Power transmission line galloping detection method based on environmental energy capture

Abstract

The invention discloses a power transmission line galloping detection method based on environmental energy capture, which belongs to the technical field of power transmission line monitoring, and comprises the following steps: synchronously acquiring mechanical energy induction electric signals and spatial attitude data in the galloping process of a power transmission line, performing multi-modal fusion processing to extract energy distribution characteristics and dynamic attitude characteristics, in combination with environment parameters obtained in real time, through calculation of a galloping analysis model, a galloping state classification result and a galloping amplitude prediction result are obtained, and the results are analyzed and output; the method does not need to depend on an external power supply or a high-cost sensor, remarkably reduces the equipment deployment and maintenance cost, has high environmental adaptability and stability, and can be widely applied to the field of intelligent power grids.

Description

Technical Field

[0001] This invention belongs to the field of power transmission line monitoring technology, specifically relating to a method for detecting power transmission line galloping based on triboelectric nanoenergy capture and six-axis attitude sensing. Background Technology

[0002] With the continuous development of dynamic monitoring technology for transmission lines, research on transmission line galloping has gradually become an important direction in the power system field. Transmission lines are prone to low-frequency, large-amplitude galloping under the influence of external environmental factors such as wind and icing, which may cause changes in conductor spacing or mechanical fatigue, leading to electrical faults and potential safety hazards in power grid operation. Therefore, developing efficient and accurate galloping detection methods is of great significance for improving the safety and stability of the power grid.

[0003] A search revealed a patent, CN118736427B, which describes a method and system for detecting transmission line galloping based on environmental energy capture, published on November 29, 2024. This patent acquires multiple frames of transmission line images and analyzes pixel offset information to determine whether galloping has occurred and calculate the amplitude of the galloping. However, this technical solution relies on image processing algorithms and has the following shortcomings: First, image acquisition equipment is susceptible to environmental factors such as lighting conditions and weather conditions, which may lead to unstable data quality; second, the image processing process requires high computing resources, and latency issues may occur, especially in scenarios with high real-time requirements; finally, this method mainly focuses on the spatial position changes of the transmission line and lacks comprehensive perception of the transmission line's attitude information (such as acceleration and angular velocity), making it difficult to accurately analyze complex galloping features.

[0004] Furthermore, a method and system for detecting icing and galloping of transmission and transformation lines based on fiber optic sensors, with publication number CN119803542B and publication date July 18, 2025, is available. This patent acquires meteorological parameters of the environment in which the transmission line is located and tension data of the suspension insulator strings using fiber optic sensors, and assesses the galloping risk by combining trend change coefficients and wind influence coefficients. However, this technical solution has the following limitations: First, the installation and maintenance costs of fiber optic sensors are high, and they are somewhat dependent on the physical structure of the transmission line, limiting their applicability in different environments; second, this method focuses on the analysis of external environmental parameters and fails to directly capture the mechanical energy changes and attitude information of the transmission line itself, which may lead to insufficient accuracy in galloping feature extraction; finally, fiber optic sensors have limited sensitivity to small vibrations, and the detection effect may not be ideal in low-frequency, large-amplitude galloping scenarios.

[0005] The aforementioned existing technologies indicate that current transmission line galloping detection methods still have certain shortcomings in terms of the comprehensiveness of data acquisition, environmental adaptability, real-time performance, and detection accuracy. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a transmission line galloping detection method based on triboelectric nanoenergy capture and six-axis attitude perception. This method converts the mechanical energy generated by the swaying of the transmission line into a signal, combines it with spatial attitude data, and uses a deep learning-based galloping analysis model to perform high-precision classification and amplitude prediction of the galloping state. Finally, it outputs the galloping type, amplitude, and potential risk level, providing decision support for power grid operation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for detecting galloping of transmission lines based on environmental energy capture includes the following steps:

[0009] S1: Synchronously acquire mechanical energy-induced electrical signals and spatial attitude data during the galloping process of transmission lines;

[0010] S2: Perform multimodal fusion processing on the collected signals and data to extract the energy distribution characteristics and dynamic attitude characteristics of the transmission line galloping;

[0011] S3: Combining the feature parameters of S2 with the real-time acquired environmental parameters, the dance analysis model is used to calculate and obtain the dance state classification results and dance amplitude prediction results.

