Abnormal state detection method for optical signals in OPGW optical cable

Abstract

The invention provides a method for detecting an abnormal state of an optical signal in an OPGW (Optical Fiber Composite Overhead Ground Wire) optical cable, which relates to the technical field of power transmission line anomaly detection and comprises the following steps: S1, hardware deployment; s2, multi-dimensional data acquisition is carried out; s3, signal preprocessing; s4, abnormal feature extraction and threshold dynamic optimization; s5, fault type identification; s6, fault positioning; and S7, performing active disposal and state evaluation. According to the method for detecting the abnormal state of the optical signal in the OPGW optical cable, the optical signal, the deformation signal, the environmental parameter and the operation parameter are synchronously collected to realize comprehensive coverage of fault influence factors, a CNN module captures local features such as polarization state sudden change amplitude and duration time, an LSTM module analyzes time sequence features such as stress concentration trend and icing periodical fluctuation, and the optical signal abnormal state detection method is used for detecting the abnormal state of the optical signal in the OPGW optical cable. A 128-dimensional multi-dimensional feature vector is formed through splicing, five types of faults can be accurately recognized in combination with a fingerprint database, meanwhile, a threshold value is dynamically corrected based on real-time parameters such as temperature and wind speed, the adaptability defect of a fixed threshold value is avoided, and the false alarm rate and the missing report rate are greatly reduced.

Description

Technical Field

[0001] This application relates to the field of power transmission line anomaly detection technology, and in particular to a method for detecting abnormal states of optical signals in OPGW optical cables. Background Technology

[0002] OPGW (Optical Fiber Composite Overhead Ground Wire) is a core component of high-voltage and ultra-high-voltage transmission lines, serving dual functions of power communication and lightning protection grounding. Its operational stability directly determines the safe and reliable power supply of the power system. As power grids develop towards ultra-high voltage, long-distance, and large-scale applications, OPGW optical cables widely traverse complex terrains and climates such as mountainous areas, forests, and high-altitude frigid regions. They are constantly exposed to risks of faults such as icing, galloping, lightning strikes, wind deflection, and snow accumulation. These faults can lead to abrupt changes in optical signal polarization, fluctuations in light intensity, and abnormal stress. If not detected and addressed promptly, they may cause communication interruptions, line tripping, or even large-scale power outages. Therefore, accurate detection, rapid identification, and efficient handling of abnormal optical signal states in OPGW optical cables have become critical requirements for power grid operation and maintenance. However, existing detection technologies have significant shortcomings. Their feature extraction capabilities are limited, and traditional algorithms or single neural networks struggle to simultaneously capture both local instantaneous features (such as polarization state abrupt changes) and temporal trend features (such as stress accumulation) of optical signals. They cannot accurately distinguish between multiple similar faults, and the anomaly judgment threshold is fixed. They do not consider the impact of environmental changes on signal characteristics and have poor adaptability under extreme conditions such as high temperature and strong winds. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for detecting abnormal states of optical signals in OPGW optical cables, thus solving the problems mentioned in the background section.

[0004] To achieve the above objectives, the present invention provides a method for detecting abnormal states of optical signals in OPGW optical cables, comprising the following steps: S1. Hardware Deployment: By utilizing the built-in optical fiber of the OPGW transmission line anomaly detection optical cable, a monitoring device is installed in the substation equipment room, and environmental parameter acquisition terminals and active processing units are deployed along the line. The feature fingerprint database and benchmark threshold of the AI ​​transmission line anomaly detection unit are initialized. S2, Multi-dimensional Data Acquisition: The monitoring device collects light signals and deformation signals, while the environmental unit collects temperature, wind speed, and load current. When abnormal data is detected, a high-frequency acquisition mode is triggered. S3, Signal Preprocessing: The original signal is denoised and the data is standardized by using circuit conditioning and filtering techniques. S4. Anomaly Feature Extraction and Dynamic Threshold Optimization: The CNN-LSTM transmission line anomaly detection model is used to extract 128-dimensional feature vectors for transmission line anomaly detection, and the anomaly judgment threshold is corrected based on environmental parameters. The anomaly detection threshold is adjusted using a linear correction formula, as follows: Stress signal threshold transmission line anomaly detection: S=S0×[1+α×(T-T0)+β×(V-V0)+γ×(I-I0)]; Polarization state abrupt change threshold transmission line anomaly detection: P=P0×[1+α×(T-T0)+β×(V-V0)]; Wherein, S0 is the stress reference threshold, P0 is the polarization state abrupt change reference threshold, T0=25℃, where 25℃ is the reference temperature, V0=3m / s, which is the reference wind speed, I0=500A, which is the reference current, α=-0.005, which is the temperature correction coefficient, reducing the threshold sensitivity at high temperatures; β=0.02, which is the wind speed correction coefficient, increasing the threshold during strong winds; γ=0.01, which is the current correction coefficient, increasing the threshold under high loads. S5. Fault type identification: The feature vector is matched with the fingerprint database to determine the fault type and severity. S6. Fault location: The fault location is calculated by combining the time difference of the abrupt polarization state signal at both ends with the optical cable parameters, and then corrected by tower information to achieve accurate fault location. Fault location uses a two-end location formula:

[0005] Among them, the abnormal detection of transmission line D is the distance from the fault to the abnormal detection of transmission line A at the substation. , Let c be the speed of light in the OPGW, and c be the speed of light for transmission line anomaly detection, i.e., 3 × 10⁻⁶. 8 m / s, t1 and t2 are the times when the sudden change signal is detected by both ends, n is the refractive index of the optical cable for OPGW transmission line anomaly detection, and L is the total length of the transmission line. S7. Proactive measures and status assessment: Based on the fault type and severity, a handling instruction is generated, and the line status is evaluated after execution.

