Pccp wire breakage risk assessment method and system based on optical fiber sensing and time sequence model

Abstract

This invention discloses a method and system for assessing PCCP (Pipeline Conduit Pipeline) wire breakage risk based on fiber optic sensing and a time-series model, belonging to the field of fiber optic sensing technology. The method includes: acquiring static attribute data of the pipeline and dynamic monitoring signals of wire breakage; establishing a basic prediction model for the static attribute data using a random forest algorithm, outputting an initial risk score and prior value for failure time; establishing an enhanced time-series model by combining a long short-term memory network to dynamically model the wire breakage characteristics, outputting a real-time updated predicted failure time; performing real-time iterative updates, calculating the feedback error based on the difference between the predicted and actual failure times when an actual wire breakage event is detected, triggering online learning, local retraining, and adaptive adjustment of feature weights; and correcting the prediction error through error-assisted regression to improve overall prediction stability and fault tolerance. This invention enables dynamic assessment and prediction of PCCP pipeline wire breakage risk, providing intelligent decision-making basis for operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing technology, and in particular to a PCCP fiber breakage risk assessment method and system based on fiber optic sensing and timing models. Background Technology

[0002] Prestressed concrete cylinder pipes (PCCPs) are widely used in large-scale water conveyance and municipal engineering projects due to their high strength and durability. However, during long-term operation, the internal prestressed steel wires of PCCPs may break due to corrosion, fatigue, or stress concentration. Once the steel wire breaks accumulate to a certain extent, it can lead to structural instability or even pipe rupture, causing serious engineering safety hazards.

[0003] Currently, the detection and assessment of PCCP broken wires mainly rely on offline methods such as acoustic emission, electromagnetic induction, or manual detection. These methods suffer from problems such as long detection cycles, limited coverage, and inability to achieve continuous monitoring, making it difficult to meet the real-time safety assessment requirements for large water transmission pipelines.

[0004] In recent years, distributed fiber acoustic sensing (DAS) technology has been widely used for online wire breakage monitoring in PCCPs due to its long-distance coverage, high spatial resolution, and sensitive response to transient events. DAS can continuously acquire wire breakage signals generated during pipeline operation, enabling real-time acquisition of the event's time, location, and characteristic frequency band. However, most existing DAS-based monitoring systems focus only on the identification and location of wire breakage events, remaining weak in areas such as the cumulative effect of wire breakage, structural degradation trend modeling, and remaining life prediction. Specifically, existing technologies cannot effectively address the correlation between monitoring data and structural mechanical behavior, and cannot use wire breakage time series to characterize its evolution process; traditional static models struggle to cope with long-term operating condition changes and noise interference, resulting in limited stability of prediction results; furthermore, these methods lack online self-updating capabilities to adapt to complex operating conditions, making it difficult to achieve consistent prediction performance across engineering scenarios.

[0005] Against this backdrop, there is an urgent need for an intelligent risk assessment method that can combine fiber optic sensing data with time-series prediction models. This method would not only enable real-time detection of wire breakage events but also allow them to be incorporated into a unified dynamic modeling framework. This would enable the inference of structural stress state, analysis of accumulated wire breakage trends, prediction of remaining life, and online error correction, providing more accurate and reliable prediction and early warning support for the health status of PCCP pipelines. Summary of the Invention

[0006] To address the problems of existing technologies, such as reliance on offline detection for broken wire monitoring, inability to dynamically assess failure trends, lagging model updates, and unstable prediction accuracy, this invention proposes a PCCP broken wire risk assessment method and system based on fiber optic sensing and a time-series model. By integrating distributed fiber acoustic sensing technology, a random forest static prediction model, a long short-term memory network (LSTM) time-series model, and an adaptive online update mechanism, an intelligent assessment system is constructed that can reflect the pipeline status in real time, dynamically predict failure time, and continuously self-calibrate, thereby improving the visualization and prediction capabilities of PCCP pipeline operation safety.

[0007] This invention adopts the following technical solution: a PCCP fiber breakage risk assessment method and system based on fiber optic sensing and timing model, comprising the following steps:

[0008] Step 1, Data Acquisition: Obtain static attribute data related to the pipeline and dynamic monitoring signals related to broken wires, perform signal preprocessing and feature extraction to obtain dynamically weighted input features;

[0009] Step 2, Predictive Model Construction: Risk modeling is performed using a combination of static feature-driven and dynamic time-series augmentation methods, including:

[0010] The basic prediction model takes static attribute data as input, uses the random forest algorithm, and establishes a nonlinear regression relationship with structural features and historical samples to output the initial risk score and prior estimate of failure time.

