Power transmission line fault detection method and system based on unmanned aerial vehicle electromagnetic detection

Abstract

This application belongs to the field of power transmission line fault detection technology, specifically a method and system for power transmission line fault detection based on UAV electromagnetic detection. The method includes: establishing a power transmission line model; establishing a three-phase power transmission line spatial magnetic field calculation model between the power transmission line current and the spatial magnetic field; calculating theoretical magnetic field data of the space surrounding the power transmission line under different fault conditions at the UAV observation point; collecting actual magnetic field data of the power transmission line under normal operating conditions using the UAV, and correcting the theoretical magnetic field data using the actual magnetic field data; fusing the corrected theoretical magnetic field data and the actual magnetic field data into a training magnetic field dataset; constructing a deep learning network and training the deep learning network using the training magnetic field dataset; and using the trained deep learning network to identify the magnetic field data of the power transmission line collected by the UAV during actual flight, thereby obtaining the fault type of the power transmission line. This achieves accurate identification of multiple fault types in power transmission lines.

Description

Technical Field

[0001] This application belongs to the field of power transmission line fault detection technology, specifically a method and system for power transmission line fault detection based on UAV electromagnetic detection. Background Technology

[0002] Transmission lines are a crucial component of the power system. Improving the stability of the power system is beneficial for the continuous production of society and industry, ensuring stable economic operation, and enhancing the efficiency of society and the quality of life for residents. In actual operation, transmission lines are constantly exposed to the natural environment and are inevitably subject to various faults such as open circuits, short circuits, and grounding due to environmental factors, equipment aging, and external damage. Failure to detect fault types in a timely and accurate manner can lead to serious economic losses.

[0003] Existing fault detection methods for transmission lines mainly include impedance measurement-based methods, traveling wave analysis methods, time-frequency analysis methods, and artificial intelligence-based fault diagnosis methods. Among them, impedance measurement-based fault detection methods typically suffer from high equipment costs, complex construction, and insufficient operational safety; traveling wave or time-frequency analysis-based methods have high requirements for sampling frequency and synchronization accuracy, and their detection performance is limited in complex transmission networks or high-noise environments.

[0004] In recent years, the application of drone technology in power transmission line inspection has gradually increased. This mainly involves using visible light or infrared imaging equipment to photograph power transmission lines and combining this with image recognition methods to detect faults. However, this method is highly dependent on lighting conditions, weather conditions, and shooting angles, and it is difficult to distinguish between various fault types with high precision. Existing research rarely combines drone electromagnetic detection with intelligent identification of power transmission line faults, and there is still a lack of an economical, safe, and efficient method for detecting and identifying power transmission line faults that does not require contact with the power transmission lines, is independent of lighting conditions, and is not dependent on lighting conditions. Summary of the Invention

[0005] This application provides a method and system for detecting transmission line faults based on electromagnetic detection by unmanned aerial vehicles (UAVs), which solves the problem that current research on transmission line fault detection relies heavily on contact measurement or manual inspection, making it difficult to quickly and accurately identify the fault status of transmission lines.

[0006] The first aspect of this application provides a method for detecting power line faults based on electromagnetic detection by unmanned aerial vehicles (UAVs), including: Establish a transmission line model to obtain the three-phase current variation characteristics of the transmission line under normal conditions and various fault conditions; Establish a three-phase transmission line spatial magnetic field calculation model between transmission line current and spatial magnetic field; The three-phase current variation characteristics are combined with the structural parameters of the transmission line and substituted into the spatial magnetic field calculation model of the three-phase transmission line to calculate the theoretical magnetic field data of the space around the transmission line under different faults at the UAV observation point. The actual magnetic field data of power transmission lines under normal operating conditions are collected by drones, and the theoretical magnetic field data is corrected by the actual magnetic field data. The corrected theoretical magnetic field data and the actual magnetic field data are fused together to form a training magnetic field dataset; A deep learning network is constructed and trained using a training magnetic field dataset. The deep learning network is used to extract multi-scale local features and temporal features of the magnetic field data. A trained deep learning network is used to identify the magnetic field data of power transmission lines collected by drones during actual flight, thereby obtaining the fault type of the power transmission lines.

