5G transmission pipeline intelligent inspection and remote monitoring system and method based on edge internet of things

Abstract

The invention relates to the technical field of intelligent inspection, in particular to a 5G transmission pipeline intelligent inspection and remote monitoring system and method based on edge Internet of Things. Various sensor units and high-definition cameras are arranged at key positions of a communication pipeline, and pipeline state data and image data are collected in real time and transmitted through a 5G network; based on the edge Internet of Things technology, a self-adaptive multi-level decomposition and reconstruction method is adopted to carry out data denoising, standardization processing is carried out on the data, and meanwhile, a legality to-be-verified identifier is generated; after the validity is verified, extracting time domain and frequency domain features, fusing multi-sensor feature information, and generating a pipeline health score; and performing feature extraction and anomaly detection on the image data in combination with a deep learning model, comprehensively evaluating the health state of the pipeline, and performing routing inspection and adjustment on the communication pipeline. According to the invention, real-time monitoring and intelligent inspection of the communication pipeline are realized, and the accuracy and efficiency of operation and maintenance and intelligent adjustment of the communication pipeline are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent inspection, and particularly relates to a 5G transmission pipeline intelligent inspection and remote monitoring system and method based on edge Internet of Things. Background Art

[0002] With the rapid development of 5G communication technology, as a key infrastructure, the safety and stability of the operating state of communication pipelines are crucial for the reliability of communication networks. However, the traditional manual inspection method has low efficiency, high cost, and it is difficult to detect potential hidden dangers in a timely manner. Especially in the case of complex and dynamically changing internal pipeline environments, it is more likely to cause significant losses due to delayed detection.

[0003] A fire pipeline anti-freezing monitoring system disclosed in a Chinese invention patent with the publication number CN108153208B includes a fire pipeline anti-freezing monitoring terminal, a gateway, a cloud server, and a monitoring platform. The fire pipeline anti-freezing monitoring terminal is installed at the monitoring point of the fire pipeline. Its MCU microcontroller wakes up the sensing and measuring unit regularly according to a preset sampling period to collect the liquid conductivity and temperature of the fire pipeline. Its communication module interacts with the gateway data based on the LoRaWAN protocol; the gateway is connected to the cloud server in the form of 3G / 4G or wired network, and the monitoring platform remotely accesses the cloud server to filter the original collected data, and draw the conductivity and temperature waveform diagrams, set the conductivity and temperature warning threshold values and alarm values. When the monitoring data at the monitoring point exceeds the warning threshold value, the monitoring platform issues a warning signal. This invention pre-warns and automatically processes potential freezing of the fire pipeline by monitoring the liquid conductivity and temperature of the fire pipeline, reducing the risk of the fire pipeline being frozen and reducing resource waste.

[0004] At present, the inspection and monitoring of communication pipelines still face multiple challenges, such as low efficiency, insufficient accuracy, and poor real-time performance of manual inspection, and it is difficult for traditional monitoring methods to handle massive data processing and rapid detection of multiple types of faults; in addition, the lack of multi-dimensional information fusion analysis means makes it more difficult to identify and locate complex anomalies; therefore, how to construct an efficient and intelligent communication pipeline inspection and monitoring mechanism to solve problems such as untimely anomaly detection and incomplete pipeline state assessment has become a key problem that urgently needs to be solved to ensure the stable operation of communication networks. Summary of the Invention

[0005] The object of the present invention is to address the problems in the background art and propose a 5G transmission pipeline intelligent inspection and remote monitoring system and method based on edge Internet of Things.

[0006] The technical solution of the present invention: A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things includes the following specific implementation steps:

[0007] S1. Collect pipeline status data and pipeline internal image data;

[0008] S2. Summarize the multi-dimensional pipeline status data and pipeline internal image data of the communication pipeline. Based on the adaptive multi-level decomposition and reconstruction denoising method, combine signal characteristics and multi-scale analysis techniques. Perform signal denoising and normalization through adaptive decomposition, noise feature recognition, key signal extraction and reconstruction, and generate a legal identification to be verified for the processed data;

[0009] S3. Verify the legality of the pipeline status data, extract the time-domain features and frequency-domain features of the sensor data, generate a health score Ⅰ through weighted fusion. Combine deep learning techniques to extract features and detect anomalies from the image data. Use the image anomaly detection model to identify the anomaly area and category and calculate the anomaly proportion, generate an image-based health score Ⅱ, and generate a comprehensive health score through the adaptive weight fusion of the sensor and image scores to judge the pipeline status. At the same time, classify and label the anomaly causes in combination with the rule base, and output the final anomaly determination result and anomaly decision rule;

[0010] S4. Combine the final anomaly determination result and the anomaly decision rule to generate a fault warning message and adjustment suggestions;

[0011] S5. Through the visualization interface, present the real-time monitoring data, comprehensive health score, final anomaly determination result and fault warning suggestions of the pipeline to the operation and maintenance personnel in the form of charts or 3D models. When an abnormal state is detected, send a real-time alarm to the operation and maintenance personnel. The operation and maintenance personnel confirm the abnormal situation and conduct further analysis and processing in combination with the image data, sensor data and adjustment suggestions.