[0011] The dynamic enhancement model, based on the prior results output by the basic prediction model, combines an LSTM temporal network to perform temporal modeling on dynamically weighted input features, and outputs the dynamically updated predicted failure time based on the number, location and evolution trend of broken wires.

[0012] Step 3, Real-time Iterative Update Mechanism: When an actual wire breakage event is detected, the time difference between the predicted failure time and the actual failure time is calculated, the most recent error samples are recorded, the average error and volatility are statistically analyzed, and an adaptive update is triggered when the set threshold is exceeded; and an auxiliary regression model is constructed to learn the nonlinear relationship between static attribute data and dynamic monitoring signals and the time difference, and to correct the prediction output of the LSTM time series network.

[0013] Step 4: System Deployment and Visualization Output. Based on the updated prediction model, generate the risk level, estimated failure time, and recommended maintenance time, and output them through visualization.

[0014] As a preferred embodiment, the static attribute data includes: pipe burial depth, material type, pipe diameter, wall thickness, steel wire structure, concrete grade, and design pressure rating;

[0015] The dynamic monitoring signal is collected in real time along the pipeline by a distributed fiber optic acoustic sensing system to collect wire breakage events, including: the number of broken wires, the breakage angle, the spatial location, and multi-dimensional sequence feature information of the wire breakage over time.

[0016] This data provides a complete information foundation for subsequent risk assessment models.

[0017] As a preferred embodiment, the signal preprocessing performs noise reduction and feature enhancement on the fiber optic monitoring signal, including three-stage processing: spectral subtraction, EMD decomposition, and Kalman filtering, to improve the transient consistency and timing stability of the broken fiber signal, thereby improving the quality of the broken fiber signal and the usability of the model.

[0018] The feature extraction process involves extracting time-domain features, frequency-domain features, or time-series statistical features related to the wire breakage event from the processed signal. Based on the occurrence frequency, signal energy, or duration of the wire breakage event, the weights of the features are adaptively calculated to obtain dynamically weighted input features.

[0019] As a preferred approach, the prediction model combines static attributes and dynamic monitoring information for risk modeling. By constructing a basic prediction model and a dynamic enhancement model, it learns the cumulative effects and change patterns of wire breakage events, providing a more accurate dynamic representation of pipeline health status.

[0020] Among them, the basic prediction model uses the random forest algorithm to establish a nonlinear regression relationship based on static attribute data, and outputs the initial risk score R0 and the prior estimate of the failure time T0.

[0021] The dynamic enhancement model employs a two-layer LSTM temporal network, with a recent time window as input. The sequence of broken wire events within the circuit is used to output the dynamically updated estimated failure time T.

[0022] As a preferred embodiment, the LSTM temporal network employs a sliding time window mechanism to handle time dependencies, is trained using a weighted MSE loss function, and the sample weights are adaptively adjusted based on the significance of the wire breakage event.

[0023] The time window length is adaptively expanded or contracted based on operational characteristics such as changes in operating pressure and the frequency of wire breakage events, ranging from 30 to 180 days. The window sliding step size is 10% to 30% of the window length to ensure the continuity of time-series information and the coverage of new information. Furthermore, a Bayesian online change point detection algorithm is used to identify local abrupt changes in the wire breakage signal, with a detection confidence level of no less than 0.95. When a sudden change in operating parameters or a change in wire breakage signal characteristics is detected, local window shortening and resampling are triggered, enabling the model to promptly capture trend changes caused by sudden events.

[0024] As a preferred embodiment, the adaptive update method in step 3 includes:

[0025] Step 3.1, Small Sample Online Learning: Select data from the most recent M broken filament samples for fine-tuning learning;

[0026] Step 3.2, Local Retraining: When the error still does not improve after multiple consecutive small-sample online learning sessions, local retraining is performed based on the monitoring data of the most recent period, using an early stopping strategy to prevent overfitting;

[0027] Step 3.3, Feature Weight Adaptive Adjustment: Recalculate the importance of the input features based on the spatial distribution and frequency characteristics of the wire breakage event. When the rate of change exceeds the preset threshold, trigger a weight update.