[0007] Furthermore, a model of the transmission line is built based on the structural parameters of the actual transmission line using a simulation platform, and a fault simulation module is set in the simulation platform to simulate the operation status of the transmission line under different fault types, different fault locations, and different line current magnitudes; the structural parameters of the transmission line include tower type, conductor arrangement, phase-to-phase distance, conductor type, height above ground, and total line length; By running the fault simulation module, the characteristic data of three-phase current variation of the transmission line under different fault types, different fault locations, and different combinations of line current magnitudes are obtained.

[0008] Furthermore, the spatial magnetic field calculation model of the three-phase transmission line takes the three-phase current variation characteristics and structural parameters of the transmission line as input variables, and the horizontal and vertical components of the magnetic induction intensity at the coordinates of the observation point as output variables. The calculation process includes: for any observation point in space, calculating the magnetic induction intensity generated by each phase conductor in the transmission line at that observation point. The magnetic induction intensity generated by the three-phase conductors is vector-superimposed in the horizontal and vertical directions to obtain the horizontal component and vertical component of the total magnetic induction intensity at the observation point.

[0009] Furthermore, the theoretical magnetic field data is corrected using actual magnetic field data, including: Calculate the deviation between the measured magnetic field data of the UAV under normal conditions and the theoretical magnetic field data under normal conditions; Based on the distribution pattern of the deviation, the type of deviation is determined, including systematic deviation and random deviation; To address the aforementioned system deviation, a correction function is established to correct the theoretical magnetic field data; To address the aforementioned random deviations, a filtering algorithm is employed to smooth the theoretical magnetic field data.

[0010] Furthermore, the correction function includes one or more of the following methods: Deviation compensation method: Calculate the average deviation between the measured magnetic field data under normal conditions and the theoretical magnetic field data under normal conditions at the same observation point, and add the average deviation as a compensation amount to the theoretical magnetic field data; Parameter calibration method: The structural parameters of the transmission line are used as the variables to be calibrated. The goal is to minimize the deviation between the measured magnetic field data under normal conditions and the theoretical magnetic field data under normal conditions. The structural parameters of the transmission line are corrected by inversion through optimization algorithm. Environmental compensation method: Collect environmental magnetic field data through auxiliary sensors carried by UAVs, establish an environmental magnetic field background model, and deduct the environmental background magnetic field from the theoretical magnetic field data or add it as an independent compensation item.

[0011] Furthermore, the deep learning network includes an input layer, a multi-scale convolutional feature extraction layer, a bidirectional long short-term memory network layer, and a classification output layer; The input layer is used to receive magnetic field sequence data, which is preprocessed training magnetic field data. The multi-scale convolutional feature extraction layer includes multiple parallel one-dimensional convolutional structures. Different one-dimensional convolutional structures use convolutional kernels of different sizes. The convolutional kernels slide along the flight path direction to extract local magnetic field features at different spatial scales. Small-sized convolutional kernels are used to capture abrupt changes in the magnetic field near the fault point, while large-sized convolutional kernels are used to capture the overall distribution trend of the magnetic field along the transmission line. The bidirectional long short-term memory network layer includes a forward long short-term memory network and a backward long short-term memory network, which are used to simultaneously learn the forward and backward temporal features of magnetic field sequence data; The classification output layer maps the feature vectors output by the bidirectional long short-term memory network layer to the fault category space through a fully connected layer, and outputs the probability values ​​corresponding to each fault type through a classification function.

[0012] Furthermore, the forward temporal features characterize the magnetic field change pattern from the flight start point to the current observation point, and the backward temporal features characterize the magnetic field change pattern from the current observation point to the flight end point. Through bidirectional learning, the temporal dependency of magnetic field changes before and after a transmission line fault is captured.