[0012] Preferably, the implementation process of the adaptive multi-level decomposition and reconstruction denoising method is as follows:

[0013] S21. Use empirical mode decomposition to perform multi-level decomposition on the collected pipeline status data signal, and decompose the complex signal into several intrinsic mode functions and a residual signal;

[0014] S22. By analyzing the power spectral density of each intrinsic mode function IMF j (t), according to the characteristic that noise is usually distributed in the high-frequency range, screen out the main noise-containing components. The power spectral density calculation formula is:

[0015]

[0016] In the formula, P j (f) represents the power spectral density of the j-th layer mode component; T represents the total signal duration; f represents the frequency; j' represents the imaginary unit;

[0017] According to P j(f) distribution, automatically set the noise frequency band threshold f noise , to distinguish the signal from the noise;

[0018] S23. For each modal component IMF j (t), calculate the signal-to-noise ratio:

[0019]

[0020] In the formula, SNR j represents the signal-to-noise ratio of the j-th layer modal component; represents the variance of the signal part of the modal component; represents the variance of the noise part of the modal component;

[0021] S24. Set the threshold SNR threshold , and screen the modal components with SNR j >SNR threshold for signal reconstruction:

[0022]

[0023] In the formula, represents the denoised signal.

[0024] Preferably, the decomposition process of decomposing the complex signal into several intrinsic modal components and a residual signal is as follows:

[0025] S31. Identify the extreme points: Find all the maximum and minimum values in the signal X k (t);

[0026] Among them, the initial signal X0(t) = x s (t);

[0027] S32. Construct the envelope: Connect all the maximum values using cubic spline interpolation to obtain the upper envelope E upper (t);

[0028] Connect all the minimum values to obtain the lower envelope E lower (t);

[0029] S33. Calculate the local mean: Calculate the mean of the upper and lower envelopes:

[0030]

[0031] S34. Remove the local mean: Subtract the local mean from the signal to obtain the initially extracted component:

[0032] h k (t) = X k (t) - m k (t);

[0033] S35. Check IMF conditions: Verify whether h k (t) satisfies two conditions of several intrinsic mode functions IMF, that is, the number of extreme points is equal to or differs by 1 from the number of zero-crossing points, and the local mean is 0;

[0034] If not satisfied, let X k (t) = h k (t), and repeat steps S31 - S35 until the conditions are met;

[0035] S36. Record IMF: If h k (t) satisfies the IMF conditions, take it as the first mode component IMF1(t), and the remaining signal is: R1(t) = x s (t) - IMF1(t);

[0036] S37. Iteratively extract IMF: For R1(t), extract the second IMF, that is, IMF2(T), to obtain the remaining signal: R2(t) = R1(t) - IMF2(t); Repeat the above steps S31 - S36 until the remaining signal R N (t) no longer contains oscillatory components, that is, a monotonic signal or less than a preset threshold;

[0037] S38. The signal x i (t) is decomposed into several intrinsic mode components IMF and a residual signal:

[0038]

[0039] In the formula, IMF j (t) represents the j - th layer intrinsic mode component, reflecting the characteristics of the signal in different frequency bands; R N (t) represents the residual signal, that is, the low - frequency trend term that has not been decomposed.

[0040] Preferably, the generation process of the legality to be verified identifier is as follows:

[0041] S41. Randomly select two different integers Calculate the first - order identifier element E1 = (DX i (t)) × g d mod p;

[0042] where p is a 1024 - bit large prime number; g represents a generator of a finite cyclic group Z p * of order p - 1; DX i (t) is the binary string of the processed pipeline state data ;

[0043] S42. Calculate the second - order identifier element E2 = (DXi (t))×g l mod p;

[0044] S43. Calculate the third-order identification element E3 = [1 - l×E1 - d×E2]×(CG) -1 mod p - 1;

[0045] where CG represents the identification generation code, satisfying CG ∈ Z p * , and satisfying gcd(CG, p - 1) = 1; gcd() represents the greatest common divisor function;

[0046] Accordingly, generate the identification parsing code CA = g CG mod p;

[0047] S44. Generate the identification Sign to be verified for legality = (E1, E2, E3).

[0048] Preferably, the verification process for verifying the legality of the pipeline status data is as follows:

[0049] S51. Calculate the first-order auxiliary verification element AE1:

[0050] where CA represents the identification parsing code, CA = g CG mod p;

[0051] S52. Calculate the second-order auxiliary verification element AE2:

[0052] S53. If AE1 = AE2, then the received is passed for legality verification.

