Pipeline risk identification generation method based on environment real-time monitoring and edge monitoring gateway

Abstract

The application relates to a pipeline risk identification generation method based on environment real-time monitoring and an edge monitoring gateway, relates to the technical field of risk identification, and comprises the following steps: acquiring a monitoring image sequence of a target pipeline area, performing alignment vibration analysis on adjacent images and eliminating background vibration to obtain vibration parameters and vibration cleaning degree reflecting vibration intensity; when the vibration parameters are abnormal, acquiring a sensing parameter sequence, identifying a first threat distance based on the vibration cleaning degree; performing visual identification on the image sequence to obtain a second threat distance; dynamically adjusting sensing and visual coefficients according to the vibration cleaning degree, fusing the first and second threat distances, and outputting a fused threat distance as a risk identification result. The application solves the technical problem that the existing pipeline risk monitoring using a chemical sensor can only passively discover risks after pipeline damage and leakage and cannot perceive and give early warnings in the external construction disturbance stage.

Description

Technical Field

[0001] This application relates to the field of risk identification technology, and in particular to a pipeline risk identification generation method and edge monitoring gateway based on real-time environmental monitoring. Background Technology

[0002] Natural gas and other long-distance pipelines are widely distributed along their routes, passing through various construction areas. When construction machinery such as excavators and drilling rigs operate near pipelines, improper operation or misalignment may cause mechanical damage to the pipelines, leading to leakage accidents. Therefore, timely monitoring and early warning of risks such as external construction along the pipeline route are crucial aspects of pipeline safety protection.

[0003] In existing technologies, pipeline risk monitoring mainly relies on deploying gas concentration or pressure sensors around the pipeline. Risk events are detected by analyzing changes in gas concentration or pressure anomalies that escape from the pipeline after a leak. This chemical or physical contact-based monitoring method means that abnormal signals can only be collected after the pipeline has been damaged and the internal medium has begun to leak. By the time the sensor triggers an alarm, the pipeline has already suffered actual damage, and a leak has occurred. The monitoring results can only support post-event emergency response and repair scheduling, and cannot detect risks and issue early warnings during the construction disturbance phase when construction machinery approaches the pipeline but has not yet caused damage. Pipelines are chronically exposed to passive damage risks due to delayed information perception, resulting in a severe lack of proactive and timely safety protection. Summary of the Invention

[0004] This application provides a pipeline risk identification generation method and edge monitoring gateway based on real-time environmental monitoring. It improves the technical problem that in the prior art, pipeline risk monitoring mainly relies on chemical sensors to detect leaking gas. When a risk event occurs, the pipeline has already suffered actual damage. The monitoring results can only support post-event emergency response and cannot achieve pre-event prevention, resulting in serious delays in early warning and extremely limited pipeline safety protection window.

[0005] This application discloses the following technical solution:

[0006] In a first aspect, this application provides a pipeline risk identification and generation method based on real-time environmental monitoring, the method comprising:

[0007] By using an image acquisition device, the monitoring image sequence of the target pipeline area is monitored, and the vibration analysis and vibration cleaning of adjacent monitoring images are performed to obtain vibration parameters and vibration cleaning degree.

[0008] When the vibration parameters meet the abnormal conditions, the sequence of sensing parameters monitored by the sensor device is acquired, and the threat distance is identified based on the vibration cleanliness to obtain the first threat distance;

[0009] The second threat distance is obtained by identifying the monitored image sequence;

[0010] Based on the vibration cleanliness, the first threat distance and the second threat distance are fused to obtain the fused threat distance, which serves as the risk identification result.

[0011] Secondly, this application provides a pipeline risk identification and generation edge monitoring gateway based on real-time environmental monitoring, the edge monitoring gateway comprising:

[0012] The data acquisition module is used to monitor the monitoring image sequence of the target pipeline area through the image acquisition device, perform vibration analysis and vibration cleaning of adjacent monitoring images, and obtain vibration parameters and vibration cleaning degree.

[0013] The sensing and identification module is used to acquire the sequence of sensing parameters monitored by the sensor device when the vibration parameters meet the abnormal conditions, and to identify the threat distance based on the vibration cleanliness to obtain the first threat distance.

[0014] A visual recognition module is used to identify the monitored image sequence to obtain a second threat distance;

[0015] The fusion output module is used to fuse the first threat distance and the second threat distance based on the vibration cleanliness level to obtain the fused threat distance as the risk identification result.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] This application's technical solution provides a pipeline risk identification generation method based on real-time environmental monitoring. First, it monitors a sequence of images of the target pipeline area using an image acquisition device. Adjacent monitoring images undergo grayscale processing and vibration analysis to extract raw vibration parameters from pixel displacements. Then, it cleans these raw vibration parameters using background vibration parameters, filtering out normal environmental vibrations such as passing vehicles. This yields vibration parameters reflecting abnormal disturbances such as those caused by construction, as well as a vibration cleanliness degree characterizing the extent of vibration cleanliness. This process transforms pixel-level micro-displacements in the image sequence into quantifiable vibration indicators. By using background cleaning, it distinguishes between normal environmental vibrations and abnormal vibrations caused by construction machinery. This allows the monitoring gateway to detect potential threats even before construction machinery touches the pipeline, addressing the problem of traditional chemical sensors only being able to passively detect and warn after pipeline damage and leakage, resulting in a significant lag.

[0018] Furthermore, when the vibration parameters exceed a preset threshold based on the minimum vibration characteristics of pipeline risk events, a sensor identification process is triggered. This process acquires the sensor parameter sequence and selects a threat distance identification agent with a corresponding amount of training data based on the range the vibration cleanliness falls into. A lower vibration cleanliness indicates greater vibration intensity and stronger sensor interference; an agent with a larger amount of training data is selected to ensure identification accuracy. Conversely, a higher vibration cleanliness selects an agent with a smaller amount of training data to improve analysis efficiency. The sensor parameter sequence is input into the selected agent to identify the first threat distance based on the sensor data. This process dynamically correlates the identification accuracy of the sensor data with the current vibration interference intensity. In strong vibration scenarios, it utilizes more extensive training experience to ensure accuracy, while in weak vibration scenarios, it uses a lightweight model for rapid response, solving the problem of unstable reliability of traditional single fixed models in complex environments.

[0019] Next, the threat distance recognition agent, constructed based on a convolutional neural network, is input frame by frame into the monitoring image sequence. This agent is trained under supervised supervision using sample monitoring images and corresponding risk events and pipeline distance annotations. After recognizing the threat distance in each frame, the average output value is taken as the second threat distance. This process directly determines the spatial relationship between construction machinery and pipelines from the images visually, extracting visual distance information between the current risk event and the pipeline. This solves the problem that traditional chemical monitoring cannot provide relative position information between construction machinery and pipelines, nor can it determine the urgency of the threat.