[0028] Step 3.4: After the update is completed, perform validation on the reserved samples. When the prediction error is lower than before the update, automatically roll back to the version before the update.

[0029] Few-shot online learning enables rapid model updates with a small number of new samples, while local retraining is used to correct long-term accumulated model biases. Adaptive adjustment of feature weights is used to further enhance the model's sensitivity to changes in key features, and the model rollback mechanism is used to avoid performance degradation caused by updates.

[0030] As a preferred embodiment, the adaptive update also includes a feature reweighting mechanism based on changes in broken wire characteristics or pipe condition labels, periodically calculating the changes in the importance of each input feature, and triggering a weight update when the change of a single feature relative to the baseline exceeds a preset threshold; the update process adjusts the feature weight vector through constraint optimization, and smooths the feature weights of spatially adjacent pipe segments to prevent local abrupt changes.

[0031] As a preferred embodiment, the auxiliary regression model learns the nonlinear relationship between static attribute data and dynamic monitoring signals and time difference ΔT, which can dynamically correct the prediction output of the LSTM time series network, further enhancing the model's fault tolerance and stability under complex operating conditions.

[0032] The present invention also provides: a PCCP filament breakage risk assessment system for implementing the above method, comprising:

[0033] The fiber optic signal acquisition unit includes a static attribute data acquisition module and a fiber optic sensing and monitoring module, which are used to acquire static attribute data related to the pipeline and dynamic monitoring signals related to broken wires, respectively.

[0034] The feature extraction unit is used to perform feature processing on the dynamic monitoring signal, including: denoising the fiber optic sensing monitoring signal to reduce the impact of environmental noise and unstructured disturbances; extracting time-domain features, frequency-domain features, or time-series statistical features related to the fiber breakage event from the processed signal; and adaptively calculating the weights of the features according to the occurrence frequency, signal energy, or duration of the fiber breakage event to provide weighted input features for the risk assessment unit.

[0035] The risk assessment unit is used to build a prediction model and conduct risk assessment. It takes static attribute data as input, uses the random forest algorithm, and establishes a nonlinear regression relationship with structural features and historical samples to output the initial risk score and the prior estimate of the failure time. Based on the prior results output by the basic prediction model, it combines the LSTM time series network to perform time series modeling on the dynamic features related to the number and location of broken wires, and outputs the dynamically updated expected failure time.

[0036] The model update unit includes a feedback error acquisition module and an error-assisted regression module. The feedback error acquisition module calculates the difference between the predicted failure time and the actual failure time, and determines whether the model needs to be updated based on the average value and volatility of the most recent error samples. The error-assisted regression module learns the nonlinear relationship between static attribute data and dynamic monitoring signals and the time difference, and refines the prediction output of the LSTM time series network.

[0037] The results output unit is used to generate the risk level, estimated failure time, and recommended maintenance time, and to provide a visual output.

[0038] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0039] 1. The method of this invention integrates distributed fiber acoustic sensing technology, random forest static prediction model, long short-term memory network (LSTM) time series model and adaptive online update mechanism to build an intelligent evaluation system that can reflect the pipeline status in real time, dynamically predict failure time and continuously self-calibrate, which can significantly improve the visualization and prediction capabilities of PCCP pipeline operation safety.

[0040] 2. The system of this invention has strong real-time performance, high prediction accuracy, strong self-learning ability, and sensitivity to abnormal changes. It can run on local servers, cloud platforms, or monitoring centers. It has high real-time performance, high robustness, and scalability. It is widely used in water conservancy, municipal and industrial water transmission fields, providing continuous and reliable intelligent support for PCCP pipeline health assessment and operation and maintenance decisions. Attached Figure Description

[0041] Figure 1 This is a flowchart of the PCCP wire breakage risk assessment method of the present invention;

[0042] Figure 2 This is a pipeline operating pressure field modeling diagram according to an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of pipeline risk scoring according to an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the application will be further described in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in this invention. All non-innovative embodiments based on these embodiments by other researchers in the art are within the protection scope of this invention. Furthermore, the step numbers in the embodiments of this invention are only set for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0045] Example 1

[0046] A method for assessing PCCP fiber breakage risk based on fiber sensing and timing models is provided, the process of which is as follows: Figure 1 As shown, the specific steps include the following:

[0047] Step 1: Data Acquisition: Obtain static attribute data related to the pipeline and dynamic monitoring signals related to broken wires, and perform signal preprocessing.

