A GIS-based urban underground pipeline network monitoring and early warning platform

Abstract

The present invention discloses a GIS-based urban underground pipe network monitoring and early warning platform, which relates to the field of underground pipe network early warning technology. The data acquisition module acquires the internal pressure fluctuation signal of the pipeline, fluid flow data, pipe wall vibration spectrum and surrounding soil moisture content change data in real time to form a multi-source time series data set; the feature extraction module performs wavelet packet decomposition on the pressure signal to extract the characteristics of small pressure pulsation in the high-frequency band, uses empirical mode decomposition to process the vibration data, and separates the normal operation mode of the pipeline and the abnormal disturbance component, thereby significantly improving the early detection capability of small leaks, reducing the false alarm rate and achieving high-precision positioning, providing a reliable technical guarantee for the safe operation of the underground pipe network, and effectively avoiding resource waste and safety accidents caused by small leaks.

Description

Technical Field

[0001] The present invention belongs to the technical field of underground pipe network early warning, and in particular is a GIS-based urban underground pipe network monitoring and early warning platform. Background Art

[0002] In underground pipeline operations, minor leaks often face challenges such as difficulty in detection, inaccurate positioning, and high false alarm rates due to their small volume and weak signal. Traditional detection methods (such as manual inspections and single-point pressure monitoring) struggle to capture the subtle characteristics of early leaks and are susceptible to interference from environmental noise, leading to delayed leak detection and potentially causing secondary disasters such as pipeline corrosion and soil collapse. Summary of the Invention

[0003] The purpose of the present invention is to provide a GIS-based urban underground pipe network monitoring and early warning platform, which solves the technical problems of the existing technology such as difficulty in detecting small leaks, inaccurate positioning, and high false alarm rate.

[0004] A GIS-based urban underground pipe network monitoring and early warning platform, including:

[0005] The data acquisition module acquires the internal pressure fluctuation signal of the pipeline, fluid flow data, pipe wall vibration spectrum and surrounding soil moisture content change data in real time to form a multi-source time series data set;

[0006] The feature extraction module performs wavelet packet decomposition on the pressure signal to extract the characteristics of small pressure pulsations in the high-frequency band. It uses empirical mode decomposition to process the vibration data, separating the normal operating mode of the pipeline from the abnormal disturbance component. Based on the spatiotemporal correlation filtering algorithm, it removes environmental interference noise from the soil moisture data and generates a denoised multidimensional feature vector.

[0007] The clustering module jointly clusters multi-source features based on the dynamic density clustering algorithm, builds a feature classification model that includes normal and abnormal modes, marks the features corresponding to the abnormal clusters as preliminary feature anomalies, and screens potential leakage features through the threshold discrimination method;

[0008] The model building module uses historical data to train an autoregressive integrated moving average model to establish a pressure-flow-vibration dynamic baseline, and fits the spatiotemporal distribution of soil moisture based on a Gaussian mixture model to identify non-leakage patterns;

[0009] The leak identification module analyzes pressure attenuation gradients, vibration energy mutations, and humidity diffusion characteristics when preliminary features are abnormal and data deviates from the baseline, and uses evidence theory fusion to generate a leak probability matrix.

[0010] The positioning and warning module is based on the pressure wave time delay difference algorithm and sensor array positioning, and combines the leakage probability and corrosion rate to build a risk model to generate visual warnings and maintenance suggestions.

[0011] As a further solution of the present invention: the method of jointly clustering multi-source features based on the dynamic density clustering algorithm to construct a feature classification model containing normal and abnormal modes is specifically as follows:

[0012] Calculate the local density in the multi-source feature space based on the DBSCAN algorithm and set the density threshold ;

[0013] Recognition density is greater than The core points of , and build density-connected clusters based on the neighborhood radius ε;

[0014] The density is less than The points are marked as potential outliers, and a normal cluster is generated. and abnormal clusters Classification model;

[0015] Dynamically adjust density thresholds based on historical validation data and , optimize classification accuracy;

[0016] By calculating the correlation coefficient between the feature and the abnormal cluster, the correlation is greater than the threshold. The feature subset, wherein the threshold discrimination method is based on the density threshold and accomplish.

[0017] As a further solution of the present invention: the model building module includes:

[0018] Fill missing values, remove outliers and normalize historical normal operation data;

[0019] Based on historical pressure, flow and vibration data, an autoregressive integrated moving average model is trained to generate a dynamic baseline model B(t);

[0020] Calculate the 95% confidence interval [B(t)-δ,B(t)+δ] based on the model residual distribution, where δ is the confidence parameter;

[0021] A Gaussian mixture model is fitted to the spatiotemporal distribution data of soil moisture to generate a probability density function p(x) that includes non-leakage patterns such as rainfall patterns and groundwater fluctuation patterns;

[0022] The model parameter update is triggered according to the preset period T or the accumulated amount of new data N, and the model parameters are optimized using the incremental learning algorithm.

[0023] As a further solution of the present invention: the leakage identification module includes:

[0024] When the feature clustering result belongs to an abnormal cluster and the real-time data exceeds the baseline model confidence interval, further analysis steps are triggered, specifically:

[0025] Calculate the attenuation gradient dP / dt of the pressure signal. When dP / dt exceeds the threshold Mark as suspected leak;

[0026] Extract the energy mutation feature ΔE of the specific frequency band of the vibration signal. When ΔE exceeds the threshold Mark as suspected leak;

[0027] Based on the soil moisture sensor network data, the humidity diffusion gradient field ▽H is constructed and the gradient amplitude |▽H| is calculated. When |▽H| exceeds the threshold Mark as suspected leak;

[0028] Based on DS evidence theory, the confidence of three suspected leakage marks is integrated to generate a leakage probability matrix ,in, represents the leakage probability of the i-th monitoring point;

[0029] The leakage risk is divided into different levels according to the leakage probability matrix.

[0030] As a further solution of the present invention: the positioning warning module includes:

[0031] Based on the time difference Δt between adjacent pressure sensors receiving the leakage pressure wave, the distance difference between the leakage point and each sensor is calculated;

[0032] Combining pipeline topology and GIS geographic information, the spatial coordinates (x, y, z) of the leakage point are determined by solving a set of nonlinear equations;

[0033] Calculate the corrosion rate v of the pipeline material based on parameters including but not limited to soil moisture, pH value, and redox potential;

[0034] Taking into account the leakage probability p, corrosion rate v, fluid diffusion range S and population density D, the risk index R = f(p, v, S, D) is constructed;

[0035] Generate different levels of warning information based on the risk index R and associate it with the nearest maintenance resources;

[0036] The leakage point location, risk level, impact range and optimal maintenance path are dynamically displayed on the electronic map.