[0006] Furthermore, the specific process of step S1 is as follows: Utilizing the existing OPGW optical cable on the transmission line, fiber optic connectors are spliced ​​at substations A and B at both ends of the cable, connecting to the optical interfaces of distributed optical sensing monitoring devices. Distributed optical sensing monitoring devices are fixed in the equipment rooms of substations A and B, connected to the OPGW optical cable connectors via fiber optic patch cords, and simultaneously connected to the AI ​​analysis and processing unit via industrial Ethernet. One set of environmental parameter acquisition terminals is deployed every 5km along the transmission line. These terminals include temperature and wind speed sensors, and load current transformers are installed at the substation outgoing lines. The terminals upload data to the AI ​​analysis and processing unit via a wireless network. Two sets of electric de-icing teeth are installed on each tower in areas prone to icing or galloping, symmetrically arranged on both sides of the OPGW optical cable. Adjustable vibration damping strips are installed every 2km in areas with excessive vibration. The electric de-icing teeth and adjustable vibration damping strips serve as active response units, receiving control commands from the AI ​​unit via 4G or 5G modules. The basic information of the line is entered into the remote monitoring platform, including the length L, the number of towers N, and the OPGW optical cable parameters, i.e., the refractive index n=1.468. The abnormal feature fingerprint database of the AI ​​analysis and processing unit is initialized. Specifically, 1,000 sets of historical fault data are imported, including the feature parameters of five types of faults: icing, galloping, lightning strike, wind deflection, and snow accumulation. The baseline thresholds are set, i.e., the stress signal baseline threshold S0=150MPa and the polarization state change baseline threshold P0=0.5rad.

[0007] Furthermore, the specific process of step S2 is as follows: The OTDR module of the distributed optical sensing monitoring device collects backscattered light distribution data of the OPGW optical cable once per second, with a sampling point interval of 0.1m. The DAS module collects abnormal optical signal transmission data 100 times per second, including polarization state and light intensity fluctuation. The raw data is stored locally on the device in real time. Abnormal data are those with light intensity fluctuation exceeding ±15% of the normal operating range and polarization state change rate exceeding 0.5° / s. The environmental parameter acquisition terminal collects temperature T and wind speed V every 5 minutes, and the load current transformer collects load current I every second. The data is uploaded to the AI ​​analysis and processing unit via LoRa or 4G. When the monitoring device detects that the optical signal fluctuation amplitude is greater than S0×50% or the polarization state change is greater than P0×50%, it automatically triggers the high-frequency acquisition mode. At this time, the OTDR sampling rate is increased to 1kHz and the DAS sampling rate is increased to 1kHz, and the acquisition continues for 5 minutes to ensure that the complete abnormal signal is captured.

[0008] Furthermore, the specific process of step S3 is as follows: The optical signal is converted into an electrical signal by a distributed optical sensing monitoring device and preprocessed; electromagnetic interference is filtered out by a low-pass filter, and the effective signal below 50kHz is retained. Using the db4 wavelet basis function, the signal is decomposed into four levels to remove high-frequency noise components, and the signal state equation and observation equation are established. The state equation is as follows: XP = A × X k-1 +W k Observation equation: Z k =H×X k +V k ; Where A is the state transition matrix with a value of 0.98, and H is the observation matrix with a value of 1; W k V k The noise is Gaussian white noise with variances of 0.01 and 0.02, respectively. The signal smoothing is achieved through the above equation. The filtered signal and environmental parameters T, V, and I are standardized to the [0,1] interval to eliminate dimensional differences. The standardized formula is as follows: X std =(Xt-X min ) ÷ (X max- X min ) Among them, X std This represents the normalized result; Xt is the original data value; X min It is the minimum value in the original data; X max It is the maximum value in the original data. This formula maps the original data X to the interval [0,1], thus achieving data normalization.

[0009] Furthermore, the specific process of step S4 is as follows: The AI ​​analysis and processing unit calls the CNN-LSTM fusion algorithm model to extract features from the standardized data: CNN module processing: Convolution operation performs local calculations on the input data by sliding convolution kernels to capture local features of the amplitude and duration of polarization state abrupt changes in the signal; after the convolutional layer, two max pooling layers are connected with a pooling kernel size of 2×2. Max pooling selects the maximum value within the pooling kernel region to reduce the data dimensionality while retaining the main features, preventing model overfitting, and further refining local features; LSTM module processing: Two LSTM layers are used, with 64 hidden units set in each layer. For transmission line monitoring data, the LSTM layer can analyze the trend changes of stress concentration signals, including the trend of gradual increase of stress caused by long-term stress on the towers; it can also capture the periodic fluctuations of icing signals, including the periodic icing melting process caused by temperature changes in winter. The local feature output extracted by the CNN module is concatenated with the temporal feature output extracted by the LSTM module. Since each layer of LSTM has 64 hidden units, the two layers have a total of 128 dimensions. Through this concatenation method, a 128-dimensional feature vector containing light signal features, deformation features, and environmental correlation features is finally obtained. Historical fault data is collected, preprocessed, and feature extracted to form a fingerprint database containing feature vectors of various fault types. The similarity between the extracted feature vectors and existing fault features in the fingerprint database is then calculated. The process is as follows: Let the extracted feature vector be W=[a1, a2, ..., a... n The fault feature vector in the fingerprint database is M=[b1, b2, ..., b n ], then calculate the dot product of W and M:

[0010] Calculate the magnitude of vector W:

[0011] Calculate the magnitude of vector M:

[0012] Similarity:

[0013] The calculated cosine similarity is compared with the 80% threshold. If it is less than 80%, it is determined to be a new feature. It needs to be added to the fingerprint database after manual annotation, i.e., the fault type is verified by the operation and maintenance personnel, so as to realize model iteration. Based on real-time environmental parameters T, V, and I and historical data of normal signals under the same operating conditions over the past year, the anomaly detection threshold is adjusted using a linear correction formula.

[0014] Furthermore, the specific process of step S5 is as follows: The output 128-dimensional feature vector is compared with the five types of fault features in the abnormal feature fingerprint database by cosine similarity calculation, that is, it is compared with the existing fault features. The fault type with the highest similarity and >80% is taken as the judgment result. Based on the comparison between the amplitude of the feature vector and historical fault data, the faults are divided into three levels: amplitude < 1.2 × threshold is a mild level; 1.2 × threshold ≤ amplitude < 1.5 × threshold is a moderate level; and amplitude ≥ 1.5 × threshold is a severe level.