[0013] Furthermore, the magnetic field sequence data is constructed according to the observation points along the flight path of the UAV. The data for each observation point includes the horizontal component of the magnetic induction intensity, the vertical component of the magnetic induction intensity, and the corresponding spatial coordinate information, which includes the horizontal position coordinates and the vertical height coordinates.

[0014] A second aspect of this application provides a power transmission line fault detection system based on unmanned aerial vehicle (UAV) electromagnetic detection, comprising: a UAV flight platform; a magnetic field acquisition module mounted on the UAV flight platform for acquiring magnetic field data of the space surrounding the power transmission line along a preset flight path; a positioning module mounted on the UAV flight platform for acquiring the spatial coordinates of the observation point; a data preprocessing module for spatiotemporally synchronizing the magnetic field data with the spatial coordinates and converting it into sequential data; and a fault identification module internally deploying a deep learning network for identifying the preprocessed sequential data and outputting the fault type of the power transmission line.

[0015] Compared with existing technologies, this application offers the following advantages: Addressing the simultaneous existence of spatial and temporal characteristics in the magnetic field response signals of transmission lines, this application designs an end-to-end deep learning network composed of a multi-scale convolutional neural network and a bidirectional long short-term memory network. This solves the problem that current research on transmission line fault detection often relies on contact measurements or manual inspections, making it difficult to achieve rapid and accurate identification of transmission line fault states. This method enables accurate identification of multiple fault types in transmission lines. It is less dependent on lighting conditions, improving the safety of fault detection without requiring contact with the transmission line, and significantly improving the accuracy of identifying similar fault types under complex operating conditions such as different current magnitudes and observation heights. It is also more suitable for fault detection in areas where traditional technicians cannot access, providing an efficient method for rapid diagnosis of transmission lines in complex environments. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for detecting transmission line faults based on electromagnetic detection by unmanned aerial vehicles (UAVs) according to an embodiment of this application. Figure 2 This is a schematic diagram of an end-to-end deep learning network architecture composed of a multi-scale convolutional neural network and a bidirectional long short-term memory network provided in the embodiments of this application. Figure 3 This is a block diagram of a power transmission line fault detection system based on UAV electromagnetic detection provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is 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.

[0018] See Figure 1 The flowchart shown illustrates a method for detecting transmission line faults based on electromagnetic detection by unmanned aerial vehicles (UAVs). This application provides a method for detecting transmission line faults based on electromagnetic detection by unmanned aerial vehicles (UAVs), comprising: S101, Establish a transmission line model and obtain the three-phase current variation characteristics of the transmission line under normal conditions and various fault conditions; S102, Establish a three-phase transmission line spatial magnetic field calculation model between transmission line current and spatial magnetic field; S103, combining the three-phase current variation characteristics with the structural parameters of the transmission line, is substituted into the spatial magnetic field calculation model of the three-phase transmission line to calculate the theoretical magnetic field data of the space around the transmission line under different faults at the UAV observation point. S104: Collects actual magnetic field data of power transmission lines under normal operating conditions using drones, and corrects theoretical magnetic field data using actual magnetic field data. S105, the corrected theoretical magnetic field data and the actual magnetic field data are fused into a training magnetic field dataset; S106, Construct a deep learning network and train the deep learning network using a training magnetic field dataset. The deep learning network is used to extract multi-scale local features and temporal features of the magnetic field data. S107 uses a trained deep learning network to identify the magnetic field data of power transmission lines collected during actual drone flights, thereby obtaining the fault type of the power transmission lines.

[0019] In step S101, a model of the transmission line is built based on the structural parameters of the actual transmission line using a simulation platform. A fault simulation module is set in the simulation platform to simulate the operation status of the transmission line under different fault types, different fault locations, and different line current magnitudes. The structural parameters of the transmission line include tower type, conductor arrangement, phase-to-phase distance, conductor type, height above ground, and total line length. By running the fault simulation module, the characteristic data of three-phase current variation of the transmission line under different fault types, different fault locations, and different combinations of line current magnitudes are obtained.

