An intelligent diagnosis and identification method for urban sewer network disease hidden dangers

Abstract

The application discloses an intelligent diagnosis and identification method for urban drainage pipe network disease hidden dangers, relates to the technical field of urban drainage pipe network monitoring, and comprises the following steps: S1, pipe network basic data acquisition and preprocessing; S2, real-time acquisition of multi-source monitoring data; S3, sample set construction, wherein normal operation state data and synthesized abnormal operation state data are combined, a training set, a verification set and a test set are divided, and a complete diagnosis training sample set is constructed; S4, intelligent diagnosis model building and training; S5, real-time diagnosis and identification and secondary verification; and S6, diagnosis result feedback and model iteration.The application adopts the above-mentioned intelligent diagnosis and identification method for urban drainage pipe network disease hidden dangers, combines deep learning and multi-source data fusion technology, realizes real-time, efficient and accurate diagnosis of drainage pipe network disease hidden dangers, reduces artificial cost, and provides a scientific basis for pipe network operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of urban drainage network monitoring technology, and in particular to an intelligent diagnostic and identification method for potential defects in urban drainage networks. Background Technology

[0003] Currently, the diagnosis and identification of potential defects in urban drainage pipe networks mainly rely on traditional methods such as manual inspection and CCTV testing. Manual inspection is inefficient, labor-intensive, and highly dependent on personnel experience and environmental conditions, making it difficult to achieve large-scale, comprehensive pipe network monitoring, and it is also prone to overlooking well-hidden defects. While CCTV testing can provide direct observation of the internal condition of pipes, it is costly, time-consuming, and requires pre-treatment such as sealing and drainage, which affects the normal operation of the pipe network. Furthermore, the test results require manual analysis and judgment, resulting in insufficient identification accuracy and low efficiency.

[0004] In recent years, with the development of intelligent technologies, some studies have adopted a single monitoring parameter combined with simple algorithms for pipeline network disease diagnosis. However, this approach suffers from problems such as limited data dimensions, weak model generalization ability, and inability to comprehensively cover various potential diseases, making it difficult to meet the needs of actual pipeline network operation and maintenance. Furthermore, existing technologies lack effective supplementation of abnormal data, resulting in insufficient model training samples and further affecting diagnostic accuracy. Simultaneously, the diagnostic results lack an effective secondary verification mechanism, making them prone to misjudgment and missed diagnoses. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent diagnostic and identification method for potential defects in urban drainage pipe networks, which solves the technical problems of low efficiency, insufficient identification accuracy, reliance on manual operation, and inability to achieve full-process intelligentization in existing drainage pipe network defect diagnosis methods.

[0006] To achieve the above objectives, the present invention provides an intelligent diagnostic and identification method for potential defects in urban drainage pipe networks, comprising the following steps: S1. Pipeline network basic data acquisition and preprocessing: RTK and total station integrated technology are used to clarify the pipeline network topology and assign unique numbers to each pipe section and manhole; the acquired basic data is cleaned, abnormal data is removed by box plots, and missing data is filled by interpolation to obtain a standardized pipeline network basic database. S2. Real-time acquisition of multi-source monitoring data: Online monitoring equipment is deployed at each monitoring point in the pipeline network to collect cumulative flow of pipe sections, water level of inspection wells, water quality parameters of the pipeline network and vibration parameters of pipeline structure, forming a multi-source monitoring data sequence. S3. Sample Set Construction: Select the diversion drainage zones where pipeline dredging and repair have been completed, and collect multi-source monitoring data of the area as the normal operation status data of the pipeline network; construct a generator and discriminator based on Generative Adversarial Network (GAN); randomly select some data from the normal operation status data and modify them to obtain abnormal data samples; iteratively train the generator and discriminator until the training objective is met; use the trained GAN model to synthesize a large amount of abnormal operation status data of the pipeline network; combine the normal operation status data with the synthesized abnormal operation status data, divide them into training set, validation set and test set, and construct a complete diagnostic training sample set; S4. Building and training an intelligent diagnostic model: Build an intelligent diagnostic model based on a neural network, train the intelligent diagnostic model using the classification cross-entropy loss function, adjust the hyperparameters of the intelligent diagnostic model using the validation set, and verify the performance of the intelligent diagnostic model through the test set until the diagnostic accuracy of the intelligent diagnostic model reaches the preset threshold. S5. Real-time diagnosis and identification and secondary verification: Input the real-time multi-source monitoring data of each monitoring point in the pipeline network into the trained intelligent diagnostic model to preliminarily determine the type, location and severity of pipeline defects. S6. Diagnostic Result Feedback and Model Iteration: The final diagnostic results are pushed to the pipeline operation and maintenance management platform, marking the specific location, type, and severity of defects and hidden dangers, and generating operation and maintenance suggestions. After the abnormal pipeline section is repaired, the operation data of the repaired pipeline section is collected, the sample set is updated, and the intelligent diagnostic model is iteratively trained to continuously improve the model's diagnostic accuracy.