[0015] Furthermore, the specific process of step S6 is as follows: The distributed optical sensing monitoring devices at substations A and B record the arrival times t1 and t2 of the polarization state abrupt change signal, respectively; based on the refractive index n of the OPGW optical cable and the speed of light c, the distance D from the fault location to substation A is calculated using the double-ended positioning formula.

[0016] The positioning results are then corrected by combining the tower location information, that is, the distance D from substation A is corrected to the distance D from the tower with the specific number, so as to facilitate accurate inspection by maintenance personnel. The distributed optical sensing monitoring devices of substations A and B are synchronized in time through the timing clock module.

[0017] Furthermore, the specific process of step S7 is as follows: The AI ​​analysis and processing unit generates targeted handling instructions based on the fault type and severity. After receiving the instructions, the proactive handling unit executes the operation and provides real-time feedback on the handling progress to the AI ​​analysis and processing unit. The AI ​​unit assesses the line status based on signal changes 10 minutes after handling, combined with the trend of abnormal feature spectrum curves.

[0018] Furthermore, the specific steps for assessing the line status by combining the trend of abnormal feature spectrum curves are as follows: If the signal amplitude after treatment is less than 0.8 × threshold and remains stable, it is considered normal. If the signal amplitude after treatment is between 0.8 × threshold and 1.0 × threshold, and continuous monitoring is required, it is assessed as needing attention. If the signal amplitude after processing is between 1.0 × threshold and 1.2 × threshold, manual inspection is required, and an early warning is deemed necessary. If the signal amplitude after handling is ≥1.2×th threshold, and an emergency power outage is required for maintenance, then the fault is still assessed. The assessment results are uploaded to the remote monitoring platform in real time, generating a "Transmission Line Operation Status Assessment Report", which includes fault type, location, handling effect, and recommended measures.

[0019] This invention provides a method for detecting abnormal states of optical signals in OPGW optical cables, which has the following beneficial effects: 1. The abnormal state detection method for optical signals in this OPGW optical cable achieves comprehensive coverage of fault influencing factors by synchronously collecting optical signals, deformation signals, environmental parameters, and operating parameters. It adopts a CNN-LSTM fusion algorithm. The CNN module captures local features such as the amplitude and duration of polarization state abrupt changes, while the LSTM module analyzes temporal features such as stress concentration trends and icing periodic fluctuations. The two modules are spliced ​​together to form a 128-dimensional multi-dimensional feature vector. Combined with a fingerprint database containing 1,000 sets of historical data, it can accurately identify 5 types of faults. The cosine similarity matching mechanism ensures the recognition accuracy. At the same time, the threshold is dynamically adjusted based on real-time parameters such as temperature and wind speed to avoid the adaptability defects of fixed thresholds and significantly reduce the false alarm rate and false negative rate.

[0020] 2. The abnormal state detection method for optical signals in this OPGW optical cable calculates the fault distance by monitoring the time difference of signals at both ends and the optical cable parameters, and corrects it by combining tower position information to achieve precise location of the fault, thereby clarifying the operation and maintenance inspection targets. Furthermore, by deploying active handling units such as electric de-icing teeth and adjustable vibration damping belts, handling instructions are automatically generated based on the fault type and severity, realizing automated operation from detection and identification to handling. This eliminates the need for immediate human intervention, significantly shortens fault handling time, and prevents fault escalation. Moreover, unknown features with a similarity of less than 80% are manually labeled and added to the fingerprint database to continuously optimize the model's recognition capabilities, adapt to newly emerging fault types, and generate an evaluation report containing fault details and handling effects, providing comprehensive support for operation and maintenance decisions and ensuring the long-term stable operation of the OPGW optical cable. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this drawing or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this drawing. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating an abnormal state detection method for optical signals in an OPGW optical cable according to the present invention.

[0023] The purpose, features, and advantages of this accompanying drawing will be further explained in conjunction with the embodiments and with reference to the accompanying drawing. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following description and illustration are provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0025] Obviously, the following description is merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios without any inventive effort. Furthermore, it is understood that although the effort involved in such development may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0026] Unless otherwise specified, the terms "comprising" and "including" as used in this application can be open-ended or closed-ended. For example, "comprising" and "including" can mean that other components not listed may also be included, or that only the listed components may be included.

[0027] Unless otherwise specified, the term "or" is inclusive in this application. For example, the phrase "A or B" means "A, B, or both A and B". More specifically, the condition "A or B" is satisfied by any of the following conditions: A is true (or exists) and B is false (or does not exist); A is false (or does not exist) and B is true (or exists); or both A and B are true (or exist).