[0053] Preferably, the generation process of generating the health score I through weighted fusion is as follows:

[0054] S61. Extract the data of each sensor in the pipeline status data and extract the time-domain features and frequency-domain features of the data:

[0055] S6101. Time-domain feature extraction: By calculating the mean, standard deviation and trend index of the data:

[0056] Mean:

[0057] Variance:

[0058] Trend index:

[0059] In the formula, μ is(t) represents the mean value of the sth sensor data in the ith sensor unit; σ is (t) represents the standard deviation of the sth sensor data in the ith sensor unit; T is (t) represents the trend index of the sth sensor data in the ith sensor unit; w represents the sliding window size;

[0060] S6102. Frequency-domain feature extraction: Extract vibration data, and use the fast Fourier transform to extract the frequency-domain energy distribution and the main frequency:

[0061]

[0062] In the formula, E is (f) represents the total low-frequency energy, which is used to characterize the main vibration energy distribution; f max,is represents the main frequency, which is used to detect changes in the vibration mode; the low-frequency range is [0, 50 Hz]; F is (f) represents the frequency-domain component, which is obtained by FFT;

[0063] S62. Fuse the features of different sensors by weighting to obtain a comprehensive index:

[0064]

[0065] In the formula, S i (t) represents the comprehensive health score of the ith sensor unit; w j represents the feature weight, which reflects the contribution degree of the feature to the comprehensive score; φ ij (t) represents the feature normalization value; f ij (t) represents the original feature value, that is, the mean value, the standard deviation, the frequency-domain energy distribution, and the main frequency; f ij,min and f ij,max respectively represent the minimum value and the maximum value of the feature;

[0066] S63. Fuse the comprehensive scores of all sensor units to obtain the overall health status of the pipeline:

[0067]

[0068] In the formula, H1(t) represents the health score Ⅰ based on sensor data; m represents the total number of sensor units.

[0069] Preferably, the generation process of the health score Ⅱ based on the image is as follows:

[0070] S71. Perform normalization processing on the collected original image:

[0071] I'(x, y) = CLAHE(I(x, y));

[0072] Wherein, I(x,y) represents the original image; I'(x,y) represents the image after contrast-limited histogram equalization; CLAHE() represents the contrast-limited histogram equalization processing model;

[0073] S72. Extract image features using a pre-trained convolutional neural network, and the feature vector is:

[0074] F img = CNN(I'(x,y));

[0075] Wherein, F img represents the high-dimensional feature vector of the image; CNN() represents the pre-trained model, and the ResNet model is adopted;

[0076] S73. Based on the extracted image features, identify the pipeline anomaly type through a classification and detection model. An image anomaly detection model is adopted to directly identify the abnormal regions in the image and generate the anomaly category and location: D = {(b j ,c j ,s j )|j = 1,2,…,n'};

[0077] Wherein, D represents the detection result set; b j represents the bounding box of the j-th abnormal region; c j represents the anomaly category; s j represents the detection confidence score;

[0078] S74. Calculate the proportion of the abnormal region according to the object detection result:

[0079]

[0080] Wherein, P(t) represents the proportion of the abnormal region in the total image; A(b j ) represents the area of the bounding box b j ; A total represents the total area of the image;

[0081] If P(t)>Ta, it is determined that the current image has a significant anomaly;

[0082] Wherein, Ta represents the set anomaly threshold;

[0083] S75. Calculate the health score II based on the image data: H2(t) = 1 - P(t);

[0084] Wherein, H2(t) represents the health score II based on the image data.

[0085] Preferably, the comprehensive health score combines the pipeline status score H1(t) reflected by sensor data and the health score H2(t) based on image data, and adopts an adaptive weight fusion method:

[0086] H(t) = β · H1(t) + (1 - β) · H2(t);

[0087] In the formula, H(t) represents the final comprehensive health score; β represents the adaptive weight.

[0088] Preferably, the image anomaly detection model is based on the YOLOv8 network architecture. A BotNet structure is added to the end of the backbone network of YOLOv8 to enhance the network's feature extraction of abnormal regions in the image, and an EMA attention mechanism is introduced at the end of the head of YOLOv8 to prevent drastic changes in weights.

[0089] The technical solution of the present invention: A 5G transmission pipeline intelligent inspection and remote monitoring system based on edge Internet of Things, which is used to execute the above-mentioned 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things, includes:

[0090] A data acquisition module, which is used to monitor pipeline status data in real time and transmit the pipeline status data to the data processing module through a 5G network;

[0091] Among them, the data acquisition module includes several sensor units and high-definition cameras;

[0092] A data processing module, that is, the set edge module, which is used to preprocess the acquired pipeline status data;

[0093] An anomaly detection module, which is used to analyze the health status of the pipeline using AI algorithms and identify potential abnormal situations;

[0094] An intelligent inspection and decision support module, which is used to provide real-time decision support in combination with the analysis results, generate a fault warning report, and guide maintenance work;

[0095] A remote monitoring and management module, which is used to display pipeline monitoring data in real time, for managers to monitor and manage, and supports a visual interface to display the health status, inspection progress and maintenance situation of the pipeline.

[0096] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0097] The present invention designs a 5G transmission pipeline intelligent inspection and remote monitoring system and method based on edge Internet of Things, which not only improves the accuracy, intelligence and security of communication pipeline monitoring, but also reduces the operation and maintenance costs, and has significant technical advantages and application value:

[0098] (1) Real-time and efficient pipeline status monitoring: By deploying a variety of sensors (including but not limited to temperature, humidity, pressure, vibration sensors and high-definition cameras) at key positions, comprehensive monitoring of the communication pipeline status is achieved, and using the high rate and low latency characteristics of 5G communication technology, the collected data is transmitted to the data analysis module in real time, improving the monitoring efficiency and timeliness;