[0020] Finally, sensing and visual coefficients are configured based on vibration cleanliness. Higher vibration cleanliness indicates higher image acquisition quality. The vibration cleanliness is directly used as the visual coefficient. The sensing coefficient is set to 1 minus the negative correlation value of the visual coefficient. The sensing and visual coefficients are then used to weight and fuse the first and second threat distances to obtain the fused threat distance as the risk identification result. This process uses vibration cleanliness as a confidence metric for both identification methods. When image quality is reduced due to strong vibration, the fusion weight of the sensing data is automatically increased. When the image is clear with weak vibration, more trust is given to the visual data. This achieves dynamic optimization and fusion of sensing and visual identification results, solving the technical problem of large fluctuations in output results and difficulty in stable early warning under changing environmental interference in traditional single identification methods.

[0021] In summary, this application extracts vibration parameters from monitoring image sequences and filters out background vibration interference. Upon sensing abnormal vibration, it triggers dual-channel threat distance identification using sensor data and visual images. Then, it dynamically adjusts the fusion weights of sensing and vision based on vibration cleanliness to obtain the fused threat distance as the risk identification result. This solves the technical problem that existing pipeline risk monitoring relies on chemical sensors, which can only passively detect pipeline damage and leakage after it occurs, and cannot provide early warnings during construction disturbances. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the pipeline risk identification and generation method based on real-time environmental monitoring provided in this application embodiment.

[0024] Figure 2 A flowchart illustrating the sensor threat distance classification identification process of the pipeline risk identification generation method based on real-time environmental monitoring provided in this application embodiment.

[0025] Figure 3 This is a schematic diagram of the structure of the pipeline risk identification and edge monitoring gateway based on real-time environmental monitoring provided in this application embodiment.

[0026] In the attached diagram, Figure 3 The components represented by each number are described as follows: data acquisition module 11, sensor recognition module 12, vision recognition module 13, and fusion output module 14. Detailed Implementation

[0027] This application provides a pipeline risk identification generation method and edge monitoring gateway based on real-time environmental monitoring. It is used to solve the technical problem that the pipeline area lacks effective means to actively perceive and identify external construction and excavation activities in advance, and cannot issue timely warnings when construction machinery approaches the pipeline but has not yet caused damage. This results in the pipeline being exposed to passive damage risks due to information lag for a long time.

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. It should be noted that the numerical values ​​in the embodiments are for illustrative purposes only and do not constitute a limitation on this application.

[0029] Example 1, as shown in the appendix Figure 1 As shown, this application provides a pipeline risk identification and generation method based on real-time environmental monitoring, the method comprising the following steps:

[0030] S100: Through the image acquisition device, monitor the monitoring image sequence of the target pipeline area, perform vibration analysis and vibration cleaning of adjacent monitoring images, and obtain vibration parameters and vibration cleaning degree.

[0031] In this embodiment, during construction monitoring along a natural gas pipeline, an edge monitoring gateway continuously captures images of the pipeline area using an image acquisition device. When construction machinery performs excavation, drilling, or other operations on the ground, the resulting ground vibrations are transmitted to the mounting base of the image acquisition device, causing minute pixel displacements between adjacent frames. By analyzing these pixel displacements, the presence and intensity of construction activities can be detected. Simultaneously, interference from normal environmental vibrations, such as passing vehicles, needs to be eliminated to address the problem that traditional chemical sensors can only passively detect leaks after pipeline damage, failing to detect risks in advance during construction disturbances.

[0032] Step S100 of the method provided in this application embodiment includes: monitoring and acquiring a sequence of monitoring images of a target pipeline area using an image acquisition device; combining multiple sets of adjacent monitoring images within the monitoring image sequence and performing grayscale processing to obtain multiple sets of grayscale monitoring images; performing aligned vibration analysis on the multiple sets of grayscale monitoring images to obtain original vibration parameters; and performing vibration cleaning on the original vibration parameters to obtain vibration parameters and vibration cleaning degree. Detailed explanation follows:

[0033] In this embodiment, the monitoring image sequence refers to a time-series image obtained by continuously capturing images of the target pipeline area at a fixed frame rate using an image acquisition device. The image acquisition device is installed on a pole or support along the pipeline, covering the pipeline area to be monitored from a fixed viewing angle. Due to environmental vibrations, the pixel position of the same target object in the image may slightly shift between adjacent monitoring images.

[0034] Grayscale monitoring images are generated by converting color monitoring images into grayscale values ​​using a weighted formula. The grayscale value of each pixel reflects the light intensity information at that location. After conversion, each pixel retains only one grayscale channel, effectively reducing the computational complexity in subsequent alignment vibration analysis. The weighting formula can employ a luminance-weighted method, converting the red component R, green component G, and blue component B of each pixel in the color image into a grayscale value Gray using the following formula: Gray = 0.299 × R + 0.587 × G + 0.114 × B. This formula, based on the differences in human eye sensitivity to different colors, assigns a higher weight to the green component and a lower weight to the blue component, making the converted grayscale image more consistent with visual perception characteristics.

[0035] The original vibration parameters refer to the set of vibration displacements obtained directly from the alignment and analysis of adjacent grayscale images. This includes both abnormal vibration components caused by construction machinery and normal environmental vibration components such as passing vehicles. Vibration cleanliness refers to the proportion of background vibration parameters in the original vibration parameters; that is, the ratio of background vibration parameters to original vibration parameters. A smaller ratio indicates a higher proportion of abnormal vibration in the total vibration, a greater intensity of the current risk event, and a more severe impact of vibration on the image acquisition device, resulting in lower image acquisition quality.

[0036] In this step, in order to extract vibration indicators that can reflect the intensity of construction activities and the subsequent risk assessment from the monitoring image sequence, the image acquisition device first takes continuous pictures of the target pipeline area at a preset frame rate to obtain a temporally continuous monitoring image sequence, which provides raw visual data for all subsequent analyses.

[0037] Then, adjacent frames within the monitoring image sequence are grouped, with each group containing two consecutive frames. Each color monitoring image within a group is then converted to a grayscale monitoring image. Grayscale conversion eliminates interference from color information on pixel matching, simplifying each pixel from a multi-channel color value to a single grayscale value. This reduces the complexity of similarity calculations in alignment vibration analysis and improves matching stability. After grayscale processing, multiple sets of grayscale monitoring images are obtained, each providing a pair of grayscale images for direct pixel-level comparison in subsequent alignment vibration analysis.

