Intelligent pipe network leakage active prediction and early warning system based on multi-technology fusion

Abstract

The invention discloses an intelligent pipe network leakage active prediction and early warning system based on multi-technology fusion, and the system comprises a multi-source heterogeneous data fusion collection module, a spatial-temporal feature depth extraction module, a degradation trend prediction and residual life evaluation module, and a multi-stage early warning and decision generation module. The multi-source heterogeneous data fusion acquisition module acquires and fuses ultrasonic guided wave signals, pressure flow time sequence data and environmental factor data; the spatial-temporal feature depth extraction module extracts spatial-temporal fusion features through continuous wavelet transform and a CNN-LSTM hybrid network; the degradation trend prediction and residual life evaluation module determines a degradation level, predicts residual life and quantifies a pipe explosion risk probability; the multi-stage early warning and decision generation module generates graded early warning signals and maintenance strategy suggestions, the technology crossing from post-event detection to pre-event prediction is realized, and the scientificity and refinement level of operation and maintenance management of a pipe network are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and predictive maintenance technology for urban water supply networks, specifically to an intelligent network leakage proactive prediction and early warning system based on the integration of multiple technologies. Background Technology

[0002] As an important component of urban infrastructure, urban water supply networks play a crucial role in delivering clean water to every household.

[0003] Existing pipeline leakage detection technologies mainly include manual inspection, area metering, acoustic detection, and pressure monitoring. Manual inspection relies on personnel using listening rods or instruments to inspect pipes section by section. This method is labor-intensive, inefficient, and the results heavily depend on the operator's experience. Area metering assesses leakage levels by dividing the pipeline into independent metering areas and comparing the difference between inflow and outflow rates. This method can only determine the general area of ​​leakage, not the precise location, and is poorly sensitive to minor leaks. Acoustic detection uses the sound signals generated by leaking water for detection and location. While it offers high accuracy, it is susceptible to environmental noise interference and struggles to detect leaks in plastic pipes. Pressure monitoring identifies anomalies by analyzing pipeline pressure fluctuations, but it is insensitive to slowly developing leaks and suffers from detection lag.

[0004] The aforementioned existing technologies all fall under the category of post-incident detection, meaning that detection and location can only be carried out after leakage has occurred. This passive response mode cannot predict the deterioration trend of pipelines and makes it difficult to take preventive maintenance measures before leakage occurs. Pipeline management departments can often only carry out emergency repairs after a pipe burst accident, which not only causes a large waste of water resources but may also trigger secondary disasters such as road collapses and traffic disruptions, seriously affecting residents' normal water use and the safety of urban operations.

[0005] In recent years, with the rapid development of IoT, big data, and AI technologies, data-driven pipeline health status assessment and fault prediction technologies have gradually become research hotspots. Ultrasonic guided wave detection technology, due to its ability to propagate long distances along pipe walls and its sensitivity to changes in pipe wall thickness, has shown promising application prospects in the field of pipeline structural health monitoring. Breakthroughs in deep learning technology in time-series data analysis and pattern recognition have provided new technical pathways for extracting pipeline degradation characteristics and predicting remaining service life from multi-source monitoring data.

[0006] However, existing technologies still have the following shortcomings: First, a single sensor type is insufficient to comprehensively characterize the multidimensional health status of pipelines. Ultrasonic signals reflect structural damage, pressure and flow reflect hydraulic characteristics, and environmental data reflect external corrosion conditions. Accurate assessment requires the fusion of multi-source heterogeneous data. Second, existing deep learning models are mostly designed for static classification tasks and lack the ability to model the dynamic evolution of pipeline deterioration processes over time. Third, remaining life prediction often uses statistical models and fails to fully utilize the rich information contained in monitoring data. Fourth, the early warning mechanism is disconnected from maintenance decisions, lacking a closed-loop mechanism from risk assessment to maintenance strategy generation. Summary of the Invention

[0007] To address the technical problems of passive pipeline leakage detection methods, difficulty in predicting deterioration trends, and disconnect between early warning mechanisms and maintenance decisions in existing technologies, this invention provides an intelligent pipeline leakage proactive prediction and early warning system based on the integration of multiple technologies.

