchical execution architecture of the hierarchical execution module. The decision layer generates the global pressure control sequence, the regional coordination layer uses the sliding window optimization algorithm to dynamically match the local pressure parameters, and the execution layer realizes the adjustment of valve opening and pump station power based on the fuzzy proportional integral control algorithm.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] In terms of detection accuracy, the data acquisition module utilizes multi-source sensors, such as pressure transmitters, electromagnetic flowmeters, acoustic sensors, and dissolved oxygen sensors, to collect a wealth of real-time pipeline network operation data, covering pressure, flow, acoustic signals, and water quality parameters. Data is processed through techniques such as spatiotemporal alignment and wavelet packet decomposition, and with the help of a dual-branch feature fusion network and a bidirectional long-short-term memory network, data features are deeply mined to generate a feature matrix that accurately reflects the state of the pipeline network. The anomaly detection module uses a multi-level graph neural network structure and a dynamic graph attention mechanism, fully considering the pipeline network topology and environmental factors. It can accurately locate potential leakage areas, greatly improving the accuracy and reliability of detection compared to traditional detection methods, and reducing missed detections and false detections.

[0051] In terms of optimization and control, the optimization analysis module constructs a multi-constrained dynamic optimization model, using pressure balance adjustment and leakage location accuracy as optimization objectives. It employs an adaptive particle swarm algorithm to globally optimize pipeline network pressure parameters. This algorithm integrates inertia weight adjustment with a constraint repair strategy, dynamically adjusting the particle search range based on real-time pipeline network load. This effectively solves the multi-constrained nonlinear programming problem in pressure optimization, achieves precise control of pipeline network pressure, ensures stable operation, and reduces the risk of leakage caused by pressure anomalies.

[0052] The hierarchical execution architecture of the hierarchical execution module boasts efficient command response capabilities. The decision-making layer generates a global pressure control sequence, providing macroscopic control over pipeline network operations. The regional coordination layer constructs a time-varying pipeline network pressure propagation model, employing a sliding window optimization algorithm that comprehensively considers pressure fluctuation suppression, energy consumption minimization, and leakage location error correction. It dynamically matches local pressure parameters, improving the flexibility and accuracy of local control. The execution layer, based on a fuzzy proportional-integral control algorithm, employs a multi-level fuzzy inference system and an adaptive proportional-integral parameter adjustment mechanism to precisely adjust valve openings and pump station power based on pressure and flow errors, achieving refined control of pipeline network equipment, effectively reducing energy consumption and improving energy efficiency.

[0053] From the perspective of overall system operation, this invention enables intelligent management of the entire urban sewage network, from data collection, feature extraction, anomaly detection, to optimized regulation. This not only enables timely detection of potential leakage risks, but also reduces leakage rates through optimized regulation, minimizing water waste and environmental pollution, lowering maintenance costs for urban sewage networks, and improving the intelligent management of urban infrastructure. It ensures the safe, stable, and efficient operation of urban sewage networks, providing strong support for sustainable urban development. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a working principle diagram of a leakage detection system for urban sewage pipe networks according to the present invention;

[0055] Figure 2 This is the working principle diagram of data collection and feature extraction;

[0056] Figure 3 This is the working principle diagram of the adaptive particle swarm algorithm;

[0057] Figure 4 This is a diagram of the working principle of the execution layer. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] See also Figures 1-4 The present invention provides a technical solution: a leakage detection system for urban sewage pipe network, the system comprising:

[0060] Data Acquisition Module: This module utilizes multiple sensors to collect real-time pipe network operating data, including pressure, flow, acoustic signals, and water quality parameters. These sensors, such as pressure transmitters, electromagnetic flowmeters, acoustic sensors, and dissolved oxygen sensors, are strategically installed at key locations within the sewage network, such as pipeline nodes and pump station inlets and outlets, ensuring comprehensive and accurate information on the network's operating status.

[0061] Feature Extraction Module: This module, based on a multi-scale convolutional neural network, extracts spatiotemporal features from the real-time operational data acquired by the Data Acquisition Module. Through a series of complex data processing and feature mining operations, it ultimately generates a pipeline network status feature matrix. This matrix effectively reflects various characteristic information during pipeline network operation, providing key data support for subsequent anomaly detection.

