A classification and early warning method and system for ice thickness of overhead transmission lines

Abstract

The present invention discloses a classification and early warning method for ice thickness of overhead transmission lines and a system thereof, comprising the following steps: S1, using a phase-sensitive optical time domain reflectometer to collect data of overhead transmission lines and classify the overhead transmission lines; S2, framing the ice data of different thicknesses in the overhead transmission lines, performing phase demodulation to obtain a phase difference spectrum, and using a Gram angle field to generate a two-dimensional image; S3, creating a CNN model, and using the two-dimensional image to train the CNN model; S4, using the CNN model to classify whether the overhead transmission lines are iced, and performing risk assessment and early warning on the overhead transmission lines that are iced. The early warning system is not easily affected by the environment, has a long service life, reliable data transmission, and can accurately and comprehensively monitor in the monitoring of ice on overhead transmission lines.

Description

Technical Field

[0001] The present invention relates to the fields of optical fiber sensing technology and computer vision, and in particular to a classification and early warning method and system for ice thickness of overhead power transmission lines. Background Art

[0002] Icing of transmission lines can easily lead to a sharp decline in the mechanical and electrical performance of transmission lines, which not only causes huge economic losses, but also seriously affects the safe and stable operation of the power system. It is very important to monitor the operating status of transmission lines to assess the threat caused by icing and take appropriate measures in time to suppress the harm caused by icing. Traditional icing monitoring methods for transmission lines, such as weighing method, image method, and conductor inclination method, are easily affected by the environment, have a short service life, poor data transmission reliability, and cannot accurately and comprehensively monitor the status of transmission lines. Distributed fiber optic sensing technology has good electrical insulation, strong anti-electromagnetic interference ability and high sensitivity. At the same time, it can theoretically monitor the entire length of the line. It is very suitable for working in harsh environments such as high voltage, strong electromagnetic interference, and strong corrosion of transmission lines. Compared with other methods, it can better guarantee the accuracy of measurement data and the stability of the monitoring system.

[0003] In the early stage of pattern recognition research, the threshold method or simple feature extraction was mainly used to identify different vibration categories, and the classification algorithm was relatively simple. Due to the influence of environmental and regional differences, the pattern recognition of distributed optical fiber vibration sensors has a high false alarm rate and poor timeliness, which cannot meet the actual needs of current industrial production. Since 2011, researchers have begun to propose a variety of pattern recognition methods for distributed optical fiber vibration sensors based on machine learning. Various feature extraction methods have emerged, such as wavelet packet analysis, wavelet information entropy, Hilbert transform, short-time energy method, etc. Pattern recognition classifiers are also constantly updated, such as adaptive threshold, support vector machine, Bayesian classifier, BP neural network, radial basis function neural network, etc. In the past five years, the accuracy of pattern recognition of distributed optical fiber vibration sensors has been further improved, but from the perspective of the features to be classified, a large number of existing methods use feature vectors (one-dimensional feature vectors in the time domain or frequency domain) in the form of. Although deep learning methods (1D CNN, RNN, LSTM, etc.) can directly process one-dimensional data, these networks are often difficult to train, some studies are difficult to apply, and there are no existing pre-trained networks.

[0004] The present invention starts from the perspective of the impact of ice on the vibration characteristics of transmission lines. Based on the principle that the thicker the ice, the smaller the natural frequency of each mode of the overhead wire, and there is a monotonically decreasing mapping relationship between the natural frequency of each mode of the overhead wire and the ice thickness, the time series is converted into GAFs images, which are classified using the advantages of CNN in machine vision, and finally a mapping relationship between GAFs images and ice thickness is established. Summary of the invention

[0005] The technical problem to be solved by the present invention is: to provide a method and system for classifying and warning the thickness of ice coating on overhead transmission lines, based on convolutional neural networks and distributed optical fiber sensing, aiming to solve the problems in ice coating monitoring of overhead transmission lines that are easily affected by the environment, have a short service life, have poor data transmission reliability, and cannot be accurately and comprehensively monitored.