[0099] (2) Precise data preprocessing and denoising: Using the adaptive multi-level decomposition and reconstruction denoising method, combined with empirical mode decomposition (EMD) and noise characteristic analysis technology, it can effectively distinguish signals from noise, extract key signals, and based on the screening mechanism of power spectral density and signal-to-noise ratio, achieve precise signal denoising, improving the reliability of sensor data and analysis accuracy;

[0100] (3) Comprehensive data standardization and legality verification: Data normalization eliminates the influence of sensor measurement range differences on data analysis, improving data comparability, and the legality identification generation and verification mechanism provides additional security for pipeline monitoring data, helping to prevent data tampering and attacks during transmission;

[0101] (4) Intelligent anomaly detection: Through time-domain feature and frequency-domain feature extraction, combined with deep learning models for image and data analysis, high-precision identification of pipeline anomalies can be achieved, and using the object detection ability of the image anomaly detection model, directly identify the abnormal areas in the image, improving the automation and intelligence level of detection;

[0102] (5) Multi-dimensional data fusion and comprehensive health assessment: Using a weighted method to perform feature fusion on multi-sensor data, generate a comprehensive health score, quantify the pipeline health status, and further fuse the scores of each sensor into an overall health score, achieving a global assessment of the overall operation status of the pipeline, providing a scientific basis for operation and maintenance decisions;

[0103] (6) Enhanced anomaly recognition ability: Combining image data with sensor data, based on edge Internet of Things technology, through deep learning models to extract features and classify and detect images, more complex anomaly types can be identified, and the combination of the proportion of abnormal areas and health scores provides an intuitive basis for evaluating the severity of pipeline anomalies. Brief Description of the Drawings

[0104] Figure 1 It is a system architecture diagram of a 5G transmission pipeline intelligent inspection and remote monitoring system based on edge Internet of Things proposed by the present invention;

[0105] Figure 2 It is a method flow diagram of a 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things proposed by the present invention. Detailed Description of the Invention

[0106] Example 1, as Figure 1 shown, a 5G transmission pipeline intelligent inspection and remote monitoring system based on edge Internet of Things proposed by the present invention includes: a data acquisition module, a data processing module, an anomaly detection module, an intelligent inspection and decision support module, and a remote monitoring and management module.

[0107] The data acquisition module monitors the pipeline status data in real time, and the data is transmitted to the data processing module through the 5G network;

[0108] Among them, the pipeline status data includes but is not limited to: the temperature, humidity, pressure, vibration environment data of the pipeline and the internal image data of the pipeline;

[0109] The data processing module, that is, the set edge module, is used to preprocess the acquired pipeline status data, including but not limited to denoising and normalization processing;

[0110] The anomaly detection module analyzes the health status of the pipeline through AI algorithms, identifies potential anomalies, including but not limited to cracks, corrosion and water leakage, and outputs the analysis results;

[0111] The intelligent inspection and decision support module combines the analysis results, provides real-time decision support, generates a fault warning report, and guides the maintenance work;

[0112] The remote monitoring and management module displays the pipeline monitoring data in real time, and the management personnel conduct monitoring and management, and supports a visual interface to display the health status, inspection progress and maintenance situation of the pipeline.

[0113] Example 2, as Figure 2 shown, a 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things proposed by the present invention is applied to a 5G transmission pipeline intelligent inspection and remote monitoring system proposed in Example 1, and its specific implementation steps are as follows:

[0114] S1. Deploy several types of sensor units at key positions of the communication pipeline, including but not limited to temperature sensors, humidity sensors, pressure sensors, vibration sensors and high-definition cameras;

[0115] Among them, the sensor unit is responsible for collecting the pipeline status data in real time, including but not limited to the temperature, humidity, pressure fluctuation and vibration signal of the pipeline;

[0116] The high-definition camera collects the internal image data of the pipeline in real time;

[0117] Accordingly, the data acquisition module transmits the real-time collected pipeline status data and internal image data of the pipeline to the data analysis module through the 5G communication network.

[0118] S2. The data analysis module preprocesses the pipeline status data collected, and its specific implementation process is as follows:

[0119] S21. Aggregate the multi-dimensional pipeline status data X i (t) = {x1(t), x2(t), …, x s (t), …, x n (t)} and the internal pipeline image data image i (t);

[0120] Among them, X i (t) represents the pipeline status data set of the i-th sensor unit at time t; x j (t) represents the original data signal collected by the s-th sensor unit in the i-th sensor unit at time t; n represents the total number of sensor units set in the i-th sensor unit;

[0121] S22. Based on the adaptive multi-level decomposition and reconstruction denoising method, combined with signal characteristics and multi-scale analysis technology, signal denoising is achieved through adaptive decomposition, noise feature recognition, key signal extraction and reconstruction. The specific denoising process is as follows:

[0122] S2201. Use empirical mode decomposition (EMD) to perform multi-level decomposition on the signal, and decompose the complex signal into several intrinsic mode functions (IMFs) and a residual signal. The decomposition process is as follows:

[0123] (1) Identify extreme points: Find all maximum and minimum values in the signal X k (t) (the initial signal X0(t) = x s (t));

[0124] (2) Construct the envelope: Connect all maximum values using an interpolation method (cubic spline interpolation in this embodiment) to obtain the upper envelope E upper (t);

[0125] Connect all minimum values to obtain the lower envelope E lower (t);