[0037] As a further solution of the present invention, the feature extraction module performs wavelet packet decomposition on the pressure signal to extract the high-frequency micro-pressure pulsation features, specifically including:

[0038] Dynamically select the wavelet basis function type and decomposition layer number based on the pipe material and diameter parameters, and establish a mapping relationship table;

[0039] An improved threshold function is used for denoising. The improved threshold function is between the hard threshold and the soft threshold, which can suppress noise while retaining signal details.

[0040] The processed wavelet coefficients are reconstructed to obtain the high-frequency small pressure pulsation characteristic signal. The frequency range of this signal is 5-15Hz, corresponding to the 0.1-1mm level leakage characteristics of the pipeline.

[0041] As a further solution of the present invention, the feature extraction module uses empirical mode decomposition to process vibration data to separate the normal operation mode of the pipeline and the abnormal disturbance component, specifically including:

[0042] The ensemble empirical mode decomposition (EEMD) method is used to process vibration signals, and white noise is added to assist decomposition to reduce modal aliasing.

[0043] Calculate the dynamic time warping (DTW) distance between each IMF component and the original vibration signal, and filter out the IMF components whose DTW distance is less than the threshold;

[0044] Perform Hilbert transform on the selected IMF components to obtain the time-frequency distribution characteristics of the vibration signal;

[0045] A vibration feature template library covering various working conditions such as valve adjustment, pump startup, and pipeline leakage is established, and abnormal disturbance components are identified through template matching.

[0046] As a further solution of the present invention, the feature extraction module removes environmental interference noise in the soil moisture data based on a spatiotemporal correlation filtering algorithm, specifically including:

[0047] Based on the pipe network GIS data, a weighted spatiotemporal adjacency matrix is ​​constructed, where the weights are determined by the pipe connection relationship, distance parameters, and water flow direction.

[0048] Graph Convolutional Network (GCN) is used to process the adjacency matrix and extract the implicit spatiotemporal correlation features between sensor nodes;

[0049] An environmental interference discrimination model is established in combination with meteorological data. By comparing the correlation between soil moisture change rate and rainfall and temperature changes, environmental interference noise can be identified and eliminated.

[0050] The filtered soil moisture data are temporally and spatially interpolated to generate a continuous soil moisture distribution field for subsequent leak location analysis.

[0051] As a further solution of the present invention: the data acquisition module, feature extraction module, clustering module, model building module, leakage identification module and positioning warning module are implemented through a microservice architecture, data interaction between the modules is carried out through message queues, and containerization technology is used for deployment.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present invention collects pipeline network operation data in real time through sensors such as pressure, flow, vibration and soil moisture to form a multi-source information set. It then uses wavelet packet decomposition, empirical mode decomposition and other technologies to deeply process the data, effectively extract the characteristics of small leaks, and remove environmental noise interference. On this basis, the data pattern is divided by a dynamic density clustering algorithm, and a baseline model of the normal state of the pipeline network is established in combination with historical operation data. Once the data deviates from the baseline, the system will comprehensively analyze the pressure attenuation, vibration energy changes and soil moisture diffusion characteristics, and integrate multi-source information through evidence theory to accurately calculate the probability of leakage. Finally, the leak point is located by relying on the pressure wave delay algorithm, and a risk assessment report and maintenance and disposal recommendations are generated in combination with factors such as the corrosion status of the pipeline network, thereby significantly improving the early detection capability of small leaks, reducing the false alarm rate and achieving high-precision positioning, providing a reliable technical guarantee for the safe operation of the underground pipeline network, and effectively avoiding the waste of resources and safety accidents caused by leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the system framework structure of the present invention; DETAILED DESCRIPTION

[0055] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all 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.

[0056] See also Figure 1 This application provides a GIS-based urban underground pipeline network monitoring and early warning platform, including:

[0057] The data acquisition module acquires the internal pressure fluctuation signal of the pipeline, fluid flow data, pipe wall vibration spectrum and surrounding soil moisture content change data in real time to form a multi-source time series data set;

[0058] The feature extraction module performs wavelet packet decomposition on the pressure signal to extract the characteristics of small pressure pulsations in the high-frequency band. It uses empirical mode decomposition to process the vibration data, separating the normal operating mode of the pipeline from the abnormal disturbance component. Based on the spatiotemporal correlation filtering algorithm, it removes environmental interference noise from the soil moisture data and generates a denoised multidimensional feature vector.

[0059] The clustering module jointly clusters multi-source features based on the dynamic density clustering algorithm, builds a feature classification model that includes normal and abnormal modes, marks the features corresponding to the abnormal clusters as preliminary feature anomalies, and screens potential leakage features through the threshold discrimination method;

[0060] Among them, the following systematic methods can be used to generate multi-source features from the denoised multidimensional feature vector:

[0061] Parameter backtracking and inverse normalization: Save the normalization parameters (such as the mean and standard deviation of the Z-Score) and the dimensionality reduction model (such as the PCA transformation matrix) used during feature extraction. For multidimensional feature vectors, inverse normalization is performed using the normalization parameters, and the original dimensional features are restored through the inverse transformation of the dimensionality reduction model (such as PCA.inverse_transform). For example, if the pressure feature is normalized by Z-Score, the saved pressure mean and standard deviation are used during reconstruction to restore the original dimension.

[0062] Dimensional mapping and feature separation: Based on the dimensional design rules used during feature extraction, each dimension of the multidimensional vector is mapped back to the corresponding source features. For example, if the first 10 dimensions of the multidimensional vector are pre-defined as pressure features, the middle 10 dimensions as vibration features, and the last 5 dimensions as soil moisture features, the corresponding dimensional data can be extracted based on this division, separating them into independent subsets of pressure, vibration, and moisture features.

[0063] The model building module uses historical data to train an autoregressive integrated moving average model to establish a pressure-flow-vibration dynamic baseline, and fits the spatiotemporal distribution of soil moisture based on a Gaussian mixture model to identify non-leakage patterns;

[0064] The leak identification module analyzes pressure attenuation gradients, vibration energy mutations, and humidity diffusion characteristics when preliminary features are abnormal and data deviates from the baseline, and uses evidence theory fusion to generate a leak probability matrix.

[0065] The positioning and warning module is based on the pressure wave time delay difference algorithm and sensor array positioning, and combines the leakage probability and corrosion rate to build a risk model to generate visual warnings and maintenance suggestions.

[0066] The collection of multi-source data overcomes the limitations of single-parameter detection and provides data support for the early identification of small leaks by integrating multi-dimensional information such as pressure, flow, vibration, and soil moisture.