[0008] The technical solution adopted in this invention is as follows:

[0009] The intelligent pipeline network leakage proactive prediction and early warning system based on multi-technology integration includes a multi-source heterogeneous data fusion acquisition module, a spatiotemporal feature deep extraction module, a degradation trend prediction and remaining life assessment module, and a multi-level early warning and decision generation module.

[0010] The multi-source heterogeneous data fusion acquisition module is used to collect ultrasonic guided wave signals, pressure time series data, flow time series data and environmental factor data during pipeline operation, and to align the collected multi-source heterogeneous data with timestamps and spatial locations to generate a fused data tensor.

[0011] The spatiotemporal feature depth extraction module is used to receive the fused data tensor, perform continuous wavelet transform on the ultrasonic guided wave signal to generate a time-frequency energy map, extract spatial feature vectors from the time-frequency energy map through a convolutional neural network, and perform temporal modeling on the spatial feature vectors through a long short-term memory network to capture the evolution law of the pipeline network state, and output the spatiotemporal fused feature vector.

[0012] The degradation trend prediction and remaining service life assessment module is used to receive spatiotemporal fusion feature vectors, determine the degradation status level of the pipeline based on the spatiotemporal fusion feature vectors, calculate the remaining service life of the pipeline based on the degradation status level and the pipeline's historical operating data, and quantify the probability of pipeline burst risk within a preset time window based on the degradation status level and remaining service life.

[0013] The multi-level early warning and decision generation module is used to receive the degradation status level, remaining service life and pipe burst risk probability. When the pipe burst risk probability exceeds the preset first risk threshold, a first early warning level signal is generated. When the pipe burst risk probability exceeds the preset second risk threshold and the remaining service life is lower than the preset service life threshold, a second early warning level signal is generated. Based on the early warning level signal and degradation status level, decision information including maintenance priority ranking and maintenance strategy suggestions is generated.

[0014] Preferably, the multi-source heterogeneous data fusion acquisition module includes an ultrasonic guided wave signal acquisition unit, a pressure and flow time series acquisition unit, and an environmental factor acquisition unit. The ultrasonic guided wave signal acquisition unit is used to deploy a piezoelectric sensor array on the outer wall of the pipeline to excite and receive ultrasonic guided wave signals propagating along the pipe wall. The pressure and flow time series acquisition unit is used to deploy pressure and flow sensors at key nodes in the pipeline to acquire pressure and flow time series data within the pipeline. The environmental factor acquisition unit is used to deploy environmental sensors in the pipeline burial area to acquire environmental factor data such as soil corrosivity index, soil temperature, soil moisture, and groundwater level.

[0015] Preferably, the spatiotemporal feature depth extraction module includes a continuous wavelet transform processing unit, a convolutional neural network feature extractor, and a long short-term memory network temporal modeler. The continuous wavelet transform processing unit uses Morlet wavelets as the mother wavelet to perform continuous wavelet transform on the ultrasonic guided wave signal, generating a time-frequency energy map characterizing the distribution of signal energy in the time and frequency dimensions. The convolutional neural network feature extractor includes multiple sequentially connected convolutional layers, batch normalization layers, and pooling layers, used to extract spatial feature vectors of pipeline structural damage from the time-frequency energy map. The long short-term memory network temporal modeler includes a multi-layer bidirectional long short-term memory network, used to receive spatial feature vectors and model the temporal evolution of the pipeline network state, outputting a spatiotemporal fused feature vector that integrates spatial and temporal features.

[0016] Preferably, the degradation trend prediction and remaining service life assessment module includes a degradation state classification unit, a remaining service life prediction unit, and a pipe burst risk quantification unit. The degradation state classification unit inputs a spatiotemporal fusion feature vector into a pre-trained degradation classification network and outputs the pipeline's degradation state level, which includes five levels: normal, slightly degraded, moderately degraded, severely degraded, and critically degraded. The remaining service life prediction unit calculates the pipeline's remaining service life based on the degradation state level and the pipeline's historical fault records, using an improved Weibull distribution model. The pipe burst risk quantification unit calculates the probability of pipe burst risk within a preset time window by comprehensively considering the degradation state level, remaining service life, pipeline material properties, pipe diameter parameters, and burial years.

