A municipal pipeline third-party damage monitoring and early warning method based on existing communication optical cable

Abstract

A kind of municipal pipeline third party damage monitoring and early warning method based on existing communication cable, distributed optical fiber acoustic sensing system is accessed to selected existing communication cable along municipal pipeline, on the basis of obtaining the geographic information data of municipal pipeline and the geographic information data of existing communication cable, the mutual spatial position relationship between the geographic coordinates of the existing communication cable and the geographic coordinates of municipal pipeline is established, the optical fiber acoustic vibration signal is collected and processed in real time by the distributed optical fiber acoustic sensing system, the root mean square value of optical fiber acoustic vibration signal is calculated and compared, suspected construction signal is identified and screened out, for suspected construction signal, channel identification model based on deep neural network is used to screen out channel-by-channel construction signal, event tracking module is used for event tracking analysis, the conversion from channel-by-channel result to independent construction event is realized, third party construction event is identified and its safety threat level to municipal pipeline is evaluated, early warning and visual presentation are carried out, so as to complete the online monitoring and early warning of third party damage on the municipal pipeline.The present application improves efficiency, reduces false positive rate, improves geographic position recognition accuracy and improves system reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the maintenance and management of municipal facilities, in particular to a municipal pipeline third-party damage monitoring and early warning method based on existing communication optical cables, belonging to the technical field of municipal engineering safety monitoring. BACKGROUND

[0002] In recent years, third-party damage caused by construction has caused damage to municipal pipelines such as gas pipelines and water supply pipelines, affecting urban production and people's lives, and thus the establishment of corresponding early warning technology is needed. At present, the municipal pipeline third-party damage monitoring technology based on machine vision has been widely used in engineering practice, but a single sensor can only cover a single point and cannot achieve full-line detection. The third-party construction damage monitoring technology based on acoustic signals has been reported in a large number of documents, but its stability and anti-interference performance need to be verified due to the influence of noise caused by vehicles, pedestrians, etc. Distributed optical fiber sensing technology has wide monitoring range, high sensitivity, and good application prospects.

[0003] In the prior art, Chinese patent application "Third-party construction monitoring method and system based on distributed optical fiber sensor" (application publication number: CN121278457A) proposes a distributed optical fiber third-party construction vibration signal acquisition system to accurately monitor the identification and early warning method of construction behavior mode in the region. The technology obtains the denoised optical fiber vibration signal, screens the suspected construction signal through the time domain feature, calculates the short-time energy and power spectral density of the suspected signal, and identifies the construction behavior based on artificial neural network, convolutional neural network or long short-term memory gate.

[0004] However, the prior art is to lay distributed optical fibers above the municipal pipeline to monitor third-party damage behavior, so it has the following defects: first, the prior art lays distributed optical fibers above the pipeline, which can only identify the type of third-party construction event, but it is difficult to determine the spatial position relationship between the construction position and the pipeline, so it cannot evaluate the threat level of the construction event to the safety of the municipal pipeline; second, the prior art relies on the short-time energy and power spectral density of the optical fiber vibration signal, and when the vibration amplitude caused by the third-party construction behavior is close to that caused by external noise such as vehicle vibration, the method has the risk of false positives; third, the technology uses the fast Fourier transform algorithm to extract the power spectral density feature from the vibration signal, which has the problem of low calculation efficiency, making it risky to delay the alarm when using long-distance communication optical cables for municipal pipeline safety monitoring. SUMMARY

[0005] This invention addresses the shortcomings of existing distributed optical fiber-based monitoring methods, which struggle to efficiently and accurately identify third-party construction damage over long-distance pipelines. It provides a more efficient and accurate method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables. This method utilizes the widely deployed existing communication optical cables in cities to acquire sensor signals, employs deep neural networks to automatically extract features to distinguish construction activities from external noise, and combines the geographical coordinates of the existing communication optical cables with the geographical coordinates of the municipal pipeline to determine the spatial relationship between the third-party damage and the municipal pipeline. This improves early warning accuracy, enhances computational efficiency, and avoids delayed or false alarms.

[0006] The objective of this invention is achieved through the following technical solution: A method for monitoring and early warning of third-party damage to municipal pipelines based on existing optical fiber cables is proposed. This method involves connecting a distributed optical fiber acoustic wave sensing system to the existing optical fiber cables along a selected municipal pipeline route. Based on the acquired geographic information data of both the municipal pipeline and the existing optical fiber cables, a spatial relationship is established between the geographic coordinates of the existing optical fiber cables and the municipal pipeline. The distributed optical fiber acoustic wave sensing system collects and processes optical fiber acoustic vibration signals in real time, calculates and compares the root mean square (RMS) values ​​of the signals, identifies and filters suspected construction signals, and uses a channel identification model based on a deep neural network to filter out channel-by-channel construction signals. An event tracking module is used for event tracking analysis, converting channel-by-channel results into independent construction events. Third-party construction events are identified, and their safety threat level to the municipal pipeline is assessed. Early warnings are then provided and visualized, thus completing online monitoring and early warning of third-party damage to the municipal pipeline.