[0048] First, the structural static properties of the PCCP are collected, including key parameters such as pipe diameter (typical range Φ1600–Φ4000 mm), wall thickness (70–125 mm), burial depth (2–8 m), concrete grade (C40–C60), tensile strength of steel wires, steel wire spacing, and design working pressure (1.0–2.5 MPa). This information is then normalized and encoded into a static feature vector, represented as follows:

[0049] ;

[0050] in, These represent the burial depth, inner radius, outer radius, wall thickness, design working pressure, and concrete grade, respectively.

[0051] Then, the dynamic monitoring section uses a distributed fiber optic acoustic sensing system to continuously collect wire breakage events along the pipeline.

[0052] The DAS system deploys optical fibers along the pipeline and reflects local strain disturbances by analyzing the spatiotemporal variations of Rayleigh scattering signals generated by laser pulses, recording the occurrence time t of fiber breakage events. k axial position z k Circumferential angle Dynamic disturbance amplitude Information such as the duration of the event constitutes a dynamic input sequence that updates over time, represented as:

[0053] ;

[0054] Each wire breakage event data point can reflect the temporal characteristics of local structural damage, providing a basis for judging the degree of wire breakage accumulation, regional fatigue characteristics, and structural risk change trends, as well as dynamic characteristic information.

[0055] Since DAS signals are susceptible to environmental noise, system noise, and unstructured disturbances, this embodiment also constructs a systematic signal preprocessing mechanism before model input, including three-level processing: spectral subtraction, EMD decomposition, and Kalman filtering.

[0056] The spectral subtraction method uses a 512-point noise estimation window, a 50% overlap rate, and a noise compensation coefficient of 1.1 to improve the clarity of the original signal through frequency domain denoising.

[0057] EMD decomposition breaks down the signal into 5–8 intrinsic mode functions (IMFs) and automatically selects the IMFs with a dominant frequency of 8–30 kHz and an energy percentage of more than 15% for wire breakage event reconstruction.

[0058] Kalman filter models use state matrices A Observation matrix H =(1,0), process noise covariance Q =0.01 and observation noise covariance R =0.1 is used for dynamic smoothing to make the transient characteristics of broken wires more recognizable in the time domain.

[0059] After the above processing, the signal-to-noise ratio of this embodiment can typically be improved by more than 10dB, which can significantly enhance the stability and reliability of subsequent modeling inputs.

[0060] Step 2, Prediction Model Construction: Risk modeling is carried out by combining static feature-driven and dynamic time-series enhancement.

[0061] Specifically, the static model uses the random forest algorithm, with 200–500 decision trees, a maximum depth of 10–20, and a minimum number of split samples of 2. It uses structural features to establish a nonlinear mapping relationship with historical samples.

[0062] Predict and output the initial failure risk score. Compared with the expected failure time , denoted as:

[0063] ;

[0064] Among them, the function This represents the Random Forest algorithm.

[0065] Specifically, the static model part uses static feature vectors. The multidimensional combined feature vector is constructed based on the stress and durability degradation mechanism of PCCP structure. It includes geometric and structural constraint features: pipe diameter, wall thickness, burial depth, steel wire spacing and their ratio relationship; material degradation sensitivity features: concrete strength grade, steel wire tensile strength, design service life; and structural safety redundancy characterization features: steel wire density to design pressure ratio, wall thickness to inner diameter ratio, etc.

[0066] By normalizing and combining the above features, a static feature input space is formed to characterize the initial structural safety boundary of the pipeline. A random forest model is then used to learn the nonlinear mapping relationship between static structural properties and historical wire breakage failure samples, outputting an initial risk score. and prior estimate of failure time .

[0067] Meanwhile, through the internal feature importance analysis of the random forest, structural dominant factors that contribute significantly to the initial failure risk are identified, and the importance ranking results are used as the prior basis for the subsequent feature weighting and initialization of the dynamic model.

[0068] Based on this, the dynamic augmentation model constructs a time series prediction model based on a long short-term memory network, taking a recent time window as input. Dynamic input sequence of broken wire events within Output the corrected estimated failure time T , represented as:

[0069] ;

[0070] in, For sample weights, This is the timing processing function for LSTM.