[0028] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0029] like Figure 1 As shown, the present invention provides a technical solution: a method for detecting abnormal states of optical signals in OPGW optical cables, comprising the following steps: S1. Hardware Deployment: Utilizing the built-in optical fiber of the OPGW transmission line anomaly detection cable, monitoring devices are installed in the substation equipment room. Environmental parameter acquisition terminals and active processing units are deployed along the transmission line. The feature fingerprint database and benchmark threshold of the AI ​​transmission line anomaly detection unit are initialized. Specifically: Utilizing the existing OPGW optical cable on the transmission line, fiber optic connectors are spliced ​​at substations A and B at both ends of the cable, connecting to the optical interfaces of distributed optical sensing monitoring devices. Distributed optical sensing monitoring devices are fixed in the equipment rooms of substations A and B, connected to the OPGW optical cable connectors via fiber optic patch cords, and simultaneously connected to the AI ​​analysis and processing unit via industrial Ethernet. One set of environmental parameter acquisition terminals is deployed every 5km along the transmission line. These terminals include temperature and wind speed sensors, and load current transformers are installed at the substation outgoing lines. The terminals upload data to the AI ​​analysis and processing unit via a wireless network. Two sets of electric de-icing teeth are installed on each tower in areas prone to icing or galloping, symmetrically arranged on both sides of the OPGW optical cable. Adjustable vibration damping strips are installed every 2km in areas with excessive vibration. The electric de-icing teeth and adjustable vibration damping strips serve as active response units, receiving control commands from the AI ​​unit via 4G or 5G modules. The basic information of the line is entered into the remote monitoring platform, including the length L, the number of towers N, and the OPGW optical cable parameters, i.e., the refractive index n=1.468. The abnormal feature fingerprint database of the AI ​​analysis and processing unit is initialized. Specifically, 1,000 sets of historical fault data are imported, including the feature parameters of five types of faults: icing, galloping, lightning strike, wind deflection, and snow accumulation. The baseline thresholds are set, i.e., the stress signal baseline threshold S0=150MPa and the polarization state change baseline threshold P0=0.5rad. S2, Multi-dimensional Data Acquisition: The monitoring device collects optical and deformation signals, while the environmental unit collects temperature, wind speed, and load current. When abnormal data is detected, a high-frequency acquisition mode is triggered. Specifically: The OTDR module of the distributed optical sensing monitoring device collects backscattered light distribution data of the OPGW optical cable once per second, with a sampling point interval of 0.1m. The DAS module collects abnormal optical signal transmission data 100 times per second, including polarization state and light intensity fluctuation. The raw data is stored locally on the device in real time. Abnormal data are those with light intensity fluctuation exceeding ±15% of the normal operating range and polarization state change rate exceeding 0.5° / s. The environmental parameter acquisition terminal collects temperature T and wind speed V every 5 minutes, and the load current transformer collects load current I every second. The data is uploaded to the AI ​​analysis and processing unit via LoRa or 4G. When the monitoring device detects that the optical signal fluctuation amplitude is greater than S0×50% or the polarization state change is greater than P0×50%, it automatically triggers the high-frequency acquisition mode. At this time, the OTDR sampling rate is increased to 1kHz and the DAS sampling rate is increased to 1kHz, and the acquisition continues for 5 minutes to ensure that the complete abnormal signal is captured. S3, Signal Preprocessing: The original signal is denoised using circuit conditioning and filtering techniques, and the data is standardized. Specifically: The optical signal is converted into an electrical signal by a distributed optical sensing monitoring device and preprocessed; electromagnetic interference is filtered out by a low-pass filter, and the effective signal below 50kHz is retained. Using the db4 wavelet basis function, the signal is decomposed into four levels to remove high-frequency noise components, and the signal state equation and observation equation are established. The state equation is as follows: XP = A × X k-1 +W k Observation equation: Z k =H×X k +V k ; Where A is the state transition matrix with a value of 0.98, and H is the observation matrix with a value of 1; W k V k The noise is Gaussian white noise with variances of 0.01 and 0.02, respectively. The signal smoothing is achieved through the above equation. The filtered signal and environmental parameters T, V, and I are standardized to the [0,1] interval to eliminate dimensional differences. The standardized formula is as follows: X std =(Xt-X min ) ÷ (X max- X min ) Among them, X std This represents the normalized result; Xt is the original data value; X min It is the minimum value in the original data; X max It is the maximum value in the original data. This formula maps the original data X to the interval [0,1] to achieve data normalization. S4. Anomaly Feature Extraction and Dynamic Threshold Optimization: The CNN-LSTM transmission line anomaly detection model is used to extract 128-dimensional feature vectors for transmission line anomaly detection. Specifically: The AI ​​analysis and processing unit calls the CNN-LSTM fusion algorithm model to extract features from the standardized data: CNN module processing: Convolution operation performs local calculations on the input data by sliding convolution kernels to capture local features of the amplitude and duration of polarization state abrupt changes in the signal; after the convolutional layer, two max pooling layers are connected with a pooling kernel size of 2×2. Max pooling selects the maximum value within the pooling kernel region to reduce the data dimensionality while retaining the main features, preventing model overfitting, and further refining local features; The specific workflow for using the CNN module in a given scenario is as follows: Input: The data matrix after preprocessing in step S3. It contains 3 types of core data (all standardized to the [0,1] interval): Optical signal data: amplitude of polarization state abrupt change, light intensity fluctuation value; Deformation signal data: stress amplitude, optical cable deformation displacement; Environmental / operating parameters: time series of temperature (T), wind speed (V), and load current (I); Concatenate multi-dimensional data ,in 300 is the time step. (5 types of parameters: polarization state, light intensity, stress, temperature, and wind speed); use convolution kernel (Window covers 3 time steps and all 5 feature classes), Number of output channels The following formula is used to capture local features such as the amplitude and duration of polarization state abrupt changes and the peak value of light intensity fluctuations by sliding the convolution kernel;

[0030] in This indicates the time step index of the output feature map. This represents the feature dimension index (fixed to 0 here because the feature dimension is compressed to 1). Indicates the output channel index. For the first The bias of each channel outputs a feature map. ; Pooling core configuration as Pooling kernels, adapted to time step dimensionality reduction while preserving feature dimensions, are defined by the formula:

[0031] Take the maximum value of the window and output it. ; Second convolutional layer: using convolution kernel Number of output channels Output after calculation ; Second-level pooling: Similarly Pooling kernel, output ,Will Flattened into a 1-dimensional vector The first 64 dimensions are taken and adapted to the 64-dimensional temporal features output by the LSTM, ensuring that the concatenated result is 128 dimensions, and the final output is... Local feature vectors (containing instantaneous features such as polarization state abrupt changes and light intensity fluctuations) are used to concatenate with the temporal features of the LSTM module; LSTM module processing: Two LSTM layers are used, with 64 hidden units set in each layer. For transmission line monitoring data, the LSTM layer can analyze the trend changes of stress concentration signals, including the trend of gradual increase of stress caused by long-term stress on the towers; it can also capture the periodic fluctuations of icing signals, including the periodic icing melting process caused by temperature changes in winter. The specific LSTM module processing scenario is as follows: First, determine the gating calculation (input gate, forget gate, output gate), let the... Step input is ( (same as the number of input features for the CNN), the previous hidden state is (64 hidden units), cell state is Then the calculation formula for the 3 doors is: Forget Gate (controls the preservation of historical cell states):