[0007] Furthermore, the monitoring and early warning method includes the following steps: Step 1: Install the distributed fiber optic acoustic wave sensing system on-site and set the acquisition parameters. 1) Select qualified spare optical fibers from the existing communication optical cables along the municipal pipeline as sensing optical fibers, and connect the selected sensing optical fibers to the distributed optical fiber acoustic wave sensing system. 1. 2) Obtain the geographic information data of the municipal pipeline and the geographic information data of the existing communication optical cable, and determine the spatial relationship between the two to provide a geographic reference for subsequent event location; the geographic information data of the municipal pipeline includes, but is not limited to, pipeline location, direction, burial depth and kilometer marker coordinates, the geographic information data of the existing communication optical cable includes, but is not limited to, optical cable route, manhole location and fiber optic cable length, and the spatial relationship between the two includes, but is not limited to, parallel segments, intersections and vertical distances; 1. 3) Set the spatial sampling interval of the distributed optical fiber acoustic wave sensing system, determine the channel number of the sensing optical fiber, establish a one-to-one correspondence between the optical fiber channel number and the geographical coordinates using events with known geographical locations, and form a calibration table so that the geographical location of the subsequently detected events can be accurately located. 1. 4) Set the sampling frequency and gauge length of the distributed fiber optic acoustic wave sensing system; Step two involves real-time acquisition of fiber optic acoustic vibration signals through the fiber optic acoustic wave sensing system, subsequent data processing, and channel-by-channel signal identification and event tracking using a deep neural network to assess the safety threat level to the municipal pipeline, thereby completing online monitoring and early warning of third-party damage to the municipal pipeline. II. 1) Acquisition and processing of optical fiber acoustic vibration signals — 2. 1-1) The distributed optical fiber acoustic wave sensing system acquires the optical fiber acoustic vibration signal of each optical fiber channel on the sensing optical fiber in the existing communication optical cable in real time; 2.1-2) Set the time interval according to experience. For the latest time period of the acquired fiber optic acoustic vibration signal, calculate the root mean square value of the fiber optic acoustic vibration signal of each fiber optic channel obtained in step 2.1-1), and set the root mean square threshold of each fiber optic channel. 2) Perform channel identification — 2.2-1) Establish a channel recognition model based on a deep neural network and train the channel recognition model; 2.2-2) Use the trained channel recognition model to perform reasoning and identify and filter out the construction signals for each channel; 2.3) Conduct event tracking and analysis— 2. 3-1) Initialize the event tracking module, generate an empty third-party construction event pool, and configure and set channel thresholds and time interval thresholds for subsequent event determination; 2.3-2) Input the channel-by-channel identification results output in step 2.2-2) into the event tracking module to realize the conversion from channel-by-channel results to independent construction events, including the following specific contents: 2. 3-2a) In the channel-by-channel recognition results of the input, the recognition results with a channel interval less than the channel threshold are merged to form the same potential event at the current time; 2. 3-2b) Based on the channel threshold and time interval threshold, match the potential events at the current moment with the historical events in the third-party construction event pool: 2. 3-2b-1) Select historical events that are less than the time interval threshold from the current time to participate in the matching. Potential events with a channel interval less than the channel threshold are merged into the corresponding historical events. Update the termination time and the range of the affected channels of the corresponding historical events. 2. In step 2.3-2b-1), potential events that were not merged into the corresponding historical events are treated as new independent events and added to the third-party construction event pool. 2. 3-2b-3) For historical events whose time interval from the termination time to the current time exceeds the time interval threshold, if their influence range is greater than the channel threshold, they are saved as ended events and will not be updated again, and will no longer participate in subsequent potential event matching; otherwise, they are regarded as non-significant events and discarded. 2. 3-2b-4) After the event matching is completed at each moment, all historical events are output for display on the graphical platform, wherein the new historical events obtained in step 2. 3-2b-2) trigger alarm push; 2.4) Based on the spatial relationship between the geographical coordinates of the existing communication optical cable and the geographical coordinates of the municipal pipeline described in step 1.2), analyze the degree of safety threat posed by the construction event to the municipal pipeline, and generate a corresponding early warning level based on the analysis results; Step 3: Perform multi-dimensional statistics on the channel-by-channel identification results obtained in Step 2.2) and the event tracking results obtained in Step 2.3) to generate visualized image data for display, supporting operation and maintenance personnel to view and analyze.

[0008] Furthermore, in step one, 1), the method for selecting the sensing optical fiber is to conduct quality tests on the spare optical fiber in the existing communication optical cable, including but not limited to the fiber's total loss, the presence of breaks or macro bends, connector loss, and polarization mode dispersion. Spare optical fibers that meet the quality test requirements for distributed acoustic wave sensing are selected as the sensing optical fiber.