[0126] (3) Calculate the local mean: Calculate the mean of the upper and lower envelopes:

[0127]

[0128] (4) Remove the local mean: Subtract the local mean from the signal to obtain the initially extracted component:

[0129] h k(t) = X k (t) - m k (t);

[0130] (5) Check the IMF condition: Verify whether h k (t) satisfies the two conditions of the IMF, that is, the number of extreme points is equal to or differs by 1 from the number of zero crossings, and the local mean is 0;

[0131] If not satisfied, let X k (t) = h k (t), and repeat steps (1) to (5) until the condition is satisfied;

[0132] (6) Record the IMF: If h k (t) satisfies the IMF condition, take it as the first mode component IMF1(t), and the remaining signal is: R1(t) = x s (t) - IMF1(t);

[0133] (7) Iteratively extract the IMF: For the remaining signal R1(t), repeat the above steps (1) to (6) to extract the second IMF (IMF2(T)): R2(t) = R1(t) - IMF2(t);

[0134] Repeat this way until the remaining signal R N (t) no longer contains oscillatory components, that is, a monotonic signal or less than a preset threshold;

[0135] (8) The signal x i (t) is decomposed into several IMFs and a residual signal:

[0136]

[0137] In the formula, IMF j (t) represents the j-th layer intrinsic mode component, reflecting the characteristics of the signal in different frequency bands; R N (t) represents the residual signal, that is, the low-frequency trend term that has not been decomposed;

[0138] S2202. By analyzing the power spectral density (PSD) of each IMF j (t), according to the characteristic that noise is usually distributed in the high-frequency range, filter out the main noise-containing components. The power spectral density calculation formula is:

[0139]

[0140] In the formula, P j (f) represents the power spectral density of the j-th layer mode component; T represents the total signal duration; f represents the frequency; j' represents the imaginary unit;

[0141] According to Pj Distribution of (f), automatically set the noise frequency band threshold f noise , to distinguish the signal from the noise;

[0142] S2203. For each modal component IMF j (t), calculate the signal-to-noise ratio (SNR):

[0143]

[0144] In the formula, SNR j represents the signal-to-noise ratio of the j-th layer modal component; represents the variance of the signal part of the modal component; represents the variance of the noise part of the modal component;

[0145] S2204. Set the threshold SNR threshold , and screen the modal components with SNR j >SNR threshold for signal reconstruction:

[0146]

[0147] In the formula, represents the denoised signal;

[0148] S23. Normalize the data, and the normalization adopts the Min-Max method:

[0149]

[0150] In the formula, represents the normalized data; x s,max represents the maximum value of the sensor data, obtained by statistical analysis of the sensor historical data; x s,min represents the minimum value of the sensor data, obtained by statistical analysis of the sensor historical data;

[0151] S24. Generate a legitimacy pending verification flag for the processed pipeline status data , and its generation process is as follows:

[0152] S2401. Randomly select two different integers Calculate the first-order flag element E1 = (DX i (t)) × g d mod p;

[0153] Among them, p is a 1024-bit large prime number; g represents the generator of the finite cyclic group Z p * with order p - 1; DX i(t) is the processed pipeline status data binary string;

[0154] S2402. Calculate the second-order identification element E2 = (DX i (t)) × g l mod p;

[0155] S2403. Calculate the third-order identification element E3 = [1 - l × E1 - d × E2] × (CG) -1 mod p - 1;

[0156] where CG represents the identification generation code, satisfying CG ∈ Z p * , and satisfying gcd(CG, p - 1) = 1; gcd() represents the greatest common divisor function;

[0157] Accordingly, generate the identification resolution code CA = g CG mod p;

[0158] S2404. Generate the identification to be verified for legality Sign = (E1, E2, E3);

[0159] S25. Transmit to the anomaly detection module.

[0160] S3. The anomaly detection module receives and verifies and, based on artificial intelligence and machine learning algorithms, analyzes sensor data and image data in real time to identify abnormal states in the communication pipeline. The specific implementation process is as follows:

[0161] S31. Perform a legality verification on the received . The verification process is as follows:

[0162] S3101. Calculate the first-order auxiliary verification element AE1:

[0163] S3102. Calculate the second-order auxiliary verification element AE2:

[0164] S3103. If AE1 = AE2, the legality verification of the received passes;

[0165] S32. Extract the sensor data in the pipeline status data and extract the time-domain features and frequency-domain features of the data:

[0166] S3201. Time-domain feature extraction: By calculating the mean, standard deviation, and trend index of the data:

[0167] Mean:

[0168] Variance:

[0169] Trend index:

[0170] Wherein, μ is (t) represents the mean value of the s-th sensor data in the i-th sensor unit; σ is (t) represents the standard deviation of the s-th sensor data in the i-th sensor unit; T is (t) represents the trend index of the s-th sensor data in the i-th sensor unit; w represents the sliding window size;

[0171] S3202. Frequency domain feature extraction: Extract vibration data, and use the fast Fourier transform (FFT) to extract the frequency domain energy distribution and the main frequency:

[0172]