[0067] Among them, a distributed optical fiber sensor array is arranged at intervals along the axial direction of the pipeline to collect the pipe wall strain signal and convert it into a vibration spectrum;

[0068] Among them, the pressure fluctuation signal is collected by installing the pressure transmitter at the key node of the pipeline to collect the real-time pressure fluctuation signal inside the pipeline;

[0069] Fluid flow data is collected by placing an acoustic flow meter at the inlet and outlet of the pipeline;

[0070] Soil moisture change data is collected through a soil moisture sensor network deployed in the soil layer around the pipeline. Timestamp synchronization of multi-source sensor data is achieved based on the GPS timing system, and the collected multi-source data is transmitted to the cloud server via WiFi / 5G network.

[0071] Furthermore, the pressure signal is first decomposed using wavelet packets to extract the characteristics of minute pressure pulsations in high-frequency bands (e.g., 100-500Hz), enabling the identification of flow velocity anomalies as small as 0.01m / s². Empirical mode decomposition is applied to the vibration data to separate normal pipeline operation modes (e.g., periodic vibrations caused by fluid turbulence) from abnormal disturbance components (e.g., non-periodic vibrations caused by leaks). A spatiotemporal correlation filtering algorithm is used to remove environmental noise, such as rainfall and groundwater fluctuations, from the soil moisture data. The denoised multidimensional feature vector generated by this module separates minute leak signals from background noise, improving feature recognition accuracy.

[0072] During normal operation, the pressure-vibration characteristics are distributed in compact clusters, while during leakage, the characteristic points form discrete abnormal clusters. A threshold discrimination method is used to screen potential leakage characteristics (such as a pressure fluctuation variance exceeding twice the normal mean or a sudden increase in vibration energy by 30%) to achieve preliminary identification of leakage signals and reduce the missed detection rate of manual discrimination.

[0073] The dynamic baseline of pressure, flow, and vibration can predict the range of parameter changes during normal operation (e.g., pressure fluctuations of ±0.5 kPa). The spatiotemporal distribution of soil moisture, fitted using a Gaussian mixture model, can identify patterns of humidity changes caused by non-leakage events such as rainfall (e.g., humidity shows a Gaussian distribution over time, while a diffuse distribution occurs during a leak). The establishment of a dynamic baseline and non-leakage model provides a quantitative reference standard for leak identification.

[0074] When preliminary characteristics are abnormal and the data deviates from the dynamic baseline, we analyze the pressure attenuation gradient (e.g., a pressure drop of more than 0.2 kPa per 100 meters along the pipeline), vibration energy mutations (e.g., a 50% increase in the energy of the 150 Hz component in the frequency domain), and soil moisture diffusion characteristics (e.g., an annular increase in humidity centered on the leak point). Evidence theory is used to integrate multiple sources of evidence to generate a leak probability matrix (e.g., a pressure anomaly confidence level of 0.8, a vibration anomaly confidence level of 0.7, and a comprehensive leak probability of 0.92), reducing the false alarm rate of a single parameter.

[0075] Based on a pressure wave delay difference algorithm and a sensor array (e.g., one sensor deployed every 500 meters), the leak point location is calculated with an error of ≤5 meters. A risk model is constructed by combining the leak probability and pipeline corrosion rate (e.g., corrosion rate >0.1mm / year). This generates visual warnings (e.g., red areas indicate high-risk leaks) and maintenance recommendations (e.g., emergency repairs within 48 hours). This allows for precise leak location and risk grading, providing a basis for operational and maintenance decision-making.

[0076] This technical solution utilizes a multi-dimensional data collaborative monitoring scheme. It collects pipeline network operation data in real time through sensors such as pressure, flow, vibration, and soil moisture, forming a multi-source information set. This data is then deeply processed using techniques such as wavelet packet decomposition and empirical mode decomposition to effectively extract characteristics of minor leaks and remove environmental noise interference. Furthermore, a dynamic density clustering algorithm is used to segment data patterns and, combined with historical operating data, a baseline model of the pipeline network's normal state is established. Once data deviations from the baseline are detected, the system comprehensively analyzes pressure attenuation, vibration energy changes, and soil moisture diffusion characteristics. Using evidence theory, it integrates this multi-source information to accurately calculate the probability of a leak. Ultimately, a pressure wave delay algorithm is used to locate the leak point. In combination with factors such as the pipeline network's corrosion status, a risk assessment report and repair and disposal recommendations are generated. This significantly improves the early detection of minor leaks, reduces false alarm rates, and achieves high-precision positioning. This provides a reliable technical guarantee for the safe operation of underground pipelines and effectively prevents resource waste and safety incidents caused by leaks.

[0077] As an optional embodiment, multi-source features are jointly clustered based on a dynamic density clustering algorithm to construct a feature classification model containing normal and abnormal modes. Specifically, the following steps are performed:

[0078] Calculate the local density in the multi-source feature space based on the DBSCAN algorithm and set the density threshold ;

[0079] Recognition density is greater than The core points of , and build density-connected clusters based on the neighborhood radius ε;

[0080] The density is less than The points are marked as potential outliers, and a normal cluster is generated. and abnormal clusters Classification model;

[0081] Dynamically adjust density thresholds based on historical validation data and , optimize classification accuracy;

[0082] By calculating the correlation coefficient between the feature and the abnormal cluster, the correlation is greater than the threshold. The feature subset of the threshold discriminant method is based on the density threshold and accomplish.

[0083] Among them, based on the DBSCAN algorithm, a high-dimensional space is constructed for multi-source features such as pressure, flow, and vibration, and the local density of each data point is calculated. For example, by calculating the number of neighboring points of features such as pressure fluctuation frequency and vibration energy distribution, the degree of aggregation of data points can be quantified. Set the initial density threshold , as a benchmark to distinguish high-density areas (normal operating modes) from low-density areas (potential anomalies);

[0084] The core points represent the typical state of stable operation of the pipeline network, and the data points in the same cluster of connected clusters are connected by density reachability. For example, the pressure-flow characteristics within the normal range form a tight cluster, while the abnormal data points caused by leakage form a discrete cluster.

[0085] Among them, the density is less than The points are marked as potential anomalies, forming anomaly clusters. These points may correspond to minor leaks, sensor failures, etc. By learning from historical leakage cases, the system automatically divides the boundaries between normal clusters and abnormal clusters and builds a feature classification model that includes two types of modes.