[0038] Furthermore, step S100 in the method provided in this application embodiment further includes: in the first set of grayscale monitoring images, randomly selecting a first pixel point within the first grayscale monitoring image, and dividing a first pixel point window within a preset window surrounding the first pixel point; traversing and selecting pixels and dividing pixel point windows within the second grayscale monitoring image in the first set of grayscale monitoring images, calculating the similarity with the first pixel point window, selecting the pixel point with the highest similarity as the mapped pixel point, calculating the distance with the first pixel point, and obtaining the first vibration distance; continuing to perform aligned vibration analysis on multiple sets of grayscale monitoring images to obtain multiple vibration distance sets; and fusing the multiple vibration distance sets to obtain the original vibration parameters. Detailed explanation follows:

[0039] In this embodiment, the first pixel is a reference pixel randomly selected in the first frame of grayscale monitoring image, used as the reference position for alignment and matching of the subsequent two frames. The first pixel window is a rectangular pixel region of a preset size, centered on the first pixel. The window size is a preset fixed pixel value to cover the local texture features of the target object. The preset size refers to the side length of the pre-set rectangular pixel region, for example, 7 pixels × 7 pixels. The mapped pixel is the pixel position in the second frame of grayscale monitoring image that has the highest similarity to the first pixel window, found through traversal search. This pixel represents the new position of the same target point in the second frame image. The similarity refers to constructing a feature vector from the grayscale values ​​of all pixels within the two pixel windows, and then calculating the feature vector similarity. The feature vector similarity can be calculated using cosine similarity. The first vibration distance is the Euclidean pixel distance between the first pixel and its mapped pixel, reflecting the displacement of the target point due to vibration between the two frames, in pixels.

[0040] In this step, to accurately quantify and extract pixel displacement caused by vibration from two adjacent grayscale monitoring images, aligned vibration analysis needs to be performed in each group of grayscale monitoring images. First, several pixels are randomly selected within the first frame of each group as the first pixel, providing a reference position for subsequent displacement calculations. A rectangular pixel region of a preset size is then divided around each first pixel as a first pixel window. This window contains the local texture features surrounding the pixel, providing sufficient feature information for searching for matching positions in the second frame image.

[0041] Then, the second frame of the grayscale monitoring image in the same group is traversed pixel by pixel. At each traversal position, a pixel window of the same size is divided, and the similarity between the traversed window and the first pixel window is calculated one by one. The purpose of similarity calculation is to quantify the consistency of pixel grayscale distribution within the two windows, in order to find the corresponding position of the same target point in the second frame image. Among all traversed pixels, the pixel with the highest similarity, i.e., the best window match, is selected as the mapping pixel. This mapping pixel represents the new position reached by the same target point in the second frame image due to vibration.

[0042] The pixel distance between the first pixel and the mapped pixel is calculated as the first vibration distance. This distance directly quantifies the displacement of the target point due to vibration between two frames. The above process is repeated for each set of grayscale monitoring images to obtain multiple sets of vibration distances. Each set contains the vibration distance values ​​of multiple sampling points within that set of images. Finally, all vibration distances are statistically fused. The fusion process aggregates the vibration distances of multiple sampling points into a comprehensive parameter representing the overall vibration level of the current monitoring period, yielding the original vibration parameter. Statistical fusion processing includes, but is not limited to, taking the arithmetic mean, median, or weighted average of all vibration distances. For example, taking the arithmetic mean of all vibration distances is used as the original vibration parameter.

[0043] Furthermore, step S100 in the method provided in this application embodiment further includes: obtaining background vibration parameters of the target pipeline area when no risk event exists; using the background vibration parameters to clean the original vibration parameters to obtain vibration parameters; and calculating the cleaning ratio of the background vibration parameters in the original vibration parameters as the vibration cleaning degree. A detailed explanation follows:

[0044] In this embodiment, the background vibration parameter refers to the baseline vibration level of the target pipeline area under normal conditions, excluding construction or other risk events, caused by environmental factors such as vehicle passage and wind. This parameter can be obtained by analyzing and statistically processing the monitoring image sequence using the same method as the original vibration parameter during periods when no construction activity is confirmed, and can be updated periodically to reflect changes in the environmental vibration baseline caused by seasonality or road condition changes. Cleaning refers to removing the normal environmental vibration components corresponding to the background vibration parameter from the original vibration parameter, retaining the abnormal vibration components caused by risk events such as construction machinery.

[0045] Vibration cleanliness is a quantitative indicator that measures the proportion of background vibration components in the original vibration, with a value ranging from 0 to 1. When the vibration cleanliness is close to 0, it indicates that the background vibration component in the original vibration is extremely small, abnormal vibration dominates, the current construction disturbance is strong, and the image acquisition device experiences significant pixel shifts in the monitored image due to severe vibration, resulting in decreased image acquisition quality and reduced reliability of visual recognition. When the vibration cleanliness is close to 1, it indicates that the original vibration is mainly composed of background vibration, the abnormal vibration component is weak, the current construction disturbance is slight or non-existent, the image acquisition device is stable, the image clarity is high, and the visual recognition results are reliable. Vibration cleanliness will serve as a key parameter for weight configuration in the subsequent agent selection and fusion processing in the sensing and recognition channel.

[0046] In this step, to eliminate the interference of normal environmental vibration on risk identification from the original vibration parameters, the background vibration parameters of the target pipeline area are first obtained when no construction or other risk events are confirmed. The purpose of the background vibration parameters is to establish a baseline for normal environmental vibration in the area, quantifying the vibration level caused by routine factors such as vehicle passage and wind force into a comparable reference value. Then, the original vibration parameters obtained in the current monitoring period are cleaned using these background vibration parameters. The purpose of cleaning is to subtract the normal environmental vibration components corresponding to the background vibration parameters from the original vibration parameters, retaining the abnormal vibration components that exceed the normal background level.

[0047] If the difference is greater than zero, it is retained as a vibration parameter, indicating the presence of abnormal disturbances such as construction work. If the difference is less than or equal to zero, the vibration parameter is set to zero, indicating that the current vibration is within the normal range. Finally, the ratio of the background vibration parameter to the original vibration parameter is calculated as the vibration cleanliness. This ratio quantifies the degree of abnormality in the current vibration: a cleanliness closer to 1 indicates that the original vibration is closer to the normal background, the intensity of the risk event is lower, and the image acquisition quality is better; a cleanliness closer to 0 indicates a higher proportion of abnormal vibration, a stronger risk event, and a decrease in image acquisition quality due to increased vibration. The vibration cleanliness will serve as a key parameter for weight configuration in the subsequent agent selection and fusion processing in the sensing and recognition channel.

[0048] For example, consider a construction monitoring scenario along a natural gas pipeline. An image acquisition device continuously captures images of the pipeline area at a frame rate of 25 frames per second, acquiring 100 sets of adjacent grayscale monitoring images within a certain monitoring period. Ten first pixels are randomly selected from each set of images for aligned vibration analysis, with a window size of 7 pixels by 7 pixels. The 100 sets of images yield 1000 vibration distance values, and the arithmetic mean yields an original vibration parameter of 3.8 pixels. The background vibration parameter for this pipeline area during periods without construction is 0.6 pixels. After cleaning, the vibration parameter equals 3.8 minus 0.6, which equals 3.2 pixels. The vibration cleanliness is approximately 0.16 (0.6 divided by 3.8). This low cleanliness value indicates the presence of strong abnormal vibrations, suggesting intense construction activity. Furthermore, the image acquisition device is significantly affected by vibration, potentially leading to a decrease in image quality.