[0007] Preferably, the basic data of the pipeline network in S1 includes pipe segments, manhole coordinates, pipe diameter data, pipe segment length, pipe material type, laying years, pipe segment burial depth, interface type and surrounding geological conditions. In the process of constructing the pipeline network topology, the correspondence between pipe segments and manholes is associated, and the flow direction and connection method of the pipe segments are clarified.

[0008] Preferably, the online monitoring equipment in S2 includes a flow sensor, a radar level sensor, a water quality sensor, and a pipeline health monitoring sensor. The flow sensor is installed at the bottom of the pipe downstream of each monitoring point, and the radar level sensor is installed at the top of the inspection well at each monitoring point. The water quality sensor collects water quality parameters including COD, BOD, ammonia nitrogen, and suspended solids concentration. The pipeline health monitoring sensor collects pipeline structural vibration parameters to determine whether there are potential defects or disconnections in the pipeline.

[0009] Preferably, the discriminator loss function in the Generative Adversarial Network (GAN) in S3 is: Loss = E ( logD ( x ))+ E { log (1-D ( G ( z The generator loss function is: Loss = E { log (1- D ( G ( z )))};in E Expressing expectations, D ( x The ) indicates the probability that the discriminator evaluates the true probability that the anomalous data is true. G ( z () represents the synthesized anomalous data output by the generator. D ( G ( z )) indicates the probability that the discriminator evaluates the synthetic anomalous data as real.

[0010] Preferably, the output format of the abnormal operating status data synthesized in S3 includes two types: {time series, cumulative flow of abnormal pipe section, water level difference between adjacent manholes in normal pipe section, pipe section location, pipe section damage} and {time series, cumulative flow of normal pipe section, water level difference between adjacent manholes in abnormal pipe section, pipe section location, pipe section siltation}.

[0011] Preferably, the neural network in S4 includes an input layer, a hidden layer, and an output layer. The input layer receives multi-source monitoring data and basic pipeline data, and the output layer outputs {pipeline segment number, disease type, severity}.

[0012] Preferably, the classification cross-entropy loss function in S4 is expressed as: J =-1 / m ∑∑ Y ( i ) log ( A ( i )),in m Indicates the number of training samples. N Indicates the number of neurons in the output layer. Y (i) is the first i The actual label of each sample log ( A ( i )) is the label distribution predicted by the model.

[0013] Preferably, the types of defects and potential hazards in S5 include pipe section damage, disconnection, siltation, blockage, corrosion, and interface leakage. The severity is divided into mild, moderate, and severe, based on the extent to which monitoring data deviates from the normal range and the degree of impact of the defects on the operation of the pipeline network.

[0014] Preferably, for pipe sections initially identified as damaged in S5, the surface soil conductivity is measured within a preset range before and after the damaged section. If the conductivity measurement result deviates from the preset value, it is confirmed that the pipe section has damage. For pipe sections initially identified as silted or blocked, a second verification is performed by combining the flow rate change trend of the pipe section and the water level difference of the inspection well to confirm the potential for damage and its severity.

[0015] Therefore, the present invention employs the above-mentioned intelligent diagnosis and identification method for potential defects in urban drainage pipe networks, which has the following beneficial effects: (1) Achieve multi-source data fusion diagnosis, comprehensively cover all kinds of disease and hidden dangers. Combining basic pipeline data and multi-source monitoring data, including parameters such as flow rate, water level, water quality and pipeline structure vibration, it can comprehensively reflect the operating status of the pipeline network and effectively identify various diseases and hidden dangers such as pipe section damage, disconnection, siltation, blockage, corrosion and interface leakage, thus solving the problem of single data dimension and limited coverage of existing technologies.