[0071] Specifically, the LSTM network in this embodiment has a two-layer structure, with a first-layer hidden unit 64 and a second-layer hidden unit 32. The dropout is set to 0.2, and the input filament breakage event sequence is a sliding time window (length 30–180 days, step size 10–30% of the window length).

[0072] This invention employs a hierarchical modeling structure that combines a static prediction model with a dynamic enhancement model. The static prediction model characterizes the initial failure risk baseline of the PCCP pipeline under structural parameters and design constraints, and outputs an initial risk score. and prior estimate of failure time This is used to reflect the long-term structural safety level without considering the wire breakage evolution process.

[0073] Based on this, the dynamic enhancement model does not directly predict the failure time from scratch. Instead, it uses the prior results output by the static model as the initial reference for time prediction. It models the cumulative characteristics, frequency of occurrence and evolution trend of wire breakage events over time through an LSTM time series network, and dynamically corrects and updates the prior failure time, thereby achieving a sensitive response to short-term abnormal behavior and sudden degradation process.

[0074] The LSTM network model uses a sliding window mechanism to handle time dependencies and is trained using a weighted MSE loss function, with sample weights... w i Adaptive adjustment based on the significance of the wire breakage event:

[0075] ;

[0076] in, The significance of a wire breakage event is determined based on the wire breakage signal energy, duration, or characteristic frequency, using a weighting adjustment coefficient. We set the value to 0.1–0.3 to ensure that high-energy, more significant wire breakage events have a greater influence during model training.

[0077] Furthermore, The formula used to measure the deviation between the predicted failure time and the actual failure time is as follows:

[0078] ;

[0079] in, and represent the actual failure time and the corresponding failure time predicted by the model for the i-th wire breakage event, respectively.

[0080] Furthermore, an online change point detection algorithm is used to identify local mutations.

[0081] The online change point detection algorithm has been improved to address the characteristics of high noise, strong randomness, and statistical characteristics that change with operating conditions in wire breakage monitoring signals. The online change point detection algorithm uses the statistical features extracted from the wire breakage monitoring signal within a continuous time window as the observation input to characterize the changing trend of wire breakage events over time, thereby reducing the impact of instantaneous noise on the identification of sudden changes.

[0082] When a local mutation is detected, the local time window is shortened and the sampling density of the corresponding time period is increased. After the operating status stabilizes again, the time window length is gradually restored to ensure the stability of the long-term risk assessment results.

[0083] Finally, the LSTM outputs the real-time updated failure time T, which is compared with the static prior estimate. Together, these factors constitute a comprehensive assessment result, enabling the model to simultaneously possess the ability to identify long-term structural trends and respond to short-term emergencies. This mechanism significantly enhances the model's ability to respond to sudden evolution patterns of wire breakage, especially in complex noise backgrounds or when wire breakage events occur sporadically, effectively identifying pipeline risk change trends and high-risk precursors.

[0084] Step 3: Real-time iterative update mechanism.

[0085] To enable the model to adapt to changes in operating conditions, wire breakage behavior patterns, and noise fluctuations during long-term operation, this embodiment designs an adaptive online update mechanism. When an actual wire breakage event is detected, the time difference between the predicted failure time and the actual failure time is calculated, the most recent error samples are recorded, and the average error and fluctuation are statistically analyzed. When the error exceeds a set threshold, an adaptive update is triggered.

[0086] Specifically, after detecting a real wire breakage event, the system calculates the prediction error. Record the most recent 20–50 error samples. When the mean error exceeds 0.1T or the error variance exceeds... At that time, the online fine-tuning mechanism is automatically triggered, using the learning rate. By updating some parameters of the LSTM through 5-10 iterations, short-term offsets can be quickly corrected.

[0087] When the error continues to deviate, the system extracts a local training set from the monitoring data of the most recent 3–6 months for local retraining and adopts an early stopping strategy with a patience value of 3 to prevent overfitting.

[0088] Meanwhile, this embodiment incorporates an adaptive feature weight adjustment mechanism, which determines whether the change threshold of 20-40% is exceeded based on the change in the importance of the input features. When triggered, the weights are updated through constraint optimization to ensure that the weight change does not exceed 30% of the original value and the sum remains 1. Smoothing constraints are also applied to adjacent pipe segments in space to avoid local abrupt distortion.