[0006] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0007] A classification and early warning method for ice thickness of overhead transmission lines comprises the following steps:

[0008] S1. Using a phase-sensitive optical time domain reflectometer to collect data from overhead transmission lines and classify the overhead transmission lines;

[0009] S2, dividing the ice data of different thicknesses in the overhead transmission line into frames, performing phase demodulation to obtain a phase difference spectrum, and using a Gram angle field to generate a two-dimensional image;

[0010] S3, creating a CNN model, and using the two-dimensional image to train the CNN model;

[0011] S4. Classify whether the overhead transmission line is covered with ice by using the CNN model, and conduct risk assessment and early warning on the overhead transmission line covered with ice.

[0012] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0013] An overhead power transmission line ice thickness classification early warning system, comprising:

[0014] A data acquisition module, used to collect data of overhead power transmission lines using a phase-sensitive optical time domain reflectometer, and classify the overhead power transmission lines;

[0015] A data analysis and preprocessing module is used to divide the ice data of different thicknesses in the overhead transmission line into frames, perform phase demodulation to obtain a phase difference spectrum, and generate a two-dimensional image using a Gram angle field;

[0016] A model training module, used for creating a CNN model, and using the two-dimensional image to train the CNN model;

[0017] The processing and early warning module is used to classify whether the overhead transmission line is covered with ice through the CNN model, and to perform risk assessment and early warning on the overhead transmission line covered with ice.

[0018] The beneficial effects of the present invention are as follows: the present invention uses a convolutional neural network to process the GAFs image generated after demodulation, making full use of the advantages of machine vision, and realizing real-time detection of icing conditions of overhead power lines while ensuring the accuracy of classification results. The convolutional neural network model proposed in the present invention does not require complex parameter adjustments and can be classified quickly and accurately. The early warning system in the present invention is not easily affected by the environment in the monitoring of icing of overhead power transmission lines, has a long service life, reliable data transmission, and can accurately and comprehensively monitor. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flowchart of an overhead transmission line ice thickness classification and early warning method according to an embodiment of the present invention;

[0020] Figure 2 A framework diagram of an overhead transmission line ice thickness classification and early warning system according to an embodiment of the present invention;

[0021] Figure 3 A structural diagram of a neural network model of an overhead transmission line ice thickness classification and early warning method according to an embodiment of the present invention and its corresponding parameters;

[0022] Figure 4 An amplitude-frequency diagram of the same amplitude and different frequencies and its corresponding GAFs image for training of an overhead transmission line ice thickness classification and early warning method according to an embodiment of the present invention;

[0023] Figure 5 Amplitude-frequency diagrams of different amplitudes and the same frequency and their corresponding GAFs images for training in an overhead transmission line ice thickness classification and early warning method according to an embodiment of the present invention;

[0024] Figure 6 The present invention provides an amplitude-frequency diagram under the influence of interference frequency of an overhead transmission line ice thickness classification and early warning method and its corresponding GAFs image for training. DETAILED DESCRIPTION

[0025] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0026] Glossary:

[0027] RBS: Radio base station, used to provide a wireless interface between the mobile station and the system, mainly composed of a wireless transceiver.

[0028] GAFs: Gram Angular Fields, can convert time series data into spatial data, that is, image-like data, and convolutional neural networks can be used for feature extraction later.

[0029] φ-OTDR: Phase-sensitive optical time-domain reflectometer is a precision optoelectronic integrated instrument made by using the backscattering caused by Rayleigh scattering and Fresnel reflection when light is transmitted in optical fiber.

[0030] CNN: A deep learning method with an architecture of convolutional, pooling, and fully connected layers.

[0031] BatchNormal: Batch normalization, normalizes the input batch data and maps it to a normal distribution with a mean of 0 and a variance of 1.

[0032] ReLU activation function: Rectified linear unit, a piecewise linear function that changes all negative values ​​to 0, while keeping positive values ​​unchanged. This operation is called unilateral inhibition.