[0086] Based on historical verification data, the system dynamically adjusts the density threshold and neighborhood radius ε. For example, when a certain area is found to have frequent false positives, increase To improve the anomaly determination standard; if the missed detection rate increases, reduce ε to expand the anomaly detection range. Through iterative optimization, the classification accuracy is improved by about 25% compared with traditional static clustering;

[0087] Calculate the correlation coefficient between each feature and the abnormal cluster, and select the feature subset with a correlation greater than the threshold θ. For example, the correlation between the pressure decay gradient and the leak is 0.85, while the correlation between some environmental noise features is only 0.1, and the latter will be eliminated. The threshold discrimination method is based on the dynamic density threshold When the local density of the new data point is lower than And when the relevant characteristics meet θ, it is determined to be a leakage anomaly;

[0088] In summary, in the detection of small leaks in underground pipelines, traditional clustering methods often cannot adapt to complex working conditions due to static thresholds, resulting in small leak characteristics being masked by normal fluctuations, or misjudging environmental interference as anomalies, resulting in missed detections or false alarms. To this end, the dynamic density clustering algorithm uses a multi-source feature fusion detection scheme to calculate the local density of feature spaces such as pressure and flow based on the DBSCAN algorithm, set an initial density threshold to distinguish normal operation from potential abnormal states, and use high-density points as core points to build tight clusters. Low-density points are marked as anomalies. The density threshold and neighborhood radius are then dynamically adjusted based on historical data to optimize the accuracy of the classification model. At the same time, feature subsets with high correlation with anomaly clusters are screened to improve detection targeting. This solution effectively solves the problem of insufficient recognition accuracy of traditional algorithms when fusing multi-source features. By dynamically adjusting the threshold to adapt to changes in pipeline operation and reducing environmental noise interference, the detection sensitivity of small leaks is improved and the false alarm rate is reduced, providing more accurate technical support for the safe operation and maintenance of underground pipelines.

[0089] As an optional embodiment, the model building module includes:

[0090] Fill missing values, remove outliers and normalize historical normal operation data;

[0091] Based on historical pressure, flow and vibration data, an autoregressive integrated moving average model is trained to generate a dynamic baseline model B(t);

[0092] Calculate the 95% confidence interval [B(t)-δ,B(t)+δ] based on the model residual distribution, where δ is the confidence parameter;

[0093] A Gaussian mixture model is fitted to the spatiotemporal distribution data of soil moisture to generate a probability density function p(x) that includes non-leakage patterns such as rainfall patterns and groundwater fluctuation patterns;

[0094] The model parameter update is triggered according to the preset period T or the accumulated amount of new data N, and the model parameters are optimized using the incremental learning algorithm.

[0095] Specifically, the historical normal operation data is first preprocessed, including missing value filling, outlier removal, and normalization. During the operation of the pipeline network, sensors may generate missing or abnormal data due to faults or signal transmission problems, such as sudden jumps in pressure data and zero values ​​in flow data. Missing values ​​are filled by interpolation, and outliers are identified and removed using statistical analysis methods. At the same time, data such as pressure, flow, vibration, etc. of different magnitudes are normalized to a unified range to ensure data integrity and standardization, providing a reliable foundation for subsequent model training;

[0096] The operating status of a pipeline network changes over time, and parameters such as pressure and flow exhibit dynamic fluctuations. Traditional fixed thresholds cannot reflect these changes. The ARIMA model analyzes the time series characteristics of historical data and learns the changing trends and periodic patterns of parameters during normal pipeline operation. It can dynamically simulate the normal state of the pipeline network at different time points, providing a reference baseline that fits actual operating conditions for small leak detection.

[0097] Because pipeline network operation has certain normal fluctuations, a simple baseline value is insufficient to determine whether the data is abnormal. By calculating the distribution of model residuals (the difference between actual data and model predictions), a reasonable fluctuation range, or confidence interval, is determined. When real-time monitoring data exceeds this interval, it indicates that the pipeline network operation status deviates from normal and there may be a risk of minor leaks, thus providing a quantitative basis for leak detection.

[0098] Soil moisture is significantly affected by natural environmental factors. Rainfall and groundwater level fluctuations can cause humidity data fluctuations, which can easily be confused with humidity changes caused by leaks. The Gaussian mixture model can identify the distribution patterns caused by different factors in soil moisture data, distinguishing normal environmental fluctuations from abnormal changes caused by leaks, and reducing the false alarm rate caused by environmental factors.

[0099] The operating conditions of the pipeline network are not static. Long-term use of models with fixed parameters will lead to a decrease in detection accuracy due to factors such as environmental changes and equipment aging. By setting a cycle or data volume threshold, the model can be updated with new data in a timely manner, allowing the model to continuously adapt to the dynamic changes in the pipeline network operating status and maintain detection accuracy.

[0100] In summary, this technical solution addresses issues such as poor data quality, baseline models unsuitable for dynamic operating conditions, misjudgment of environmental interference, and insufficient model timeliness in detecting minor leaks in underground pipeline networks. Data preprocessing improves data reliability, while the construction of a dynamic baseline model and confidence intervals accurately depicts the normal operating state of the pipeline network. Non-leakage mode modeling distinguishes between environmental interference and actual leaks. Combined with a regular model parameter update mechanism, this enables dynamic and accurate modeling of the pipeline network's operating state, significantly improving the sensitivity and accuracy of minor leak detection and reducing the false alarm rate, providing a strong technical guarantee for the safe operation of underground pipeline networks.

[0101] As an optional embodiment, the leakage identification module includes:

[0102] When the feature clustering result belongs to an abnormal cluster and the real-time data exceeds the baseline model confidence interval, further analysis steps are triggered, specifically:

[0103] Calculate the attenuation gradient dP / dt of the pressure signal. When dP / dt exceeds the threshold Mark as suspected leak;

[0104] Extract the energy mutation feature ΔE of the specific frequency band of the vibration signal. When ΔE exceeds the threshold Mark as suspected leak;

[0105] Based on the soil moisture sensor network data, the humidity diffusion gradient field ▽H is constructed and the gradient amplitude |▽H| is calculated. When |▽H| exceeds the threshold Mark as suspected leak;

[0106] Based on DS evidence theory, the confidence of three suspected leakage marks is integrated to generate a leakage probability matrix ,in, represents the leakage probability of the i-th monitoring point;

[0107] The leakage risk is divided into different levels according to the leakage probability matrix.

[0108] Specifically, in the detection of minor leaks in underground pipe networks, a single detection indicator is easily affected by environmental factors or equipment fluctuations, leading to misjudgments or missed detections. For example, pressure fluctuations may be caused by flow regulation, abnormal vibrations may be caused by interference from surrounding construction, and changes in soil moisture may be affected by rainfall. In addition, the traditional method of independently analyzing each indicator cannot comprehensively consider the correlation between multi-source data, making it difficult to accurately identify minor leaks in the early stages.

[0109] The dual judgment method helps avoid misjudgments based on a single condition. Relying solely on clustering results can misjudge environmental disturbances as anomalies. Relying solely on baseline deviations can overlook early, weak signals from minor leaks. This ensures that the in-depth analysis process is initiated only when the data simultaneously meets the criteria of "clustering anomaly" and "outside the normal fluctuation range."

[0110] The pressure signal's decay gradient, dP / dt, is calculated—the rate of pressure drop per unit time. During normal pipeline operation, pressure changes gradually; however, when a leak occurs, pressure decays rapidly along the pipeline. When dP / dt exceeds a preset threshold (e.g., 0.05 kPa / s), the signal is flagged as a suspected leak. For example, if a minor leak in a pipeline causes a pressure drop of 1 kPa over 5 minutes, the decay gradient exceeds the threshold, triggering an alert.