[0049] In summary, this step extracts the original vibration parameters by performing pixel-level aligned vibration analysis on adjacent monitoring images, and then cleans them using background vibration parameters under no-risk events to filter out vibration interference from the normal environment. This yields vibration parameters reflecting abnormal construction activities and vibration cleanliness characterizing vibration intensity, providing a quantitative basis for the dynamic weight configuration of subsequent sensing recognition, visual recognition, and fusion processing.

[0050] S200: When the vibration parameters meet the abnormal conditions, acquire the sequence of sensing parameters monitored by the sensor device, and based on the vibration cleanliness, identify the threat distance to obtain the first threat distance.

[0051] In this embodiment, based on the obtained vibration parameters and vibration cleanliness, when the vibration parameters exceed a preset threshold, it indicates that there is abnormal disturbance such as construction in the current pipeline area. Further use of sensor parameters such as pressure, vibration acceleration, and sound waves acquired by the sensor device is needed to quantitatively identify the threat distance between the construction machinery and the pipeline. Traditional chemical sensors can only detect leaks after pipeline damage and cannot provide threat distance information during construction disturbances. Furthermore, single-sensor data analysis, when facing different vibration interference intensities, struggles to balance identification accuracy and computational efficiency with fixed-parameter models. Reliability decreases under strong vibration interference, while computational resources are wasted in weak vibration scenarios. The process of graded identification of sensor threat distance in this step is as follows: Figure 2 As shown.

[0052] Step S200 of the method provided in this application embodiment includes: determining whether the vibration parameter is greater than or equal to a vibration parameter threshold, wherein the vibration parameter threshold is set according to the minimum vibration parameter when a pipeline risk event occurs; if not, continuing monitoring; if yes, acquiring the sensing parameter sequence monitored by the sensor device; acquiring multiple qualified sensing threat distance identification agents, wherein the multiple sensing threat distance identification agents correspond to multiple vibration cleanliness intervals, and the vibration cleanliness intervals are negatively correlated with the amount of training data of the sensing threat distance identification agents; selecting the corresponding sensing threat distance identification agent according to the vibration cleanliness interval into which the vibration cleanliness falls, inputting the sensing parameter sequence, and identifying and outputting the first threat distance. Detailed explanation follows:

[0053] In this embodiment, the vibration parameter threshold is a preset judgment threshold value based on the minimum vibration parameter at the time of a pipeline risk event. When the vibration parameter is greater than or equal to the threshold, it indicates that the current vibration intensity has reached a level that may threaten pipeline safety, triggering the sensing and identification process for further identification. When the vibration parameter is less than the threshold, it indicates that the current vibration is still within a safe range, and monitoring continues through the image acquisition device. The sensing parameter sequence is a set of multi-dimensional parameters collected by the sensor device in a time series, including but not limited to physical quantities reflecting the activities of construction machinery such as pressure, vibration acceleration, and sound waves. The sensor device can be installed on the surface or in shallow soil along the pipeline.

[0054] The threat distance identification agent is a machine learning model that takes a sequence of sensor parameters as input and threat distance as output. It is trained using historical sensor data and corresponding threat distance annotations to identify the estimated distance between construction machinery and pipelines from sensor data. The vibration cleanliness interval divides the vibration cleanliness range from 0 to 1 into multiple continuous intervals. Each interval corresponds to a vibration interference intensity and image acquisition quality level, used to select the matching threat distance identification agent.

[0055] The vibration cleanliness range is negatively correlated with the amount of training data for the threat distance identification agent. That is, a smaller vibration cleanliness indicates stronger vibration interference, and an agent with a larger amount of training data is selected to ensure identification accuracy; conversely, a larger vibration cleanliness indicates weaker vibration interference, and an agent with a smaller amount of training data is selected to improve computational efficiency. The first threat distance is the estimated distance between the construction machinery and the pipeline, measured in meters, obtained based on the sensor parameter sequence.

[0056] In this step, to further quantify the threat posed by construction machinery to the pipeline after confirming the presence of abnormal vibration, it is first determined whether the current vibration parameter is greater than or equal to a vibration parameter threshold. The vibration parameter threshold is set based on the minimum vibration parameter recorded in historical pipeline risk events to ensure that an effective response to potential risk events can be triggered. If the vibration parameter is less than the threshold, vibration monitoring continues; if the vibration parameter is greater than or equal to the threshold, the sensor identification process is triggered to obtain the sensor parameter sequence collected by the sensor device in the current time period.

[0057] To balance accuracy and computational efficiency when identifying threat distances, a suitable identification model needs to be dynamically selected based on the intensity of the current vibration interference. Multiple qualified threat distance identification agents are pre-trained using training samples of varying amounts of data, and a correspondence is established between these agents and different vibration cleanliness ranges. The vibration cleanliness ranges are defined as follows: when the vibration cleanliness is between 0.8 and 1.0, the vibration interference is weak, the sensor data signal-to-noise ratio is high, and an agent trained with 20% of the training data is sufficient to meet the accuracy requirements; when the vibration cleanliness is between 0.6 and 0.8, an agent with 40% of the training data is used; when it is between 0.4 and 0.6, an agent with 60% of the training data is used; when it is between 0.2 and 0.4, an agent with 80% of the training data is used; when it is less than 0.2, the vibration interference is extremely strong, requiring an agent with 100% of the full training data to ensure identification reliability. Based on the range in which the currently obtained vibration cleaning degree falls, a corresponding sensing threat distance identification agent is selected. The sensing parameter sequence is input into the agent, and after forward inference calculation, the first threat distance is output. Here, 20%, 40%, 60%, 80%, and 100% are just examples; those skilled in the art can set other quantities with different training data for training.