[0033] The embodiment of the present invention provides a classification and early warning method for ice thickness of overhead transmission lines, which is characterized by comprising the following steps:

[0034] S1. Using a phase-sensitive optical time domain reflectometer to collect data from overhead transmission lines and classify the overhead transmission lines;

[0035] S2, dividing the ice data of different thicknesses in the overhead transmission line into frames, performing phase demodulation to obtain a phase difference spectrum, and using a Gram angle field to generate a two-dimensional image;

[0036] S3, creating a CNN model, and using the two-dimensional image to train the CNN model;

[0037] S4. Classify whether the overhead transmission line is covered with ice by using the CNN model, and conduct risk assessment and early warning on the overhead transmission line covered with ice.

[0038] From the above description, it can be seen that the beneficial effects of the present invention are: the present invention uses a convolutional neural network to process the GAFs image generated after demodulation, making full use of the advantages of machine vision, and realizing real-time detection of icing conditions of overhead power lines while ensuring the accuracy of classification results. The convolutional neural network model proposed in the present invention does not require complex parameter adjustments and can be classified quickly and accurately. The early warning system in the present invention is not easily affected by the environment in the monitoring of icing of overhead power transmission lines, has a long service life, reliable data transmission, and can accurately and comprehensively monitor.

[0039] Furthermore, the classification of the overhead transmission line data in step S1 is specifically as follows:

[0040] The overhead transmission line data is divided into three categories: no ice coverage, ice coverage below a risk threshold, and ice coverage exceeding a risk threshold, wherein the risk threshold is set as an axial stress value, and the axial stress value is less than the yield strength of the steel;

[0041] The ice load F on the transmission tower node i in the transmission line i for:

[0042]

[0043] Where n is the number of components, ρ is the ice density, and h is 2 represents the coefficient of variation of ice diameter with height, l j represents the length of a single component, g represents the ice-covered gravity of the ground wire;

[0044] The ice-covered gravity of the ground wire g=9.8×0.9πδ(d+δ)×10 -3 , the unit is electric charge, where d represents the outer diameter of the conductor and ground wire, and δ represents the ice thickness.

[0045] From the above description, it can be seen that the collected ice coverage data of overhead transmission lines with different thicknesses are framed, demodulated to obtain phase difference spectra, and GAFs are used to generate two-dimensional images to provide data samples for the deep learning model.

[0046] Furthermore, the phase demodulation in step S2 is specifically as follows:

[0047] S201, performing intermediate frequency filtering on the original signal of the wireless base station collected by the data acquisition card to suppress broadband noise and obtain an intermediate frequency signal;

[0048] S202, performing IQ demodulation on the intermediate frequency signal, constructing a complex intermediate frequency, and extracting the amplitude and phase of the complex intermediate frequency;

[0049] S203, performing sliding average of the amplitude along the time line with a window of appropriate width;

[0050] S204, dividing the n sampling points of each time row in step S203 into phase-detection intervals of equal width connected end to end, and finding and saving the column index of the position with the largest amplitude in all intervals corresponding to the time row in the result of S203;

[0051] S205, extracting the phases in the two adjacent left and right intervals from the phase as the estimation of the RBS phase at the optical fiber position corresponding to the sampling points of the two left and right intervals at the current moment, calculating the first phase difference, tracing back the phases at a pair of column indices of the two adjacent left and right intervals in the previous row that are the same as the current moment, as the estimation of the RBS phase at the optical fiber position corresponding to the sampling points of the two intervals at the previous moment, calculating the second phase difference, subtracting the phase difference at the previous moment from the phase difference at the current moment, and obtaining the change in the phase difference;

[0052] S206, repeat step S205, start traversing from the second row, the second interval, and calculate the change in the phase difference between the left and right intervals of each interval within one detection pulse cycle;

[0053] Along the time axis, the matrix is ​​accumulated and summed, and then unwound to restore the time-varying optical fiber expansion and contraction between adjacent intervals caused by external vibration.

[0054] From the above description, it can be seen that the GAFs image generated after demodulation is convenient to be further processed in the convolutional neural network.

[0055] Furthermore, the step S2 of using the Gram angle field to generate a two-dimensional image is specifically as follows:

[0056] S207, scaling the data range of the phase difference spectrum to [0, 1] using the following formula:

[0057]

[0058] S208, convert the scaled sequence data into polar coordinates, regard the numerical value as the cosine value of the angle, and regard the time axis as the radius:

[0059]

[0060] S209, obtaining the angle difference between the corresponding angles at different time points according to the following formula:

[0061]

[0062] In the formula, Represents a one-dimensional time series.