[0016] (2) Using GAN model to supplement abnormal samples and improve the generalization ability of the model. In view of the problem that it is difficult to obtain abnormal data in actual pipeline network, a large amount of high-quality abnormal operation status data is synthesized by GAN model, which solves the problem of sample imbalance, so that the model can fully learn the characteristics of normal and abnormal states, and improve the generalization ability and diagnostic accuracy of the model.

[0017] (3) Introduce a secondary verification mechanism to ensure the reliability of the diagnosis results. For the preliminary diagnosis results, adopt targeted secondary verification methods, such as surface soil conductivity measurement, flow rate and water level trend analysis, etc., to effectively avoid misjudgment and omission, improve the reliability of the diagnosis results, and provide accurate basis for pipeline operation and maintenance.

[0018] (4) The whole process is intelligent and reduces labor costs. This invention realizes the whole process of intelligent data collection, sample construction, model training, real-time diagnosis to result feedback and model iteration. It does not require a lot of manual intervention, greatly improves the efficiency of disease diagnosis, reduces the intensity of manual labor and detection costs, and is suitable for large-scale, full-coverage urban drainage network monitoring.

[0019] (5) The model can be iteratively optimized to meet long-term operation and maintenance needs. By collecting the operation data of the repaired pipeline network, the sample set is continuously updated and the model is iteratively trained, so that the model can adapt to the changes in the operation status of the pipeline network and maintain high diagnostic accuracy in the long term, providing continuous technical support for the operation and maintenance of the pipeline network throughout its entire life cycle.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1This is a schematic diagram of an embodiment of the intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0024] Example Please see Figure 1 This invention provides an intelligent diagnostic and identification method for potential defects in urban drainage pipe networks, comprising the following steps: S1. Pipeline network basic data acquisition and preprocessing: Collect basic parameters of the urban drainage pipeline network, including coordinates of pipe sections and manholes, pipe diameter data, pipe section length, pipe material type and laying period. Use RTK and total station integrated technology to clarify the pipeline network topology and assign unique numbers to each pipe section and manhole. Clean the collected basic data, remove abnormal data through box plots, and fill in missing data using interpolation to obtain a standardized pipeline network basic database.

[0025] In this step, the basic pipeline data is the foundation for disease diagnosis. By constructing accurate pipeline topology using technologies such as RTK and total station, the connection relationships and flow directions between pipe sections can be clearly defined, providing support for subsequent disease location. The data cleaning process can remove outliers and supplement missing values, avoiding invalid data from affecting the accuracy of diagnostic results and ensuring the reliability of the basic data.

[0026] S2. Real-time acquisition of multi-source monitoring data: Online monitoring equipment is deployed at various monitoring points in the pipeline network. The online monitoring equipment includes flow sensors, radar level sensors, water quality sensors, and pipeline health monitoring sensors, which respectively collect the cumulative flow of the pipeline section, the water level of the inspection well, the water quality parameters of the pipeline network, and the vibration parameters of the pipeline structure to form a multi-source monitoring data sequence.

[0027] Multi-source monitoring data can comprehensively reflect the operating status of the pipeline network. Among them, flow and water level data can be used to determine whether there are functional defects such as siltation and blockage in the pipeline section, water quality data can reflect whether there are structural defects such as damage and leakage in the pipeline network, and pipeline structural vibration parameters can directly reflect whether there are hidden dangers such as damage and disconnection in the pipeline. The reasonable deployment of various sensors can realize real-time monitoring of the operating status of the pipeline network and provide sufficient data support for defect diagnosis.

[0028] S3. Sample Set Construction: Select the diversion drainage zones where pipeline dredging and repair have been completed, and collect multi-source monitoring data of the area as the normal operation status data of the pipeline network. Construct a generator and discriminator based on a generative adversarial network (GAN). Randomly select some data from the normal operation status data and modify them to obtain abnormal data samples. Iterate the generator and discriminator alternately until the training objective is met. Use the trained GAN model to synthesize a large amount of abnormal operation status data of the pipeline network. Combine the normal operation status data with the synthesized abnormal operation status data, divide them into training set, validation set and test set, and construct a complete diagnostic training sample set.