[0017] Preferably, the multi-level early warning and decision generation module includes an early warning level determination unit and a maintenance strategy generation unit. The early warning level determination unit receives the probability of pipe burst risk, remaining service life, and degradation status level, and determines the early warning level based on a preset risk threshold matrix. The early warning levels include four levels: blue, yellow, orange, and red. The maintenance strategy generation unit calculates the maintenance priority score for each pipeline based on the early warning level, degradation status level, and pipeline topology, and generates maintenance strategy recommendations that include maintenance time windows, maintenance methods, and maintenance resource configurations.

[0018] Preferably, the system further includes a closed-loop feedback optimization unit, which receives feedback data on the pipeline status after actual maintenance operations, compares the feedback data with the prediction results of the degradation trend prediction and remaining life assessment module, calculates the prediction error, and updates the model parameters of the degradation trend prediction and remaining life assessment module based on the prediction error using an incremental learning method.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] First, by integrating ultrasonic guided wave signals, pressure and flow time-series data, and environmental factor data, this invention achieves a multi-dimensional and comprehensive characterization of pipeline health status, overcoming the limitations of single-sensor detection and significantly improving the accuracy and reliability of pipeline condition assessment.

[0021] Second, the present invention adopts the technical route of continuous wavelet transform combined with CNN-LSTM hybrid deep learning network, which can simultaneously extract spatial damage features and temporal evolution law in ultrasonic signals, realizing a technical leap from post-detection to pre-prediction, and transforming pipeline management from a passive response mode to an active prevention mode.

[0022] Third, this invention establishes a complete technical chain from deterioration state classification, remaining life prediction, pipe burst risk quantification to multi-level early warning and maintenance strategy generation, realizing closed-loop linkage between risk assessment and maintenance decision-making, providing pipeline management departments with operable maintenance priority ranking and strategy suggestions, and effectively improving the scientific and refined level of pipeline operation and maintenance management.

[0023] Fourth, this invention introduces a closed-loop feedback optimization mechanism to continuously update the prediction model based on the actual maintenance effect, so that the prediction accuracy of the system continues to improve with the accumulation of running time, and has the ability to learn and evolve on its own. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall architecture of the intelligent pipeline network leakage proactive prediction and early warning system based on the fusion of multiple technologies of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0026] This invention provides an intelligent pipeline network leakage proactive prediction and early warning system based on multi-technology integration. The system includes four core modules: a multi-source heterogeneous data fusion acquisition module, a spatiotemporal feature deep extraction module, a degradation trend prediction and remaining life assessment module, and a multi-level early warning and decision generation module. The modules are tightly coupled through data flow to form a complete processing link from data acquisition to early warning decision.

[0027] The multi-source heterogeneous data fusion acquisition module is the data input end of the entire system. It is responsible for collecting various types of sensor data from the pipeline network site and fusing these heterogeneous data to provide high-quality input data for subsequent feature extraction and predictive analysis.

[0028] In practical implementation, this module comprises three sub-units: an ultrasonic guided wave signal acquisition unit, a pressure and flow time-series acquisition unit, and an environmental factor acquisition unit. The ultrasonic guided wave signal acquisition unit deploys a piezoelectric sensor array on the outer wall of the pipe. A pulse signal is applied to the piezoelectric sensors via an excitation circuit, generating ultrasonic guided waves that propagate along the pipe wall. When the guided waves encounter defects such as pipe wall thinning, cracks, or corrosion pits, reflection and mode conversion occur. The receiving sensor captures these changed signals and performs digital sampling. In this embodiment, the ultrasonic guided wave signal sampling rate is set to 2MHz, acquiring 8192 sampling points per acquisition cycle, effectively capturing weak signal changes caused by pipe structural damage.

[0029] The pressure and flow time-series acquisition unit installs high-precision pressure sensors and electromagnetic flowmeters at key nodes in the pipeline to monitor changes in water pressure and flow rate in real time. The pressure sensor's range is set to 0 to 1.6 MPa with an accuracy class of 0.1; the electromagnetic flowmeter's range is selected according to the pipeline diameter, with an accuracy class of 0.5. The sampling period is set to 1 second to ensure the capture of rapid fluctuations in pressure and flow rate. When a pipeline leaks, the pressure upstream of the leak point will drop locally, and the flow rate will exhibit abnormal changes inconsistent with normal operating conditions. This information is of significant reference value in assisting in judging the health status of the pipeline.