[0063] From the above description, we can know that GAFs is used to convert the scaled one-dimensional sequence data from rectangular coordinates to polar coordinates, and then the time correlation of different time points is identified by calculating the angle difference between different points, and the time series is converted into a GAFs image for subsequent recognition and processing on the CNN model.

[0064] Furthermore, the CNN model in step S3 is specifically composed of an input layer, 1 layer of first convolution module, 3 layers of second convolution modules and an output module.

[0065] From the above description, we can see that the CNN model has a shared convolution kernel, which is stress-free for processing high-dimensional data and has good feature classification effect.

[0066] Furthermore, the convolution module is composed of a convolution layer-BatchNormal layer-ReLU activation function layer.

[0067] From the above description, we can know that the role of the convolution module is to automatically extract features. The BatchNormal layer can make the data distribution of each batch consistent and avoid the vanishing gradient. The unilateral inhibition property of the ReLU activation function is used to make the neurons in the neural network sparsely activated, especially in deep neural network models (such as CNN). When the model adds N layers, the activation rate of the ReLU neurons will theoretically decrease by 2 to the Nth power. Therefore, the convolution module composed of the convolution layer-BatchNormal layer-ReLU activation function layer can improve the gradient flowing through the network, allowing a larger learning rate and greatly improving the training speed.

[0068] Furthermore, the output module is composed of a linear layer and an activation function. When the output value of the activation function is 1, the corresponding ice thickness category is no ice; when the output value of the activation function is 2, the corresponding ice thickness category is ice thickness below a threshold; when the output value of the activation function is 3, the corresponding ice thickness category is ice thickness above a threshold.

[0069] From the above description, we can see that the size of the activation function output dimension is consistent with the category of ice thickness, and the value of each dimension represents the possibility of different categories.

[0070] Furthermore, the activation function is a softmax activation function.

[0071] From the above description, we can see that using the exponential form of the Softmax function can make the numerical distance between large differences larger. In deep learning, back propagation is usually used to solve the gradient and then gradient descent is used to update the parameters. The exponential function is more convenient when deriving.

[0072] Furthermore, in step S4, the user is prompted through the human-computer interaction interface.

[0073] From the above description, it can be seen that the human-computer interaction interface is adopted to facilitate users to obtain information and perform operations conveniently and intuitively.

[0074] An overhead power transmission line ice thickness classification early warning system, comprising:

[0075] A data acquisition module, used to collect data of overhead power transmission lines using a phase-sensitive optical time domain reflectometer, and classify the overhead power transmission lines;

[0076] A data analysis and preprocessing module is used to divide the ice data of different thicknesses in the overhead transmission line into frames, perform phase demodulation to obtain a phase difference spectrum, and generate a two-dimensional image using a Gram angle field;

[0077] A model training module, used for creating a CNN model, and using the two-dimensional image to train the CNN model;

[0078] The processing and early warning module is used to classify whether the overhead transmission line is covered with ice through the CNN model, and to perform risk assessment and early warning on the overhead transmission line covered with ice.

[0079] The above-mentioned method and system for classifying and warning ice thickness of overhead transmission lines of the present invention can solve the problems that the monitoring of ice coverage of overhead transmission lines is easily affected by the environment, the equipment has a short service life, the data transmission reliability is poor, and the monitoring cannot be accurate and comprehensive. The following is an explanation through specific implementation methods:

[0080] Embodiment 1

[0081] Please refer to Figure 1 The present invention discloses a classification and early warning method for ice thickness of overhead transmission lines, which specifically comprises the following steps:

[0082] Step 1: Prepare data

[0083] Use φ-OTDR to collect data on overhead transmission lines and classify the acquired data.

[0084] In this embodiment, the φ-OTDR based on heterodyne coherent detection has a carrier signal frequency of 200MHz emitted by its internal RF source, and a balanced photodetector detects the Rayleigh backscattered light (RBS) signal continuously generated during the propagation of the incident pulse light along the optical fiber. After the photoelectric conversion is completed, a band-limited intermediate frequency signal with a central frequency of 200MHz and a very narrow bandwidth is output, and the sampling frequency of the data acquisition card is 250MHz.