[0029] The discriminator loss function of a Generative Adversarial Network (GAN) is: Loss=E(logD(x))+E{log(1-D(G(z)))} The generator loss function is: Loss=E(logD(x))+E{log(1-D(G(z)))} ,in E Expressing expectations, D ( x The ) indicates the probability that the discriminator evaluates the true probability that the anomalous data is true. G ( z () represents the synthesized anomalous data output by the generator. D ( G ( z )) indicates the probability that the discriminator evaluates the synthetic anomalous data as real.

[0030] The output format of the synthesized abnormal operating status data includes two types: {time series, cumulative flow of abnormal pipe section, water level difference between adjacent manholes in normal pipe section, pipe section location, pipe section damage} and {time series, cumulative flow of normal pipe section, water level difference between adjacent manholes in abnormal pipe section, pipe section location, pipe section siltation}.

[0031] Since it is difficult to obtain a large amount of abnormal state data in actual pipeline networks, synthesizing abnormal data through GAN models can effectively supplement the number of samples and solve the problem of sample imbalance. The generated abnormal data has a high similarity with real abnormal data, which can improve the generalization ability of the model and ensure that the model can accurately identify various diseases and hidden dangers in practical applications.

[0032] S4. Intelligent Diagnostic Model Construction and Training: Build an intelligent diagnostic model based on a neural network. The neural network includes an input layer, a hidden layer, and an output layer. The input layer receives multi-source monitoring data and basic pipeline network data, and the output layer outputs {pipeline segment number, disease type, severity}. The intelligent diagnostic model is trained using a classification cross-entropy loss function. The model hyperparameters are adjusted using a validation set, and the model performance is verified using a test set until the model's diagnostic accuracy reaches a preset threshold.

[0033] The classification cross-entropy loss function is expressed as: J =-1 / m ∑∑ Y ( i ) log ( A ( i ), where m represents the number of training samples, Y ( i ) is the first i The actual label of each sample log ( A ( i )) is the label distribution predicted by the model.

[0034] Neural network models can achieve deep fusion of multi-source data, uncover the intrinsic correlation between data, and accurately identify different types of disease risks and their severity; the classification cross-entropy loss function can effectively measure the difference between the model's prediction results and the actual labels, and improve the model's diagnostic accuracy and stability by iteratively training and optimizing the model parameters.

[0035] S5. Real-time diagnosis and secondary verification: Input real-time multi-source monitoring data from each monitoring point in the pipeline network into the trained intelligent diagnostic model to initially determine the type, location, and severity of pipeline defects. For pipeline sections initially identified as damaged, measure the electrical conductivity of the surface soil within a preset range before and after the damaged section. If the electrical conductivity measurement result deviates from the preset value, the presence of damage in the pipeline section is confirmed. For pipeline sections initially identified as silted or blocked, perform secondary verification by combining the pipeline flow rate change trend and the water level difference in the inspection well to confirm the defects and their severity.

[0036] The secondary verification mechanism can effectively avoid model misjudgment and omission, and improve the reliability of diagnostic results. By adopting corresponding verification methods for different types of disease hazards, the characteristics of various types of data can be fully utilized to ensure the accuracy of diagnostic results.

[0037] S6. Diagnostic Result Feedback and Model Iteration: The final diagnostic results are pushed to the pipeline operation and maintenance management platform, marking the specific location, type, and severity of the defects and potential hazards, and generating operation and maintenance suggestions. After the abnormal pipeline section is repaired, the operation data of the repaired pipeline section is collected, the sample set is updated, and the intelligent diagnostic model is iteratively trained to continuously improve the model's diagnostic accuracy.

[0038] A method for intelligent diagnosis and identification of potential defects in urban drainage pipe networks, the specific implementation steps of which are as follows: S1. Pipeline Network Basic Data Acquisition and Preprocessing: A separate drainage zone in a city was selected as a pilot area. Basic parameters of the drainage pipeline network in this area were collected, including coordinates of pipe sections and manholes (obtained using GPS positioning), pipe diameter data (DN300-DN1200mm), pipe section length (50-200m), pipe material type (PVC pipe, cast iron pipe, concrete pipe), and laying period (5-30 years). The pipeline network was mapped using a combination of RTK and total station technologies to clarify the connection relationship between pipe sections and manholes, the flow direction of pipe sections, and to construct the pipeline network topology. Each pipe section and each manhole was uniquely numbered (e.g., pipe section number GD-XX-XXX, manhole number JC-XX-XXX). The collected basic data was cleaned by removing outliers in parameters such as pipe diameter and length (e.g., data deviating from the average by 3 times the standard deviation) using box plots. Missing coordinates, laying period, and other data were filled using linear interpolation to obtain a standardized pipeline network basic database, which was stored on a cloud server.