[0012] S4: Based on the results of S3, generate and output the galloping type, galloping amplitude, and potential risk level of the transmission line.

[0013] In the above technical solution, the dancing analysis model is a parallel hybrid neural network architecture, including:

[0014] A convolutional neural network branch for extracting spatial correlations in fused features contains three one-dimensional convolutional layers, each followed by a batch normalization layer and a modified linear unit activation function layer.

[0015] A branch of the Long Short-Term Memory (LSTM) network for capturing the time-series dependencies of fusion features consists of two stacked LSM network layers, with the second LSM network layer followed by a randomly deactivated layer.

[0016] In the above technical solution, constructing a data input model includes: segmenting the dimensionality-reduced fused feature vector sequence according to the window length and sliding step size to form a dimensionless model. The input tensor is N, where N is the number of samples, T is the time step, and F is the dimension of the fused features.

[0017] In the above technical solution, a fusion and output model is constructed, which includes concatenating the outputs of the convolutional neural network branch and the outputs of the long short-term memory network branch in the feature dimension, and inputting the concatenated feature dimension into two independent fully connected output heads.

[0018] In the above technical solution, the fully connected output head includes:

[0019] A dance state classification head consists of a fully connected layer with 64 neurons and an output layer. The number of neurons in the output layer is equal to the preset number of dance types. An activation function is used to output the probability of each category.

[0020] The dancing amplitude prediction head consists of a fully connected layer with 32 neurons and an output layer. The output layer contains one neuron and uses a linear activation function to output the predicted peak-to-peak amplitude of the dancing motion.

[0021] In the above technical solution, a serial fusion strategy is adopted to concatenate the energy feature vector and the dynamic attitude feature vector to form a high-dimensional fused feature vector.

[0022] Principal component analysis is performed on the fused feature vectors, and dimensionality reduction is carried out. The principal components with a cumulative contribution rate of 95% are selected as the output fused features.

[0023] In the above technical solution, the energy feature vector is extracted from the energy distribution characteristics of the mechanical energy induced electrical signal, including the following steps:

[0024] A1: Calculate the short-time energy and short-time zero-crossing rate of the preprocessed mechanical energy induced electrical signal sequence, and extract its time-domain features;

[0025] A2: Perform a short-time Fourier transform on the preprocessed mechanical energy induced electrical signal sequence to obtain its time-frequency spectrum. Extract the spectral centroid, spectral spread, and spectral entropy from the time-frequency spectrum as frequency domain features.

[0026] A3: Wavelet packet decomposition is used to decompose the signal into four frequency bands, and the energy proportion and energy entropy of each frequency band are calculated.

[0027] A4: The instantaneous energy of the signal is calculated using a nonlinear operator, and the rapid changes in energy are characterized by a discrete form:

[0028]

[0029] in: It is a sequence of mechanically induced electrical signals. The instantaneous energy of the signal.

[0030] A5: Construct a feature vector from the features extracted from A1 to A3.

[0031] In the above technical solution, the dynamic attitude feature vector is extracted from the dynamic attitude features of spatial attitude data, including the following steps:

[0032] B1: Convert the real-time attitude quaternion into an Euler angle sequence;

[0033] B2: After removing gravity components and transforming the coordinate system from the preprocessed three-dimensional acceleration data, the displacement trajectory sequence of the monitoring points on the transmission line in three-dimensional space is reconstructed;

[0034] B3: Calculate the maximum amplitude, root mean square value, and ratio of the major and minor axes of the trajectory ellipse in the vertical, horizontal, and axial directions of the displacement trajectory sequence;

[0035] B4: Calculate the multi-order statistical values ​​of Euler angle sequences and angular velocity sequences;

[0036] B5: The results calculated from B1 to B4 form a dynamic attitude feature vector.

[0037] In the above technical solution, the synchronization signal and data include:

[0038] C1: The original mechanical energy induced electrical signal on the transmission line is input to the pre-charge for amplification with a gain of 40-60 dB. The amplified signal is then filtered with a passband frequency set to 0.1 Hz to 10 Hz. The filtered signal is then sampled after analog-to-digital conversion.