[0009] Optionally, in step one, 1), the distributed fiber optic acoustic wave sensing system is based on the principle of a phase-sensitive optical time-domain reflectometer.

[0010] Furthermore, in step 1.3), the known geographical location events include manual knocking, known construction points, and optical time domain reflectometer event points. The distance coordinates of the optical time domain reflectometer event points are known by precise determination by the optical time domain reflectometer, including fiber optic connectors, bends, and break points.

[0011] Optionally, in step 2.1-2), the method for setting the root mean square threshold of each optical fiber channel is as follows: through continuous detection for a certain period of time in the early stage, the root mean square value distribution of optical fiber acoustic vibration signal data of each optical fiber channel during quiet periods along the line is statistically analyzed to obtain the histogram of the root mean square value interval of each optical fiber channel, and then the 95th percentile value is taken as the root mean square threshold of each optical fiber channel.

[0012] Further, in step 2.2-1), the training method of the channel recognition model is as follows: the output module is a classification head, including a global average pooling layer, a fully connected layer, and a binary classification output layer; the training samples are slices of single event samples with a fixed number of channels and a fixed number of time samples; the channel recognition model learns the ability to recognize third-party construction signals through backpropagation and stochastic gradient descent; and the trained channel recognition model is verified using a test sample set to ensure the accuracy and precision of the channel recognition model.

[0013] Further, in step 2.2-2), the reasoning method of the channel identification model is as follows: the output module is switched to a segmentation head, including a fixed-step average pooling layer and two 1×1 convolutional layers; the fiber acoustic vibration signal data of all fiber channels obtained in steps 2.1-1) and 2.1-2) are directly input into the channel identification model for calculation; the root mean square value of the fiber acoustic vibration signal data of each fiber channel is compared with the root mean square threshold of the fiber channel one by one along the direction of the sensing fiber channel; the fiber acoustic vibration signal exceeding the root mean square threshold is a suspected construction signal, and the location of the fiber channel where the suspected construction signal is located is the geographical location of the third-party construction event; for fiber channels whose root mean square value does not exceed the root mean square threshold, the fill value is set to 0, and subsequent model calculations are skipped.

[0014] Furthermore, the channel recognition model adopts a dual-mode design. The stride of the average pooling layer is set to an integer multiple of the application range of the global average pooling layer of the channel recognition model during the training phase. The weights of the convolutional layer inherit the weights of the fully connected layer of the channel recognition model during the training phase, and the weight structure is transformed through dimensionality broadcasting.

[0015] Optionally, in step 2.3-1), the channel threshold is set according to the range of the number of optical fiber channels affected by a single construction signal, and consecutive channel events with a channel interval less than the channel threshold are merged into the same potential event; the time interval threshold is selected empirically and is an integer multiple of the update time interval of the optical fiber acoustic vibration signal.

[0016] Optionally, the visualization image data generated in step three includes: 1) GIS map – Overlaying display of the routes of the municipal pipelines and existing communication optical cables; when an alarm occurs, clicking the alarm icon displayed at the corresponding location will allow you to view event details including time, location, scope of impact, and processing status; 2) Waterfall chart of all-day monitoring results - intuitively displays the distribution of all construction events along the municipal pipeline at any given time; the horizontal axis represents the location of the fiber optic channel, the vertical axis represents time, and the color of the pixel indicates the construction status at that spatiotemporal point; 3) Hourly alarm duration statistics chart - with hours as the horizontal axis, the cumulative duration of construction alarms in each hour is displayed in the form of a bar chart to identify high-incidence periods of construction. 4) Alarm Event Push List - Displays all current alarm push information, including the event number, start time, stop time, duration, center location, channel width, and whether the processing has been completed.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) It has improved real-time processing capabilities and met the needs of ultra-long-distance monitoring.

[0018] This invention achieves efficient real-time inference through multi-level optimization: First, a large number of channels without anomalies are filtered out based on the root mean square value, reducing the computational load of subsequent models and improving computational efficiency; Second, a dual-mode segmentation model design is adopted, which can directly input and calculate all channel data and output channel-by-channel results during the inference stage, avoiding the cumbersome process of data slicing, segment-by-segment inference, and result summarization in traditional methods, thus shortening the processing cycle; In addition, the core computational content is encapsulated in a single neural network, which is conducive to the quantitative deployment of practical applications and further reduces the consumption of computing resources and the latency of inference.

[0019] (2) Improved recognition accuracy and significantly reduced false alarm rate.

[0020] While enhancing real-time processing capabilities, this invention employs a more complex deep convolutional neural network for construction signal identification. The convolutional neural network can automatically extract features without time-frequency domain transformation, solving the computational inefficiency problem of existing technologies. The channel identification model can learn signal features with higher dimensions than short-time energy and power spectral density, including the signal's temporal patterns, frequency distribution characteristics, and spatial correlation. Compared to existing technologies that rely on only two features and shallow neural networks for identification, the model of this invention has stronger nonlinear fitting capabilities and can effectively distinguish environmental interference with amplitudes close to construction signals (such as heavy vehicle passage or subway operation), effectively differentiating construction activities from external noise, thereby reducing false alarm rates and improving the reliability of early warnings.