[0058] Furthermore, step S200 in the method provided in this application embodiment further includes: collecting a set of sample sensor parameter sequences based on historical pipeline monitoring data, and collecting a set of sample first threat distances corresponding to different sample sensor parameter sequences; repeatedly extracting the set of sample sensor parameter sequences and the set of sample first threat distances to obtain multiple sets of training and testing data, wherein multiple extractions are performed according to different data volumes; constructing multiple sensor threat distance recognition agents, and using multiple sets of training and testing data for supervised training and testing respectively, and obtaining multiple sensor threat distance recognition agents after passing the test. Detailed explanation follows:

[0059] In this embodiment, the sample sensor parameter sequence set is a collection of multiple sets of sensor parameter sequences collected from historical pipeline monitoring data, each set corresponding to a construction disturbance event record. The sample first threat distance set is the threat distance label value corresponding to the sample sensor parameter sequence, obtained by on-site measurement of the actual distance between construction machinery and the pipeline or by expert evaluation. Multiple training and test data sets are multiple training and test sets obtained by repeatedly sampling from the sample set according to different data proportions, for example, randomly sampling at 20%, 40%, 60%, 80%, and 100% of the total, respectively, for training agents of different sizes. A test pass means that the prediction error of the trained agent on the test set is less than the preset accuracy requirement. An agent that reaches this state can be put into use. The preset accuracy requirement can be determined according to the actual business needs of pipeline safety protection. For example, the accuracy threshold can be set as an average absolute error of less than 1 meter or a root mean square error of less than 1.5 meters. When the pipeline is in a high-consequence zone or other scenarios with stricter safety distance requirements, a stricter accuracy threshold can be set, such as an average absolute error of less than 0.5 meters.

[0060] In this step, to obtain multiple sensing threat distance identification agents matching different vibration cleaning intensity ranges, the sensor parameter sequences recorded in past construction disturbance events are first extracted from the pipeline historical monitoring database. Based on the measured or assessed distance between the construction machinery and the pipeline in each event, the corresponding threat distance is labeled for each set of sensor parameter sequences, constructing a sample sensor parameter sequence set and a sample first threat distance set. Then, multiple random samplings are performed from the sample set at different proportions to generate five training and testing datasets, representing 20%, 40%, 60%, 80%, and 100% of the total data. Each training and testing dataset contains independent training and testing sets. The selection of data proportions matches the vibration cleaning intensity range; a larger data volume indicates a stronger generalization ability of the model under strong disturbances.

[0061] Next, multiple threat distance recognition agents with identical structures but different initial parameter values ​​were constructed. These agents could employ multi-layer fully connected networks or one-dimensional convolutional networks. Supervised training was performed on each of the five agents using five different datasets of training and testing data. During training, the sensor parameter sequence was used as input, and the labeled threat distance was used as the output label. The model parameters were updated by minimizing the mean squared error between the predicted and labeled distances. After training, the prediction accuracy was evaluated on their respective test sets. If the prediction error was less than a preset accuracy threshold, the test was considered successful. Finally, five successful threat distance recognition agents corresponding to different vibration cleanliness ranges were obtained.

[0062] For example, consider a construction monitoring scenario along a natural gas pipeline. The vibration parameter threshold is 2.0 pixels, and the current vibration parameter is 3.2 pixels, exceeding the threshold and triggering sensor recognition. The current vibration cleanliness is 0.16, falling within the range of less than 0.2, corresponding to the selection of a sensor threat distance recognition agent based on 100% of the full training data. The sensor parameter sequence collected by the sensor device during the current time period is obtained, including the root mean square value of vibration acceleration (0.8 m / s²), sound intensity (75 dB), and soil pressure change (12 kPa), etc. This sequence is input into the selected agent, and the model outputs a first threat distance of 8.5 meters, indicating that the current construction machinery is approximately 8.5 meters from the pipeline.

[0063] In summary, this step triggers the sensing and identification process when the vibration parameters exceed a threshold. It dynamically selects the sensing threat distance identification agent with the corresponding amount of training data based on the interval into which the vibration cleanliness falls. Under strong vibration interference, a large data volume model is used to ensure identification accuracy, while under weak vibration, a small data volume model is used to improve computational efficiency. Finally, the first threat distance based on the sensing data is output, achieving a dynamic balance between identification accuracy and computational resources.

[0064] S300: Identify the monitored image sequence to obtain the second threat distance.

[0065] In this embodiment, after obtaining the first threat distance through sensor identification, in order to further utilize the monitoring image sequence to independently determine the distance between the construction machinery and the pipeline from a visual dimension, it is necessary to perform threat distance identification on the collected monitoring images to provide second-dimensional distance information for subsequent fusion identification. Traditional chemical sensors can only detect changes in gas concentration after a leak and cannot provide spatial distance information between the construction machinery and the pipeline; while relying solely on single sensor data for distance estimation results in unstable identification accuracy when strong vibration interference causes a decrease in the signal-to-noise ratio of the sensor signal, lacks cross-validation methods in the visual dimension, and makes it difficult to obtain reliable distance estimates supported by multi-dimensional data.

[0066] Step S300 of the method provided in this application embodiment includes: collecting a set of sample monitoring images based on historical pipeline monitoring data; labeling the distance between risk events and pipelines within each sample monitoring image to obtain a sample second threat distance set; constructing a visual threat distance recognition agent based on a convolutional neural network; using the sample monitoring image set and the sample second threat distance set, supervising and training the visual threat distance recognition agent until the test is qualified; sequentially inputting the monitoring image sequence into the visual threat distance recognition agent, calculating the output mean, and obtaining the second threat distance. Detailed explanation follows:

[0067] In this embodiment, the sample monitoring image set consists of multiple monitoring images containing construction risk events collected from historical pipeline monitoring data. The relative positions of the construction machinery and the pipeline differ in each image, covering scenarios with varying distances and types of construction machinery. The sample second threat distance set comprises threat distance labels corresponding to the sample monitoring images, obtained through on-site measurement of the actual distance between the construction machinery and the pipeline or by expert evaluation based on the proportion of reference objects in the image, with units in meters. The threat distance recognition agent based on a convolutional neural network is a deep learning model that takes monitoring images as input and threat distance as output. It extracts spatial features and the relative positional relationship between the construction machinery and the pipeline layer by layer through convolutional layers, and maps the extracted features to output distance values ​​through fully connected layers. The output mean is the arithmetic average of multiple distance prediction values ​​obtained after inputting each frame of the monitoring image sequence into the agent. This is used to smooth out possible random errors in single-frame recognition and improve the stability and reliability of the second threat distance.

[0068] In this step, to establish the visual recognition capability from monitoring images to threat distances, monitoring images containing construction risk events are first collected from the pipeline historical monitoring database as a sample monitoring image set. The actual distance between the construction machinery and the pipeline in each sample image is labeled. This labeling can be done through on-site measurement records or by experts estimating based on the proportions of known reference objects in the image, thus obtaining the sample second threat distance corresponding to each sample image and constructing a sample second threat distance set. The diversity of the sample images needs to cover different types of construction machinery, different lighting conditions, and different distance ranges to ensure that the trained agent has strong generalization ability.

[0069] Then, a threat distance recognition agent based on a convolutional neural network is constructed. The network structure of the threat distance recognition agent can use multiple convolutional layers and pooling layers stacked alternately for image spatial feature extraction. Each convolutional layer uses multiple convolutional kernels to slide on the image for local feature perception, and the pooling layer downsamples the feature map to reduce the number of parameters and enhance the translation invariance of features. The feature map extracted by the convolutional layers is flattened and fed into a fully connected layer. The fully connected layer performs non-linear combination of features and finally outputs a continuous distance value as the threat distance prediction value.