[0085] Deep learning models often require a large number of data samples. Neural networks can only be trained with a large number of samples.

[0086] Only after optimization can we find stable and reliable parameters and models. Here we divide the collected ice data of overhead transmission lines with different thicknesses into frames, demodulate them to obtain phase difference spectra, use GAFs to generate two-dimensional images, and filter and annotate the collected image data, including manual filtering and annotation.

[0087] The acquired data are classified into three categories: no ice, ice below the risk threshold, and ice exceeding the risk threshold. The threshold is set as the axial stress value less than the yield strength of the steel. The ice load F on the transmission tower node i is i for:

[0088]

[0089] Where n is the number of components; ρ is the ice density; h 2 is the coefficient of variation of ice diameter with height; l j is the length of a single component; g is the unit charge of ice-covered ground wire: g = 9.8 × 0.9πδ (d + δ) × 10 -3 , where d is the outer diameter of the conductor and ground wire, and δ is the ice thickness.

[0090] Since different wind speeds affect the axial stress, a model of wind speed and maximum ice thickness is established, and the risk threshold is dynamically adjusted according to the local weather forecast for the day. In this embodiment, when the wind speed is 20m / s, the ice thickness between 0 and 12mm is defined as not exceeding the risk threshold, and the ice thickness exceeding 12mm is defined as exceeding the risk threshold.

[0091] Step 2: Data preprocessing

[0092] The collected ice data of overhead transmission lines with different thicknesses are divided into frames, and their phases are demodulated to obtain phase difference spectra, and GAFs are used to generate two-dimensional images.

[0093] Among them, the phase demodulation is specifically:

[0094] S201, designing a bandpass filter to perform intermediate frequency filtering on the RBS original signal collected by the data acquisition card, suppressing broadband noise, and obtaining an intermediate frequency signal;

[0095] S202, performing IQ demodulation on the intermediate frequency signal, constructing a complex intermediate frequency, and extracting its amplitude A and phase Φ;

[0096] S203, performing sliding average of the amplitude A along the time line with a window of appropriate width;

[0097] S204, divide the n sampling points in each row in step S203 into phase-detection intervals of equal width, and find and save the column index of the position with the largest amplitude in all intervals in the result corresponding to S203;

[0098] S205, extract the phases in the two adjacent left and right intervals from Φ as the estimation of the RBS phase at the fiber position corresponding to the sampling points of the two intervals at the current moment, and calculate the phase difference. Trace back the phases at a pair of column indices of the two adjacent left and right intervals in the previous row with the same index as the current moment, and use them as the estimation of the RBS phase at the fiber position corresponding to the sampling points of the two intervals at the previous moment, and calculate the phase difference. Subtract the phase difference at the previous moment from the phase difference at the current moment, and calculate the change in the phase difference;

[0099] S206, repeat step S205, starting from the second row, the second interval, and calculate the change in phase difference between the left and right intervals of each interval within one detection pulse cycle; along the time axis, accumulate and sum the matrix, and untwist it to restore the change in optical fiber expansion and contraction between adjacent intervals caused by external vibration over time.

[0100] Use GAFs to generate a two-dimensional image, that is, convert the scaled one-dimensional sequence data from the rectangular coordinate system to the polar coordinate system through GAFs, and then identify the time correlation of different time points by considering the angle difference between different points. Specifically:

[0101] S207, scaling the phase difference spectrum data range to [0,1] using the following formula:

[0102]

[0103] S208, convert the scaled sequence data into polar coordinates, regard the numerical value as the cosine value of the angle, and regard the time axis as the radius. The specific formula is as follows:

[0104]

[0105] S209, obtaining the angle difference between the corresponding angles at different time points according to the following formula:

[0106]

[0107] In practical applications, the accuracy of network feature extraction may be low and the model robustness may be poor due to wind speed or other uncontrollable factors. Therefore, in order to solve this problem, the collected data set is expanded by enhancing sample diversity. The expanded data set is mainly obtained by adding random signal noise of different intensities to the converted GAFs image.