[0021] (3) The positioning capability is accurate, realizing the association between the event location and geographic information.

[0022] This invention acquires GIS data of municipal pipelines and existing communication optical cables in advance and establishes a calibration mapping between optical fiber channel numbers and GIS coordinates. This enables the detected third-party construction events to be accurately located to real geographical coordinates (such as mileage markers or latitude and longitude), thus accurately determining the distance between the construction damage activities and the municipal pipelines. When an alarm is pushed, the system also provides the specific geographical location information of the event, which can be displayed intuitively on the GIS map. This facilitates maintenance personnel to quickly rush to the scene for handling, avoiding the shortcomings of existing technologies that can only report approximate mileage but cannot associate geographical location.

[0023] In summary, this invention improves efficiency, shortens the early warning cycle, avoids environmental interference, reduces the false alarm rate, improves the accuracy of geographic location identification, and enhances the reliability of the system. Attached Figure Description

[0024] Figure 1 This is a system overview diagram of the present invention.

[0025] Figure 2 This is a flowchart illustrating the real-time operation of the present invention.

[0026] Figure 3 To identify the learning network structure diagram of the model.

[0027] Figure 4 This is a flowchart of the event tracking module's operation. Detailed Implementation

[0028] The present invention will now be described in detail with reference to embodiments in order to provide a clearer understanding of the invention, but this should not be construed as limiting the scope of protection of the invention.

[0029] This invention is used for monitoring and early warning of damage to municipal pipelines caused by third-party construction. The aforementioned municipal pipeline third-party damage monitoring and early warning method based on communication optical cables uses sensing optical fibers deployed along the measured road to monitor in real time the acoustic vibration signals of the road surface under different water accumulation states caused by vehicle traffic events. The signal is compared with the correspondence model between the acoustic vibration signals of the sensing optical fibers and the water accumulation state of the road surface established by machine learning methods. The actual water accumulation state of the measured road surface is analyzed and judged and output, thereby realizing real-time and accurate perception and monitoring of the water accumulation state along the road.

[0030] Please see Figure 1 The monitoring and early warning method includes the following steps: Step 1, Preliminary Preparations, please refer to the following: Figure 1 and Figure 2 : 1) Obtain GIS data of municipal pipelines (including pipeline route, burial depth, and mileage marker coordinates) and GIS data of existing communication optical cables along the municipal pipeline (including optical cable route, manhole location, and fiber optic cable length); determine the spatial relationship between the two, including parallel sections, intersections, and vertical distances, to provide a geographic benchmark for subsequent event location.

[0031] 1. 2) Conduct quality checks on the spare optical fibers or idle fiber cores in the existing communication optical cables intended for use as sensors; the tests include: the total loss of the optical cable, whether there are any breaks or macro bends, the loss of the connectors, and the polarization mode dispersion of the optical fiber; the communication optical cable can only be used for subsequent monitoring if its quality meets the requirements of distributed acoustic sensing.

[0032] 1. 3) Connect the distributed fiber optic acoustic sensing system to the spare fiber or idle fiber core in the existing communication optical cable. Utilize events with known geographical locations (such as manual knocking, known construction points) or optical time domain reflectometer (OTDR) event points (i.e., fiber optic connectors, bends, breaks, etc., whose distance coordinates can be accurately determined by the OTDR, and the ground location is usually known) to establish a one-to-one correspondence between fiber optic channel numbers (or distance domain sampling points) and real geographical coordinates (mileage, latitude and longitude), forming a calibration table so that subsequently detected vibration events can be accurately converted into GIS location information.

[0033] Step 2, online monitoring, please refer to [link / reference]. Figure 2 : After completing the above preliminary preparations, the real-time monitoring phase begins: Based on the resolution of the distributed fiber optic acoustic sensing system, channels are divided along the fiber optic cable length (e.g., with a resolution of 10m, one channel is set every 10m along the cable length). The system collects the average vibration magnitude within the monitoring range of the fiber optic channel in real time. The update time interval for the monitoring results of the real-time monitoring system is selected. For the real-time sensing signal obtained from the communication fiber optic cable within each update time interval, the following steps are performed: 2.1) Calculate the root mean square (RMS) of the raw data of the real-time sensing signals obtained by each fiber optic channel, and screen the suspected channels with suspected third-party construction signals based on their RMS values. For long-distance monitoring, construction behavior is an occasional event. The RMS threshold can be set as the 95th percentile of the RMS value statistics at quiet times (such as night) along the line (i.e., the RMS signal of 95% of the samples is less than this threshold). Signals exceeding this RMS threshold are considered suspected construction signals.