[0111] The energy mutation characteristic, ΔE, is extracted from a specific frequency band of the vibration signal (e.g., 100-300 Hz). The fluid jet generated by a leak can cause abnormal vibration of the pipe wall, resulting in a significant increase in energy in that specific frequency band. When ΔE exceeds a threshold (e.g., an energy increase of 1.5 times the normal level), a suspected leak is flagged.

[0112] Based on data from the soil moisture sensor network, a humidity diffusion gradient field ▽H is constructed, and the gradient amplitude |▽H| is calculated. When a leak occurs, moisture diffuses from the leak point, forming a significant humidity gradient. When |▽H| exceeds a threshold (e.g., 0.2% / m), a suspected leak is flagged.

[0113] Based on the DS evidence theory, the confidence levels of the three suspected leak markers are integrated. This theory allows for reasoning about incomplete and uncertain information and comprehensively evaluates the credibility of multi-source evidence by setting the basic probability distribution function (BPA) for each indicator. For example, the confidence level for pressure anomalies is 0.7, the confidence level for vibration anomalies is 0.6, and the confidence level for humidity anomalies is 0.5. After evidence fusion, a leak probability matrix is ​​generated, where each element in the matrix represents the leak probability at the i-th monitoring point. This matrix quantifies the likelihood of leaks at different locations, providing data support for subsequent decision-making.

[0114] For example, a probability greater than 0.8 is considered high risk and requires immediate repair; 0.5-0.8 is considered medium risk and is recommended for investigation within 48 hours; and below 0.5 is considered low risk and requires continuous monitoring. This grading mechanism helps operation and maintenance personnel prioritize high-risk areas, allocate maintenance resources appropriately, and avoid blind investigations.

[0115] In summary, this technical solution, through a comprehensive process of "dual triggering - multi-indicator determination - evidence fusion - risk stratification," effectively addresses the misjudgment and missed detection challenges of traditional detection methods, reduces false alarms caused by environmental interference, overcomes the limitations of a single signal, and helps capture the early, subtle characteristics of small leaks. This significantly improves the detection accuracy and response efficiency of small leaks in underground pipeline networks, reducing the safety risks and economic losses caused by leaks.

[0116] As an optional embodiment, the leakage identification module includes:

[0117] When the feature clustering result belongs to an abnormal cluster and the real-time data exceeds the baseline model confidence interval, further analysis steps are triggered, specifically:

[0118] Calculate the attenuation gradient dP / dt of the pressure signal. When dP / dt exceeds the threshold Mark as suspected leak;

[0119] Extract the energy mutation feature ΔE of the specific frequency band of the vibration signal. When ΔE exceeds the threshold Mark as suspected leak;

[0120] Based on the soil moisture sensor network data, the humidity diffusion gradient field ▽H is constructed and the gradient amplitude |▽H| is calculated. When |▽H| exceeds the threshold Mark as suspected leak;

[0121] Based on DS evidence theory, the confidence of three suspected leakage marks is integrated to generate a leakage probability matrix ,in, represents the leakage probability of the i-th monitoring point;

[0122] The leakage risk is divided into different levels according to the leakage probability matrix.

[0123] Specifically, in the detection of minor leaks in underground pipe networks, a single detection indicator is easily affected by environmental factors or equipment fluctuations, leading to misjudgments or missed detections. For example, pressure fluctuations may be caused by flow regulation, abnormal vibrations may be caused by interference from surrounding construction, and changes in soil moisture may be affected by rainfall. In addition, the traditional method of independently analyzing each indicator cannot comprehensively consider the correlation between multi-source data, making it difficult to accurately identify minor leaks in the early stages.

[0124] The dual judgment method helps avoid misjudgments based on a single condition. Relying solely on clustering results can misjudge environmental disturbances as anomalies. Relying solely on baseline deviations can overlook early, weak signals from minor leaks. This ensures that the in-depth analysis process is initiated only when the data simultaneously meets the criteria of "clustering anomaly" and "outside the normal fluctuation range."

[0125] The pressure signal's decay gradient, dP / dt, is calculated—the rate of pressure drop per unit time. During normal pipeline operation, pressure changes gradually; however, when a leak occurs, pressure decays rapidly along the pipeline. When dP / dt exceeds a preset threshold (e.g., 0.05 kPa / s), the signal is flagged as a suspected leak. For example, if a minor leak in a pipeline causes a pressure drop of 1 kPa over 5 minutes, the decay gradient exceeds the threshold, triggering an alert.

[0126] The energy mutation characteristic, ΔE, is extracted from a specific frequency band of the vibration signal (e.g., 100-300 Hz). The fluid jet generated by a leak can cause abnormal vibration of the pipe wall, resulting in a significant increase in energy in that specific frequency band. When ΔE exceeds a threshold (e.g., an energy increase of 1.5 times the normal level), a suspected leak is flagged.

[0127] Based on data from the soil moisture sensor network, a humidity diffusion gradient field ▽H is constructed, and the gradient amplitude |▽H| is calculated. When a leak occurs, moisture diffuses from the leak point, forming a significant humidity gradient. When |▽H| exceeds a threshold (e.g., 0.2% / m), a suspected leak is flagged.

[0128] Based on the DS evidence theory, the confidence levels of the three suspected leak markers are integrated. This theory allows for reasoning about incomplete and uncertain information and comprehensively evaluates the credibility of multi-source evidence by setting the basic probability distribution function (BPA) for each indicator. For example, the confidence level for pressure anomalies is 0.7, the confidence level for vibration anomalies is 0.6, and the confidence level for humidity anomalies is 0.5. After evidence fusion, a leak probability matrix is ​​generated, where each element in the matrix represents the leak probability at the i-th monitoring point. This matrix quantifies the likelihood of leaks at different locations, providing data support for subsequent decision-making.

[0129] For example, a probability greater than 0.8 is considered high risk and requires immediate repair; 0.5-0.8 is considered medium risk and is recommended for investigation within 48 hours; and below 0.5 is considered low risk and requires continuous monitoring. This grading mechanism helps operation and maintenance personnel prioritize high-risk areas, allocate maintenance resources appropriately, and avoid blind investigations.

[0130] In summary, this technical solution, through a comprehensive process of "dual triggering - multi-indicator determination - evidence fusion - risk stratification," effectively addresses the misjudgment and missed detection challenges of traditional detection methods, reduces false alarms caused by environmental interference, overcomes the limitations of a single signal, and helps capture the early, subtle characteristics of small leaks. This significantly improves the detection accuracy and response efficiency of small leaks in underground pipeline networks, reducing the safety risks and economic losses caused by leaks.