[0030] The environmental factor acquisition unit deploys multi-parameter soil sensors in the pipeline burial area to collect environmental factors that significantly affect pipeline corrosion, including soil pH, soil resistivity, soil moisture content, soil temperature, and groundwater level. The higher the soil corrosivity, moisture content, and temperature, the faster the pipeline corrosion rate and the more rapid the deterioration process. The sampling period for environmental factor data is set to 1 hour, automatically switching to a 10-minute high-frequency sampling mode under special weather conditions such as rainfall.

[0031] The core technical challenge of the multi-source heterogeneous data fusion acquisition module lies in the spatiotemporal alignment of heterogeneous data. Since different types of sensors have different sampling frequencies and sampling times, a unified time reference needs to be established. This invention uses a GPS timing module to provide a synchronous clock signal for each sensor, ensuring that all acquired data has a unified timestamp. Regarding spatial correlation, each sensor records its installation location's GIS coordinates within the pipeline network during installation, and establishes the correspondence between the sensor and the pipeline through a pipeline network topology model.

[0032] The specific implementation process of spatiotemporal alignment and fusion of multi-source heterogeneous data is as follows: First, the data of each channel is timestamped and normalized, and data with different sampling frequencies are uniformly interpolated onto a common time axis; then, a spatial index is established based on pipeline GIS information and sensor locations to associate multi-source data of the same pipeline segment; finally, the aligned multi-source data are stacked according to the channel dimension to form a multi-channel fused data tensor. The formula for generating the fused data tensor is:

[0033] ,

[0034] in, Represents the fused data tensor. This represents the resampled ultrasonic guided wave signal matrix. This represents the resampled pressure time series data vector. This represents the resampled traffic time-series data vector. This represents the resampled environmental factor data vector. This indicates a splicing operation along the channel dimension.

[0035] The spatiotemporal feature deep extraction module receives fused data tensors and uses a technical approach that combines signal processing and deep learning to extract high-level feature representations that can characterize the health status of pipelines from the original sensor data.

[0036] This module first performs time-frequency analysis on the ultrasonic guided wave signal using a continuous wavelet transform processing unit. Ultrasonic guided wave signals are typical non-stationary signals, with their frequency components varying over time. Traditional Fourier transforms can only obtain global spectral information and cannot reveal the evolution of signal frequency components over time. Continuous wavelet transforms, by scaling and shifting the mother wavelet function, can analyze signals with variable time-frequency resolution, making them particularly suitable for processing transient signal changes caused by pipeline structural damage. This invention selects the Morlet wavelet as the mother wavelet, which possesses good time-frequency localization characteristics and high frequency resolution. The mathematical expression of the continuous wavelet transform is:

[0037] ,

[0038] in, Represents wavelet coefficients, Indicates the scale factor. Indicates the translation factor. This indicates an ultrasonic guided wave signal. Represents the complex conjugate function of the Morlet mother wavelet. Scale factor. The scaling of the wavelet corresponds to the change in frequency; the translation factor... Controls the position of the wavelet on the time axis. Wavelet coefficients. modulus Indicates the signal at time. ,scale Energy distribution at a location.

[0039] The output of the continuous wavelet transform is a two-dimensional time-frequency energy map matrix. ,in This represents the number of discretizations of the scale factor. This represents the number of discretizations on the time axis. In this embodiment, the frequency range corresponding to the scale factor is set to 10kHz to 500kHz, with a total of 128 scale levels; the time axis is divided at equal intervals according to the signal sampling length, with a total of 256 time points. The time-frequency energy map visually displays the distribution of signal energy in both time and frequency dimensions in image form. Damage to the pipeline structure will cause energy accumulation or attenuation in specific frequency regions, exhibiting characteristic patterns that differ from normal signals.