[0034] 2.2) For the fiber optic channels filtered out in step 2.1), fill in the default value (0); for the suspected channels that have been selected, input the segmentation model based on the convolutional neural network module to determine the identification result of the third-party construction behavior of the suspected channels; the channel identification model adopts a dual-mode design: II. 2-1) Channel Recognition Model Training Phase: The output module is a classification head, including a global average pooling layer, a fully connected layer, and a binary classification output layer, see... Figure 3 and Figure 4 The training samples are slices of single event samples with a fixed number of channels and a fixed number of samples at a fixed time. The model learns its ability to identify third-party construction signals through backpropagation and stochastic gradient descent.

[0035] 2.2) Channel Recognition Model Inference Stage: The output module switches to a segmentation head, including a fixed-stride average pooling layer and two 1×1 convolutional layers, see... Figure 3 and Figure 4 The following design ensures computational consistency between the training and inference phases: 2.2-2a) Set the average pooling layer stride to an integer multiple of the application range of the global average pooling layer during the training phase; 2.2-2b) The weights of the convolutional layers in the segmentation head are inherited from the weights of the fully connected layers in the classification head during the training phase, and the weight structure is transformed through dimensionality broadcasting.

[0036] This dual-mode design enables the channel recognition model to directly input full-channel fiber optic data for calculation during the inference phase and output channel-by-channel recognition results, avoiding data slicing and result aggregation, and completely encapsulating the core calculation process in a single neural network.

[0037] II. 2-3) Event tracking analysis of channel-by-channel identification results: Input the channel-by-channel identification results output in step II. 2-2) into the event tracking module to realize the conversion from channel-by-channel results to independent construction events. The specific implementation of this event tracking module includes: 2. 2-3a) When starting up, an empty third-party construction event pool is generated, and channel thresholds and time interval thresholds are set for event determination in subsequent steps. The channel thresholds should be set with reference to the channel range affected by a single construction signal in the specific scenario, and the time interval thresholds should be selected based on experience and should be an integer multiple of the selected monitoring result update interval. 2. 2-3b) For the channel-by-channel recognition results of the new input, the recognition results with a channel interval smaller than the channel threshold are merged to form the potential event at the current moment; 2.2-3c) Based on the set channel threshold and time interval threshold, match the potential events at the current moment with historical events in the third-party construction event pool: 2.2-3c-1) Historical events are selected from those less than the time interval threshold from the current time for matching. Potential events with channel intervals less than the channel threshold are merged into the corresponding historical events. The termination time and the range of the affected channels of the corresponding historical events are updated. II. Sub-step 2-3c-2): Potential events not merged into historical events in step 2-3c-1) are treated as new independent events and added to the third-party construction event pool; 2.2-3c-3) For historical events whose time interval from the end time to the current time exceeds the preset time interval threshold, if their impact range is greater than the channel threshold, they will be saved and no longer updated; otherwise, they will be regarded as non-significant events and discarded. 2. 2-3c-4) After each time period is matched, all historical events are output for display on the graphical platform, and newly added historical events trigger alarm push notifications.

[0038] 2.4) Further statistically analyze the results of steps 2.2) and 2.3) into a waterfall chart of the all-day monitoring results and hourly alarm duration, as multi-dimensional historical information to support maintenance personnel in viewing and analyzing.

[0039] Example (1) System deployment and parameter configuration A distributed optical fiber acoustic sensing system (DAS) based on the principle of phase-sensitive optical time domain reflectometer (Φ-OTDR) is connected to one core of the spare optical fiber in the communication optical cable; the system sampling frequency is set to 2kHz, and 10080 channels are monitored simultaneously; for long-term third-party damage monitoring tasks, the time interval for updating the communication optical fiber sensing signal is preferably set to 1 to 5 minutes. In this embodiment, the time interval for updating the monitoring results is set to 5 minutes, that is, the system acquires the latest real-time raw data of the optical fiber every 5 minutes, and then performs the following processing flow.

[0040] (2) Channel-level initial screening based on root mean square (rms) value First, through a week of continuous observation, the distribution of RMS values ​​for each fiber optic channel along the route during the night (00:00-05:00) was statistically analyzed, resulting in histograms of each RMS value interval. The 95th percentile value was then used as the RMS threshold for each channel. For example, if the nighttime RMS statistical value of a typical channel is 0.05, and its 95th percentile value is 0.08, then the RMS threshold for that channel is set to 0.08.

[0041] Please see Figure 2 For each update period, the RMS value of all channels in the original data is compared one by one along the channel direction: for channels with an RMS value exceeding 0.08, they are marked as "suspected construction channels" and enter the subsequent fine identification; for fiber optic channels with an RMS value not exceeding the RMS threshold, the fill value is 0 and the subsequent model calculation is skipped.