[0020] In one example, a transmission line model is established based on the structural parameters of an actual 220kV single-circuit transmission line on the same tower. The transmission line simulation model is built in the MATLAB-Simulink simulation platform, with the total length of the transmission line set to 32km. A three-phase circuit breaker module is used to simulate a circuit breaker fault. By changing the fault type, fault location, and transmission line current magnitude, the three-phase current variation characteristics of the transmission line under normal conditions and various fault conditions are obtained. Various fault conditions include, but are not limited to, single-phase circuit breaker, two-phase circuit breaker, single-phase ground fault, two-phase ground fault, and two-phase phase-to-phase fault conditions. A single-phase circuit breaker refers to the disconnection of one phase conductor in a three-phase power system, causing an electrical interruption in that phase while the other two phases maintain normal operation, resulting in an asymmetrical operating state. Two-phase circuit break refers to a severe asymmetrical fault state in a three-phase power system where any two phase conductors break simultaneously, leaving only one phase maintaining normal operation. A single-phase ground fault refers to a fault in a three-phase power system where an abnormal electrical connection occurs between a conductor of one phase and the ground. A two-phase ground fault refers to a fault in a three-phase power system where two phase conductors simultaneously experience an abnormal electrical connection with the ground. A two-phase short circuit refers to a fault in a three-phase power system where an abnormal electrical connection occurs between two phase conductors.

[0021] The three-phase current variation characteristics include the currents of phases A, B, and C, and the current characteristics include amplitude, phase, and waveform variation characteristics. Specifically, the current amplitude includes instantaneous value, effective value, and peak value; the current phase includes the phase difference between each phase; and the waveform variation characteristics include waveform shape, distortion degree, and harmonic content. Under normal conditions, the three-phase current amplitudes are equal, the phases differ by 120 degrees, and the waveform is a sine wave. Under fault conditions, the amplitude, phase, and waveform of the three-phase currents change accordingly, with different fault types corresponding to different current variation patterns. These three-phase current variation characteristics serve as input parameters for the spatial magnetic field calculation model of the three-phase transmission line in step S102, used to calculate the magnetic field distribution around the transmission line under different fault conditions.

[0022] The spatial magnetic field calculation model of the three-phase transmission line established in step S102 takes the three-phase current variation characteristics and structural parameters of the transmission line as input variables, and the horizontal and vertical components of the magnetic induction intensity at the coordinates of the observation point as output variables. The calculation process includes: for any observation point in space, calculating the magnetic induction intensity generated by each phase conductor in the transmission line at that observation point. The magnetic induction intensity generated by the three-phase conductors is vector-superimposed in the horizontal and vertical directions to obtain the horizontal component and vertical component of the total magnetic induction intensity at the observation point.

[0023] In one example, the Biot-Savart law is introduced to establish a three-phase transmission line spatial magnetic field calculation model between the transmission line current and the spatial magnetic field. The magnetic field distribution around the transmission line under different faults is calculated. The Biot-Savart law is the fundamental law in electromagnetism describing the magnetic field generated by current. It is a method for calculating the magnetic induction intensity generated by a current-carrying conductor of arbitrary shape at a certain point in space.

[0024] For any observation point in space, calculate the magnetic flux density generated by each phase conductor in the transmission line at that observation point. Specifically, based on the Biot-Savart law, each phase conductor is considered as an infinitely long straight conductor. The magnitude of the magnetic flux density generated by any phase conductor at the observation point is directly proportional to the magnitude of the current in that phase conductor and inversely proportional to the distance from that phase conductor to the observation point. The direction of the magnetic flux density is perpendicular to the radial direction from the observation point to the conductor.

[0025] The magnetic field strength generated by each phase conductor is decomposed into horizontal and vertical components. The horizontal component is related to the vertical offset of the observation point relative to the conductor, and the vertical component is related to the horizontal offset of the observation point relative to the conductor.

[0026] The horizontal and vertical components of the magnetic flux density generated by the three-phase conductors are vector-superimposed to obtain the horizontal and vertical components of the total magnetic flux density at the observation point. The horizontal component of the total magnetic flux density is the sum of the horizontal components generated by the A, B, and C phase conductors, and the vertical component of the total magnetic flux density is the sum of the vertical components generated by the A, B, and C phase conductors.