[0039] C2: Real-time state estimation and noise filtering are performed on the acquired raw data sequences of three-dimensional acceleration and three-dimensional angular velocity. The state vector of the filter is defined as follows:

[0040]

[0041] The filtered angular velocity data is then integrated using quaternion differential equations to update the attitude quaternions in real time.

[0042]

[0043] in: Represents the original pose quaternion.

[0044] For gyroscope bias estimation,

[0045] The filtered angular velocity direction, It is by The constructed antisymmetric matrix, Let be the filter state vector. For the updated pose quaternion;

[0046] C3: Using a timestamp mechanism based on a network time protocol, the mechanical energy sensing electrical signal sequence is aligned in time with the spatial attitude data sequence.

[0047] In the above technical solution, a triboelectric nanogenerator with a contact-separation mode structure is constructed to capture the mechanical energy generated by the galloping of transmission lines and convert it into an electrical signal induced by the mechanical energy, including:

[0048] A flexible substrate, a first electrode layer deposited on the substrate, a second electrode layer disposed opposite to the first electrode layer, a friction layer spin-coated on the first electrode layer and the second electrode layer respectively, and a spacer disposed between the friction layer on the first electrode and the friction layer on the second electrode.

[0049] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0050] This invention achieves direct capture of mechanical energy during the galloping of power transmission lines by introducing a triboelectric nanogenerator, avoiding the dependence of traditional sensors on external power sources and improving the sensitivity of minute vibration signals.

[0051] This invention combines the attitude sensing capability of a six-axis inertial measurement unit to comprehensively analyze the spatial motion characteristics of a conductor, breaking through the limitations of a single sensor in terms of data dimensions.

[0052] This invention does not rely on complex image processing or high-cost fiber optic sensors, significantly reducing equipment deployment and maintenance costs. It also has strong environmental adaptability and can work stably under conditions such as insufficient light and harsh weather.

[0053] The multimodal data fusion technology of this invention effectively enhances the ability to identify low-frequency, large-amplitude galloping characteristics, providing a more accurate and real-time solution for the safety monitoring of transmission lines, and can be widely applied in the field of smart grids. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0055] Figure 1 This is a schematic diagram of the overall process of this embodiment;

[0056] Figure 2 A schematic diagram of the data collection and synchronization process;

[0057] Figure 3This is a schematic diagram of the data preprocessing workflow;

[0058] Figure 4 This is a schematic diagram of the feature extraction process;

[0059] Figure 5 This is a comparison chart of the astronomical amplitude monitoring value after data fusion and the monitoring results based on motion sensors. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0061] Example 1

[0062] This implementation example Figure 1 As shown, the overall process of this embodiment includes the following key steps: energy harvesting from the triboelectric nanogenerator, attitude data acquisition from the six-axis inertial measurement unit, multimodal data fusion processing, construction of a deep learning-based dance analytical model, and analysis of the output results. In practical applications, these steps are completed through an integrated device, which includes a triboelectric nanogenerator module, a six-axis inertial measurement unit module, a data processing module, and a communication module.

[0063] The specific process is as follows:

[0064] S1: Using a monitoring device installed on the power transmission line, the mechanical energy induced electrical signal and spatial attitude data of the power transmission line galloping process are collected synchronously; the monitoring device integrates a triboelectric nanogenerator and a six-axis inertial measurement unit; the triboelectric nanogenerator is used to capture the mechanical energy generated by the galloping of the power transmission line and convert it into the mechanical energy induced electrical signal; the six-axis inertial measurement unit is used to collect the three-dimensional acceleration and three-dimensional angular velocity data of the power transmission line as the spatial attitude data.

[0065] S2: Perform multimodal fusion processing on the collected mechanical energy induced electrical signals and the spatial attitude data to extract the energy distribution characteristics and dynamic attitude characteristics of the transmission line galloping.

[0066] S3: The extracted energy distribution features and dynamic posture features, along with the real-time acquired environmental parameters, are input into a preset deep learning galloping analysis model. The model then performs calculations to obtain the classification results of the galloping state and the prediction results of the galloping amplitude of the transmission line.

[0067] S4: Based on the galloping state classification results and the galloping amplitude prediction results, generate and output the galloping type, galloping amplitude, and potential risk level of the transmission line.