[0042] (3) Segmentation model recognition based on deep convolutional neural networks exist Figure 3 Based on the network structure shown, the details of the deep convolutional neural network structure used in this embodiment are as follows: Feature extraction module: Consists of five convolutional downsampling layers with strides of 5, 5, 4, 4, and 4, and several residual convolutional blocks stacked alternately. In this embodiment, each convolutional downsampling layer is followed by 1, 1, 4, 4, and 6 residual convolutional blocks, respectively. Each residual convolutional block contains two combined layers of "convolution-batch normalization-linear rectification units". The input signal's time window length is 40,000 sampling points (approximately 20 seconds), and the number of channels is arbitrary to accommodate the different modes required during the module's training and inference phases.

[0043] 1) Model training phase: The output module is a classification header, which includes a global average pooling layer, a fully connected layer, and a binary classification output layer.

[0044] The training samples consist of positive samples (third-party construction signals) and negative samples (environmental background noise, passing vehicles, etc.), totaling approximately 30,000 samples. The sample slices are fixed at 1 channel × 40,000 sampling points. After approximately 100 iterations using the stochastic gradient descent method, the model performance tends to stabilize. The trained third-party construction damage identification and classification model was tested on a test set. On the validation set, the model achieved an accuracy of 96.2%, a recall of 99.9%, and a mean precision (AP) of 99.7%.

[0045] 2) Model inference stage: Replace the trained classification head with a segmentation head. The segmentation head structure includes: ① Average pooling layer with a stride of 25 (corresponding to 40,000 sampling points covered by the global average pooling layer during the training phase); ② Two 1×1 convolutional layers. The weights are inherited from the two fully connected layers of the classification head during the training phase, and are obtained through dimensionality broadcasting and reshaping (input channel × output channel → input channel × output channel × 1×1).

[0046] During inference, the model can directly input two-dimensional data with any channel (10080 without RMS filtering) × 40000 sampling points. A single inference outputs a channel-by-channel recognition result corresponding to channel × 2 (each channel outputs two probability values, and the category corresponding to the higher probability is taken as the recognition result). In actual testing on a consumer-grade CPU Intel® Alder Lake-N N97, for 20 seconds of raw fiber optic signal data, a single inference takes approximately 5 seconds, meeting the real-time requirements.

[0047] (4) Event tracking and alarm integration The parameters for the event tracking module are configured as follows: 1) Channel merging threshold: 5 channels with a channel interval of less than 5 consecutive alarm channels are merged into the same potential event; 2) Event matching threshold: If the minimum distance between the channel range of a potential event and the channel range of a historical event in the latest hour is less than the channel merging threshold (5 in this embodiment), they are considered as the same event; 3) Historical event expiration threshold: Historical events that have not been updated for 1 hour are considered non-significant events and discarded if their impact range is less than 5 channels (channel merging threshold); if their impact range is greater than or equal to 5 channels (channel merging threshold), they are saved as "ended events" and will not be updated again.

[0048] like Figure 4 As shown, the specific execution flow of the event tracking module is as follows: 1) Assume that at time t=0 (initial state), the third-party construction event pool is empty. The system outputs the channel-by-channel identification results for the first time. In channels 10080, channels 2277~2282 are identified as "construction," while the other channels are identified as "non-construction." The channels identified as "construction" are merged to form a potential event E1, located at channel 2280, with a channel range of 2277~2282. Since the event pool is empty, E1 is added to the event pool as a new event, triggering an alarm push. 2) At t=1 minute, the new identification results for channels 2280-2289 show "Construction". This potential event overlaps with E1 on the channel, and the time interval is less than 1 hour, so it is determined to be the same event. The termination time of E1 is updated to the current time, expanding the affected channel range to 2277~2289. Since this event already exists, no new alarm push is triggered.

[0049] 3) At t=1 hour 01 minute, if no new matching results have been found for this event for one hour, it is determined to be a terminated event. Since the impact exceeds the channel merging threshold, it is saved as a historical event and will not be updated further. The system marks this event as a "terminated event" in the historical record and it will no longer participate in subsequent matching with potential events.

[0050] (5) Statistical analysis of multi-dimensional monitoring results The system updates the monitoring results every minute and outputs visualized data: 1) GIS map display. The routes of pipelines and communication optical cables are overlaid on the GIS map. When an alarm occurs, an alarm icon is displayed at the corresponding location. Clicking on the icon allows you to view event details (such as time, location, affected area, and processing status).

[0051] 2) Waterfall chart of all-day monitoring results: The horizontal axis represents the channel location, and the vertical axis represents time. The color of each pixel indicates whether the spatiotemporal point is under construction (red indicates construction). This chart visually displays the distribution of all construction events in any time period.

[0052] 3) Hourly alarm duration statistics chart: With hours as the horizontal axis, the cumulative duration (in minutes) of construction alarms occurring in each hour is counted and displayed in the form of a bar chart to help maintenance personnel identify high-incidence periods of construction.

[0053] 4) Alarm Event Push List: Displays all current alarm push information, including the number assigned to the event (reset daily), start time, stop time, duration, center location, channel width, and whether the processing is complete.