[0027] By incorporating the three-phase current variation characteristics and the structural parameters of the transmission line into the above-mentioned spatial magnetic field calculation model for three-phase transmission lines, theoretical magnetic field data of the space surrounding the transmission line under different faults at the UAV observation point are calculated.

[0028] For a single-phase transmission line, neglecting line sag and treating it as an infinitely long conductor, assume the transmission line current flows along... Directional propagation, only at any position direction and Directional magnetic field, observation point of three-phase transmission line The magnetic field component at the observation point is the vector superposition of the magnetic fields of the three conductors, and can be calculated from this component. The formula for calculating the magnetic field strength is: , for Point magnetic induction intensity, for point Directional magnetic flux density for point Directional magnetic flux density The permeability of free space, For transmission line current, The straight-line distance from the traverse to the observation point. and They are respectively shaft and Unit vector along the axial direction, The height at which the conductor is suspended. Let be the radial unit vector; The abscissa of the phase conductor is 0, and the measurement point is... The horizontal and vertical components of the magnetic field strength are represented as follows: , , In the above formula, For observation point Directional magnetic flux density For observation point Directional magnetic field strength, coordinate axes Represents the horizontal position of the observation point. This represents the vertical height of the observation point. , , These represent the magnetic induction intensities generated by the A-phase, B-phase, and C-phase conductors at the observation point, respectively. The directional component, i.e., the horizontal component of the magnetic flux density. , , These represent the magnetic induction intensities generated by the A-phase, B-phase, and C-phase conductors at the observation point, respectively. The directional component, i.e., the perpendicular component of the magnetic flux density. , , These represent the conductor heights of phases A, B, and C, respectively. , These represent the distances of phase A and phase C conductors from the origin of the horizontal axis, respectively. The permeability of free space, , , These represent the currents in phases A, B, and C, respectively.

[0029] Substitute the three-phase current characteristics under different fault conditions into the above measurement points. The formulas for the horizontal and vertical components of magnetic flux density are given, and the characteristic form of the three-phase current is as follows: Combined with the structural parameters of the transmission line This allows for the calculation of the horizontal and vertical components of the magnetic flux density at the observation point under different fault conditions, enabling quantitative calculation of the magnetic field distribution and obtaining the magnetic field distribution around the transmission line under different fault conditions. Different fault types can be obtained by simply changing the three-phase current characteristics to obtain the corresponding magnetic field distribution.

[0030] The deviations in theoretical magnetic field data mainly stem from multiple factors, including model assumptions, parameter inputs, and the external environment.

[0031] Regarding model assumptions, the transmission line is treated as an infinitely long straight conductor, and the influence of conductor sag is ignored. This simplification introduces certain model biases. In reality, transmission lines have sag in the middle of the span, the conductors are not perfectly straight, and the towers themselves also have a certain impact on the magnetic field distribution. These factors are not fully considered in the transmission line model.

[0032] Regarding the structural parameters of power transmission circuits, the actual height and horizontal position of the conductors above ground in transmission lines may deviate from the design values. Such deviations in structural parameters directly affect the accuracy of magnetic field calculations. Furthermore, measurement deviations in parameters such as conductor type and phase-to-phase distance will also be reflected in the calculation results.

[0033] Regarding current parameters, the three-phase current variation characteristics obtained from the transmission line model may differ from the current under actual fault conditions, including deviations in current amplitude, phase, and waveform characteristics. These deviations will be directly reflected in the magnetic field calculation results.

[0034] Regarding the external environment, factors such as the geomagnetic background, the additional magnetic field generated by nearby energized equipment, and environmental electromagnetic interference are not considered in the transmission line model. These environmental magnetic fields will be superimposed on the magnetic field generated by the transmission line, causing deviations between theoretical calculations and actual measurements.

[0035] Regarding the measurement platform, the motors and electronic devices carried by the drone itself will generate certain magnetic interference. This magnetic field will be superimposed on the acquired signal, and the power transmission line model cannot include this interference.