[0032] Input gate (controls the updating of new cell states):

[0033]

[0034] Output gate (controls the output hiding state):

[0035] Cell state and hidden state updates are as follows, cell state (long-term memory) update:

[0036] Hidden state (short-term memory, output features) update:

[0037] in, The gating weight matrix (dimension adapted to 64 hidden units and 5 input features); This is a gated bias term; This is a sigmoid activation function (outputs 0~1, controlling the gating switch). Hyperbolic tangent activation function (output -1 to 1, controlling the range of feature values); For element-wise multiplication (multiplying element by element to update cell state and hidden state); Same preprocessed data as CNN modules 300 time steps, 5 types of features, initial hidden state of the first LSTM layer. Initial cell state The same applies to the second layer. , ; For each time step arrive Substitute the formulas for the three gates above to calculate the forget gate. Input gate Cell state Hidden state ; pass Long-term memory of stress amplitude temporal changes captures the stress increase trend caused by long-term stress on the tower, and outputs the first-level hidden state sequence. ; The hidden state sequence output from the first layer As the input for the second layer (the input at each time step is...) Similarly, substituting the formulas for the three gates, we calculate the second-layer gating and state. The core objective is to... Capture the periodic fluctuations in icing signals (e.g., the icing thickness increases from 3mm to 5mm every 30 minutes and then melts, corresponding to a light intensity fluctuation period of 0.3Hz). Take the hidden state of the last time step of the second layer. As the final output of the LSTM module Final output ; The local feature output extracted by the CNN module is concatenated with the temporal feature output extracted by the LSTM module. Since each layer of LSTM has 64 hidden units, the two layers have a total of 128 dimensions. Through this concatenation method, a 128-dimensional feature vector containing light signal features, deformation features, and environmental correlation features is finally obtained. 64-dimensional local features output by CNN 64-dimensional temporal features of LSTM output Perform a concatenation operation to generate a 128-dimensional feature vector. :

[0038] in, This is used for fault type identification in step S5 (calculating cosine similarity with the fingerprint database). Historical fault data is collected, preprocessed, and feature extracted to form a fingerprint database containing feature vectors of various fault types. The similarity between the extracted feature vectors and existing fault features in the fingerprint database is then calculated. The process is as follows: Let the extracted feature vector be W=[a1, a2, ..., a... n The fault feature vector in the fingerprint database is M=[b1, b2, ..., b n ], then calculate the dot product of W and M:

[0039] Calculate the magnitude of vector W:

[0040] Calculate the magnitude of vector M:

[0041] Similarity:

[0042] The calculated cosine similarity is compared with the 80% threshold. If it is less than 80%, it is determined to be a new feature. It needs to be added to the fingerprint database after manual annotation, i.e., the fault type is verified by the operation and maintenance personnel, so as to realize model iteration. Based on real-time environmental parameters T, V, and I and historical data of normal signals under the same working conditions over the past year, the anomaly judgment threshold is adjusted using a linear correction formula. The formula for adjusting the anomaly detection threshold using the linear correction formula is as follows: Stress signal threshold transmission line anomaly detection: S=S0×[1+α×(T-T0)+β×(V-V0)+γ×(I-I0)]; Polarization state abrupt change threshold transmission line anomaly detection: P=P0×[1+α×(T-T0)+β×(V-V0)]; Wherein, S0 is the stress reference threshold, P0 is the polarization state abrupt change reference threshold, T0=25℃, where 25℃ is the reference temperature, V0=3m / s, which is the reference wind speed, I0=500A, which is the reference current, α=-0.005, which is the temperature correction coefficient, reducing the threshold sensitivity at high temperatures; β=0.02, which is the wind speed correction coefficient, increasing the threshold during strong winds; γ=0.01, which is the current correction coefficient, increasing the threshold under high loads. S5. Fault type identification: The feature vector is matched with the fingerprint database to determine the fault type and severity. Specifically: The output 128-dimensional feature vector is compared with the five types of fault features in the abnormal feature fingerprint database by cosine similarity calculation, that is, it is compared with the existing fault features. The fault type with the highest similarity and >80% is taken as the judgment result. Based on the comparison between the amplitude of the feature vector and historical fault data, the faults are divided into three levels: amplitude < 1.2 × threshold is a mild level; amplitude 1.2 × threshold ≤ amplitude < 1.5 × threshold is a moderate level; and amplitude ≥ 1.5 × threshold is a severe level. S6. Fault location: The fault location is calculated by combining the time difference of the abrupt polarization state change signal at both ends with optical cable parameters, and then corrected using tower information to achieve precise fault location. Specifically: The distributed optical sensing monitoring devices at substations A and B record the arrival times t1 and t2 of the polarization state abrupt change signal, respectively; based on the refractive index n of the OPGW optical cable and the speed of light c, the distance D from the fault location to substation A is calculated using the double-ended positioning formula.

[0043] Among them, the abnormal detection of transmission line D is the distance from the fault to the abnormal detection of transmission line A at the substation. , Let c be the speed of light in the OPGW, and c be the speed of light for transmission line anomaly detection, i.e., 3 × 10⁻⁶. 8 m / s, t1 and t2 are the times when the sudden change signal is detected by both ends, n is the refractive index of the optical cable for OPGW transmission line anomaly detection, and L is the total length of the transmission line. The positioning results are then corrected by combining the tower location information, that is, the distance D from substation A is corrected to the distance D from the tower with the specific number, so as to facilitate accurate inspection by maintenance personnel. The distributed optical sensing monitoring devices of substations A and B are synchronized in time through the timing clock module. S7. Proactive measures and status assessment: Based on the fault type and severity, a handling instruction is generated, and after execution, the line status is assessed. Specifically: The AI ​​analysis and processing unit generates targeted handling instructions based on the fault type and severity. The proactive handling unit receives the instructions, executes the operation, and provides real-time feedback on the handling progress to the AI ​​analysis and processing unit. The AI ​​unit then assesses the line status based on signal changes over 10 minutes after the handling, combined with the trend of the abnormal characteristic spectrum curve. If the signal amplitude after treatment is less than 0.8 × threshold and remains stable, it is considered normal. If the signal amplitude after treatment is between 0.8 × threshold and 1.0 × threshold, and continuous monitoring is required, it is assessed as needing attention. If the signal amplitude after processing is between 1.0 × threshold and 1.2 × threshold, manual inspection is required, and an early warning is deemed necessary. If the signal amplitude after handling is ≥1.2×th threshold, and an emergency power outage is required for maintenance, then the fault is still assessed. The assessment results are uploaded to the remote monitoring platform in real time, generating a "Transmission Line Operation Status Assessment Report", which includes fault type, location, handling effect, and recommended measures.