[0089] Furthermore, to improve the efficiency of the correction, an auxiliary regression model is introduced, which learns... With error The nonlinear relationship between them is used to refine the prediction output of the LSTM time series network.

[0090] Specifically, the auxiliary regression model is used to correct the predicted failure time through error compensation, including:

[0091] The time difference between the predicted failure time output by the LSTM time series network and the actual observed failure time is obtained as the feedback error. Using the static attribute data of the pipeline and the dynamic monitoring features of the wire breakage event as input, a regression relationship between the feedback error and the input features is established. Based on the regression relationship, the predicted failure time output by the LSTM time series network is corrected by error compensation to reduce the systematic prediction bias generated by the model during long-term operation.

[0092] In addition, this embodiment supports a weighting mechanism based on features such as geological labels, construction year, and broken wire density, which helps to improve the model's generalization ability and enable cross-engineering application.

[0093] Step 4: System Deployment and Visualization Output. Based on the updated prediction model, generate the risk level, estimated failure time, and recommended maintenance time, and output them through visualization.

[0094] In terms of system deployment, this embodiment constructs a graphical visualization output to facilitate engineers and technicians in quickly identifying high-risk areas.

[0095] In this embodiment, the pressure field of the pipeline operation is modeled, such as... Figure 2 As shown, the model displays the stress field distribution of the pipeline using a cloud map approach. Darker areas correspond to higher stress concentration levels, indicating locations with a higher risk of failure. Based on the current risk score R, four risk level intervals are then set. Pipeline risk scoring results are as follows Figure 3 As shown.

[0096] Finally, a visual interface for the pipeline early warning platform was built, which can display risk scores and estimated failure time windows in real time. It provides information such as recommended maintenance periods and inspection priority ranking, and supports historical data review and playback of broken wire distribution trends, so that maintenance personnel can conduct long-term structural evolution analysis and strategy evaluation.

[0097] In summary, the method of this invention forms a closed-loop workflow covering monitoring, prediction, updating, and display, while emphasizing the overall closed-loop adaptive characteristics. Whenever a new wire breakage event occurs, it can automatically complete the entire process of signal acquisition, risk prediction, error comparison, parameter correction, and result output, ensuring that the prediction model always closely matches the actual operating state and can promptly reflect the impact of complex factors such as structural fatigue, environmental disturbances, or changes in operating conditions on the risk.

[0098] Example 2

[0099] A PCCP fiber breakage risk assessment system is provided, including an optical fiber signal acquisition unit, a feature extraction unit, a risk assessment unit, a model update unit, and a result output unit. These units are connected via a data bus to form a complete closed-loop feedback system, capable of outputting the fiber breakage risk level, estimated failure time, and maintenance recommendations.

[0100] This embodiment of the system is based on a deep understanding of the structural mechanical characteristics of PCCP, the mechanism of wire breakage evolution, and the behavior of distributed optical fiber acoustic sensing. It systematically integrates multi-source data acquisition, signal processing, time series modeling, error feedback-driven adaptive update mechanism, and visualized risk output to form an intelligent wire breakage risk prediction system that can operate online for a long time.

[0101] Specifically, the fiber optic signal acquisition unit obtains static attribute data and dynamic monitoring signals related to the pipeline through the system data acquisition module. Static attribute data includes basic engineering parameters such as pipeline burial depth, material type, pipe diameter, wall thickness, steel wire structure, concrete grade, and design pressure rating. Then, the fiber optic sensing and monitoring module acquires dynamic monitoring signals in real time, including the number of broken wires, fracture angle, and spatial location, providing a complete information foundation for subsequent risk assessment models.

[0102] The feature extraction unit preprocesses the dynamic monitoring signal through the signal preprocessing module. It uses spectral subtraction, empirical mode decomposition (EMD), and Kalman filtering to denoise and enhance the fiber optic monitoring signal, thereby reducing the impact of environmental noise and unstructured disturbances and improving the signal's transient consistency and stability. The feature extraction module extracts multidimensional sequence features of the signal changing over time at the broken section of the pipeline based on the frequency of the broken wire event, signal energy, or duration. The feature weights are adaptively calculated and input into the subsequent risk assessment unit.