[0068] In this embodiment, as shown Figure 2 As shown, the synchronous acquisition of mechanical energy-induced electrical signals and spatial attitude data during the galloping process of transmission lines includes the following steps:

[0069] S11: Construct a triboelectric nanogenerator, which adopts a contact-separation structural design, including a flexible polyethylene terephthalate (PET) substrate, an aluminum (Al) electrode layer deposited on the PET substrate, a polytetrafluoroethylene (PTFE) friction layer spin-coated on the Al electrode layer, and a second Al electrode / friction layer disposed opposite to the PTFE friction layer; a polydimethylsiloxane (PDMS) spacer with a thickness of 1-3 mm is disposed between the PTFE friction layer and the second Al electrode / friction layer to ensure a fixed contact gap.

[0070] The core material of the triboelectric nanogenerator consists of two layers with different materials at their contact surfaces: a flexible polymer film and a metal electrode. When the conductor vibrates, the mechanical vibration causes relative displacement between the two layers, inducing electrostatic induction and generating an electrical signal. To improve signal sensitivity, the triboelectric nanogenerator employs microstructure surface treatment technology, increasing the contact area and enhancing the coefficient of friction. Furthermore, the output voltage of the triboelectric nanogenerator is positively correlated with the amplitude and frequency of the conductor's vibration. Measuring the output voltage indirectly reflects the intensity of the conductor's vibration. In practical deployments, the triboelectric nanogenerator is installed at critical nodes on power transmission lines, such as areas with large spans or those susceptible to wind, to ensure comprehensive capture of the conductor's vibration characteristics.

[0071] S12: Preprocessing the original mechanical energy-induced electrical signal output from the triboelectric nanogenerator.

[0072] Signal amplification: The original mechanical energy induced electrical signal is input to a pre-charge amplifier for processing, and the gain of the charge amplifier is set to 40-60 dB.

[0073] Signal filtering: The amplified signal is passed through a fourth-order Butterworth bandpass filter with a passband frequency range of 0.1 Hz to 10 Hz to filter out DC bias and high-frequency noise interference.

[0074] Signal acquisition: A 24-bit analog-to-digital converter (ADC) is used to digitally sample the filtered signal, with the sampling frequency set to 100 Hz.

[0075] S13: Preprocess the raw spatial attitude data acquired by the six-axis inertial measurement unit.

[0076] The six-axis inertial measurement unit (IMU) is used to synchronously acquire three-dimensional acceleration and three-dimensional angular velocity data of a conductor, obtaining information on the conductor's attitude changes in space. This module integrates a three-axis accelerometer and a three-axis gyroscope, used to measure the acceleration and angular velocity components of the conductor in the X, Y, and Z directions, respectively. By integrating these data, the spatial displacement and rotation angle of the conductor can be further calculated, thus reconstructing the conductor's trajectory. Specifically, the displacement component is obtained after two integrations of the acceleration component, while the rotation angle component is obtained after one integration of the angular velocity component. These data not only reflect the instantaneous motion state of the conductor but also reveal its long-term dynamic behavior characteristics. In practical applications, the six-axis IMU module is usually integrated with a triboelectric nanogenerator module in the same housing to reduce external interference and improve data synchronization.

[0077] The processing procedure is as follows:

[0078] Sensor calibration: The six-axis inertial measurement unit is calibrated using the six-plane method before deployment to compensate for its inherent zero bias error, scaling factor error and inter-axis cross-coupling error.

[0079] Data filtering: For the acquired raw data sequences of three-dimensional acceleration and three-dimensional angular velocity, an extended Kalman filter (EKF) is applied for real-time state estimation and noise filtering. The state vector x(t) of the extended Kalman filter is defined as:

[0080]

[0081] in: Represents the original pose quaternion.

[0082] This is the bias estimate for the gyroscope.

[0083] Attitude calculation: Using filtered angular velocity data, attitude integration is performed through quaternion differential equations, and the attitude quaternion q(t) is updated in real time.

[0084]

[0085] in: The filtered angular velocity direction, It is by The resulting antisymmetric matrix.

[0086] Data synchronization: Using a timestamp mechanism based on the Network Time Protocol (NTP), the mechanical energy sensing electrical signal sequence and the spatial attitude data sequence are strictly aligned in time, and the time synchronization error is controlled within 1 millisecond.