[0070] Next, the constructed agent is trained under supervised supervision using a set of sample monitoring images and a set of sample second threat distances. During training, the sample monitoring images are used as input, and the corresponding labeled threat distances are used as output labels. The mean squared error between the predicted distance and the labeled distance is used as the loss function. The weight parameters of each layer of the network are iteratively updated through the backpropagation algorithm, enabling the model to gradually learn the ability to accurately determine the distance between construction machinery and pipelines from images. During training, the prediction accuracy is periodically evaluated on the validation set. When the prediction error continuously decreases and stabilizes within the preset accuracy range, the test is considered passed, and the trained visual threat distance recognition agent is obtained.

[0071] Finally, during the real-time monitoring phase, each frame of the monitoring image sequence collected within the current monitoring period is sequentially input into a trained visual threat distance recognition agent. For each frame, the agent outputs a distance prediction value. After traversing all frames in the sequence, the arithmetic mean of the distance prediction values ​​for all frames is taken to obtain the second threat distance. The purpose of averaging is to reduce the accidental recognition errors caused by factors such as instantaneous occlusion, sudden changes in lighting, or image noise in a single frame, making the output second threat distance more stably reflect the actual spatial distance between the construction machinery and the pipeline.

[0072] In summary, this step constructs a threat distance recognition agent based on a convolutional neural network, identifies each frame of the monitored image sequence and takes the average value, and independently outputs the second threat distance from the visual dimension. This provides distance estimation from the image perspective for subsequent fusion and recognition with the first threat distance, making up for the lack of visual cross-validation in complex environments by single-sensor recognition.

[0073] S400: Based on the vibration cleanliness, the first threat distance and the second threat distance are fused to obtain the fused threat distance, which is used as the risk identification result.

[0074] In this embodiment, after obtaining the first threat distance through sensor recognition and the second threat distance through visual recognition, it is necessary to fuse the distance estimates from the two dimensions to obtain a more reliable threat distance. However, vibration cleanliness reflects the intensity of the current vibration interference and its impact on image acquisition quality. When the vibration cleanliness is low, the vibration interference is strong, the image quality deteriorates, and the reliability of visual recognition decreases accordingly. When the vibration cleanliness is high, the vibration interference is weak, the image is clear, and the accuracy of visual recognition is high. If a fixed weight is used for fusion, it is impossible to adaptively adjust the confidence ratio of the two distance estimates in the fusion according to the change of vibration cleanliness. This results in the fusion result being more affected by visual errors during strong vibrations and failing to fully utilize the advantages of visual accuracy during weak vibrations, making it difficult to stably output accurate risk identification results in different environments.

[0075] Step S400 of the method provided in this application embodiment includes: configuring a sensing coefficient and a visual coefficient based on the vibration cleaning degree; and fusing the first threat distance and the second threat distance based on the sensing coefficient and the visual coefficient to obtain a fused threat distance as the risk identification result. A detailed explanation follows:

[0076] In this embodiment, the sensing coefficient is a weighting coefficient assigned to the first threat distance in the fusion processing, with a value between 0 and 1. It is determined in a negative correlation with the vibration cleanliness level; that is, the smaller the vibration cleanliness level, the larger the sensing coefficient, indicating that when strong vibration interference causes image quality degradation, the sensing data has a higher reliable weight in the fusion. The visual coefficient is a weighting coefficient assigned to the second threat distance in the fusion processing, with a value between 0 and 1. The fused threat distance is a weighted comprehensive distance value obtained by multiplying the first threat distance by the sensing coefficient and adding the second threat distance by the visual coefficient, which is output as the final risk identification result in meters.

[0077] In this step, to integrate the threat distances obtained independently from sensor recognition and visual recognition into a more reliable risk assessment result, sensor coefficients and visual coefficients are first configured based on vibration cleanliness. Vibration cleanliness directly reflects the intensity of current vibration interference and the level of image acquisition quality. Higher vibration cleanliness indicates weaker vibration interference, clearer and more stable images, and higher accuracy of the second threat distance obtained from visual recognition; therefore, it should have a dominant weight in the fusion process. Conversely, lower vibration cleanliness indicates stronger vibration interference, blurrier images, and greater visual recognition errors; therefore, its weight should be reduced in the fusion process, and the weight of sensor data should be increased accordingly. Based on the configured sensor coefficients and visual coefficients, the first and second threat distances are weighted and fused. The fused threat distance equals the sensor coefficient multiplied by the first threat distance plus the visual coefficient multiplied by the second threat distance.

[0078] Furthermore, step S400 of the method provided in this application embodiment further includes: using the vibration cleaning degree as a visual coefficient; and calculating a negatively correlated sensing coefficient based on the visual coefficient. A detailed explanation follows:

[0079] In this embodiment, the visual coefficient is directly taken as the value of vibration cleanliness. Physically, a higher vibration cleanliness level indicates higher image acquisition quality and more reliable visual recognition results, thus giving it a greater weight in the fusion process. The sensing coefficient and the visual coefficient are negatively correlated, calculated as sensing coefficient = 1 - visual coefficient. The sum of both is always 1, ensuring that the fusion result is numerically within the same dimension as the distances estimated by the two independent measurements.

[0080] In this step, to determine the specific values ​​of the two coefficients, the currently calculated vibration cleanliness is directly used as the visual coefficient. When the vibration cleanliness is close to 1, vibration interference is very weak, the visual coefficient is close to 1, and the corresponding sensing coefficient is equal to 1 minus the visual coefficient, which is close to 0. The fusion result is mainly determined by the second threat distance recognized by vision. When the vibration cleanliness is close to 0, vibration interference is extremely strong, the visual coefficient is close to 0, and the sensing coefficient is close to 1. The fusion result is mainly determined by the first threat distance recognized by sensing. In this way, the configuration of the coefficients is entirely driven by the vibration cleanliness, without the need for manually setting thresholds or rules, thus achieving adaptive adjustment of the fusion weights.

[0081] For example, continuing the pipeline construction monitoring example described above. The current vibration cleanliness level is 0.16, the visual coefficient is 0.16, and the sensing coefficient equals 1 minus 0.16, which equals 0.84. The first threat distance obtained in the previous steps is 8.5 meters, and the second threat distance, identified by the visual agent, has an average value of 11.2 meters. The fused threat distance equals 0.84 multiplied by 8.5 plus 0.16 multiplied by 11.2, which equals 7.14 plus 1.79, which equals 8.93 meters. Because the current vibration cleanliness level is low, vibration interference is strong, and image acquisition quality is poor, the low value of the visual coefficient reduces the influence of visual estimation in the fusion process. The high value of the sensing coefficient makes the fusion result closer to the 8.5 meters given by the sensing data. This fused distance is output as the risk identification result.