[0044] Taking a 220kV transmission line (length L=50km, OPGW optical cable model: OPGW-24B1-120, passing through a mountainous area prone to icing and lightning strikes, tower spacing 500m, optical cable refractive index n=1.468) as an example, the detection method of this invention is fully implemented, and the specific process is as follows: 08:00: Environmental terminal collects temperature T=-8℃ and wind speed V=6m / s; load current transformer collects current I=600A, and data is uploaded to AI unit in real time; 08:00-08:14: Monitoring device's normal acquisition mode: OTDR module collects backscattered light data once per second (sampling interval 0.1m), DAS module collects optical signal data 100 times per second (including polarization state and light intensity), and raw data is stored locally; 8:15:00: The DS-OFW-01 device detected a polarization state change amplitude of 0.65 rad (>P0=0.5 rad), triggering the high-frequency acquisition mode: the OTDR sampling rate was increased to 1 kHz, and the DAS sampling rate was increased to 1 kHz; 08:15:00-08:20:00: The stress signal amplitude of 180 MPa and the light intensity fluctuation frequency of 0.3 Hz were continuously acquired (consistent with the typical signal characteristics of icing faults), and the data was uploaded to the AI ​​unit in real time; The weak optical signal voltage (original value approximately 5 × 10⁻⁶) was acquired using an OPA227 operational amplifier. -4 The voltage is amplified to 2.5V and filtered by a 50kHz low-pass filter to remove 100kHz electromagnetic interference (power line interference), retaining the effective signal from 1-50kHz. The filtered stress signal (180MPa), polarization state change (0.65rad), and environmental parameters (T=-8℃, V=6m / s, I=600A) are then standardized to the [0,1] interval using a formula to eliminate dimensional differences (e.g., stress signals Xmin=50MPa, Xmax=500MPa, after standardization: (180 50)÷(500 50)≈0.289 AI unit calls CNN-LSTM fusion model: CNN module: Captures local features of polarization state abrupt changes (abrupt change amplitude 0.65 rad, duration 12 s) through sliding convolution kernels, refines features through two 2×2 max pooling layers to reduce data dimensionality; LSTM module: 2 LSTM layers (64 hidden units per layer) analyze the temporal trend of stress signal (180MPa shows a slow upward trend, consistent with the characteristics of icing accumulation) and capture the periodicity of light intensity fluctuation (0.3Hz, matching the periodicity of icing thickness change). Feature concatenation: Concatenate the local features of CNN with the temporal features of LSTM to generate a 128-dimensional feature vector containing light signal, deformation, and environmental features; The anomaly detection threshold is adjusted based on environmental parameters, and the stress signal threshold S is substituted into the parameters:

[0045]

[0046] Substitute the polarization state abrupt change threshold P into the parameters:

[0047] The cosine similarity between the 128-dimensional feature vector and the five types of fault features in the fingerprint database was calculated. The results showed that the similarity with the icing fault was 92% (>80% threshold), so there was no need to add new fingerprint database features. Comparing the eigenvector amplitude with the dynamic threshold, the measured stress amplitude of 180MPa < 1.2×S (1.2×333.75=400.5MPa), which is determined to be light icing (thickness of about 3-5mm). Substation A detected the moment of polarization state change. Substation B detected Time difference ; Calculation of the speed of light:

[0048] Fault distance calculation:

[0049]

[0050] At kilometer 25.01, the distance between the line towers is 500m. 25011.49m÷500m≈50.02 towers. Therefore, the fault location is 11.49m to the right of tower No. 50. Based on the information about the light icing and tower number 50, the AI ​​unit generates the following instructions: the electric de-icing tooth of tower number 50 extends 20mm, reciprocates 6 times per minute, and continues to de-ice. 08:17:00: De-icing gears begin operation; 08:20:00: Feedback indicates 80% de-icing complete. 08:25:00 (10 minutes after handling): The monitoring device collected a stress signal amplitude that dropped to 120MPa, with a polarization state change amplitude of 0.3rad. The evaluation standard comparison was: 120MPa < 0.8×S (0.8×333.75=267MPa), and the signal remained stable. The final evaluation result was: the line returned to normal. The evaluation report (including fault type, location, and handling effect) was uploaded to the remote monitoring platform in real time.

[0051] Based on the above description, this invention achieves comprehensive coverage of fault influencing factors by simultaneously acquiring optical signals, deformation signals, environmental parameters (temperature, wind speed), and operating parameters (load current). It adopts a CNN-LSTM fusion algorithm, where the CNN module captures local features such as the amplitude and duration of polarization state abrupt changes, and the LSTM module analyzes temporal features such as stress concentration trends and icing periodic fluctuations. These are then combined to form a 128-dimensional multi-dimensional feature vector. Combined with a fingerprint database containing 1,000 sets of historical data, it can accurately identify 5 types of faults. The cosine similarity matching mechanism ensures recognition accuracy. At the same time, the threshold is dynamically adjusted based on real-time parameters such as temperature and wind speed to avoid the adaptive defects of fixed thresholds and significantly reduce the false alarm rate and false negative rate. Furthermore, by calculating the fault distance using the time difference of the dual-end monitoring signals and optical cable parameters, and correcting it with tower location information, the fault location can be accurately pinpointed, thus clarifying the operation and maintenance inspection targets. Moreover, by deploying active handling units such as electric de-icing teeth and adjustable vibration damping belts, handling instructions are automatically generated based on the fault type and severity, realizing automated operation from detection and identification to handling. This eliminates the need for immediate human intervention, significantly shortening fault handling time and preventing fault escalation. In addition, unknown features with a similarity of less than 80% are manually labeled and added to the fingerprint database to continuously optimize the model's recognition capabilities, adapt to newly emerging fault types, and generate an evaluation report containing fault details and handling effects, providing comprehensive support for operation and maintenance decisions and ensuring the long-term stable operation of OPGW optical cables.