[0040] The convolutional neural network (CNN) feature extractor takes the time-frequency energy map as input and extracts spatial features of pipeline structural damage through multi-layer convolutional operations. This feature extractor employs a 5-layer convolutional structure: the first layer has 32 kernels (3×3 size), a stride of 1, same padding, and ReLU activation; the second layer has 64 kernels (3×3 size) followed by a 2×2 max-pooling layer; the third layer has 128 kernels (3×3 size) followed by a 2×2 max-pooling layer; the fourth layer has 256 kernels (3×3 size) followed by a 2×2 max-pooling layer; and the fifth layer has 512 kernels (3×3 size) followed by a global average pooling layer. Each convolutional layer is followed by a batch normalization layer to accelerate training convergence and improve model generalization ability. The output of the CNN feature extractor is a 512-dimensional spatial feature vector. .

[0041] Long Short-Term Memory (LSTM) network temporal modelers receive spatial feature vector sequences and model the dynamic evolution of pipeline states over time. Pipeline degradation is a gradual, time-varying process, with a close correlation between the current state and historical states. LSTM networks, through gating mechanisms, effectively capture temporal dependencies over long time spans, avoiding the gradient vanishing problem faced by ordinary recurrent neural networks in long sequence modeling. This invention employs a bidirectional LSTM structure, modeling the sequence simultaneously from both forward and backward directions to more comprehensively capture temporal context information. The bidirectional LSTM design consists of two stacked LSTM layers, each with a hidden state dimension of 256. The concatenation of the forward and backward hidden states outputs a 512-dimensional spatiotemporal fusion feature vector. .

[0042] The core computational process of an LSTM unit includes updating the forget gate, input gate, candidate cell states, and output gate. The forget gate determines how much information is discarded from the cell state at the previous time step, and its calculation formula is as follows:

[0043] ,

[0044] in, This represents the output vector of the forget gate. This represents the sigmoid activation function. This represents the forget gate weight matrix. This represents the hidden state vector from the previous time step. This represents the input vector at the current moment, i.e., the spatial feature vector. This represents the forget gate bias vector. This indicates a vector concatenation operation.

[0045] The input gate and candidate cell states together determine how much new information is added to the cell state, calculated using the following formula:

[0046] ,

[0047] ,

[0048] in, This represents the input gate output vector. Represents the candidate cell state vector. This represents the hyperbolic tangent activation function. This represents the input gate weight matrix. This represents the candidate cell state weight matrix. This represents the input gate bias vector. This represents the candidate cell state bias vector.

[0049] Cell state update combines the outputs of the forget gate and the input gate, and the calculation formula is as follows:

[0050] ,

[0051] in, This represents the cell state vector at the current moment. This represents the cell state vector from the previous time step. This indicates element-wise multiplication.

[0052] The output gate determines the hidden state of the output at the current moment, and the calculation formula is:

[0053] ,

[0054] ,

[0055] in, This represents the output vector of the output gate. This represents the hidden state vector at the current time step. This represents the output gate weight matrix. This represents the output gate bias vector.

[0056] The degradation trend prediction and remaining life assessment module receives spatiotemporal fusion feature vectors and outputs three key assessment indicators: pipeline degradation status level, remaining life, and probability of pipe burst risk, providing a basis for subsequent early warning and decision-making.

[0057] The degradation status grading unit inputs the spatiotemporal fusion feature vector into a pre-trained degradation grading network. This network adopts a multi-layer fully connected structure, containing three hidden layers with 256, 128, and 64 neurons respectively. Finally, a 5-dimensional softmax output layer generates probability distributions for five degradation levels. The degradation status grading criteria are as follows: Normal level indicates that the pipe structure is intact with no detectable damage or degradation signs; Slight degradation level indicates that the pipe has slight surface corrosion or deposits, which does not affect normal use; Moderate degradation level indicates that the pipe has obvious wall thinning or local damage, requiring close attention; Severe degradation level indicates that the pipe has serious structural damage and there is a risk of leakage; Severe degradation level indicates that the pipe structure is severely damaged and a pipe burst accident may occur at any time. The level with the highest probability is taken as the final grading result, and the corresponding probability value is used as the grading confidence. The calculation formula of the degradation status comprehensive scoring algorithm is as follows:

[0058] ,

[0059] in, Indicates the degradation level value. Indicates the first The predicted probability at each level, Indicates the first The weight vector corresponding to each level Indicates the first The bias scalar corresponding to each level This represents the spatiotemporal fusion feature vector.