[0082] In summary, this step uses vibration cleanliness as a visual coefficient and 1 minus the visual coefficient as a sensing coefficient to perform weighted fusion of the first and second threat distances. When the image quality is reduced due to strong vibration, the fusion weight of the sensing data is automatically increased. When the image is clear with weak vibration, more trust is given to the visual data. The fused threat distance is output as the risk identification result, realizing the dynamic optimization fusion of sensing and visual identification under different environmental conditions.

[0083] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects:

[0084] This application proposes a pipeline risk identification generation method based on real-time environmental monitoring. It extracts vibration parameters from a monitoring image sequence and filters out background vibration interference to obtain vibration cleanliness. When the vibration parameter exceeds a threshold, sensor recognition is triggered. Based on the vibration cleanliness, a matching sensing threat distance identification agent is dynamically selected to output a first threat distance. Simultaneously, a convolutional neural network is used to perform visual recognition on the monitoring image sequence to output a second threat distance. Finally, the two distances are weighted and fused using vibration cleanliness as the visual coefficient and 1 minus the visual coefficient as the sensing coefficient, outputting the fused threat distance as the risk identification result. This organic combination of methods effectively solves the technical problem of existing pipeline risk monitoring methods that rely on chemical sensors and can only passively detect pipeline damage and leakage after it occurs, failing to provide early detection and warning during construction disturbances.

[0085] Example 2, as shown in the appendix Figure 3 As shown, based on the inventive concept of the pipeline risk identification generation method based on real-time environmental monitoring provided in Embodiment 1, this application also provides a pipeline risk identification generation edge monitoring gateway based on real-time environmental monitoring, specifically including:

[0086] Data acquisition module 11 is used to monitor the monitoring image sequence of the target pipeline area through the image acquisition device, perform vibration analysis and vibration cleaning of adjacent monitoring images, and obtain vibration parameters and vibration cleaning degree.

[0087] The sensing and identification module 12 is used to acquire the sequence of sensing parameters monitored by the sensor device when the vibration parameters meet the abnormal conditions, and to identify the threat distance based on the vibration cleanliness to obtain the first threat distance.

[0088] The visual recognition module 13 is used to identify the monitoring image sequence to obtain the second threat distance;

[0089] The fusion output module 14 is used to fuse the first threat distance and the second threat distance according to the vibration cleanliness to obtain the fused threat distance as the risk identification result.

[0090] In one embodiment, the data acquisition module 11 is further configured to: monitor and acquire a sequence of monitoring images of the target pipeline area using an image acquisition device; combine multiple adjacent monitoring images within the monitoring image sequence and perform grayscale processing to obtain multiple sets of grayscale monitoring images; perform aligned vibration analysis on the multiple sets of grayscale monitoring images to obtain original vibration parameters; and perform vibration cleaning on the original vibration parameters to obtain vibration parameters and vibration cleaning degree.

[0091] Furthermore, the data acquisition module 11 is also used to: randomly select a first pixel in the first grayscale monitoring image and divide the first pixel into a window within a preset window around the first pixel; traverse and select pixels and divide the pixel window in the second grayscale monitoring image in the first group of grayscale monitoring images, calculate the similarity with the first pixel window, select the pixel with the highest similarity as the mapped pixel, calculate the distance with the first pixel to obtain the first vibration distance; continue to perform aligned vibration analysis on multiple groups of grayscale monitoring images to obtain multiple vibration distance sets; and perform fusion processing on multiple vibration distance sets to obtain the original vibration parameters.

[0092] Furthermore, the data acquisition module 11 is also used to: acquire background vibration parameters of the target pipeline area when there are no risk events; use the background vibration parameters to clean the original vibration parameters to obtain vibration parameters; and calculate the cleaning ratio of the background vibration parameters in the original vibration parameters as the vibration cleaning degree.

[0093] In one embodiment, the sensing and identification module 12 is further configured to: determine whether the vibration parameter is greater than or equal to a vibration parameter threshold, wherein the vibration parameter threshold is set according to the minimum vibration parameter when a pipeline risk event occurs; if not, continue monitoring; if yes, acquire the sensing parameter sequence monitored by the sensor device; acquire multiple qualified sensing threat distance identification agents, wherein the multiple sensing threat distance identification agents correspond to multiple vibration cleanliness intervals, and the vibration cleanliness intervals are negatively correlated with the amount of training data of the sensing threat distance identification agents; select the corresponding sensing threat distance identification agent according to the vibration cleanliness interval into which the vibration cleanliness falls, input the sensing parameter sequence, and identify and output the first threat distance.

[0094] Furthermore, the sensor identification module 12 is also used to: collect a set of sample sensor parameter sequences based on historical pipeline monitoring data, and collect a set of sample first threat distances corresponding to different sample sensor parameter sequences; extract multiple sets of training and testing data from the set of sample sensor parameter sequences and the set of sample first threat distances, wherein the extraction is performed multiple times according to different data volumes; construct multiple sensor threat distance identification agents, and conduct supervised training and testing using multiple sets of training and testing data respectively, and obtain multiple sensor threat distance identification agents after passing the test.

[0095] In one embodiment, the visual recognition module 13 is further configured to: collect a set of sample monitoring images based on historical pipeline monitoring data, label the distance between risk events and pipelines within each sample monitoring image to obtain a sample second threat distance set; construct a visual threat distance recognition agent based on a convolutional neural network; use the sample monitoring image set and the sample second threat distance set to supervise and train the visual threat distance recognition agent until the test is qualified; input the monitoring image sequence into the visual threat distance recognition agent in sequence, calculate the output mean, and obtain the second threat distance.

[0096] In one embodiment, the fusion output module 14 is further configured to: configure sensing coefficients and visual coefficients according to the vibration cleanliness; and perform fusion processing on the first threat distance and the second threat distance according to the sensing coefficients and the visual coefficients to obtain a fused threat distance as a risk identification result.

[0097] Furthermore, the fusion output module 14 is also used to: use the vibration cleanliness as a visual coefficient; and calculate a negatively correlated sensing coefficient based on the visual coefficient.

[0098] The pipeline risk identification method and edge monitoring gateway based on real-time environmental monitoring provided in this application can achieve fully automated risk identification in construction monitoring scenarios along long-distance natural gas pipelines and urban gas pipeline networks, from image vibration perception to threat distance identification through sensor and visual fusion. This system can be integrated into edge monitoring gateway devices along the pipeline. Through the collaborative perception of image acquisition devices and sensor devices, it proactively detects risks at the stage where construction machinery generates ground vibrations but has not yet contacted the pipeline. This effectively improves the early warning capability of external construction threats to the pipeline and reduces the risk of delayed accident detection caused by chemical sensors only being able to passively detect leaks after they occur. Simultaneously, it provides structured fusion identification data support for proactive defense and risk-level management of pipeline safety operation and maintenance. For specific risk identification methods and implementation details, please refer to Example 1.