[0036] In terms of sensors, factors such as the accuracy limitations of magnetic field sensors, temperature drift, and measurement noise can also introduce measurement deviations.

[0037] Based on the nature and variation pattern of the deviation, the deviations in theoretical magnetic field data can be divided into two main categories: systematic deviations and random deviations.

[0038] Systematic bias refers to a bias whose value and sign remain constant or change according to a certain pattern under certain conditions. The geomagnetic field background bias is a typical example of a systematic bias. The Earth's inherent magnetic field has a relatively stable spatial distribution; its magnitude varies with geographical location but remains constant over a certain measurement period. Structural parameter bias is also a systematic bias; the deviation in magnetic field calculation caused by the discrepancy between the actual height of the conductor and the design value is fixed. Magnetic interference from UAV platforms also exhibits a systematic bias. This interference is related to the UAV's flight attitude and motor operating status and shows certain regularities.

[0039] Random bias refers to deviations whose values ​​and signs do not follow a fixed pattern and change randomly. Environmental noise is a type of random bias; magnetic field fluctuations caused by weather changes, surrounding activities, and other factors do not follow a fixed pattern. Sensor noise is also a type of random bias; the electronic noise inside a sensor exhibits high-frequency random fluctuations. Positioning bias also has randomness; the accuracy limitations of the positioning module cause slight random fluctuations in the coordinates of the observation point.

[0040] Based on the above deviation analysis results, the theoretical magnetic field data needs to be corrected to improve the accuracy of the data.

[0041] In one embodiment, step S104 involves collecting actual magnetic field data of the power transmission line under normal operating conditions using a drone, and then correcting the theoretical magnetic field data using the actual magnetic field data. Correcting theoretical magnetic field data using actual magnetic field data includes: Calculate the deviation between the measured magnetic field data of the UAV under normal conditions and the theoretical magnetic field data under normal conditions; Based on the distribution pattern of the deviation, the type of deviation is determined, including systematic deviation and random deviation; To address the aforementioned system deviation, a correction function is established to correct the theoretical magnetic field data; To address the aforementioned random deviations, a filtering algorithm is employed to smooth the theoretical magnetic field data.

[0042] Specifically, systematic biases can be addressed using various correction methods. The correction functions employed include one or more of the following: Deviation compensation method: Calculate the average deviation between the measured magnetic field data under normal conditions and the theoretical magnetic field data under normal conditions at the same observation point, and add the average deviation as a compensation amount to the theoretical magnetic field data; Parameter calibration method: The structural parameters of the transmission line are used as the variables to be calibrated. The goal is to minimize the deviation between the measured magnetic field data under normal conditions and the theoretical magnetic field data under normal conditions. The structural parameters of the transmission line are corrected by inversion through optimization algorithm. Environmental compensation method: Collect environmental magnetic field data through auxiliary sensors carried by UAVs, establish an environmental magnetic field background model, and deduct the environmental background magnetic field from the theoretical magnetic field data or add it as an independent compensation item.

[0043] Among these methods, the environmental compensation method is suitable for correcting background geomagnetic field interference and interference from nearby equipment. It involves using an UAV equipped with auxiliary sensors to collect environmental magnetic field data, establishing an environmental magnetic field background model, and subtracting the background magnetic field from the theoretical magnetic field data to obtain the magnetic field generated by the transmission line itself. The parameter calibration method is suitable for correcting structural parameter deviations. It aims to minimize the deviation between the measured magnetic field data and the theoretical magnetic field data under normal conditions. It uses an optimization algorithm to invert and correct the structural parameters of the transmission line, including the ground height and horizontal position coordinates of each phase conductor. The deviation compensation method is suitable for correcting magnetic interference from the UAV platform. It calculates the average deviation between the measured magnetic field data and the theoretical magnetic field data under normal conditions at the same observation point, and then adds this average deviation as a compensation amount to the theoretical magnetic field data.