[0062] Anomaly Detection Module: This module inputs the pipeline network status feature matrix generated by the Feature Extraction Module into a pre-trained spatiotemporal anomaly detection model. This model, trained on extensive historical data and possessing powerful pattern recognition capabilities, accurately analyzes the pipeline network status feature matrix, generating potential leakage area markers and identifying areas within the pipeline network where leakage is likely to occur.

[0063] Optimization and Analysis Module: Based on the leakage area markers generated by the Anomaly Detection Module, the Optimization and Analysis Module constructs a multi-constraint dynamic optimization model. This model uses pressure balance adjustment and leak location accuracy as optimization objectives and employs an adaptive particle swarm algorithm to globally optimize the pipeline network pressure parameters. This approach not only more accurately locates leaks but also adjusts the pipeline network pressure to achieve a balanced state, reducing leakage risk.

[0064] Hierarchical Execution Module: Based on the optimal control strategy output by the dynamic optimization model, the hierarchical execution module achieves distributed response to instructions through its unique hierarchical execution architecture. This architecture comprises a decision-making layer, a regional coordination layer, and an execution layer. The decision-making layer is responsible for generating the global pressure control sequence, planning the direction of pipeline network pressure control from a macro perspective; the regional coordination layer uses a sliding window optimization algorithm to dynamically match local pressure parameters, making local control more precise and flexible; the execution layer uses a fuzzy proportional-integral control algorithm to adjust valve opening and pump station power, thereby achieving precise control of the actual operation of the pipeline network and ensuring its stable operation.

[0065] The present invention will be further described below in conjunction with Examples 1 to 5: Example 1:

[0066] In this embodiment, the specific implementation of the data acquisition module and the related data processing flow are further described in detail. The multi-source sensors used in the data acquisition module include pressure transmitters, electromagnetic flow meters, acoustic sensors, and dissolved oxygen sensors. In the actual installation process, the pressure transmitters are installed at different locations in the pipeline to measure the pressure value at each point, which is recorded as ; The electromagnetic flowmeter is installed at a suitable location to obtain flow data, which is recorded as The data collection frequencies of these two sensors must be consistent to facilitate subsequent processing.

[0067] The pressure transmitter data and electromagnetic flowmeter data are aligned in time and space to construct the pressure-flow distribution diagram of the pipeline network. Time and space alignment means matching the data collected by different sensors at different times and spaces so that it can accurately reflect the pressure and flow status of different locations in the pipeline network at the same time. By dividing the time into equal time intervals , the space is divided into certain grid areas , the pressure and flow data within the same time interval and in the same grid area are associated to construct a pipeline network pressure-flow distribution map.

[0068] The data collected by the acoustic sensor contains rich information about the operation of the pipe network. Wavelet packet decomposition is performed on the acoustic sensor data to generate a high-frequency vibration feature sequence. Wavelet packet decomposition is a time-frequency analysis method that can decompose the signal into different frequency bands, thereby extracting more detailed features. Assume that the original signal collected by the acoustic sensor is , after wavelet packet decomposition, the high-frequency vibration characteristic sequence is obtained ,in represents discrete time points, Indicates different frequency sub-bands.

[0069] A dual-branch feature fusion network is constructed. The first branch uses a three-dimensional convolutional network to extract the spatial correlation features of the pressure-flow distribution map. The pressure-flow distribution map of the pipeline network is divided into multi-level grids, and each grid cell stores the pressure gradient. Statistics of traffic change rate (i.e., traffic mutation term and change time In the encoding stage, the dilated convolution kernel is used to expand the spatial perception range, and the dilation rate of the dilated convolution kernel is set to , the convolution kernel size is After the dilated convolution operation, the spatial dimensions of the feature map change, thereby capturing a wider range of spatially correlated features. In the decoding phase, a transposed convolution layer is used to restore local details, restoring the size and resolution of the feature map to an appropriate state. The second branch uses a temporal convolutional network to extract the time-varying characteristics of the high-frequency vibration feature sequence. Through convolution operations with different kernel sizes and strides, the temporal variation of the high-frequency vibration feature sequence is explored.

[0070] A cross-modal gating mechanism fuses spatially correlated features with time-varying features to generate a joint feature tensor. The cross-modal gating mechanism dynamically adjusts the fusion weights based on the importance of the two features, enabling the fused joint feature tensor to better reflect the operational status of the pipeline network. Dynamic pattern modeling of the joint feature tensor is performed using a bidirectional long-short-term memory network. The bidirectional long-short-term memory network simultaneously considers past and future information, effectively capturing dynamic patterns in pipeline network operation. The final output is a pipeline network status feature matrix that includes pressure fluctuations, vibration anomalies, and sudden changes in water quality. Example 2:

[0071] This example focuses on the specific construction and operation of the spatiotemporal anomaly detection model in the anomaly detection module. The spatiotemporal anomaly detection model uses a multi-layer graph neural network structure and generates leakage area markers based on a dynamic graph attention mechanism.