[0060] The remaining service life prediction unit uses an improved Weibull distribution model to calculate the pipeline's remaining service life. Traditional Weibull distribution models typically estimate shape and scale parameters based on historical failure data, making it difficult to reflect the pipeline's current actual condition. The improved method proposed in this invention uses a spatiotemporal fusion feature vector as input and dynamically predicts shape and scale parameters through a fully connected network, enabling the model to adaptively adjust based on real-time monitoring data. The reliability function of the improved Weibull distribution is:

[0061] ,

[0062] in, Indicates the time of the pipeline Reliability, This represents the scale parameter determined by the spatiotemporal fusion feature vector. This represents the shape parameters determined by the spatiotemporal fusion feature vector. The dynamic prediction formulas for the shape and scale parameters are as follows:

[0063] ,

[0064] ,

[0065] in, This represents the weight vector for shape parameter prediction. A scalar representing the bias of shape parameter prediction. The weight vector represents the prediction of the scaling parameters. This represents the bias scalar for the scale parameter prediction. This indicates the softplus activation function, used to ensure that the output parameter is a positive value.

[0066] Remaining useful life is defined as the reliability decreasing to a preset threshold. The corresponding time is calculated using the following formula:

[0067] ,

[0068] in, Indicates the remaining service life. This represents the preset reliability threshold, which is set to 0.1 in this embodiment. This indicates the current service life of the pipeline.

[0069] The pipe burst risk quantification unit comprehensively considers the deterioration state level, remaining service life, and static properties of the pipeline to calculate the probability of a pipe burst within a preset time window. The formula for calculating the pipe burst risk probability is:

[0070] ,

[0071] in, Indicates the probability of pipe bursting. This indicates the preset time window length, which is set to 365 days in this embodiment. The material risk factor is indicated as follows: cast iron pipe is set to 1.2, ductile iron pipe to 1.0, steel pipe to 0.9, and PE pipe to 0.8. This indicates the risk coefficient for pipe diameter; smaller diameter pipes have a higher risk coefficient than larger diameter pipes. This indicates the risk coefficient based on the pipe's age; the risk coefficient increases for pipes that have exceeded their designed service life.

[0072] The multi-level early warning and decision generation module receives the degradation status level, remaining service life, and probability of pipe burst risk, and generates graded early warning signals and actionable maintenance strategy suggestions.

[0073] The early warning level determination unit maps pipeline status to four warning levels based on a preset risk threshold matrix. A blue warning indicates normal or slightly deteriorated pipeline status with a burst risk probability below 0.1, allowing for routine inspections. A yellow warning indicates moderate pipeline deterioration with a burst risk probability between 0.1 and 0.3, requiring maintenance within 3 months. An orange warning indicates severe pipeline deterioration with a burst risk probability between 0.3 and 0.6, requiring maintenance within 1 month. A red warning indicates severe pipeline deterioration with a burst risk probability exceeding 0.6 or remaining service life less than 30 days, requiring emergency handling within 7 days. The rules for determining the warning level are expressed as follows:

[0074] ,

[0075] in, Indicates the warning level value. This indicates that the first risk threshold is set to 0.1. This indicates that the second risk threshold is set to 0.3. This indicates that the third risk threshold is set to 0.6. This indicates that the remaining lifespan threshold is set to 30 days.

[0076] The maintenance strategy generation unit calculates the maintenance priority score for each pipeline and generates specific maintenance strategy recommendations. The calculation of the maintenance priority score comprehensively considers the probability of pipe burst risk, the number of users the pipeline serves, pipeline repair costs, and the degree of traffic impact on the area where the pipeline is located. The formula for the comprehensive ranking algorithm for maintenance priorities is:

[0077] ,

[0078] in, This indicates the maintenance priority score. , , , The weight coefficients of each factor satisfy the following conditions: In this embodiment, it is set to , , , , Indicates the number of users the pipeline can support. This represents the maximum number of users that a single pipeline in the network can support. This represents the estimated cost of pipeline maintenance. This represents the maximum maintenance cost of a single pipeline in the pipeline network. An index indicating the degree of traffic impact in the area where the pipeline is located. This represents the maximum value of the traffic impact index.