[0087] This implementation example Figure 3As shown, the process of performing multimodal fusion processing and extracting energy distribution features and dynamic pose features includes the following steps:

[0088] S21: Extract the energy distribution characteristics of the mechanical energy induced electrical signal.

[0089] Temporal feature extraction: Extraction of preprocessed mechanical energy induced electrical signal sequences Calculate its short-time energy (STE) and short-time zero-crossing rate (ZCR).

[0090] Frequency domain feature extraction: Perform short-time Fourier transform (STFT) on the signal sequence x(n) to obtain its time-frequency spectrum S(k, m), and extract the spectral centroid, spectral spread and spectral entropy from the time-frequency spectrum as frequency domain features.

[0091] Time-frequency domain energy feature extraction: Wavelet packet decomposition (WPD) is used to decompose the signal into four frequency bands, namely 0.1-1Hz, 1-2Hz, 2-5Hz, and 5-10Hz; then the energy proportion and energy entropy of each frequency band are calculated.

[0092] Nonlinear dynamic feature extraction: The instantaneous energy of the signal is calculated using the Teager energy operator (TEO). To characterize rapid changes in energy, its discrete form calculation formula is as follows:

[0093]

[0094] in: It is a sequence of mechanically induced electrical signals.

[0095] The extracted short-time energy, zero-crossing rate, spectral centroid, spectral spread, spectral entropy, energy percentage of each frequency band, energy entropy of each frequency band, and instantaneous energy are used to construct the energy feature vector V. energy .

[0096] S22: Extract the dynamic attitude features of the spatial attitude data.

[0097] State parameter transformation: the obtained real-time attitude quaternion Convert to Euler angle sequence, namely roll angle, pitch angle and yaw angle.

[0098] Motion trajectory construction: After removing the gravity component and transforming the coordinate system of the preprocessed three-dimensional acceleration data, a second integration is performed to reconstruct the displacement trajectory sequence of the monitoring points on the transmission line in three-dimensional space.

[0099] Trajectory feature extraction: Calculate the maximum amplitude, root mean square value, and ratio of the major and minor axes of the trajectory ellipse in the vertical, horizontal, and axial directions of the displacement trajectory sequence.

[0100] Rotation feature extraction: Calculate the first and second statistical moments of the Euler angle sequence (roll, pitch, and yaw) and the angular velocity sequence, including mean, variance, kurtosis, and kurtosis.

[0101] The extracted trajectory amplitude, root mean square value, ellipse axis ratio, and statistical moments of attitude angle and angular velocity are used to form a dynamic attitude feature vector.

[0102] S23: Perform multimodal data fusion at the feature layer; adopt a serial fusion strategy to concatenate the energy feature vector and the dynamic attitude feature vector to form a high-dimensional fused feature vector.

[0103] Principal component analysis (PCA) is then performed on the fused feature vectors to reduce dimensionality, and principal components with a cumulative contribution rate of 95% are selected as the final fused features input to the deep learning model.

[0104] This implementation example Figure 4 As shown, the deep learning dance parsing model calculation for features includes the following process:

[0105] S41: Construct the deep learning dance parsing model, which employs a parallel hybrid neural network architecture comprising a convolutional neural network (CNN) branch and a long short-term memory (LSTM) branch.

[0106] CNN Branch Design: This branch is used to extract spatial correlations from fused features; its structure contains three one-dimensional convolutional layers (Conv1D), each followed by a batch normalization layer and a rectified linear unit (ReLU) activation function layer; the specific parameters are: the first convolutional layer has 64 kernels and a kernel size of 5; the second convolutional layer has 128 kernels and a kernel size of 3; the third convolutional layer has 256 kernels and a kernel size of 3; each convolutional layer is followed by a max pooling layer (MaxPooling1D) with a pooling size of 2.

[0107] LSTM branch design: This branch is used to capture the time-series dependencies of fused features; its structure contains two stacked LSTM layers, with 128 LSTM units in the first layer and 128 LSTM units in the second layer; both LSTM layers are set to return_sequences=True, and a Dropout layer is followed by the second layer with a dropout rate of 0.3.

[0108] S42: Construct the model input. The obtained dimensionality-reduced fused feature vector sequence is divided according to a window length of 20 seconds and a sliding step of 5 seconds to form an input tensor with dimensions [N, T, F], where N is the number of samples, T is the time step (calculated based on the sampling rate), and F is the fused feature dimension.