[0108] 1. Perform random data enhancement on the converted GAFs images to enrich the sample set, manually screen and label them with category labels, and store the storage paths of the labeled images in the corresponding folders to create a data set. Figures 3 to 5 The phase difference amplitude-frequency diagram and the corresponding GAFs image of different ice thicknesses under simulated conditions are shown.

[0109] 2. Shuffle the data set, randomly select 80% as the training data set, and the remaining 20% ​​as the test data set.

[0110] 3. Scale the input image to a 224*224 size image and standardize the image.

[0111] Step 3: Model building

[0112] A CNN model was created, which consists of an input layer, a 7×7 convolution module, three 3×3 convolution modules, and an output module. The stride of each convolution layer is 2. The convolution module consists of a convolution layer-BatchNormal layer-ReLU activation function layer. The output module consists of a linear layer and a softmax activation function. When the output value of the activation function is 1, the corresponding ice thickness category is no ice; when the output value of the activation function is 2, the corresponding ice thickness category is ice thickness below the threshold; when the output value of the activation function is 3, the corresponding ice thickness category is ice thickness above the threshold.

[0113] The present invention applies the Pytorch deep learning framework to the classification of ice thickness of overhead transmission lines, and the version is 1.10.0. Figure 2 The figure shows the structure diagram of the convolutional neural network model. Through continuous debugging, a convolutional neural network model suitable for the classification of ice thickness of overhead transmission lines was constructed.

[0114] The CNN model consists of an input layer, a 7×7 convolutional module, three 3×3 convolutional modules, and an output module.

[0115] The convolution layer uses the linear rectification function ReLU function as the activation function. When the input value is negative, the output is 0. When the input value is greater than 0, the output is the original value. The ReLU expression is:

[0116] f(x)=max(0,x).

[0117] The pooling layer uses the max-pooling method to take the maximum value of the feature points in the neighborhood. Max-pooling can reduce the deviation of the estimated mean caused by the error of the convolution layer parameters. The texture features of the three types of GAFs images identified by the present invention are relatively obvious, and more texture information can be retained by using max-pooling. After pooling, a local response normalization operation is performed to create a competition mechanism for the activity of local neurons, so that its corresponding relatively large value becomes relatively larger, and other neurons with smaller feedback are inhibited, thereby enhancing the generalization ability of the model.

[0118] Step 4: Model training

[0119] Training the CNN model using the two-dimensional images generated by GAFs;

[0120] 1. Randomly set the weight parameters of each layer in the network to any value close to 0 and initialize the hyperparameters;

[0121] 2. Set batch_size to obtain the train_batch (including train_data and train_label) of each batch of batch training sets in turn, send train_batch to the neural network, and perform gradient descent and iterative training with a learning rate of 0.0001;

[0122] 3. Calculate the actual output vector of the network;

[0123] 4. Compare the elements of the target vector with the elements in the output vector and calculate the output loss;

[0124] 5. Calculate the adjustment amount of each weight and the adjustment amount of the threshold in turn;

[0125] 6. Adjust weights and thresholds;

[0126] 7. After M iterations, the training is finished and the weights and thresholds are saved in the corresponding folders. At this time, it can be considered that the weights have reached stability and the classifier has been formed.

[0127] Based on the principle that the thicker the ice is, the smaller the natural frequency of each mode of the overhead wire is, and there is a monotonically decreasing mapping relationship between the natural frequency of each mode of the overhead wire and the ice thickness, the time series is converted into GAFs images, which are classified using the advantages of CNN in machine vision, and finally a mapping relationship between GAFs images and ice thickness is established.

[0128] Step 5: Model testing

[0129] The CNN model is used to classify whether an overhead transmission line is covered with ice, conduct a risk assessment on the overhead transmission line covered with ice, and prompt the user through a human-computer interaction interface.

[0130] The trained neural network model that meets the accuracy requirements needs to be tested using a test set to test its recognition accuracy. The specific steps are as follows:

[0131] 1. Send the test set sample data and labels into the model;

[0132] 2. Use the trained model parameters to obtain the output vector;

[0133] 3. Use the accuracy function to calculate the accuracy of the output vector.