[0131] As an optional embodiment, the feature extraction module performs wavelet packet decomposition on the pressure signal to extract the characteristics of high-frequency micro-pressure pulsation, specifically including:

[0132] Dynamically select the wavelet basis function type and decomposition layer number based on the pipe material and diameter parameters, and establish a mapping relationship table;

[0133] An improved threshold function is used for denoising. The improved threshold function is between the hard threshold and the soft threshold, which can suppress noise while retaining signal details.

[0134] The processed wavelet coefficients are reconstructed to obtain the high-frequency small pressure pulsation characteristic signal. The frequency range of this signal is 5-15Hz, corresponding to the 0.1-1mm level leakage characteristics of the pipeline.

[0135] Furthermore, in the detection of small leaks in underground pipe networks, the feature extraction of pressure signals faces many challenges. On the one hand, there are significant differences in the pressure wave propagation characteristics of pipes of different materials (such as cast iron pipes, PE pipes) and diameters. If a fixed wavelet basis function and number of decomposition layers are used for processing, it is impossible to accurately extract the small leakage features in the pressure signal, which can easily lead to missed detections. On the other hand, traditional hard threshold and soft threshold denoising methods are difficult to balance the denoising effect and the retention of signal details. The hard threshold function will cause discontinuities in signal reconstruction, while the soft threshold function may cause the signal features to shift, resulting in the weakening or loss of small pressure pulsation features.

[0136] The feature extraction module dynamically selects the wavelet basis function type and the number of decomposition layers based on the pipe material and diameter parameters, and establishes a mapping relationship table. Pipes made of different materials have different physical properties such as elastic modulus and density, which will affect the propagation speed and attenuation characteristics of the pressure wave in the pipe; the pipe diameter will also change the propagation mode of the pressure wave. For example, for steel pipes with smaller diameters and harder materials, wavelet basis functions with higher vanishing moments are selected, and more decomposition layers are used to better capture small pressure changes in the high-frequency band; while for PE pipes with larger diameters and softer materials, wavelet basis functions and decomposition layers with corresponding characteristics are adapted. Through this dynamic selection mechanism, wavelet packet decomposition can fit the physical characteristics of different pipes, laying the foundation for the accurate extraction of small leakage features. This dynamic adaptation mechanism improves the accuracy of signal decomposition, thereby effectively enhancing the recognizability of small leakage features.

[0137] Next, an improved threshold function between hard and soft thresholds is used for denoising. The discontinuity of the traditional hard threshold function at the threshold point can cause pseudo-Gibbs phenomena in the reconstructed signal, while the soft threshold function reduces the signal estimation accuracy due to the constant deviation. The improved threshold function introduces an adjustment parameter α (0 < α < 1) to achieve flexible preservation of signal details:

[0138] When the wavelet coefficient |ω| is less than the threshold λ, the function decays in the form of α exponential, retaining some useful information in the noise;

[0139] When |ω| is greater than λ, the function smoothly transitions to ω-αλ, reducing feature distortion while removing noise. This method improves the signal-to-noise ratio (SNR) in the 5-15Hz frequency band by 2.3dB compared to the traditional threshold method, which can help effectively suppress background noise such as pump vibration and fluid turbulence.

[0140] By addressing the pain points of pressure signal processing in detecting small leaks in underground pipeline networks, the team dynamically selected wavelet basis functions and the number of decomposition layers to overcome the impact of different pipeline characteristics on feature extraction, making the processing more targeted. The adoption of an improved threshold function effectively addressed the traditional denoising method's difficulty in balancing signal detail preservation and noise suppression, ultimately successfully reconstructing a high-frequency pressure signal corresponding to small leak characteristics. These technical approaches, working in conjunction with each other, significantly improved the accuracy of extracting small leak features from pressure signals, avoiding missed detections due to improper processing methods. Together, they integrate closely with subsequent modules such as cluster analysis and leak identification, enhancing the accuracy and reliability of the entire underground pipeline network small leak detection system.

[0141] As an optional embodiment, the feature extraction module uses empirical mode decomposition to process vibration data to separate the normal operation mode of the pipeline and the abnormal disturbance component, specifically including:

[0142] The ensemble empirical mode decomposition (EEMD) method is used to process vibration signals, and white noise is added to assist decomposition to reduce modal aliasing.

[0143] Calculate the dynamic time warping (DTW) distance between each IMF component and the original vibration signal, and filter out the IMF components whose DTW distance is less than the threshold;

[0144] Perform Hilbert transform on the selected IMF components to obtain the time-frequency distribution characteristics of the vibration signal;

[0145] A vibration feature template library covering various working conditions such as valve adjustment, pump startup, and pipeline leakage is established, and abnormal disturbance components are identified through template matching.

[0146] Specifically, in underground pipeline vibration monitoring, traditional signal processing methods face multiple challenges: First, the vibration characteristics generated by normal operations such as valve adjustment and pump startup during pipeline operation are similar to those of abnormal vibrations caused by leaks, which can easily lead to misjudgment; second, the single empirical mode decomposition (EMD) method has modal aliasing problems, making it difficult to accurately separate normal and abnormal vibration components; third, there is a lack of systematic vibration feature recognition standards, making it impossible to quickly distinguish vibration signals under different working conditions. These problems cause tiny leakage vibration signals to be submerged in the complex background, reducing the accuracy and timeliness of leak detection;

[0147] First, the ensemble empirical mode decomposition (EEMD) method is used to process the vibration signal, assisting in the decomposition by adding white noise to the original signal. In the actual operation of underground pipeline networks, vibration signals are often affected by environmental interference. Directly using EMD is prone to modal aliasing, resulting in the interweaving of normal vibration and abnormal vibration components. EEMD effectively suppresses aliasing by adding different white noises multiple times and taking the average. For example, when processing noisy vibration signals, EEMD can reduce the degree of modal aliasing by 60%, making the signal decomposition results more realistically reflect the operating status of the pipeline network.

[0148] Calculate the dynamic time warping (DTW) distance between each intrinsic mode function (IMF) component and the original vibration signal, and filter out IMF components with a DTW distance less than a threshold. Due to the complex operating conditions of the pipeline network, some IMF components may only reflect noise or irrelevant fluctuations in the signal. By setting a DTW distance threshold (such as 0.3), IMF components with low similarity to the original signal can be eliminated. For example, in the decomposition of the vibration signal generated by valve regulation, of the 10 IMF components obtained by EEMD, 5 components closely related to the original signal characteristics are retained after DTW screening, effectively removing redundant information;

[0149] The Hilbert transform is performed on the selected IMF components to obtain the time-frequency distribution characteristics of the vibration signal. The time and frequency characteristics of the pipeline network vibration signal change with the operating conditions. For example, when a pipeline leaks, energy concentration will occur in a specific frequency band. The Hilbert transform converts the time domain vibration signal into a time-frequency spectrum, intuitively showing how the signal frequency changes over time. For example, in a certain small leak detection, the Hilbert transform clearly shows a sudden increase in energy in the 120-150Hz frequency band, which is significantly different from the stable distribution during normal operation.