[0173] Wherein, E is (f) represents the total low-frequency energy, which is used to characterize the main vibration energy distribution; f max,is represents the main frequency, which is used to detect changes in the vibration mode; the low-frequency range is set to [0, 50 Hz] in this embodiment; F is (f) represents the frequency domain component, which is obtained through FFT;

[0174] S33. Fuse the features of different sensors through a weighted method to obtain a comprehensive index:

[0175]

[0176] Wherein, S i (t) represents the comprehensive health score of the i-th sensor unit; w j represents the feature weight, which reflects the contribution degree of the feature to the comprehensive score; φ ij (t) represents the feature normalization value; f ij (t) represents the original feature value (including but not limited to the mean value, standard deviation, and frequency energy); f ij,min and f ij,max respectively represent the minimum value and the maximum value of the feature, which are determined through historical data statistics;

[0177] S34. Fuse the comprehensive scores of all sensor units to obtain the overall health status of the pipeline:

[0178]

[0179] Wherein, H1(t) represents the health score based on sensor data; m represents the total number of sensor units;

[0180] S35. Feature extraction and preliminary analysis of image data are performed using a deep learning model, and the analysis process is as follows:

[0181] S3501. Standardize the collected original image:

[0182] I'(x, y) = CLAHE(I(x, y));

[0183] In the formula, I(x, y) represents the original image; I'(x, y) represents the image after contrast-limited adaptive histogram equalization (CLAHE);

[0184] S3502. Use a pre-trained convolutional neural network (CNN) to extract image features, and the feature vector is:

[0185] F img = CNN(I'(x, y));

[0186] In the formula, F img represents the high-dimensional feature vector of the image; CNN() represents the pre-trained model, and in this embodiment, the ResNet model is adopted;

[0187] S3503. Based on the extracted image features, identify the pipeline anomaly type through a classification and detection model. An image anomaly detection model is used to directly identify the abnormal regions in the image and generate the anomaly category and location: D = {(b j , c j , s j )|j = 1, 2,..., n'};

[0188] In the formula, D represents the detection result set; b j represents the bounding box of the j-th abnormal region; c j represents the anomaly category; s j represents the detection confidence score;

[0189] Among them, the image anomaly detection model is designed based on the YOLOv8 network architecture. A BotNet structure is added to the end of the backbone network of YOLOv8 to enhance the network's feature extraction of abnormal regions in the image. An EMA attention mechanism is introduced at the end of the head of YOLOv8 to prevent drastic changes in weights;

[0190] It should be noted that BotNet (botnet) is a simple but powerful backbone network architecture that incorporates self-attention into various computer vision tasks, including image classification, object detection, and instance segmentation. This method significantly improves the baseline in instance segmentation and object detection while also reducing the number of parameters, thereby minimizing latency;

[0191] S3504. Calculate the proportion of the abnormal area based on the target detection result:

[0192]

[0193] In the formula, P(t) represents the proportion of the abnormal area in the total image; A(b j ) represents the area of the bounding box b j ; A total represents the total area of the image;

[0194] If P(t) > Ta (the set abnormal threshold), it is determined that there is a significant abnormality in the current image;

[0195] S3505. Calculate the health score H2(t) based on the image data: H2(t) = 1 - P(t);

[0196] S36. Combine the pipeline status score H1(t) reflected by the sensor data and the health score H2(t) based on the image data, and adopt an adaptive weight fusion method:

[0197] H(t) = β·H1(t) + (1 - β)·H2(t);

[0198] In the formula, H(t) represents the final comprehensive health score; β represents the adaptive weight;

[0199] S37. Make a final abnormality determination through the comprehensive health score:

[0200] If H(t) < T final , it indicates that the health status of the communication pipeline is healthy;

[0201] If H(t) ≥ T final , it indicates that the health status of the communication pipeline is abnormal;

[0202] Among them, T final represents the comprehensive health score threshold;

[0203] In addition, according to the abnormal categories in the image (including but not limited to cracks, sediments) and the abnormal characteristics of the sensor (including but not limited to sharp increase in vibration, gas leakage), classify and label the causes of abnormalities in combination with the rule base;

[0204] S38. Transmit the comprehensive health score and the final abnormality determination result to the intelligent patrol inspection and decision support module and the remote monitoring and management module.

[0205] S4. The intelligent patrol inspection and decision support module combines the final abnormality determination result and the abnormal decision rule, provides real-time decision support, generates a fault warning suggestion, and transmits the fault warning information and adjustment suggestion to the remote monitoring and management module.

[0206] S5. The remote monitoring and management module presents the real-time monitoring data, comprehensive health score, final abnormal determination result, and fault warning suggestions of the pipeline to the operation and maintenance personnel in the form of charts or 3D models through a visual interface;

[0207] When an abnormal state is detected, a real-time alarm is sent to the operation and maintenance personnel in various forms including but not limited to sound and pop-up windows. The operation and maintenance personnel confirm the abnormal situation and conduct further analysis and processing in combination with image data, sensor data, and fault warning suggestions.