[0109] S43: Model fusion and output layer design, concatenate the outputs of the CNN branch and the LSTM branch along the feature dimension, and then input the concatenated features into two independent fully connected output heads.

[0110] The dancing state classification head consists of a fully connected layer with 64 neurons and an output layer. The number of neurons in the output layer is equal to the preset number of dancing types (dancing, breeze vibration, sub-segment oscillation, stillness), and the Softmax activation function is used to output the probability of each type.

[0111] The dancing amplitude prediction head consists of a fully connected layer with 32 neurons and an output layer; the output layer contains one neuron, which uses a linear activation function to directly output the predicted peak-to-peak amplitude of the dancing motion.

[0112] S44: Model training and optimization, using the multi-task learning loss function Ltotal for model training:

[0113]

[0114] Where Lclass is the cross-entropy loss function for classification tasks, Lreg is the mean squared error (MSE) loss function for regression tasks, α and β are weight coefficients, and α+β=1, initially set α=0.4, β=0.6; the Adam optimizer is used to iteratively update the model parameters, the initial learning rate is set to 0.001, and a learning rate decay strategy is adopted.

[0115] Based on this embodiment, in a practical application scenario, suppose a section of transmission line is located in a windy area, frequently subjected to strong winds that cause the conductors to gallop violently. In this case, following the implementation steps of this method, firstly, an integrated device is installed on this section of the line, including a triboelectric nanogenerator module and a six-axis inertial measurement unit module. Then, a data acquisition system is activated to record the vibration signals and attitude data of the conductors in real time, and transmits this data to a remote server for processing. After multimodal data fusion and deep learning model analysis, the system output shows that the galloping type of this section of the conductor is low-frequency, small-amplitude galloping, such as... Figure 5 As shown.

[0116] This embodiment achieves high-precision monitoring and prediction of transmission line galloping by utilizing the collaborative operation of a triboelectric nanogenerator and a six-axis inertial measurement unit, combined with multimodal data fusion technology and a deep learning model. Compared to traditional methods, this method does not rely on external power sources or high-cost sensors, significantly reducing equipment deployment and maintenance costs. It also possesses strong environmental adaptability and stability, making it widely applicable in the smart grid field.

[0117] Decision-making is adjusted by incorporating environmental parameters. Real-time environmental parameters, including wind speed, wind direction, temperature, and humidity, are concatenated with the fusion feature layer before the fully connected output head, so that the model can take into account the influence of environmental factors when making the final decision.

[0118] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting galloping of transmission lines based on environmental energy capture, characterized in that... Includes the following steps: S1: Synchronously acquire mechanical energy-induced electrical signals and spatial attitude data during the galloping process of transmission lines; S2: Perform multimodal fusion processing on the collected signals and data to extract the energy distribution characteristics and dynamic attitude characteristics of the transmission line galloping; S3: Combining the feature parameters of S2 with the real-time acquired environmental parameters, the dance analysis model is used to calculate and obtain the dance state classification results and dance amplitude prediction results. S4: Based on the results of S3, generate and output the galloping type, galloping amplitude, and potential risk level of the transmission line.

2. The transmission line galloping detection method based on environmental energy capture according to claim 1, characterized in that... The dancing analytical model is a parallel hybrid neural network architecture, including: A convolutional neural network branch for extracting spatial correlations in fused features contains three one-dimensional convolutional layers, each followed by a batch normalization layer and a modified linear unit activation function layer. A branch of the Long Short-Term Memory (LSTM) network for capturing the time-series dependencies of fusion features consists of two stacked LSM network layers, with the second LSM network layer followed by a randomly deactivated layer.

3. The transmission line galloping detection method based on environmental energy capture according to claim 1, characterized in that... Constructing a data input model includes: splitting the dimensionality-reduced fused feature vector sequence according to the window length and sliding step size to form a model with dimension . The input tensor is N, where N is the number of samples, T is the time step, and F is the dimension of the fused features.

4. The transmission line galloping detection method based on environmental energy capture according to claim 1, characterized in that... Construct a fusion and output model, which involves concatenating the outputs of the convolutional neural network branch and the long short-term memory network branch along the feature dimension, and inputting the concatenated feature dimension into two independent fully connected output heads.