[0079] Based on the maintenance priority score, the system generates maintenance strategy recommendations including maintenance time windows, maintenance methods, and maintenance resource configurations. The maintenance method is determined according to the level of deterioration: for lightly deteriorated pipelines, trenchless repair techniques such as lining repair are recommended; for moderately deteriorated pipelines, partial replacement or reinforcement measures are recommended; and for severely and critically deteriorated pipelines, a complete section replacement is recommended. Maintenance resource configuration includes the required number of construction personnel, type of construction equipment, specifications of pipe materials and spare parts, and the estimated construction period.

[0080] To ensure the system's prediction accuracy continuously improves over time, this invention introduces a closed-loop feedback optimization unit. This unit receives feedback data on the pipeline's condition after actual maintenance, including the actual pipeline degradation level before and after maintenance, the actual remaining wall thickness measurement, and the maintenance effectiveness evaluation results. The feedback data is compared with the system's prediction results to calculate the prediction error. The closed-loop feedback model incremental update algorithm uses online learning to update the model parameters of the degradation trend prediction and remaining life assessment modules. The parameter update formula is:

[0081] ,

[0082] in, This represents the updated set of model parameters. This represents the set of model parameters before the update. This represents the learning rate, which is set to 0.001 in this embodiment. This represents the gradient of the feedback loss function with respect to the model parameters. The feedback loss function is defined as:

[0083] ,

[0084] in, The cross-entropy loss represents the prediction of the degradation level. This represents the mean squared error loss in the remaining lifetime prediction. and The weighting coefficients representing the two losses are set to [value] in this embodiment. and .

[0085] This invention utilizes a technical approach involving multi-source heterogeneous data fusion, deep spatiotemporal feature extraction, degradation trend prediction, and closed-loop feedback optimization to achieve proactive prediction and tiered early warning of urban water supply network leakage. This transforms the traditional post-event detection and passive repair model into a pre-event prediction and proactive prevention model, effectively reducing the network leakage rate and improving the level of water supply security.

[0086] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A smart pipeline network leakage proactive prediction and early warning system based on multi-technology integration, characterized in that, include: The multi-source heterogeneous data fusion acquisition module is used to collect ultrasonic guided wave signals, pressure time series data, flow time series data and environmental factor data during pipeline operation, and to align the collected multi-source heterogeneous data with timestamps and spatial locations to generate a fused data tensor. The spatiotemporal feature depth extraction module is used to receive the fused data tensor, perform continuous wavelet transform on the ultrasonic guided wave signal to generate a time-frequency energy map, extract spatial feature vectors from the time-frequency energy map through a convolutional neural network, perform time-series modeling on the spatial feature vectors through a long short-term memory network to capture the evolution law of the pipeline network state, and output spatiotemporal fusion feature vectors. The degradation trend prediction and remaining service life assessment module is used to receive the spatiotemporal fusion feature vector, determine the degradation state level of the pipeline based on the spatiotemporal fusion feature vector, calculate the remaining service life of the pipeline based on the degradation state level and the pipeline's historical operating data, and quantify the probability of pipeline burst risk within a preset time window based on the degradation state level and the remaining service life. A multi-level early warning and decision generation module is used to receive the degradation status level, the remaining service life, and the probability of pipe bursting. When the probability of pipe bursting exceeds a preset first risk threshold, a first early warning level signal is generated. When the probability of pipe bursting exceeds a preset second risk threshold and the remaining service life is lower than a preset service life threshold, a second early warning level signal is generated. Based on the early warning level signal and the degradation status level, decision information including maintenance priority ranking and maintenance strategy suggestions is generated.

2. The system according to claim 1, characterized in that, The multi-source heterogeneous data fusion and acquisition module includes: An ultrasonic guided wave signal acquisition unit is used to deploy a piezoelectric sensor array on the outer wall of a pipe to excite and receive ultrasonic guided wave signals propagating along the pipe wall, wherein the sampling rate of the ultrasonic guided wave signals is not less than 1MHz. The pressure and flow time series acquisition unit is used to deploy pressure sensors and flow sensors at key nodes of the pipeline to collect pressure time series data and flow time series data in the pipeline. The sampling period of the pressure time series data and the flow time series data is no more than 1 second. The environmental factor acquisition unit is used to deploy environmental sensors in the pipeline burial area to collect environmental factor data such as soil corrosivity index, soil temperature, soil moisture and groundwater level.