[0044] Random deviations can be addressed using filtering algorithms. For example, moving average filtering smooths high-frequency random fluctuations by averaging adjacent sampling points. Kalman filtering provides an optimal estimate of random noise based on the system state equation and observation equation. Wavelet denoising effectively eliminates random noise while preserving the main characteristics of the signal by decomposing the signal into different scales and thresholding the noise components.

[0045] After the above correction process, the deviation between the theoretical magnetic field data output and the measured data will be significantly reduced, making the theoretical magnetic field data consistent with the measured data in terms of statistical characteristics, thus providing high-quality data support for the training of subsequent deep learning models.

[0046] In one example, the drone's flight altitude and observation path are set within a 20-meter range above the transmission line to acquire actual magnetic field data under normal operating conditions at the corresponding location. Theoretical magnetic field data is calculated through steps S101 to S103. The transmission line model simulates magnetic field data under 20 current magnitudes, 17 observation altitudes, 3 fault locations, 15 fault types, and 1 normal state. The magnetic field data obtained from the theoretical calculation is used to construct a dataset of main fault samples. At the same time, actual magnetic field data under normal operating conditions of the transmission line is collected through actual drone flight to correct the theoretical magnetic field data. The data is then fused with the corrected theoretical magnetic field data to finally construct a training magnetic field dataset containing multiple fault types and normal states.

[0047] In step S106, see Figure 2 As shown, the deep learning network includes an input layer, a multi-scale convolutional feature extraction layer, a bidirectional long short-term memory network layer, and a classification output layer; The input layer receives magnetic field sequence data, which is preprocessed training magnetic field data. The magnetic field sequence data is constructed according to the observation points along the UAV's flight path. The data for each observation point includes the horizontal and vertical components of the magnetic induction intensity, as well as the corresponding spatial coordinates, including horizontal position coordinates and vertical height coordinates. The multi-scale convolutional feature extraction layer includes multiple parallel one-dimensional convolutional structures. Different one-dimensional convolutional structures use convolutional kernels of different sizes. The convolutional kernels slide along the flight path to extract local magnetic field features at different spatial scales. Small-sized convolutional kernels are used to capture abrupt changes in the magnetic field near the fault point, while large-sized convolutional kernels are used to capture the overall distribution trend of the magnetic field along the transmission line. The bidirectional long short-term memory network layer includes a forward long short-term memory network and a backward long short-term memory network, which are used to simultaneously learn the forward and backward temporal features of magnetic field sequence data; The classification output layer maps the feature vectors output by the bidirectional long short-term memory network layer to the fault category space through a fully connected layer, and outputs the probability values ​​corresponding to each fault type through a classification function.

[0048] In one example, the magnetic field sequence data is constructed in time series format as follows: The sequence input data, where This refers to the number of sampling points or observation points. The multi-scale convolutional feature extraction layer consists of multiple parallel one-dimensional convolutional structures. Different convolutional structures use convolutional kernels of different sizes to perform convolution operations on the magnetic field sequence data. In this example, the convolutional kernel sizes are 1×3 and 1×5, used to extract local magnetic field features within different receptive fields, and the convolution results are nonlinearly mapped using an activation function. The multi-scale feature sequence obtained after convolution is input to the bidirectional long short-term memory (LSTM) network layer. The bidirectional LSTM network layer includes two sub-network structures: a forward LSTM network and a backward LSTM network. These are used to simultaneously learn the forward and backward temporal features of the magnetic field sequence data, thereby capturing the time dependency of magnetic field changes during transmission line faults. The forward temporal features characterize the magnetic field change pattern from the flight start point to the current observation point, and the backward temporal features characterize the magnetic field change pattern from the current observation point to the flight end point. Through bidirectional learning, the temporal dependency of magnetic field changes before and after transmission line faults is captured. The feature vector output by the bidirectional long short-term memory network layer is input to the fully connected layer, and the probability value corresponding to each fault category is obtained through the classification function. The category with the highest probability is taken as the final fault identification result.