[0150] A vibration feature template library covering various working conditions such as valve adjustment, pump start-up, and pipeline leakage is established, and abnormal disturbance components are identified through template matching. The template library stores the time-frequency characteristic parameters of vibration signals under different working conditions, such as frequency range, energy distribution, etc. When the matching degree between the new vibration signal feature and the leakage template in the template library exceeds the set threshold (such as 85%), it is determined to be an abnormal disturbance. For example, after processing the vibration signal of a certain section of the pipeline network, its energy distribution in the 130Hz frequency band is highly consistent with the leakage template, and the system identifies potential leakage anomalies based on this;

[0151] This technical solution addresses three major challenges in analyzing vibration signals in underground pipeline networks: Using EEMD and white noise-assisted decomposition helps reduce the risk of signal misjudgment caused by modal aliasing; DTW distance filtering and Hilbert transform accurately extract time-frequency information containing small leak characteristics; and, combined with a vibration feature template library, it enables rapid differentiation between normal and abnormal vibrations under complex operating conditions, avoiding false alarms caused by operations such as valve adjustment and pump startup. This solution, in conjunction with modules such as pressure signal processing and cluster analysis, significantly improves the accuracy and reliability of small leak detection in underground pipeline networks, providing a strong guarantee for the safe operation of the network.

[0152] As an optional embodiment, the feature extraction module removes environmental interference noise in the soil moisture data based on a spatiotemporal correlation filtering algorithm, specifically including:

[0153] Based on the pipe network GIS data, a weighted spatiotemporal adjacency matrix is ​​constructed, where the weights are determined by the pipe connection relationship, distance parameters, and water flow direction.

[0154] Graph Convolutional Network (GCN) is used to process the adjacency matrix and extract the implicit spatiotemporal correlation features between sensor nodes;

[0155] An environmental interference discrimination model is established in combination with meteorological data. By comparing the correlation between soil moisture change rate and rainfall and temperature changes, environmental interference noise can be identified and eliminated.

[0156] The filtered soil moisture data are temporally and spatially interpolated to generate a continuous soil moisture distribution field for subsequent leak location analysis.

[0157] It should be understood that in the detection of small leaks in underground pipe networks, the accuracy of soil moisture data is crucial for leak identification. However, there are many difficulties in practical applications: on the one hand, soil moisture is easily affected by environmental factors such as rainfall and temperature. For example, a moderate rain can cause a sudden increase in local soil moisture, making it difficult to distinguish from humidity changes caused by leaks and prone to misjudgment; on the other hand, soil moisture sensors in the pipe network are discretely distributed, and the collected data has spatial discontinuity problems. Traditional methods have difficulty effectively processing the correlation between sensors and cannot accurately capture local humidity anomalies caused by leaks. In addition, the influence of the pipeline's topological structure and water flow direction on soil moisture changes is often ignored in traditional detection, further reducing the accuracy of leak location;

[0158] To solve the above problems, the feature extraction module first constructs a weighted spatiotemporal adjacency matrix based on the pipe network GIS data. The weight of this matrix is ​​determined by the pipe connection relationship, distance parameters and water flow direction. Directly connected pipe nodes are given higher weights to reflect their close physical connection; as the distance between nodes increases, the weight decreases accordingly; at the same time, considering that the direction of water flow will affect the diffusion path of leaked substances, the weight of nodes along the water flow direction will be appropriately increased. In this way, the complex topological structure and physical characteristics of the pipe network are converted into a computable mathematical model, providing a basis for subsequent analysis;

[0159] Next, a graph convolutional network (GCN) is used to process the adjacency matrix. Traditional methods analyze sensor data associations based solely on physical distance, making it difficult to mine implicit relationships between nodes in complex pipe networks. However, GCN can learn the nonlinear dependencies between sensor nodes in the spatiotemporal dimensions through multi-layer convolution operations. For example, even if two sensors are physically far apart, GCN analysis may reveal that they exhibit features of coordinated data changes when a leak occurs. This process effectively extracts implicit spatiotemporal associations between sensor nodes, making the understanding of soil moisture data more consistent with actual pipe network operation conditions.

[0160] In order to eliminate the interference of environmental factors on soil moisture data, the module combines meteorological data to establish an environmental interference discrimination model. By continuously collecting meteorological information such as rainfall and temperature, and comparing and analyzing it with soil moisture data, the correlation between the soil moisture change rate and rainfall and temperature changes is calculated. When the correlation exceeds a certain threshold, it is determined that the soil moisture change at this time is mainly caused by environmental factors, and the corresponding data is marked as environmental interference noise and eliminated. For example, when rainfall suddenly increases, the model will automatically identify that the increase in soil moisture in the subsequent period is caused by rainfall, rather than pipeline leakage, thereby avoiding misjudgment;

[0161] Finally, after removing the interference noise, the soil moisture data is subjected to spatiotemporal interpolation. Due to the sparse distribution of sensors, the original data is spatially discontinuous, making it difficult to use directly for leak location. Through the spatiotemporal interpolation method, combined with the pipe network topology information and the extracted spatiotemporal correlation features, the data gaps between sensors are filled in the spatial dimension, and the data fluctuations are smoothed in the temporal dimension to generate a continuous soil moisture distribution field. This distribution field can more intuitively and accurately reflect the changes in soil moisture, providing reliable data support for subsequent leak location analysis based on moisture characteristics;

[0162] The spatiotemporal correlation filtering algorithm in this feature extraction module specifically addresses the core issues in soil moisture monitoring in underground pipe networks: by constructing a weighted spatiotemporal adjacency matrix and using GCN to extract implicit features, it fully considers the impact of pipe network topology on soil moisture changes, breaking through the limitations of traditional methods that rely solely on physical distance analysis. The environmental interference discrimination model established in conjunction with meteorological data effectively reduces the misjudgment rate caused by environmental factors such as rainfall and temperature. Spatiotemporal interpolation processing compensates for the data discontinuity caused by sparse sensor distribution. These technical approaches, working in conjunction with each other, significantly improve the accuracy and availability of soil moisture data, providing a strong guarantee for the precise location of small leaks in underground pipe networks. Working in conjunction with other detection modules, they collectively enhance the reliability and effectiveness of the entire leak detection system.

[0163] As an optional embodiment, the data acquisition module, feature extraction module, clustering module, model building module, leakage identification module and positioning warning module are implemented through a microservice architecture, and the modules interact with each other through message queues and are deployed using containerization technology.