[0134] Embodiment 2

[0135] Please refer to Figure 2 , an overhead transmission line ice thickness classification warning system, characterized by comprising:

[0136] A data acquisition module, used to collect data of overhead power transmission lines using a phase-sensitive optical time domain reflectometer, and classify the overhead power transmission lines;

[0137] A data analysis and preprocessing module is used to divide the ice data of different thicknesses in the overhead transmission line into frames, perform phase demodulation to obtain a phase difference spectrum, and generate a two-dimensional image using a Gram angle field;

[0138] A model training module, used for creating a CNN model, and using the two-dimensional image to train the CNN model;

[0139] The processing and early warning module is used to classify whether the overhead transmission line is covered with ice through the CNN model, and to perform risk assessment and early warning on the overhead transmission line covered with ice.

[0140] In summary, the present invention provides an overhead transmission line ice thickness classification and early warning method and system, which utilizes convolutional neural networks and distributed optical fiber sensors to establish a CNN model to classify whether the overhead transmission line is iced, and perform risk assessment and early warning, thereby solving the problems of overhead transmission line ice monitoring being easily affected by the environment, having a short service life, poor data transmission reliability, and being unable to accurately and comprehensively monitor.

[0141] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0142] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0143] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A classification and early warning method for ice thickness of overhead transmission lines, characterized in that: The following steps are involved: S1. Using a phase-sensitive optical time domain reflectometer to collect data from overhead transmission lines and classify the overhead transmission lines; S2, dividing the ice data of different thicknesses in the overhead transmission line into frames, performing phase demodulation to obtain a phase difference spectrum, and using a Gram angle field to generate a two-dimensional image; S3, creating a CNN model, and using the two-dimensional image to train the CNN model; S4. Classifying whether the overhead transmission line is covered with ice by using the CNN model, and conducting risk assessment and early warning for the overhead transmission line covered with ice; The phase demodulation in step S2 is specifically as follows: S201, performing intermediate frequency filtering on the original signal of the wireless base station collected by the data acquisition card to suppress broadband noise and obtain an intermediate frequency signal; S202, performing IQ demodulation on the intermediate frequency signal, constructing a complex intermediate frequency, and extracting the amplitude and phase of the complex intermediate frequency; S203, performing sliding average of the amplitude along the time line with a window of appropriate width; S204, dividing the n sampling points of each time row in step S203 into phase-detection intervals of equal width connected end to end, and finding and saving the column index of the position with the largest amplitude in all intervals corresponding to the time row in the result corresponding to S203; S205, extracting the phases in the two adjacent left and right intervals from the phase as the estimation of the RBS phase at the optical fiber position corresponding to the sampling points of the two left and right intervals at the current moment, calculating the first phase difference, tracing back the phases at a pair of column indices of the two adjacent left and right intervals in the previous row that are the same as the current moment, as the estimation of the RBS phase at the optical fiber position corresponding to the sampling points of the two intervals at the previous moment, calculating the second phase difference, subtracting the phase difference at the previous moment from the phase difference at the current moment, and obtaining the change in the phase difference; S206, repeat step S205, start traversing from the second row, the second interval, and calculate the change in the phase difference between the left and right intervals of each interval within one detection pulse cycle; Along the time axis, the matrix is ​​accumulated and summed, and then unwound to restore the time-varying optical fiber expansion and contraction between adjacent intervals caused by external vibration. The method of using the Gram angle field to generate a two-dimensional image in step S2 is specifically as follows: S207, scaling the data range of the phase difference spectrum to [0, 1] using the following formula: S208, convert the scaled sequence data into polar coordinates, regard the numerical value as the cosine value of the angle, and regard the time axis as the radius: S209, obtaining the angle difference between the corresponding angles at different time points according to the following formula: In the formula, Represents a one-dimensional time series.