[0052] It should be noted that this application is not limited to the above-described embodiments. The above embodiments are merely examples, and any embodiments with the same structure and effect as the technical concept within the scope of this application are included in the technical scope of this application. Furthermore, various modifications that can be conceived by those skilled in the art to the embodiments, and other ways of constructing by combining some of the constituent elements of the embodiments, without departing from the spirit of this application, are also included in the scope of this application.

Claims

1. A method for detecting abnormal states of optical signals in OPGW optical cables, characterized in that: Includes the following steps: S1. Hardware Deployment: By utilizing the built-in optical fiber of the OPGW transmission line anomaly detection optical cable, a monitoring device is installed in the substation equipment room, and environmental parameter acquisition terminals and active processing units are deployed along the line. The feature fingerprint database and benchmark threshold of the AI ​​transmission line anomaly detection unit are initialized. S2, Multi-dimensional Data Acquisition: The monitoring device collects light signals and deformation signals, while the environmental unit collects temperature, wind speed, and load current. When abnormal data is detected, a high-frequency acquisition mode is triggered. S3, Signal Preprocessing: The original signal is denoised and the data is standardized by using circuit conditioning and filtering techniques. S4. Anomaly Feature Extraction and Dynamic Threshold Optimization: The CNN-LSTM transmission line anomaly detection model is used to extract 128-dimensional feature vectors for transmission line anomaly detection, and the anomaly judgment threshold is corrected based on environmental parameters. The anomaly detection threshold is adjusted using a linear correction formula, as follows: Stress signal threshold transmission line anomaly detection: S=S0×[1+α×(T-T0)+β×(V-V0)+γ×(I-I0)]; Polarization state abrupt change threshold transmission line anomaly detection: P=P0×[1+α×(T-T0)+β×(V-V0)]; Wherein, S0 is the stress reference threshold, P0 is the polarization state abrupt change reference threshold, T0=25℃, where 25℃ is the reference temperature, V0=3m / s, which is the reference wind speed, I0=500A, which is the reference current, α=-0.005, which is the temperature correction coefficient, reducing the threshold sensitivity at high temperatures; β=0.02, which is the wind speed correction coefficient, increasing the threshold during strong winds; γ=0.01, which is the current correction coefficient, increasing the threshold under high loads. S5. Fault Type Identification: The feature vector is matched with the fingerprint database to determine the fault type and severity. S6. Fault location: The fault location is calculated by combining the time difference of the double-ended polarization state abrupt signal with optical cable parameters, and then corrected by tower information to achieve accurate fault location. Fault location uses a two-end location formula: Among them, the abnormal detection of transmission line D is the distance from the fault to the abnormal detection of transmission line A at the substation. , Let c be the speed of light in the OPGW, and c be the speed of light for transmission line anomaly detection, i.e., 3 × 10⁻⁶. 8 m / s, t1 and t2 are the times when the sudden change signal is detected by both ends, n is the refractive index of the optical cable for OPGW transmission line anomaly detection, and L is the total length of the transmission line. S7. Proactive measures and status assessment: Based on the fault type and severity, a handling instruction is generated, and the line status is evaluated after execution.

2. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S1 is as follows: Utilizing the existing OPGW optical cable on the transmission line, fiber optic connectors are spliced ​​at substations A and B at both ends of the cable, connecting to the optical interfaces of distributed optical sensing monitoring devices. Distributed optical sensing monitoring devices are fixed in the equipment rooms of substations A and B, connected to the OPGW optical cable connectors via fiber optic patch cords, and simultaneously connected to the AI ​​analysis and processing unit via industrial Ethernet. One set of environmental parameter acquisition terminals is deployed every 5km along the transmission line. These terminals include temperature and wind speed sensors, and load current transformers are installed at the substation outgoing lines. The terminals upload data to the AI ​​analysis and processing unit via a wireless network. Two sets of electric de-icing teeth are installed on each tower in areas prone to icing or galloping, symmetrically arranged on both sides of the OPGW optical cable. Adjustable vibration damping strips are installed every 2km in areas with excessive vibration. The electric de-icing teeth and adjustable vibration damping strips serve as active response units, receiving control commands from the AI ​​unit via 4G or 5G modules. Enter basic line information into the remote monitoring platform, including length L, number of towers N, and OPGW optical cable parameters, i.e., refractive index n=1.

468. Initialize the abnormal feature fingerprint database of the AI ​​analysis and processing unit, specifically by importing 1000 sets of historical fault data, including the feature parameters of five types of faults: icing, galloping, lightning strike, wind deflection, and snow accumulation. Set the benchmark thresholds, i.e., stress signal benchmark threshold S0=150MPa and polarization state change benchmark threshold P0=0.5rad. The specific thresholds are set according to actual needs.

3. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S2 is as follows: The OTDR module of the distributed optical sensing monitoring device collects backscattered light distribution data of the OPGW optical cable once per second, with a sampling point interval of 0.1m. The DAS module collects abnormal optical signal transmission data 100 times per second, including polarization state and light intensity fluctuation. The raw data is stored locally on the device in real time. Abnormal data are those with light intensity fluctuation exceeding ±15% of the normal operating range and polarization state change rate exceeding 0.5° / s. The specific thresholds are set according to actual needs. The environmental parameter acquisition terminal collects temperature T and wind speed V every 5 minutes, and the load current transformer collects load current I every second. The data is uploaded to the AI ​​analysis and processing unit via LoRa or 4G. When the monitoring device detects that the optical signal fluctuation amplitude is greater than S0×50% or the polarization state change is greater than P0×50%, it automatically triggers the high-frequency acquisition mode. At this time, the OTDR sampling rate is increased to 1kHz and the DAS sampling rate is increased to 1kHz. The acquisition continues for 5 minutes to ensure that the complete abnormal signal is captured. The specific threshold and sampling rate are set according to actual needs.

4. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S3 is as follows: The optical signal is converted into an electrical signal by a distributed optical sensing monitoring device and preprocessed; electromagnetic interference is filtered out by a low-pass filter, and the effective signal below 50kHz is retained. Using the db4 wavelet basis function, the signal is decomposed into four levels to remove high-frequency noise components, and the signal state equation and observation equation are established. The state equation is as follows: XP=A×X k-1 +W k Observation equation: With k =H×X k +V k Where A is the state transition matrix with a value of 0.98, and H is the observation matrix with a value of 1; W k V k The noise is Gaussian white noise with variances of 0.01 and 0.02, respectively. The above equation is used to smooth the signal. The specific values ​​are set according to actual needs. The filtered signal and environmental parameters T, V, and I are standardized to the [0,1] interval to eliminate dimensional differences. The standardized formula is as follows: X std =(Xt-X min )÷(X max- X min ) Among them, X std This represents the normalized result; Xt is the original data value; X min It is the minimum value in the original data; X max It is the maximum value in the original data. This formula maps the original data X to the interval [0,1], thus achieving data normalization.

5. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S4 is as follows: The AI ​​analysis and processing unit calls the CNN-LSTM fusion algorithm model to extract features from the standardized data: CNN module processing: Convolution operation performs local calculations on the input data by sliding convolution kernels to capture local features of the amplitude and duration of polarization state abrupt changes in the signal; after the convolutional layer, two max pooling layers are connected with a pooling kernel size of 2×2. Max pooling selects the maximum value within the pooling kernel region to reduce the data dimensionality while retaining the main features, preventing model overfitting, and further refining local features; LSTM module processing: Two LSTM layers are used, with 64 hidden units set in each layer. For transmission line monitoring data, the LSTM layer can analyze the trend changes of stress concentration signals, including the trend of gradual increase of stress caused by long-term stress on the towers; it can also capture the periodic fluctuations of icing signals, including the periodic icing melting process caused by temperature changes in winter. The local feature output extracted by the CNN module is concatenated with the temporal feature output extracted by the LSTM module. Since each layer of LSTM has 64 hidden units, the two layers have a total of 128 dimensions. Through this concatenation method, a 128-dimensional feature vector containing light signal features, deformation features, and environmental correlation features is finally obtained. Historical fault data is collected, preprocessed, and feature extracted to form a fingerprint database containing feature vectors of various fault types. The similarity between the extracted feature vectors and existing fault features in the fingerprint database is then calculated. The process is as follows: Let the extracted feature vector be W=[a1, a2, ..., a... n The fault feature vector in the fingerprint database is M=[b1, b2, ..., b n ], then calculate the dot product of W and M: Calculate the magnitude of vector W: Calculate the magnitude of vector M: Similarity: The calculated cosine similarity is compared with the 80% threshold. If it is less than 80%, it is determined to be a new feature. It needs to be added to the fingerprint database after manual annotation, i.e., the fault type is verified by the operation and maintenance personnel, so as to realize model iteration. Based on real-time environmental parameters T, V, and I and historical data of normal signals under the same operating conditions over the past year, the anomaly detection threshold is adjusted using a linear correction formula.

6. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S5 is as follows: The output 128-dimensional feature vector is compared with the five types of fault features in the abnormal feature fingerprint database by cosine similarity calculation, that is, it is compared with the existing fault features. The fault type with the highest similarity and >80% is taken as the judgment result. Based on the comparison between the amplitude of the feature vector and historical fault data, the faults are divided into three levels: amplitude < 1.2 × threshold is a mild level; 1.2 × threshold ≤ amplitude < 1.5 × threshold is a moderate level; and amplitude ≥ 1.5 × threshold is a severe level.

7. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S6 is as follows: The distributed optical sensing monitoring devices at substations A and B record the arrival times t1 and t2 of the polarization state abrupt change signal, respectively; based on the refractive index n of the OPGW optical cable and the speed of light c, the distance D from the fault location to substation A is calculated using the double-ended positioning formula. The positioning results are then corrected by combining the tower location information, that is, the distance D from substation A is corrected to the distance D from the tower with the specific number, so as to facilitate accurate inspection by maintenance personnel. The distributed optical sensing monitoring devices of substations A and B are synchronized in time through the timing clock module.

8. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 1, characterized in that: The specific process of step S7 is as follows: The AI ​​analysis and processing unit generates targeted handling instructions based on the fault type and severity. After receiving the instructions, the proactive handling unit executes the operation and provides real-time feedback on the handling progress to the AI ​​analysis and processing unit. The AI ​​unit assesses the line status based on signal changes 10 minutes after handling, combined with the trend of abnormal feature spectrum curves.

9. The method for detecting abnormal states of optical signals in an OPGW optical cable according to claim 8, characterized in that: The specific steps for assessing the line status by combining the trend of abnormal feature spectrum curves are as follows: If the signal amplitude after treatment is less than 0.8 × threshold and remains stable, it is considered normal. If the signal amplitude after treatment is between 0.8 × threshold and 1.0 × threshold, and continuous monitoring is required, it is assessed as needing attention. If the signal amplitude after processing is between 1.0 × threshold and 1.2 × threshold, manual inspection is required, and an early warning is deemed necessary. If the signal amplitude after handling is ≥1.2×th threshold, and an emergency power outage is required for maintenance, then the fault is still assessed. The assessment results are uploaded to the remote monitoring platform in real time, generating a "Transmission Line Operation Status Assessment Report", which includes fault type, location, handling effect, and recommended measures.

Images