[0039] S2. Real-time Acquisition of Multi-Source Monitoring Data: Online monitoring equipment is deployed at key monitoring points in the pilot area's pipeline network (one monitoring point is set every 200m, and monitoring points are also set at manholes). Among them, the flow sensor (model: LDG-1000) is installed at the bottom of the downstream pipe opening to collect real-time flow data and calculate the hourly cumulative flow; the radar level sensor (model: Radar-200) is installed at the top of the manhole to collect real-time water level data; the water quality sensor (model: WS-500) is installed inside the manhole to collect water quality parameters such as COD, BOD, ammonia nitrogen, and suspended solids concentration; and the pipeline health monitoring sensor (model: Vibration-100) is attached to the outer wall of the pipe section to collect pipeline structural vibration parameters. All monitoring equipment transmits monitoring data to the cloud server in real time via wireless communication modules (5G / 4G) to form a multi-source monitoring data sequence, with a sampling frequency set to once every 10 minutes.

[0040] S3. Sample Set Construction: A 3km section of pipe within the pilot area that has undergone dredging and repair was selected as the normal sample collection area. Multi-source monitoring data was continuously collected for one month as the normal operating status data of the pipeline network. A generator and discriminator were constructed based on a generative adversarial network (GAN). The generator adopted a convolutional neural network (CNN) structure, and the discriminator adopted a fully connected neural network structure. 30% of the data from the normal operating status data was randomly selected, and abnormal data samples were obtained by modifying parameters such as flow rate, water level, and vibration (e.g., modifying the cumulative flow rate to 1.5 times the normal range to simulate potential blockage; modifying the vibration parameters to the normal range). The generator and discriminator are trained iteratively (twice the size of the original model to simulate potential damage). The number of training iterations is set to 1000, and the learning rate is set to 0.001, until the discriminator can no longer distinguish between real abnormal data and synthetic abnormal data, and the training stops. The trained GAN model is used to synthesize 10,000 sets of abnormal operation status data of the pipeline network, covering abnormal types such as damage, siltation, blockage, and corrosion. The normal operation status data (5,000 sets) and the synthetic abnormal operation status data (10,000 sets) are combined and divided into training set, validation set and test set in a ratio of 7:2:1 to construct a complete diagnostic training sample set.

[0041] S4. Intelligent Diagnostic Model Construction and Training: An intelligent diagnostic model based on a combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network is constructed. The model includes an input layer, three hidden layers (two convolutional hidden layers and one LSTM hidden layer), and an output layer. The input layer receives multi-source monitoring data (flow rate, water level, water quality, vibration parameters) and basic pipeline network data (pipe diameter, length, pipe material type, laying age). The output layer outputs {pipe segment number, disease type, severity}, where disease types are divided into 6 categories and severity is divided into 3 categories. The intelligent diagnostic model is trained using a classification cross-entropy loss function, which is expressed as: J =-1 / m ∑∑ Y ( i ) log ( A ( i )),in m =10500 (number of training set samples) N =18 (6 types of diseases × 3 severity levels); Adjust the model hyperparameters (number of hidden layer neurons, number of iterations, learning rate) using the validation set (3000 samples), verify the model performance using the test set (1500 samples), and stop training when the model's diagnostic accuracy reaches 95% or higher, thus obtaining a well-trained intelligent diagnostic model.

[0042] S5. Real-time Diagnosis and Secondary Verification: Real-time multi-source monitoring data from various monitoring points in the pilot area's pipeline network are input into the trained intelligent diagnostic model. The model initially determines the type, location, and severity of potential pipeline defects. For pipeline sections initially identified as damaged (such as section number GD-01-005), an electrical conductivity meter is used to measure the conductivity of the surface soil within a 50m radius before and after the pipeline section. The preset normal conductivity range is 50-200 μS / cm. If the measured result is 350 μS / cm... If the flow rate deviates from the normal range, the pipe section is confirmed to have damage, with a severity level of moderate. For pipe sections initially judged to be silted up (such as pipe section number GD-02-012), the 24-hour cumulative flow rate change trend of the pipe section is analyzed (if the flow rate continues to decrease and the decrease exceeds 30%) and the water level difference between adjacent inspection wells (if the difference exceeds 50cm), the pipe section is confirmed to have siltation, with a severity level of mild. All pipe sections initially diagnosed as abnormal are subject to secondary verification to form the final diagnosis.