[0208] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things, characterized in that It includes the following specific implementation steps: S1. Collect pipeline status data and internal pipeline image data; S2. Summarize the multi-dimensional pipeline status data and internal pipeline image data of the communication pipeline. Based on the adaptive multi-level decomposition and reconstruction denoising method, combined with signal characteristics and multi-scale analysis techniques, perform signal denoising and normalization through adaptive decomposition, noise feature recognition, key signal extraction and reconstruction, and generate a legal identification to be verified for the processed data; S3. Verify the legality of the pipeline status data, extract the time-domain features and frequency-domain features of the sensor data, generate a health score Ⅰ through weighted fusion, combine deep learning techniques to extract features and detect anomalies from the image data, use the image anomaly detection model to identify the anomaly area and category and calculate the anomaly ratio, generate an image-based health score Ⅱ, and generate a comprehensive health score through the adaptive weight fusion of the sensor and image scores to judge the pipeline status. At the same time, classify and label the anomaly causes in combination with the rule library, and output the final anomaly determination result and anomaly decision rule; S4. Combine the final anomaly determination result and the anomaly decision rule to generate a fault warning message and adjustment suggestions; S5. Through the visualization interface, present the real-time monitoring data, comprehensive health score, final anomaly determination result and fault warning suggestions of the pipeline to the operation and maintenance personnel in the form of charts or 3D models. When an abnormal state is detected, send a real-time alarm to the operation and maintenance personnel. The operation and maintenance personnel confirm the abnormal situation and conduct further analysis and processing in combination with the image data, sensor data and adjustment suggestions.

2. The intelligent inspection and remote monitoring method for a 5G transmission pipeline based on edge Internet of Things according to claim 1, characterized in that, The implementation process of the adaptive multi-level decomposition and reconstruction denoising method is as follows: S21. Use empirical mode decomposition to perform multi-level decomposition on the collected pipeline status data signal, and decompose the complex signal into several intrinsic mode components and a residual signal; S22. By analyzing the power spectral density of each intrinsic mode function IMF j (t), according to the characteristic that noise is usually distributed in the high-frequency range, the main noise-containing components are screened out. The power spectral density calculation formula is as follows: where P j (f) represents the power spectral density of the j-th layer modal component; T represents the total signal duration; f represents the frequency; j' represents the imaginary unit; According to the distribution of P j automatically set the noise frequency band threshold f according to the distribution of (f) noise to distinguish the signal from the noise; S23. Calculate the signal-to-noise ratio for each modal component IMF j (t): where, SNR j represents the signal-to-noise ratio of the j-th layer modal component; represents the variance of the signal part of the modal component; represents the variance of the noise part of the modal component; S24. Set the threshold SNR threshold , and filter the modal components with SNR j >SNR threshold for signal reconstruction: In the formula, represents the denoised signal.

3. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things according to claim 2, characterized in that, The decomposition process of decomposing the complex signal into several intrinsic mode components and a residual signal is as follows: S31. Identify extreme points: Find all the maxima and minima in the signal X k (t); Among them, the initial signal X0(t) = x s (t); S32. Construct the envelope: Connect all the maxima using cubic spline interpolation to obtain the upper envelope E upper (t); Connect all the minima to obtain the lower envelope E lower (t); S33. Calculate the local mean: Calculate the mean of the upper and lower envelope lines: S34. Remove the local mean: Subtract the local mean from the signal to obtain a preliminarily extracted component: h k h(t) = X k h(t) - m k h(t); S35. Check IMF conditions: Verify whether h k (t) satisfies the two conditions of the intrinsic mode function IMF, that is, the number of extreme points is equal to or differs by 1 from the number of zero crossings, and the local mean is 0; If not satisfied, let X k (t) = h k (t), and repeat steps S31 - S35 until the condition is satisfied; S36. Record the IMF: If h k (t) meets the conditions of the IMF, take it as the first modal component IMF1(t), and the remaining signal is: R1(t) = x s (t) - IMF1(t); S37. Iteratively extract the IMF: For R1(t), extract the second IMF, i.e., IMF2(T), to obtain the remaining signal: R2(t) = R1(t) - IMF2(t); repeat the above steps S31 - S36 until the remaining signal R N (t) no longer contains oscillatory components, i.e., a monotonic signal or is less than a preset threshold; S38. Signal x i (t) is decomposed into a number of intrinsic mode functions IMF and a residual signal: where, IMF j (t) represents the j-th layer of intrinsic mode components, reflecting the characteristics of the signal in different frequency bands; R N (t) represents the residual signal, that is, the low-frequency trend term that has not been decomposed.

4. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things according to claim 1, characterized in that, The generation process of the legal identification to be verified is as follows: S41. Randomly select two different integers Calculate the first-order identification element E1 = (DX i (t)) × g d mod p; Among them, p is a large prime number of 1024 bits; g represents the generator of the finite cyclic group Z p * of order p - 1; DX i (t) is the processed pipeline state data as a binary string; S42. Calculate the second-order identification element E2 = (DX i (t)) × g l mod p; S43. Calculate the third-order identification element E3 = [1 - l×E1 - d×E2]×(CG) -1 mod p - 1; Among them, CG represents the identification generation code, satisfying CG ∈ Z p * , and satisfying gcd(CG, p - 1) = 1; gcd() represents the greatest common divisor function; Accordingly, generate the identification resolution code CA = g CG mod p; S44. Generate a legal identification to be verified Sign=(E1, E2, E3).

5. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things according to claim 1, characterized in that, The verification process of verifying the legality of the pipeline status data is as follows: S51. Calculate the first-order auxiliary verification element AE1: where CA represents the identifier resolution code, and CA = g CG mod p; S52. Calculate the second-order auxiliary verification element AE2: S53. If AE1 = AE2, then the legitimacy verification is passed.

6. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things according to claim 1, characterized in that, The generation process of generating a health score Ⅰ through weighted fusion is as follows: S61. Extract pipeline status data from the data of each sensor, and extract the time-domain features and frequency-domain features of the data: S6101. Time-domain feature extraction: By calculating the mean, standard deviation and trend index of the data: Mean: Variance: Trend indicator: where μ is (t) represents the mean value of the s-th sensor data in the i-th sensor unit; σ is (t) represents the standard deviation of the s-th sensor data in the i-th sensor unit; T is (t) represents the trend index of the s-th sensor data in the i-th sensor unit; w represents the sliding window size; S6102. Frequency-domain feature extraction: Extract vibration data, use the fast Fourier transform to extract the frequency-domain energy distribution and main frequency: where E is (f) represents the total low-frequency energy, which is used to characterize the main vibration energy distribution; f max,is represents the main frequency, which is used to detect changes in the vibration mode; the low-frequency range is [0, 50 Hz]; F is (f) represents the frequency-domain component, which is obtained by FFT; S62. Fuse the features of different sensors in a weighted manner to obtain a comprehensive index; where S i (t) represents the comprehensive health score of sensor unit i; w j represents the feature weight, reflecting the contribution of the feature to the comprehensive score; φ ij (t) represents the feature normalization value; f ij (t) represents the original eigenvalue, i.e., the mean value, standard deviation, frequency domain energy distribution, and main frequency; f ij,min and f ij,max represent the minimum value and maximum value of the feature, respectively; S63. Fuse the comprehensive scores of all sensor units to obtain the overall health status of the pipeline; In the formula, H1(t) represents the health score Ⅰ based on sensor data; m represents the total number of sensor units.

7. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things according to claim 1, characterized in that The generation process of the image-based health score Ⅱ is as follows: S71. Standardize the collected original image: I'(x, y)=CLAHE(I(x, y)); Where, I(x, y) represents the original image; I'(x, y) represents the image after contrast-limited histogram equalization; CLAHE() represents the contrast-limited histogram equalization processing model; S72. Use a pre-trained convolutional neural network to extract image features, and the feature vector is: F img = CNN(I'(x,y)); where, F img represents the high-dimensional feature vector of the image; CNN() represents the pre-trained model, and the ResNet model is adopted; S73. Based on the extracted image features, identify the pipeline anomaly type through a classification and detection model. Use an image anomaly detection model to directly identify the abnormal regions in the image and generate the anomaly category and location: D = {(b j , c j , s j )|j = 1, 2, …, n'}; where D represents the set of detection results; b j represents the bounding box of the j-th abnormal area; c j represents the abnormal category; s j represents the detection confidence score; S74. Calculate the proportion of the abnormal area according to the object detection result: where P(t) represents the proportion of the abnormal area in the total image; A(b j ) represents the area of the bounding box b j ; A total represents the total area of the image; If P(t) > Ta, it is determined that the current image has a significant abnormality; where, Ta represents the set abnormality threshold; S75. Calculate the health score II based on the image data: H2(t) = 1 - P(t); where, H2(t) represents the health score II based on the image data.

8. A 5G transmission pipeline intelligent inspection and remote monitoring method based on edge Internet of Things according to claim 1, characterized in that, Integrate the pipeline status score H1(t) reflected by the sensor data and the health score H2(t) based on the image data, and adopt an adaptive weight fusion method: H(t) = β · H1(t) + (1 - β) · H2(t); In the formula, H(t) represents the final comprehensive health score; β represents the adaptive weight.

9. The intelligent inspection and remote monitoring method for a 5G transmission pipeline based on edge Internet of Things according to claim 1, wherein, The image anomaly detection model is based on the YOLOv8 network architecture. A BotNet structure is added to the end of the backbone network of YOLOv8 to enhance the network's feature extraction of abnormal areas in the image, and an EMA attention mechanism is introduced at the end of the head of YOLOv8 to prevent drastic changes in weights.

10. A 5G transmission pipeline intelligent inspection and remote monitoring system based on edge Internet of Things, which is used to execute the method for intelligent inspection and remote monitoring of a 5G transmission pipeline based on edge Internet of Things according to any one of claims 1 to 9, and is characterized in that, It includes: A data acquisition module, which is used to monitor the pipeline status data in real time and transmit the pipeline status data to the data processing module through a 5G network; Among them, the data acquisition module includes several sensor units and high-definition cameras; A data processing module, that is, the set edge module, which is used to preprocess the acquired pipeline status data; An anomaly detection module, which is used to analyze the health status of the pipeline using AI algorithms and identify potential abnormal situations; An intelligent inspection and decision support module, which is used to provide real-time decision support in combination with the analysis results, generate a fault warning report, and guide the maintenance work; A remote monitoring and management module, which is used to display the pipeline monitoring data in real time, for managers to monitor and manage, and supports a visual interface to display the health status, inspection progress and maintenance situation of the pipeline.

Images