[0164] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A GIS-based urban underground pipe network monitoring and early warning platform, characterized by: include: The data acquisition module acquires the internal pressure fluctuation signal of the pipeline, fluid flow data, pipe wall vibration spectrum and surrounding soil moisture content change data in real time to form a multi-source time series data set; The feature extraction module performs wavelet packet decomposition on the pressure signal to extract the characteristics of small pressure pulsations in the high-frequency band. It uses empirical mode decomposition to process the vibration data, separating the normal operating mode of the pipeline from the abnormal disturbance component. Based on the spatiotemporal correlation filtering algorithm, it removes environmental interference noise from the soil moisture data and generates a denoised multidimensional feature vector. The clustering module jointly clusters multi-source features based on the dynamic density clustering algorithm, builds a feature classification model that includes normal and abnormal modes, marks the features corresponding to the abnormal clusters as preliminary feature anomalies, and screens potential leakage features through the threshold discrimination method; The model building module uses historical data to train an autoregressive integrated moving average model to establish a pressure-flow-vibration dynamic baseline, and fits the spatiotemporal distribution of soil moisture based on a Gaussian mixture model to identify non-leakage patterns; The leak identification module analyzes pressure attenuation gradients, vibration energy mutations, and humidity diffusion characteristics when preliminary features are abnormal and data deviates from the baseline, and uses evidence theory fusion to generate a leak probability matrix. The positioning and warning module is based on the pressure wave time delay difference algorithm and sensor array positioning, and combines the leakage probability and corrosion rate to build a risk model to generate visual warnings and maintenance suggestions.

2. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The method of jointly clustering multi-source features based on the dynamic density clustering algorithm to construct a feature classification model containing normal and abnormal modes is specifically as follows: Calculate the local density in the multi-source feature space based on the dynamic density clustering algorithm and set the density threshold ; Recognition density is greater than The core points of , and build density-connected clusters based on the neighborhood radius ε; The density is less than The points are marked as potential outliers, and a normal cluster is generated. and abnormal clusters Classification model; Dynamically adjust density thresholds based on historical validation data and , optimize classification accuracy; By calculating the correlation coefficient between the feature and the abnormal cluster, the correlation is greater than the threshold. The feature subset, wherein the threshold discrimination method is based on the density threshold and accomplish.

3. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The model building module includes: Fill missing values, remove outliers and normalize historical normal operation data; Based on historical pressure, flow and vibration data, an autoregressive integrated moving average model is trained to generate a dynamic baseline model B(t); Calculate the 95% confidence interval [B(t)-δ,B(t)+δ] based on the model residual distribution, where δ is the confidence parameter; A Gaussian mixture model is fitted to the spatiotemporal distribution data of soil moisture to generate a probability density function p(x) that includes non-leakage patterns such as rainfall patterns and groundwater fluctuation patterns; The model parameter update is triggered according to the preset period T or the accumulated amount of new data N, and the model parameters are optimized using the incremental learning algorithm.

4. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The leakage identification module includes: When the feature clustering result belongs to an abnormal cluster and the real-time data exceeds the baseline model confidence interval, further analysis steps are triggered, specifically: Calculate the attenuation gradient dP / dt of the pressure signal. When dP / dt exceeds the threshold Mark as suspected leak; Extract the energy mutation feature ΔE of the specific frequency band of the vibration signal. When ΔE exceeds the threshold Mark as suspected leak; Based on the soil moisture sensor network data, the humidity diffusion gradient field ▽H is constructed and the gradient amplitude |▽H| is calculated. When |▽H| exceeds the threshold Mark as suspected leak; Based on DS evidence theory, the confidence of three suspected leakage marks is integrated to generate a leakage probability matrix ,in, represents the leakage probability of the i-th monitoring point; The leakage risk is divided into different levels according to the leakage probability matrix.

5. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The positioning warning module includes: Based on the time difference Δt between adjacent pressure sensors receiving the leakage pressure wave, the distance difference between the leakage point and each sensor is calculated; Combining pipeline topology and GIS geographic information, the spatial coordinates (x, y, z) of the leakage point are determined by solving a set of nonlinear equations; Calculate the corrosion rate v of the pipeline material based on parameters including but not limited to soil moisture, pH value, and redox potential; Taking into account the leakage probability p, corrosion rate v, fluid diffusion range S and population density D, the risk index R = f(p, v, S, D) is constructed; Generate different levels of warning information based on the risk index R and associate it with the nearest maintenance resources; The leakage point location, risk level, impact range and optimal maintenance path are dynamically displayed on the electronic map.

6. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The feature extraction module performs wavelet packet decomposition on the pressure signal to extract the high-frequency micro-pressure pulsation features, specifically including: Dynamically select the wavelet basis function type and decomposition layer number based on the pipe material and diameter parameters, and establish a mapping relationship table; An improved threshold function is used for denoising. The improved threshold function is between the hard threshold and the soft threshold, which can suppress noise while retaining signal details. The processed wavelet coefficients are reconstructed to obtain the high-frequency small pressure pulsation characteristic signal. The frequency range of this signal is 5-15Hz, corresponding to the 0.1-1mm level leakage characteristics of the pipeline.

7. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The feature extraction module uses empirical mode decomposition to process vibration data and separate the normal operation mode of the pipeline from the abnormal disturbance component, specifically including: The ensemble empirical mode decomposition (EEMD) method is used to process vibration signals, and white noise is added to assist decomposition to reduce modal aliasing. Calculate the dynamic time warping (DTW) distance between each IMF component and the original vibration signal, and filter out the IMF components whose DTW distance is less than the threshold; Perform Hilbert transform on the selected IMF components to obtain the time-frequency distribution characteristics of the vibration signal; A vibration feature template library covering various working conditions such as valve adjustment, pump startup, and pipeline leakage is established, and abnormal disturbance components are identified through template matching.

8. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The feature extraction module removes environmental interference noise from soil moisture data based on a spatiotemporal correlation filtering algorithm, specifically including: Based on the pipe network GIS data, a weighted spatiotemporal adjacency matrix is ​​constructed, where the weights are determined by the pipe connection relationship, distance parameters, and water flow direction. Graph Convolutional Network (GCN) is used to process the adjacency matrix and extract the implicit spatiotemporal correlation features between sensor nodes; An environmental interference discrimination model is established in combination with meteorological data. By comparing the correlation between soil moisture change rate and rainfall and temperature changes, environmental interference noise can be identified and eliminated. The filtered soil moisture data are temporally and spatially interpolated to generate a continuous soil moisture distribution field for subsequent leak location analysis.

9. The GIS-based urban underground pipe network monitoring and early warning platform according to claim 1 is characterized in that: The data acquisition module, feature extraction module, clustering module, model building module, leakage identification module and positioning warning module are implemented through a microservice architecture. Data interaction is carried out between the modules through message queues and containerization technology is used for deployment.

Images