[0054] The above are merely preferred embodiments of the present invention. It must be noted that all equivalent modifications, variations and alterations made by those skilled in the art based on the contents of this application should be included within the scope of protection of this invention.

Claims

1. A method for monitoring and early warning of third-party damage to municipal pipelines based on existing optical fiber communication cables, characterized in that: A distributed fiber optic acoustic wave sensing system is connected to existing communication optical cables along a selected municipal pipeline. Based on the acquired geographic information data of the municipal pipeline and the existing communication optical cable, the spatial relationship between the geographic coordinates of the existing communication optical cable and the municipal pipeline is established. The distributed fiber optic acoustic wave sensing system collects and processes fiber optic acoustic vibration signals in real time, calculates and compares the root mean square value of the fiber optic acoustic vibration signals, identifies and filters suspected construction signals, and uses a channel identification model based on a deep neural network to filter out channel-by-channel construction signals. An event tracking module is used for event tracking analysis to realize the conversion from channel-by-channel results to independent construction events, identify third-party construction events and assess their safety threat level to the municipal pipeline, and provide early warning and visualization, thereby completing online monitoring and early warning of third-party damage to the municipal pipeline.

2. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 1, characterized in that: The monitoring and early warning method includes the following steps: Step 1: Install the distributed fiber optic acoustic wave sensing system on-site and set the acquisition parameters. 1) Select qualified spare optical fibers from the existing communication optical cables along the municipal pipeline as sensing optical fibers, and connect the selected sensing optical fibers to the distributed optical fiber acoustic wave sensing system.

1. 2) Obtain the geographic information data of the municipal pipeline and the geographic information data of the existing communication optical cable, and determine the spatial relationship between the two to provide a geographic reference for subsequent event location; the geographic information data of the municipal pipeline includes, but is not limited to, pipeline location, direction, burial depth and kilometer marker coordinates, the geographic information data of the existing communication optical cable includes, but is not limited to, optical cable route, manhole location and fiber optic cable length, and the spatial relationship between the two includes, but is not limited to, parallel segments, intersections and vertical distances; 1. 3) Set the spatial sampling interval of the distributed optical fiber acoustic wave sensing system, determine the channel number of the sensing optical fiber, establish a one-to-one correspondence between the optical fiber channel number and the geographical coordinates using events with known geographical locations, and form a calibration table so that the geographical location of the subsequently detected events can be accurately located.

1. 4) Set the sampling frequency and gauge length of the distributed fiber optic acoustic wave sensing system; Step two involves real-time acquisition of fiber optic acoustic vibration signals through the fiber optic acoustic wave sensing system, subsequent data processing, and channel-by-channel signal identification and event tracking using a deep neural network to assess the safety threat level to the municipal pipeline, thereby completing online monitoring and early warning of third-party damage to the municipal pipeline. II. 1) Acquisition and processing of optical fiber acoustic vibration signals — 2. 1-1) The distributed optical fiber acoustic wave sensing system acquires the optical fiber acoustic vibration signal of each optical fiber channel on the sensing optical fiber in the existing communication optical cable in real time; 2.1-2) Set the time interval according to experience. For the latest time period of the acquired fiber optic acoustic vibration signal, calculate the root mean square value of the fiber optic acoustic vibration signal of each fiber optic channel obtained in step 2.1-1), and set the root mean square threshold of each fiber optic channel. 2) Perform channel identification — 2.2-1) Establish a channel recognition model based on a deep neural network and train the channel recognition model; 2.2-2) Use the trained channel recognition model to perform reasoning and identify and filter out the construction signals for each channel; 2.3) Conduct event tracking and analysis— 2. 3-1) Initialize the event tracking module, generate an empty third-party construction event pool, and configure and set channel thresholds and time interval thresholds for subsequent event determination; 2.3-2) Input the channel-by-channel identification results output in step 2.2-2) into the event tracking module to realize the conversion from channel-by-channel results to independent construction events, including the following specific contents:

2. 3-2a) In the channel-by-channel recognition results of the input, the recognition results with a channel interval less than the channel threshold are merged to form the same potential event at the current time; 2. 3-2b) Based on the channel threshold and time interval threshold, match the potential events at the current moment with the historical events in the third-party construction event pool:

2. 3-2b-1) Select historical events that are less than the time interval threshold from the current time to participate in the matching. Potential events with a channel interval less than the channel threshold are merged into the corresponding historical events. Update the termination time and the range of the affected channels of the corresponding historical events.

2. In step 2.3-2b-1), potential events that were not merged into the corresponding historical events are treated as new independent events and added to the third-party construction event pool.

2. 3-2b-3) For historical events whose time interval from the termination time to the current time exceeds the time interval threshold, if their influence range is greater than the channel threshold, they are saved as ended events and will not be updated again, and will no longer participate in subsequent potential event matching; otherwise, they are regarded as non-significant events and discarded.