3. The system according to claim 1, characterized in that, The spatiotemporal feature depth extraction module includes: The continuous wavelet transform processing unit is used to perform continuous wavelet transform on the ultrasonic guided wave signal using Morlet wavelet as the mother wavelet to generate a time-frequency energy map characterizing the distribution of signal energy in the time and frequency dimensions. A convolutional neural network feature extractor, comprising multiple convolutional layers, batch normalization layers, and pooling layers connected in sequence, is used to extract spatial feature vectors of pipeline structural damage from the time-frequency energy map. The Long Short-Term Memory Network Temporal Modeler includes a multi-layer bidirectional Long Short-Term Memory Network, which receives the spatial feature vector and models the temporal evolution of the network state, and outputs a spatiotemporal fusion feature vector that integrates spatial and temporal features.

4. The system according to claim 1, characterized in that, The degradation trend prediction and remaining life assessment module includes: The degradation state classification unit is used to input the spatiotemporal fusion feature vector into the pre-trained degradation classification network and output the degradation state level of the pipeline. The degradation state level includes five levels: normal, mild degradation, moderate degradation, severe degradation and serious degradation. The remaining service life prediction unit is used to calculate the remaining service life of the pipeline based on the degradation status level and the pipeline's historical failure records, using an improved Weibull distribution model. The pipe burst risk quantification unit is used to calculate the probability of pipe burst risk within a preset time window by comprehensively considering the degradation status level, the remaining service life, the material properties of the pipe, the pipe diameter parameters, and the number of years it has been buried.

5. The system according to claim 1, characterized in that, The multi-level early warning and decision generation module includes: The warning level determination unit is used to receive the probability of pipe bursting, the remaining service life and the level of deterioration, and determine the warning level according to the preset risk threshold matrix. The warning level includes four levels: blue warning, yellow warning, orange warning and red warning. The maintenance strategy generation unit is used to calculate the maintenance priority score of each pipeline based on the warning level, the deterioration status level and the pipeline topology, and generate maintenance strategy recommendations that include maintenance time windows, maintenance methods and maintenance resource configurations.

6. The system according to claim 1, characterized in that, The system also includes: The closed-loop feedback optimization unit is used to receive feedback data on the pipeline status after actual maintenance operations, compare the feedback data with the prediction results of the degradation trend prediction and remaining life assessment module, calculate the prediction error, and update the model parameters of the degradation trend prediction and remaining life assessment module based on the prediction error using an incremental learning method.

7. The system according to claim 4, characterized in that, In the improved Weibull distribution model used by the remaining lifetime prediction unit, the shape parameter and scale parameter are dynamically determined by the spatiotemporal fusion feature vector through a fully connected network. The input of the fully connected network is the spatiotemporal fusion feature vector, and the output is the shape parameter and the scale parameter.

8. The system according to claim 4, characterized in that, When the pipe burst risk quantification unit calculates the pipe burst risk probability, if the confidence level of the prediction result is lower than the preset confidence level threshold, it triggers the multi-source heterogeneous data fusion acquisition module to increase the sampling frequency, and re-inputs the data acquired after increasing the sampling frequency into the spatiotemporal feature depth extraction module for feature extraction.

9. The system according to claim 5, characterized in that, When calculating the maintenance priority score, the maintenance strategy generation unit comprehensively considers the probability of pipeline bursting, the number of users the pipeline carries, the pipeline maintenance cost, and the degree of traffic impact in the area where the pipeline is located, and uses a multi-objective weighted summation method to calculate the maintenance priority score.

10. The system according to claim 1, characterized in that, The multi-level early warning and decision generation module is also used to send the early warning level signal and the decision information to the pipeline network dispatch center through the communication interface, and receive the dispatch instructions returned by the pipeline network dispatch center to adjust the data acquisition strategy of the multi-source heterogeneous data fusion acquisition module.

Images