[0072] First, we build a pipeline network-environment interaction diagram, where the nodes include pipeline nodes, pump station nodes, valve nodes, and environmental monitoring nodes. Each node has rich attributes, among which the pipeline node attributes include pressure values. , flow value and sound wave energy , here Represents different pipeline nodes. Pump station node attributes include pump station power , inlet and outlet pressure difference etc.; valve node attributes include valve opening , Pressure difference before and after the valve etc.; the environmental monitoring node attributes include surrounding soil moisture , atmospheric pressure wait.

[0073] A dual-path attention mechanism is used. The first path calculates the interaction weights between pipeline nodes and adjacent nodes through the spatial graph aggregation layer. The relative distance vector between pipeline nodes and adjacent nodes is defined, including the hydraulic gradient. , Pipe material attenuation coefficient and connection direction The relative distance vector is converted into the offset parameter of the graph convolution kernel through a nonlinear mapping layer , the formula is ,in Represents a nonlinear mapping function. The offset parameter is superimposed on the standard graph convolution operation, allowing factors such as the relative position and connection relationship between nodes to be fully considered when calculating node interaction weights. The second path uses the time series graph pooling layer to perform feature filtering on historical operation data. The time series graph pooling layer aggregates and filters the historical attribute data of nodes according to chronological order, extracting key temporal features. For example, through maximum pooling or average pooling operations, it retains features such as the maximum or average pressure and flow values over a period of time to highlight data trends.

[0074] Based on the multi-head graph attention module, the node attributes are iteratively updated, and each attention head integrates the node state and the environmental monitoring parameters. Assume that the number of attention heads is , each attention head fuses node attributes and environmental monitoring parameters through different weight matrices, and the formula is

[0075] ,

[0076] in Indicates the The result after the fusion of the attention heads is Indicates the The attention head is The training process is stabilized through skip connections and layer normalization. Skip connections prevent the vanishing gradient problem, enabling the model to better learn deep features. Layer normalization normalizes the input of each layer, accelerating model convergence. Ultimately, after multiple iterative updates and processing, the model outputs leakage area markers that incorporate network topology constraints, accurately identifying potential leakage areas. Example 3:

[0077] This embodiment focuses on the specific implementation process of the adaptive particle swarm algorithm in the optimization analysis module. In the optimization analysis module, the pressure optimization problem is modeled as a multi-constrained nonlinear programming problem, and the decision variables include discrete valve state variables. Continuous pumping station power variable The discrete valve state variable represents the on / off state of the valve, for example Indicates that the valve is closed. Indicates that the valve is open; the continuous pump station power variable can change continuously within a certain range to meet different pipe network pressure requirements.

[0078] Initialize the particle swarm and calculate the Pareto front candidate solutions. The particle swarm consists of multiple particles, each of which represents a set of possible decision variable values, i.e., a combination of valve status and pump station power. Let the number of particles be , No. The position of a particle is expressed as Calculate the fitness value of each particle. The fitness value is used to measure the degree to which the decision variable combination represented by the particle satisfies the optimization goal. The optimization goal here is pressure balance adjustment and leakage positioning accuracy. Use the inertia weight adaptive mechanism to adjust the particle search range according to the real-time pipe network load. Collect the particle fitness value during the historical optimization process. Constraint Violation As training data, construct the extreme learning machine network fitting weight coefficient The mapping relationship with the pipe network status (such as pressure, flow and other parameters) is as follows: ,in Represents the fitting function of the extreme learning machine network. The network parameters are updated via online recursive least squares, adjusting the particle swarm search step in real time. When a sudden change in pipe network pressure is detected, a rapid weight reset operation is triggered, allowing the particle swarm to quickly adapt to the change in pipe network status and re-search for a better solution.