5. The transmission line galloping detection method based on environmental energy capture according to claim 4, characterized in that... The fully connected output head includes: A dance state classification head consists of a fully connected layer with 64 neurons and an output layer. The number of neurons in the output layer is equal to the preset number of dance types. An activation function is used to output the probability of each category. The dancing amplitude prediction head consists of a fully connected layer with 32 neurons and an output layer. The output layer contains one neuron and uses a linear activation function to output the predicted peak-to-peak amplitude of the dancing motion.

6. A transmission line galloping detection method based on environmental energy capture according to any one of claims 1-5, characterized in that: A serial fusion strategy is adopted to concatenate the energy feature vector and the dynamic attitude feature vector to form a high-dimensional fused feature vector. Principal component analysis is performed on the fused feature vectors, and dimensionality reduction is carried out. The principal components with a cumulative contribution rate of 95% are selected as the output fused features.

7. The transmission line galloping detection method based on environmental energy capture according to claim 6, characterized in that... The energy feature vector is extracted from the energy distribution characteristics of the mechanically induced electrical signal, including the following steps: A1: Calculate the short-time energy and short-time zero-crossing rate of the preprocessed mechanical energy induced electrical signal sequence, and extract its time-domain features; A2: Perform a short-time Fourier transform on the preprocessed mechanical energy induced electrical signal sequence to obtain its time-frequency spectrum. Extract the spectral centroid, spectral spread, and spectral entropy from the time-frequency spectrum as frequency domain features. A3: Wavelet packet decomposition is used to decompose the signal into four frequency bands, and the energy proportion and energy entropy of each frequency band are calculated. A4: The instantaneous energy of the signal is calculated using a nonlinear operator, and the rapid changes in energy are characterized by a discrete form: ， in: It is a sequence of mechanically induced electrical signals. The instantaneous energy of the signal. A5: Construct a feature vector from the features extracted from A1 to A3.

8. The transmission line galloping detection method based on environmental energy capture according to claim 6, characterized in that... The dynamic attitude feature vector is extracted from the dynamic attitude features of spatial attitude data, including the following steps: B1: Convert the real-time attitude quaternion into an Euler angle sequence; B2: After removing gravity components and transforming the coordinate system from the preprocessed three-dimensional acceleration data, the displacement trajectory sequence of the monitoring points on the transmission line in three-dimensional space is reconstructed; B3: Calculate the maximum amplitude, root mean square value, and ratio of the major and minor axes of the trajectory ellipse in the vertical, horizontal, and axial directions of the displacement trajectory sequence; B4: Calculate the multi-order statistical values ​​of Euler angle sequences and angular velocity sequences; B5: The results calculated from B1 to SB4 form a dynamic attitude feature vector.

9. A transmission line galloping detection method based on environmental energy capture according to claims 6 or 7, characterized in that... Synchronization signals and data include: C1: The original mechanical energy induced electrical signal on the transmission line is input to the pre-charge for amplification with a gain of 40-60 dB. The amplified signal is then filtered with a passband frequency set to 0.1 Hz to 10 Hz. The filtered signal is then sampled after analog-to-digital conversion. C2: Real-time state estimation and noise filtering are performed on the acquired raw data sequences of three-dimensional acceleration and three-dimensional angular velocity. The state vector of the filter is defined as follows: ， The filtered angular velocity data is then integrated using quaternion differential equations to update the attitude quaternions in real time. ， in: Represents the original pose quaternion. For gyroscope bias estimation, The filtered angular velocity direction, It is by The constructed antisymmetric matrix, Let be the filter state vector. For the updated pose quaternion; C3: Using a timestamp mechanism based on a network time protocol, the mechanical energy sensing electrical signal sequence is aligned in time with the spatial attitude data sequence.

10. The transmission line galloping detection method based on environmental energy capture according to claim 1, characterized in that... A triboelectric nanogenerator with a contact-separation mode structure is constructed to capture the mechanical energy generated by the galloping of power transmission lines and convert it into an electrical signal induced by the mechanical energy, including: A flexible substrate, a first electrode layer deposited on the substrate, a second electrode layer disposed opposite to the first electrode layer, a friction layer spin-coated on the first electrode layer and the second electrode layer respectively, and a spacer disposed between the friction layer on the first electrode and the friction layer on the second electrode.

Images