2. The method for classifying and warning ice thickness of overhead power transmission lines according to claim 1 is characterized in that: The classification of the overhead transmission lines in step S1 is specifically as follows: The overhead transmission line is divided into three categories: no ice coverage, ice coverage below a risk threshold, and ice coverage exceeding a risk threshold, wherein the risk threshold is an axial stress value, and the axial stress value is less than the yield strength of the steel; The transmission tower node in the transmission line i Ice load on F i for: ， In the formula, n Indicates the number of components, ρ represents the ice density, h z represents the coefficient of variation of ice-covered diameter with height, l j Represents the length of a single component, g Indicates the ice-covered gravity of the ground wire; The ground wire ice gravity , the unit is electric charge, where, d Indicates the outer diameter of the conductor and ground wire. δ Indicates the thickness of ice cover.

3. The method for classifying and warning ice thickness of overhead power transmission lines according to claim 1 is characterized in that: The CNN model in step S3 is specifically composed of an input layer, a first convolutional module, three second convolutional modules and an output module.

4. The method for classifying and warning ice thickness of overhead power transmission lines according to claim 3 is characterized in that: The convolution module consists of a convolution layer-BatchNormal layer-ReLU activation function layer.

5. The method for classifying and warning ice thickness of overhead power transmission lines according to claim 3 is characterized in that: The output module is composed of a linear layer and an activation function. When the output value of the activation function is 1, the corresponding ice thickness category is no ice; when the output value of the activation function is 2, the corresponding ice thickness category is ice thickness below a threshold; when the output value of the activation function is 3, the corresponding ice thickness category is ice thickness above a threshold.

6. The method for classifying and warning ice thickness of overhead power transmission lines according to claim 5, characterized in that: The activation function is a softmax activation function.

7. The method for classifying and warning ice thickness of overhead power transmission lines according to claim 1 is characterized in that: In step S4, the user is prompted through the human-computer interaction interface.

8. An overhead power transmission line ice thickness classification warning system, characterized in that: include: A data acquisition module, used to collect data of overhead power transmission lines using a phase-sensitive optical time domain reflectometer, and classify the overhead power transmission lines; A data analysis and preprocessing module is used to divide the ice data of different thicknesses in the overhead transmission line into frames, perform phase demodulation to obtain a phase difference spectrum, and generate a two-dimensional image using a Gram angle field; A model training module, used for creating a CNN model, and using the two-dimensional image to train the CNN model; A processing and early warning module, used to classify whether the overhead power transmission line is covered with ice by using the CNN model, and to conduct risk assessment and early warning for the overhead power transmission line covered with ice; The phase demodulation is specifically as follows: S201, performing intermediate frequency filtering on the original signal of the wireless base station collected by the data acquisition card to suppress broadband noise and obtain an intermediate frequency signal; S202, performing IQ demodulation on the intermediate frequency signal, constructing a complex intermediate frequency, and extracting the amplitude and phase of the complex intermediate frequency; S203, performing sliding average of the amplitude along the time line with a window of appropriate width; S204, dividing the n sampling points of each time row in step S203 into phase-detection intervals of equal width connected end to end, and finding and saving the column index of the position with the largest amplitude in all intervals corresponding to the time row in the result corresponding to S203; S205, extracting the phases in the two adjacent left and right intervals from the phase as the estimation of the RBS phase at the optical fiber position corresponding to the sampling points of the two left and right intervals at the current moment, calculating the first phase difference, tracing back the phases at a pair of column indices of the two adjacent left and right intervals in the previous row that are the same as the current moment, as the estimation of the RBS phase at the optical fiber position corresponding to the sampling points of the two intervals at the previous moment, calculating the second phase difference, subtracting the phase difference at the previous moment from the phase difference at the current moment, and obtaining the change in the phase difference; S206, repeat step S205, start traversing from the second row, the second interval, and calculate the change in the phase difference between the left and right intervals of each interval within one detection pulse cycle; Along the time axis, the matrix is ​​accumulated and summed, and then unwound to restore the time-varying optical fiber expansion and contraction between adjacent intervals caused by external vibration. The method of using the Gram angle field to generate a two-dimensional image is specifically as follows: S207, scaling the data range of the phase difference spectrum to [0, 1] using the following formula: S208, convert the scaled sequence data into polar coordinates, regard the numerical value as the cosine value of the angle, and regard the time axis as the radius: S209, obtaining the angle difference between the corresponding angles at different time points according to the following formula: In the formula, Represents a one-dimensional time series.

Images