[0079] In the optimization phase, the global problem is divided into multiple subgroup optimization problems based on the target decomposition technique. The particle swarm is divided into several subgroups, each of which focuses on a sub-goal or a local area of the optimization problem. For example, one subgroup can be mainly responsible for optimizing the pressure balance in a certain area, while another subgroup focuses on improving the accuracy of leak location. In the constraint repair phase, the relaxation factor is introduced. Balance conflicting constraints between subgroups. When the optimization objectives of different subgroups conflict, the relaxation factor is adjusted to coordinate the optimization results of each subgroup, avoiding the situation where over-optimization of a sub-objective makes the overall objective unattainable. A parallel evolutionary strategy is used to iteratively optimize the subgroups, updating the local optimal solution with each iteration and synchronizing the optimal solution through global information sharing. Each subgroup iteratively optimizes within its own search space to find the local optimal solution. Then, through the global information sharing mechanism, the optimal solutions of each subgroup are communicated and integrated, allowing the entire particle swarm to evolve towards a more optimal direction, ultimately achieving global optimization of the pipe network pressure parameters. Example 4:

[0080] This embodiment details the specific implementation of the sliding window optimization algorithm used by the regional coordination layer in the hierarchical execution module to dynamically match local pressure parameters. The regional coordination layer constructs a time-varying pipe network pressure propagation model and discretizes the fluid dynamics equation into a state transfer equation. The pressure is , the flow rate is , the pressure decay gradient is , the traffic mutation term is , the valve response delay term is , then the state transfer equation is ,in represents the spatial step length, This equation describes how pressure in a pipe network changes over time and space, taking into account factors such as pressure decay, flow rate variations, and valve response delays.

[0081] The design window optimization objective function includes the pressure fluctuation suppression term, energy consumption minimization term, and leakage location error correction term. Assume that the pressure fluctuation suppression term is ,in Indicates the The pressure fluctuation value of the time step, Represents the number of time steps in the sliding window. This term is used to measure and suppress the degree of pressure fluctuation to make the pipe network pressure more stable. The energy consumption minimization term is ,in Indicates the The power of the pumping station in each time step, Indicates the number of time steps of the pump station operation within the sliding window. This item is used to calculate the total energy consumption of the pump station within the sliding window. By optimizing this item, the energy consumption cost of the pipeline network operation can be reduced. The leakage location error correction item is ,in Indicates the actual leakage location, represents the predicted leakage location, Represents the number of leakage points, which is used to measure the accuracy of leakage location. By minimizing this term, the accuracy of leakage location can be improved. Combining these three terms, the window optimization objective function is: ,in 、 、 The weight coefficients of the three items are adjusted according to actual needs to balance the importance of different objectives. By continuously adjusting the local pressure parameters within the sliding window, the value of the objective function is minimized, thereby achieving dynamic matching of local pressure parameters and improving the stability and efficiency of pipeline network operation. Example 5:

[0082] This embodiment describes in detail the specific process of implementing valve opening and pump station power regulation based on fuzzy proportional integral control algorithm in the execution layer of the hierarchical execution module. The execution layer designs a multi-level fuzzy inference system to convert the pressure error into and flow error Mapped to fuzzy rule triggering degree. Pressure error Refers to the difference between the actual pressure value and the set pressure value, flow error This refers to the difference between the actual flow rate and the set flow rate. First, the pressure error and flow rate error are fuzzified and divided into different fuzzy sets. For example, the pressure error is divided into fuzzy sets such as "negative large," "negative medium," "negative small," "zero," "positive small," "positive medium," and "positive large." The flow rate error is similarly divided. Fuzzy rules are developed based on experience and actual operating data. For example, when the pressure error is "positive large" and the flow rate error is "positive small," the corresponding fuzzy rules are triggered to determine the direction and degree of adjustment of the valve opening and pump station power.

[0083] An adaptive proportional integral parameter adjustment mechanism is constructed to dynamically adjust the slope of the fuzzy membership function according to the error accumulation. Assume that the error accumulation is ,in It can be a pressure error or a flow error, Represents the number of time steps. When the accumulated error is large, it means that the system deviation is large and the adjustment speed needs to be accelerated. At this time, the slope of the fuzzy membership function is increased to make the fuzzy inference system more sensitive to the error and able to make a faster adjustment response; when the accumulated error is small, the slope of the fuzzy membership function is appropriately reduced to make the adjustment process smoother and avoid over-adjustment. By continuously monitoring the accumulated error and dynamically adjusting the slope of the fuzzy membership function according to its changes, accurate adjustment of the valve opening and pump station power can be achieved. For example, if the fuzzy membership function is ,in is the input variable (pressure error or flow error), is the slope, according to the error accumulation The size of The value of ,in This method allows for precise adjustment of valve opening and pump station power, ensuring stable operation of the pipeline network and effectively addressing abnormalities such as pipeline leakage.