[0043] S6. Diagnostic Result Feedback and Model Iteration: The final diagnostic results (including pipe segment number, disease type, severity, and specific location) are pushed to the pipeline network operation and maintenance management platform. The platform automatically generates a disease hazard ledger, sorts the maintenance tasks according to severity (prioritizing severe diseases), and generates targeted maintenance suggestions (such as lining repair for damaged pipe segments and high-pressure cleaning for silted pipe segments). After the abnormal pipe segment is repaired, the operation data of the pipe segment for one month after the repair is collected, the sample set is updated, and the intelligent diagnostic model is iteratively trained. The number of iterations is set to 500 times to continuously improve the model's diagnostic accuracy.

[0044] In this embodiment, the intelligent diagnostic and identification method improves the efficiency of diagnosing pipeline network defects and hidden dangers by more than 80% compared with traditional manual inspection, and the diagnostic accuracy rate reaches more than 95%. It effectively identifies various hidden defects and hidden dangers, reduces labor costs, and provides scientific and reliable technical support for pipeline network operation and maintenance.

[0045] Therefore, this invention employs the aforementioned intelligent diagnostic and identification method for urban drainage pipe network defects, achieving multi-source data fusion diagnosis and comprehensively covering various defects. This invention combines basic pipe network data with multi-source monitoring data, including parameters such as flow rate, water level, water quality, and pipe structure vibration, to comprehensively reflect the pipe network's operational status. It can effectively identify various defects such as pipe section damage, disconnection, siltation, blockage, corrosion, and interface leakage, solving the problems of single data dimensions and limited coverage in existing technologies. A GAN model is used to supplement abnormal samples, improving the model's generalization ability: Addressing the difficulty in obtaining abnormal data in actual pipe networks, a large amount of high-quality abnormal operating state data is synthesized through a GAN model, solving the problem of imbalanced samples. This allows the model to fully learn the characteristics of normal and abnormal states, improving the model's generalization ability and diagnostic accuracy. A secondary verification mechanism is introduced to ensure the reliability of diagnostic results: For preliminary diagnostic results, targeted secondary verification methods are used, such as surface soil conductivity measurement and flow rate and water level trend analysis, effectively avoiding misjudgments and omissions, improving the reliability of diagnostic results, and providing accurate basis for pipe network operation and maintenance. End-to-end intelligent process, reducing labor costs: This invention achieves end-to-end intelligent processing from data collection, sample construction, model training, real-time diagnosis to result feedback and model iteration, eliminating the need for extensive manual intervention. This significantly improves the efficiency of disease diagnosis, reduces labor intensity and testing costs, and is suitable for large-scale, comprehensive urban drainage network monitoring. Iteratively optimized model to adapt to long-term operation and maintenance needs: By collecting operational data from repaired pipelines, continuously updating the sample set and iteratively training the model, the model can adapt to changes in the pipeline's operational status, maintaining high diagnostic accuracy over the long term and providing continuous technical support for the entire lifecycle operation and maintenance of the pipeline network.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent diagnosis and identification of potential defects in urban drainage pipe networks, characterized in that, Includes the following steps: S1. Pipeline network basic data acquisition and preprocessing: RTK and total station integrated technology are used to clarify the pipeline network topology and assign unique numbers to each pipe section and manhole; the acquired basic data is cleaned, abnormal data is removed by box plots, and missing data is filled by interpolation to obtain a standardized pipeline network basic database. S2. Real-time acquisition of multi-source monitoring data: Online monitoring equipment is deployed at each monitoring point in the pipeline network to collect cumulative flow of pipe sections, water level of inspection wells, water quality parameters of the pipeline network and vibration parameters of pipeline structure, forming a multi-source monitoring data sequence. S3. Sample set construction: Select the diversion drainage zone where the pipeline network has been dredged and repaired, and collect multi-source monitoring data of the area as the normal operating status data of the pipeline network. A generator and discriminator are constructed based on a generative adversarial network (GAN). Abnormal data samples are obtained by randomly selecting a portion of the normal operating state data and modifying it. The generator and discriminator are trained iteratively and alternately until the training objective is met. A large amount of abnormal operating state data of the pipeline network is synthesized using the trained GAN model. By combining normal operating status data with abnormal operating status data of the pipeline network, and dividing them into training set, validation set and test set, a complete diagnostic training sample set is constructed. S4. Building and training an intelligent diagnostic model: Build an intelligent diagnostic model based on a neural network, train the intelligent diagnostic model using the classification cross-entropy loss function, adjust the hyperparameters of the intelligent diagnostic model using the validation set, and verify the performance of the intelligent diagnostic model through the test set until the diagnostic accuracy of the intelligent diagnostic model reaches the preset threshold. S5. Real-time diagnosis and identification and secondary verification: Input the real-time multi-source monitoring data of each monitoring point in the pipeline network into the trained intelligent diagnostic model to preliminarily determine the type, location and severity of pipeline defects. S6. Diagnostic result feedback and model iteration: The final diagnostic results are pushed to the pipeline operation and maintenance management platform, marking the specific location, type and severity of the defects and potential hazards, and generating operation and maintenance suggestions. After the abnormal pipe section is repaired, the operational data of the repaired pipe section is collected, the sample set is updated, and the intelligent diagnostic model is iteratively trained to continuously improve the diagnostic accuracy of the model.

2. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 1, characterized in that: The basic data of the pipeline network in S1 includes pipe segments, manhole coordinates, pipe diameter data, pipe segment length, pipe material type, laying years, pipe segment burial depth, interface type and surrounding geological conditions. In the process of constructing the pipeline network topology, the correspondence between pipe segments and manholes is associated, and the flow direction and connection method of pipe segments are clarified.

3. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 1, characterized in that: The online monitoring equipment in S2 includes flow sensors, radar level sensors, water quality sensors, and pipeline health monitoring sensors. The flow sensors are installed one by one at the bottom of the downstream pipe opening at each monitoring point. The radar level sensors are installed one by one at the top of the inspection well at each monitoring point. The water quality sensors collect water quality parameters including COD, BOD, ammonia nitrogen, and suspended solids concentration. The pipeline health monitoring sensors collect pipeline structural vibration parameters to determine whether there are potential pipeline damages or disconnections.

4. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 1, characterized in that: The discriminator loss function in the Generative Adversarial Network (GAN) of S3 is: Loss = E ( logD ( x ))+ E { log (1- D ( G ( z )))}; The generator loss function is: Loss = E { log (1- D ( G ( z )))};in E Expressing expectations, D ( x The ) indicates the probability that the discriminator evaluates the true probability that the anomalous data is true. G ( z () represents the synthesized anomalous data output by the generator. D ( G ( z )) indicates the probability that the discriminator evaluates the synthetic anomalous data as real.

5. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 4, characterized in that: The abnormal operating status data output format synthesized in S3 includes two types: {time series, cumulative flow of abnormal pipe section, water level difference between adjacent manholes in normal pipe section, pipe section location, pipe section damage} and {time series, cumulative flow of normal pipe section, water level difference between adjacent manholes in abnormal pipe section, pipe section location, pipe section siltation}.

6. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 1, characterized in that: The neural network in S4 includes an input layer, a hidden layer, and an output layer. The input layer receives multi-source monitoring data and basic pipeline data, and the output layer outputs {pipeline segment number, disease type, and severity}.

7. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 6, characterized in that: The classification cross-entropy loss function in S4 is expressed as: J =-1 / m ∑∑ Y ( i ) log ( A ( i )),in m Indicates the number of training samples. N Indicates the number of neurons in the output layer. Y (i) is the first i The actual label of each sample log ( A ( i )) is the label distribution predicted by the model.

8. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 1, characterized in that: S5 includes pipe section damage, disconnection, siltation, blockage, corrosion, and interface leakage. The severity is divided into mild, moderate, and severe, based on the extent to which monitoring data deviates from the normal range and the degree of impact of the defects on the operation of the pipeline network.

9. The intelligent diagnosis and identification method for potential defects in urban drainage pipe networks according to claim 8, characterized in that: For pipe sections initially identified as damaged in S5, the surface soil conductivity is measured within a preset range before and after the damaged section. If the conductivity measurement result deviates from the preset value, the pipe section is confirmed to have damage. For pipe sections initially identified as silted or blocked, a second verification is conducted by combining the flow rate change trend of the pipe section and the water level difference of the inspection well to confirm the potential damage and its severity.

Images