2. 3-2b-4) After the event matching is completed at each moment, all historical events are output for display on the graphical platform, wherein the new historical events obtained in step 2. 3-2b-2) trigger alarm push; 2.4) Based on the spatial relationship between the geographical coordinates of the existing communication optical cable and the geographical coordinates of the municipal pipeline described in step 1.2), analyze the degree of safety threat posed by the construction event to the municipal pipeline, and generate a corresponding early warning level based on the analysis results; Step 3: Perform multi-dimensional statistics on the channel-by-channel identification results obtained in Step 2.2) and the event tracking results obtained in Step 2.3) to generate visualized image data for display, supporting operation and maintenance personnel to view and analyze.

3. The method for third-party damage monitoring and early warning of municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: In step 1.1), the method for selecting the sensing optical fiber is to conduct quality tests on the spare optical fiber in the existing communication optical cable, including but not limited to the fiber's total loss, whether there are breaks or macro bends, connector loss, and polarization mode dispersion. The spare optical fiber that meets the quality test requirements for distributed acoustic wave sensing is selected as the sensing optical fiber.

4. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: In step 1.1), the distributed fiber optic acoustic wave sensing system is based on the principle of a phase-sensitive optical time-domain reflectometer.

5. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: In step 1.3), the known geographical location events include manual knocking, known construction points, and optical time domain reflectometer event points. The distance coordinates of the optical time domain reflectometer event points are known by precise determination by the optical time domain reflectometer, including fiber optic connectors, bends, and break points.

6. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: In step 2.1-2), the method for setting the root mean square threshold of each optical fiber channel is as follows: through continuous detection for a certain period of time in the early stage, the root mean square value distribution of optical fiber acoustic vibration signal data of each optical fiber channel during quiet periods along the line is statistically analyzed to obtain the histogram of the root mean square value interval of each optical fiber channel. Then, the 95th percentile value is taken as the root mean square threshold of each optical fiber channel.

7. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: In step 2.2-1), the training method of the channel recognition model is as follows: the output module is a classification head, including a global average pooling layer, a fully connected layer, and a binary classification output layer; the training samples are slices of single event samples with a fixed number of channels and a fixed number of time samples; the channel recognition model learns the ability to recognize third-party construction signals through backpropagation and stochastic gradient descent; and the trained channel recognition model is verified using a test sample set to ensure the accuracy and precision of the channel recognition model.

8. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 7, characterized in that: In step 2.2-2), the reasoning method of the channel identification model is as follows: the output module is switched to a segmentation head, including a fixed-step average pooling layer and two 1×1 convolutional layers; the fiber acoustic vibration signal data of all fiber channels obtained in steps 2.1-1) and 2.1-2) are directly input into the channel identification model for calculation; the root mean square value of the fiber acoustic vibration signal data of each fiber channel is compared with the root mean square threshold of the fiber channel one by one along the direction of the sensing fiber channel; the fiber acoustic vibration signal exceeding the root mean square threshold is a suspected construction signal, and the location of the fiber channel where the suspected construction signal is located is the geographical location of the third-party construction event; for fiber channels whose root mean square value does not exceed the root mean square threshold, the fill value is set to 0, and subsequent model calculations are skipped.

9. The method for third-party damage monitoring and early warning of municipal pipelines based on existing communication optical cables according to claim 8, characterized in that: The channel recognition model adopts a dual-mode design. The stride of the average pooling layer is set to an integer multiple of the application range of the global average pooling layer of the channel recognition model during the training phase. The weights of the convolutional layer inherit the weights of the fully connected layer of the channel recognition model during the training phase, and the weight structure is transformed through dimensionality broadcasting.

10. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: In step 2.3-1), the channel threshold is set according to the range of the number of fiber optic channels affected by a single construction signal, and consecutive channel events with a channel interval less than the channel threshold are merged into the same potential event; the time interval threshold is selected empirically and is an integer multiple of the update time interval of the fiber optic acoustic vibration signal.

11. The method for monitoring and early warning of third-party damage to municipal pipelines based on existing communication optical cables according to claim 2, characterized in that: The visualization image data generated in step three includes: 1) GIS map – Overlaying display of the routes of the municipal pipelines and existing communication optical cables; when an alarm occurs, clicking the alarm icon displayed at the corresponding location will allow you to view event details including time, location, scope of impact, and processing status; 2) Waterfall chart of all-day monitoring results - intuitively displays the distribution of all construction events along the municipal pipeline at any given time; the horizontal axis represents the location of the fiber optic channel, the vertical axis represents time, and the color of the pixel indicates the construction status at that spatiotemporal point; 3) Hourly alarm duration statistics chart - with hours as the horizontal axis, the cumulative duration of construction alarms in each hour is displayed in the form of a bar chart to identify high-incidence periods of construction. 4) Alarm Event Push List - Displays all current alarm push information, including the event number, start time, stop time, duration, center location, channel width, and whether the processing has been completed.

Images