[0084] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0085] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A leakage detection system for urban sewage pipe network, characterized in that: include: Data acquisition module: used to collect real-time operation data of the pipe network through multi-source sensors, including pressure, flow, acoustic wave signals and water quality parameters; Feature extraction module: extracts spatiotemporal features of the real-time operation data based on a multi-scale convolutional neural network to generate a pipeline network state feature matrix; Anomaly detection module: inputs the pipeline network state feature matrix into the pre-trained spatiotemporal anomaly detection model to generate potential leakage area markers; Optimization analysis module: constructs a multi-constraint dynamic optimization model based on the leakage area markers. The model takes pressure balance adjustment and leakage location accuracy as optimization goals and uses an adaptive particle swarm algorithm to globally optimize the pipe network pressure parameters. Hierarchical execution module: Based on the dynamic optimization model, it outputs the optimal control strategy and implements distributed response to instructions through a hierarchical execution architecture. The architecture includes a decision-making layer, a regional coordination layer, and an execution layer. The decision-making layer generates a global pressure control sequence, the regional coordination layer uses a sliding window optimization algorithm to dynamically match local pressure parameters, and the execution layer uses a fuzzy proportional-integral control algorithm to adjust valve opening and pump station power. The real-time operation data of the pipe network collected by multi-source sensors includes: Multi-source sensors include pressure transmitters, electromagnetic flow meters, acoustic wave sensors, and dissolved oxygen sensors; The pressure transmitter data and electromagnetic flowmeter data are aligned in time and space to construct a pressure-flow distribution map of the pipe network; the acoustic sensor data is decomposed by wavelet packets to generate a high-frequency vibration feature sequence; Constructing a dual-branch feature fusion network, wherein the first branch uses a three-dimensional convolutional network to extract the spatial correlation features of the pressure-flow distribution graph, and the second branch uses a temporal convolutional network to extract the time-varying features of the high-frequency vibration feature sequence; The spatial correlation features and time-varying features are fused through a cross-modal gating mechanism to generate a joint feature tensor. The joint feature tensor is dynamically modeled based on a bidirectional long short-term memory network to output a pipe network state feature matrix containing pressure fluctuations, vibration anomalies, and water quality mutations. The spatiotemporal anomaly detection model adopts a multi-level graph neural network structure and generates leakage area markers based on a dynamic graph attention mechanism; the structure includes: Construct a pipe network-environment interaction diagram. The nodes in the diagram include pipeline nodes, pump station nodes, valve nodes, and environmental monitoring nodes. The node attributes include pressure value, flow value, and sound wave energy. A dual-path attention mechanism is used. The first path calculates the interaction weights between pipeline nodes and adjacent nodes through a spatial graph aggregation layer, and the second path performs feature screening on historical operation data through a temporal graph pooling layer. Node attributes are iteratively updated based on a multi-head graph attention module, and each attention head fuses node status with environmental monitoring parameters. The training process is stabilized through skip connections and layer normalization mechanisms, and the final output is a leakage area label that contains the network topology constraints.

2. The urban sewage pipe network leakage detection system according to claim 1, characterized in that: The adaptive particle swarm algorithm integrates inertia weight adjustment and constraint repair strategy, including: The pressure optimization problem is modeled as a multi-constrained nonlinear programming problem, where the decision variables include discrete valve state variables and continuous pump station power variables. Initialize the particle swarm and calculate the Pareto front candidate solutions. Use the inertia weight adaptive mechanism to adjust the particle search range according to the real-time pipe network load. In the optimization phase, the global problem is divided into multiple subgroup optimization problems based on the target decomposition technique; in the constraint repair phase, relaxation factors are introduced to balance the conflicting constraints between subgroups; A parallel evolutionary strategy is used to iteratively optimize the sub-swarms, updating the local optimal solution in each iteration and synchronizing the optimal solution through global information sharing.

3. The urban sewage pipe network leakage detection system according to claim 1, characterized in that: The regional coordination layer uses a sliding window optimization algorithm to dynamically match local pressure parameters, including: A time-varying pipe network pressure propagation model is constructed, and the fluid dynamics equation is discretized into a state transition equation, which includes a pressure decay gradient, a flow mutation term, and a valve response delay term; The window optimization objective function is designed, which includes the pressure fluctuation suppression term, the energy consumption minimization term, and the leakage positioning error correction term.