[0103] The risk assessment unit is used to build a prediction model and conduct risk assessment. It takes static attribute data as input, uses the random forest algorithm, and establishes a nonlinear regression relationship with structural features and historical samples to output the initial risk score and prior estimate of failure time. Based on the prior results output by the basic prediction model, it combines the LSTM time series network to perform time series modeling on the dynamic features related to the number and location of broken wires, and outputs the dynamically updated expected failure time.

[0104] The model update unit includes a feedback error acquisition module and an error-assisted regression module. The feedback error acquisition module calculates the difference between the predicted failure time and the actual failure time, and determines whether the model needs to be updated based on the average value and volatility of the most recent error samples. The error-assisted regression module learns the nonlinear relationship between static attribute data and dynamic monitoring signals and the time difference, and refines the prediction output of the LSTM time series network.

[0105] The results output unit is used to generate the risk level, estimated failure time, and recommended maintenance time, and to provide a visual output.

[0106] The system of this invention can run on a local server, cloud platform or monitoring center, and has high real-time performance, high robustness and scalability. It combines real-time performance, scalability and robustness, and is suitable for PCCP pipeline structural health monitoring tasks of different project scales, different geological conditions and different service years. It can be widely used in water conservancy, municipal and industrial water transmission fields, and provides continuous and reliable intelligent support for PCCP pipeline health assessment and operation and maintenance decision-making.

[0107] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A PCCP wire break risk assessment method based on optical fiber sensing and time series model, characterized in that, Includes the following steps: Step 1, Data Acquisition: Obtain static attribute data related to the pipeline and dynamic monitoring signals related to broken wires, perform signal preprocessing and feature extraction to obtain dynamically weighted input features; Step 2, Predictive Model Construction: Risk modeling is performed using a combination of static feature-driven and dynamic time-series augmentation methods, including: The basic prediction model takes static attribute data as input, uses the random forest algorithm, and establishes a nonlinear regression relationship with structural features and historical samples to output the initial risk score and prior estimate of failure time. The dynamic enhancement model, based on the prior results output by the basic prediction model, combines an LSTM temporal network to perform temporal modeling on dynamically weighted input features, and outputs the dynamically updated predicted failure time based on the number, location and evolution trend of broken wires. The dynamic enhancement model employs a two-layer LSTM temporal network, with a recent time window as input. Internal filament breakage event sequence Output the dynamically updated estimated expiration time. : ； wherein, is the sample weight, is the LSTM time series processing function; Step 3, Real-time Iterative Update Mechanism: When an actual wire breakage event is detected, the time difference between the predicted failure time and the actual failure time is calculated, the most recent error samples are recorded, the average error and volatility are statistically analyzed, and an adaptive update is triggered when the set threshold is exceeded; and an auxiliary regression model is constructed to learn the nonlinear relationship between static attribute data and dynamic monitoring signals and the time difference, and to correct the prediction output of the LSTM time series network. Step 4: System Deployment and Visualization Output. Based on the updated prediction model, generate the risk level, estimated failure time, and recommended maintenance time, and output them through visualization.

2. The PCCP wire break risk assessment method of claim 1, wherein, The static attribute data includes: pipe burial depth, material type, pipe diameter, wall thickness, steel wire structure, concrete grade, and design pressure rating; The dynamic monitoring signal is collected in real time along the pipeline by a distributed fiber optic acoustic sensing system to collect wire breakage events, including: the number of broken wires, the breakage angle, the spatial location, and multi-dimensional sequence feature information of the wire breakage over time.

3. The PCCP wire break risk assessment method of claim 1, wherein, The signal preprocessing performs noise reduction and feature enhancement on the dynamic monitoring signal, including three-stage processing: spectral subtraction, EMD decomposition, and Kalman filtering, to improve the transient consistency and temporal stability of the wire breakage signal. The feature extraction process involves extracting time-domain features, frequency-domain features, or time-series statistical features related to the wire breakage event from the processed signal. Based on the occurrence frequency, signal energy, or duration of the wire breakage event, the weights of the features are adaptively calculated to obtain dynamically weighted input features.

4. The PCCP wire break risk assessment method of claim 1, wherein, The LSTM temporal network employs a sliding time window mechanism to handle time dependencies and is trained using a weighted MSE loss function, with sample weights... Adaptive adjustment based on the significance of the wire breakage event: ； in, Used to determine the significance of a wire breakage event, determined based on the wire breakage signal energy, duration, or characteristic frequency; This is the weighting adjustment coefficient, used to control the degree of influence of significance on sample weights; Used to measure the deviation between predicted failure time and actual failure time, the calculation method is as follows: ； wherein, respectively represent the actual failure time and the model predicted corresponding failure time for the ith filament breakage event.