[0099] It should be noted that the order of the embodiments in this application is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A method for identifying risks in a pipeline based on real-time monitoring of the environment, characterized in that, Applied to an edge monitoring gateway, the edge monitoring gateway connecting an image acquisition device and a sensor device, the method includes: By using an image acquisition device, the monitoring image sequence of the target pipeline area is monitored, and the vibration analysis and vibration cleaning of adjacent monitoring images are performed to obtain vibration parameters and vibration cleaning degree. When the vibration parameters meet the abnormal conditions, the sequence of sensing parameters monitored by the sensor device is acquired, and the threat distance is identified based on the vibration cleanliness to obtain the first threat distance; The second threat distance is obtained by identifying the monitored image sequence; Based on the vibration cleanliness, the first threat distance and the second threat distance are fused to obtain the fused threat distance, which serves as the risk identification result.

2. The method for identifying pipeline risks based on environmental real-time monitoring according to claim 1, wherein, Using an image acquisition device, a sequence of monitoring images of the target pipeline area is monitored. Alignment vibration analysis and vibration cleaning of adjacent monitoring images are performed to obtain vibration parameters and vibration cleaning degree, including: The image acquisition device is used to monitor and acquire a sequence of images of the target pipeline area. Multiple adjacent monitoring images within the monitoring image sequence are combined and subjected to grayscale processing to obtain multiple sets of grayscale monitoring images; Aligned vibration analysis was performed on multiple sets of grayscale monitoring images to obtain the original vibration parameters; The original vibration parameters are subjected to vibration cleaning to obtain vibration parameters and vibration cleaning degree.

3. The method for pipeline risk identification based on environmental real-time monitoring according to claim 2, characterized in that, Aligned vibration analysis was performed on multiple sets of grayscale monitoring images to obtain the original vibration parameters, including: In the first set of grayscale monitoring images, a first pixel is randomly selected within the first grayscale monitoring image, and a first pixel window is divided within a preset window around the first pixel. Within the second grayscale monitoring image in the first set of grayscale monitoring images, pixels are traversed and selected, and pixel windows are divided. The similarity with the first pixel window is calculated, and the pixel with the highest similarity is selected as the mapping pixel. The distance with the first pixel is calculated to obtain the first vibration distance. Further alignment vibration analysis was performed on multiple sets of grayscale monitoring images to obtain multiple sets of vibration distances; Multiple sets of vibration distances are fused to obtain the original vibration parameters.

4. The method for pipeline risk identification based on environmental real-time monitoring according to claim 2, characterized in that, The original vibration parameters are subjected to vibration cleaning to obtain vibration parameters and vibration cleaning degree, including: Obtain background vibration parameters of the target pipeline area when no risk events occur; The original vibration parameters are cleaned using the background vibration parameters to obtain the vibration parameters. The cleaning ratio of the background vibration parameters in the original vibration parameters is calculated and used as the vibration cleaning degree.

5. The method for pipeline risk identification based on environmental real-time monitoring according to claim 1, characterized in that, When the vibration parameters meet abnormal conditions, the sequence of sensing parameters monitored by the sensor device is acquired. Based on the vibration cleanliness, threat distance identification is performed to obtain the first threat distance, including: Determine whether the vibration parameter is greater than or equal to the vibration parameter threshold, wherein the vibration parameter threshold is set according to the minimum vibration parameter when a pipeline risk event occurs; If not, continue monitoring; if yes, acquire the sequence of sensing parameters monitored by the sensor device. Multiple sensor threat distance recognition agents that have passed the test are obtained. Among them, multiple sensor threat distance recognition agents correspond to multiple vibration cleanliness intervals. The vibration cleanliness intervals are negatively correlated with the amount of training data of the sensor threat distance recognition agents. Based on the vibration cleanliness range into which the vibration cleanliness falls, a corresponding sensing threat distance identification agent is selected, the sensing parameter sequence is input, and the first threat distance is obtained from the identification output.

6. The method for pipeline risk identification based on environmental real-time monitoring according to claim 5, characterized in that, Acquire multiple qualified threat distance recognition agents, including: Based on historical pipeline monitoring data, a set of sample sensor parameter sequences was collected, and a set of sample first threat distances corresponding to different sample sensor parameter sequences was collected. Multiple sets of training and test data are obtained by repeatedly extracting the set of sample sensing parameter sequences and the set of sample first threat distances, wherein multiple extractions are performed according to different data volumes. Multiple intelligent agents for identifying the distance of a sensor threat are constructed, and supervised training and testing are performed using multiple sets of training and testing data. After passing the tests, multiple intelligent agents for identifying the distance of a sensor threat are obtained.

7. The method for pipeline risk identification based on environmental real-time monitoring according to claim 1, characterized in that, The second threat distance is obtained by identifying the monitored image sequence, including: Based on historical pipeline monitoring data, a set of sample monitoring images is collected, and the distance between risk events and pipelines within each sample monitoring image is labeled to obtain a sample second threat distance set; Construct a visual threat distance recognition agent based on convolutional neural networks; Using the sample monitoring image set and the sample second threat distance set, supervised training and testing of the visual threat distance recognition agent is conducted until the test is passed; The monitored image sequence is sequentially input into the visual threat distance recognition agent, and the average value is calculated to obtain the second threat distance.

8. The method for pipeline risk identification based on environmental real-time monitoring according to claim 1, characterized in that, Based on the vibration cleanliness level, the first threat distance and the second threat distance are fused to obtain a fused threat distance, which serves as the risk identification result, including: Based on the vibration cleaning degree, configure the sensing coefficient and visual coefficient; Based on the sensing coefficient and the visual coefficient, the first threat distance and the second threat distance are fused to obtain the fused threat distance, which is used as the risk identification result.

9. The method for pipeline risk identification based on environmental real-time monitoring according to claim 1, characterized in that, Based on the vibration cleaning degree, the sensing coefficient and visual coefficient are configured, including: The vibration cleaning degree is used as a visual coefficient; Based on the visual coefficients, calculate the negatively correlated sensing coefficients.

10. An edge monitoring gateway, characterized by The device that connects the image acquisition device and the sensor device includes: The data acquisition module is used to monitor the monitoring image sequence of the target pipeline area through the image acquisition device, perform vibration analysis and vibration cleaning of adjacent monitoring images, and obtain vibration parameters and vibration cleaning degree. The sensing and identification module is used to acquire the sequence of sensing parameters monitored by the sensor device when the vibration parameters meet the abnormal conditions, and to identify the threat distance based on the vibration cleanliness to obtain the first threat distance. A visual recognition module is used to identify the monitored image sequence to obtain a second threat distance; The fusion output module is used to fuse the first threat distance and the second threat distance based on the vibration cleanliness level to obtain the fused threat distance as the risk identification result.

Images