4. The urban sewage pipe network leakage detection system according to claim 1, characterized in that: The execution layer implements the adjustment of valve opening and pump station power based on the fuzzy proportional integral control algorithm, including: A multi-level fuzzy inference system is designed to map pressure error and flow error into fuzzy rule triggering degrees; an adaptive proportional-integral parameter adjustment mechanism is constructed to dynamically adjust the slope of the fuzzy membership function according to the accumulated error.

5. The urban sewage pipe network leakage detection system according to claim 1, characterized in that: The three-dimensional convolutional network adopts a multi-scale dilated convolution structure to accelerate feature extraction, including: The pipeline network pressure-flow distribution map is divided into multi-level grids, and each grid cell stores the statistics of pressure gradient and flow rate change rate; In the encoding stage, dilated convolution kernels are used to expand the spatial perception range, and in the decoding stage, transposed convolution layers are used to restore local details; The spatial attention module is introduced to assign regional weights to the feature map.

6. The urban sewage pipe network leakage detection system according to claim 1, characterized in that: The spatial graph aggregation layer adopts a relative distance coding mechanism, including: Define the relative distance vector between the pipeline node and the adjacent nodes, including the hydraulic gradient, pipeline material attenuation coefficient and connection direction; The relative distance vector is converted into the offset parameter of the graph convolution kernel through a nonlinear mapping layer; The offset parameter is added to the standard graph convolution operation.

7. The urban sewage pipe network leakage detection system according to claim 2, characterized in that: The inertia weight adaptation mechanism is implemented based on an online feedback strategy, including: Collect the particle fitness values and constraint violation degrees during the historical optimization process as training data; Construct the mapping relationship between the extreme learning machine network fitting weight coefficient and the pipe network status; The network parameters are updated through online recursive least squares method, and the particle swarm search step size is adjusted in real time; When a sudden change in pipe network pressure is detected, a rapid weight reset operation is triggered.

8. A method for detecting leakage in a town sewage pipe network, applied to the system for detecting leakage in a town sewage pipe network according to any one of claims 1 to 7, characterized in that: The following steps are involved: Step 1: Using a data acquisition module to collect real-time operation data of the pipe network using multi-source sensors, the real-time operation data includes pressure, flow, acoustic wave signals and water quality parameters; Step 2: Using the feature extraction module, the real-time operation data collected in step 1 is extracted for spatiotemporal features based on a multi-scale convolutional neural network to generate a pipeline network status feature matrix. Step 3: Input the pipeline network status feature matrix generated in step 2 into the pre-trained spatiotemporal anomaly detection model, and generate potential leakage area markers through the anomaly detection module; Step 4: The optimization analysis module constructs a multi-constraint dynamic optimization model based on the leakage area markers obtained in Step 3. The model uses an adaptive particle swarm algorithm to globally optimize the pipe network pressure parameters, with pressure balance adjustment and leakage location accuracy as optimization objectives. Step 5: Based on the optimal control strategy output by the dynamic optimization model in Step 4, a distributed response to instructions is implemented through the hierarchical execution architecture of the hierarchical execution module. The decision layer generates the global pressure control sequence, the regional coordination layer uses a sliding window optimization algorithm to dynamically match local pressure parameters, and the execution layer uses a fuzzy proportional-integral control algorithm to adjust the valve opening and pump station power. The real-time operation data of the pipe network collected by multi-source sensors includes: Multi-source sensors include pressure transmitters, electromagnetic flow meters, acoustic wave sensors, and dissolved oxygen sensors; The pressure transmitter data and electromagnetic flowmeter data are aligned in time and space to construct a pressure-flow distribution map of the pipe network; the acoustic sensor data is decomposed by wavelet packets to generate a high-frequency vibration feature sequence; Constructing a dual-branch feature fusion network, wherein the first branch uses a three-dimensional convolutional network to extract the spatial correlation features of the pressure-flow distribution graph, and the second branch uses a temporal convolutional network to extract the time-varying features of the high-frequency vibration feature sequence; The spatial correlation features and time-varying features are fused through a cross-modal gating mechanism to generate a joint feature tensor. The joint feature tensor is dynamically modeled based on a bidirectional long short-term memory network to output a pipeline network state feature matrix containing pressure fluctuations, vibration anomalies, and water quality mutations.

Images