5. The PCCP wire break risk assessment method of claim 4, wherein, The dynamic enhancement model uses an online change point detection algorithm to identify local mutations. The online change point detection algorithm is improved to address the characteristics of high noise, strong randomness, and statistical characteristics that change with operating conditions in wire breakage monitoring signals. It uses the statistical features extracted from the wire breakage monitoring signal within a continuous time window as the observation input to characterize the changing trend of wire breakage events over time and reduce the impact of instantaneous noise on change identification. When a local mutation is detected, the local time window is shortened and the sampling density of the corresponding time period is increased. After the operating status stabilizes again, the time window length is gradually restored to ensure the stability of long-term risk assessment results.

6. The PCCP wire break risk assessment method of claim 1, wherein, The adaptive update described in step 3 is as follows: Step 3.1, Small Sample Online Learning: Select data from the most recent M broken filament samples for fine-tuning learning; Step 3.2, Local Retraining: When the error still does not improve after multiple consecutive small-sample online learning sessions, local retraining is performed based on the monitoring data of the most recent period, using an early stopping strategy to prevent overfitting; Step 3.3, Feature Weight Adaptive Adjustment: Recalculate the importance of the input features based on the spatial distribution and frequency characteristics of the wire breakage event. When the rate of change exceeds the preset threshold, trigger a weight update. Step 3.4: After the update is completed, perform validation on the reserved samples. When the prediction error is lower than before the update, automatically roll back to the version before the update.

7. The PCCP wire break risk assessment method of claim 1, wherein, The adaptive update described in step 3 also includes a feature reweighting mechanism based on changes in broken wire features or pipe condition labels, which periodically calculates the changes in the importance of each input feature and triggers a weight update when the change of a single feature relative to the baseline exceeds a preset threshold. The update process adjusts the feature weight vector through constraint optimization, and smooths the feature weights of spatially adjacent pipe segments to prevent local abrupt changes.

8. The PCCP wire break risk assessment method of claim 1, wherein, The auxiliary regression model is used to compensate for errors in the predicted failure time, specifically including: The time difference between the predicted failure time output by the LSTM time series network and the actual observed failure time is obtained as the feedback error. Using the static attribute data of the pipeline and the dynamic monitoring features of the wire breakage event as input, a regression relationship between the feedback error and the input features is established. Based on the regression relationship, the predicted failure time output by the LSTM time series network is corrected by error compensation to reduce the systematic prediction bias generated by the model during long-term operation.

9. A PCCP filament breakage risk assessment system for implementing the method of any one of claims 1 to 8, comprising: The fiber optic signal acquisition unit includes a static attribute data acquisition module and a fiber optic sensing and monitoring module, which are used to acquire static attribute data related to the pipeline and dynamic monitoring signals related to broken wires, respectively. The feature extraction unit is used to perform feature processing on the dynamic monitoring signal, including: denoising the fiber optic sensing monitoring signal to reduce the impact of environmental noise and unstructured disturbances; extracting time-domain features, frequency-domain features, or time-series statistical features related to the fiber breakage event from the processed signal; and adaptively calculating the weights of the features according to the occurrence frequency, signal energy, or duration of the fiber breakage event to provide weighted input features for the risk assessment unit. The risk assessment unit is used to build a prediction model and conduct risk assessment. It takes static attribute data as input, uses the random forest algorithm, and establishes a nonlinear regression relationship with structural features and historical samples to output the initial risk score and the prior estimate of the failure time. Based on the prior results output by the basic prediction model, it combines the LSTM time series network to perform time series modeling on the dynamic features related to the number and location of broken wires, and outputs the dynamically updated expected failure time. The model update unit includes a feedback error acquisition module and an error-assisted regression module. The feedback error acquisition module calculates the difference between the predicted failure time and the actual failure time, and determines whether the model needs to be updated based on the average value and volatility of the most recent error samples. The error-assisted regression module learns the nonlinear relationship between static attribute data and dynamic monitoring signals and the time difference, and refines the prediction output of the LSTM time series network. The results output unit is used to generate the risk level, estimated failure time, and recommended maintenance time, and to provide a visual output.

Images