[0049] The internal calculations of the bidirectional long short-term memory network layer are as follows: , in, This is the hidden state from the previous moment. For the current input, forget gate This determines what information the cell node loses, among which... For the sigmoid function, For the weight of the forget gate, The forget gate bias ultimately determines how much information from the previous cell is retained; the input gate... This determines what information the cell node retains. For the input gate weights, The input gate bias determines the value that needs to be updated; the output gate... The information to be output is determined. For the final output, For the output gate weights, For output gate bias, For long-term memory, it allows information to be passed through long sequences without being easily lost.

[0050] During the training process of the deep learning model, the nonlinear relationship between the magnetic field sequence data and the fault type is continuously updated; the cross-entropy loss function is used as the training objective function, and the Adam optimization algorithm is used to optimize the network parameters; the model training learning rate is set to 0.001, the batch size is 64, and the maximum number of training rounds is 100. When the loss function converges, the trained deep learning model is obtained.

[0051] On the other hand, see Figure 3 As shown in the figure, this application provides a power transmission line fault detection system based on UAV electromagnetic detection, including: Unmanned aerial vehicle (UAV) flight platform; A magnetic field acquisition module, mounted on the UAV flight platform, is used to collect magnetic field data of the space surrounding the power transmission line along a preset flight path; The positioning module, mounted on the UAV flight platform, is used to acquire the spatial coordinates of the observation point; The data preprocessing module is used to synchronize the magnetic field data with the spatial coordinates in time and space, and convert them into magnetic field sequence data; The fault identification module has a deep learning network deployed inside it to identify magnetic field sequence data and output the fault type of the transmission line.

[0052] In one example, the UAV flight platform adopts a multi-rotor UAV structure, which has vertical take-off and landing capabilities and hovering stability. It can fly close to power transmission lines under complex terrain conditions. The flight altitude is adaptively adjusted according to the voltage level and electromagnetic field distribution characteristics of the power transmission lines to ensure that the magnetic field acquisition module is at the optimal detection distance. The UAV flight platform is equipped with a high-precision inertial measurement unit and a dual-frequency GPS / BeiDou combined navigation system to achieve centimeter-level positioning accuracy and attitude stability control, effectively suppressing the interference of flight jitter on magnetic field measurement.

[0053] The magnetic field acquisition module includes a three-axis fluxgate sensor array, a signal conditioning circuit, and an analog-to-digital converter unit. The three-axis fluxgate sensor array is arranged along three orthogonal directions of the UAV's body coordinate system to simultaneously measure the three-dimensional vector components of the spatial magnetic field. The signal conditioning circuit performs low-noise amplification and anti-aliasing filtering on the sensor output signal, and the analog-to-digital converter unit completes digital acquisition at a sampling rate of not less than 10kHz to ensure the temporal resolution of the magnetic field data.

[0054] The positioning module integrates satellite navigation signals and inertial measurement data. In environments such as canyons or forests where satellite signals are blocked, it maintains positioning continuity through short-term calculations using inertial navigation. At the same time, it is equipped with lidar or visual sensors to measure the relative distance to power transmission line conductors, providing auxiliary positioning information for the spatial calculation of magnetic field data.

[0055] The data preprocessing module is deployed on the UAV's airborne edge computing unit or ground station server and performs the following processing flow: First, based on the timestamp alignment algorithm, the magnetic field data and positioning coordinates are synchronized at the millisecond level; then, a coordinate transformation method is used to transform the magnetic field data in the body coordinate system to the geographic coordinate system or the local coordinate system of the route; subsequently, according to the preset spatial sampling interval, the non-uniformly distributed magnetic field data is interpolated and resampled to generate magnetic field sequence data with equal spatial intervals; finally, detrending processing and bandpass filtering are performed to suppress geomagnetic field background drift and high-frequency noise interference, and highlight fault characteristic signals.

[0056] The deep learning network deployed in the fault identification module includes an input layer, a multi-scale convolutional feature extraction layer, a bidirectional long short-term memory network layer, and a classification output layer. The input layer is preprocessed magnetic field sequence data, and the classification output layer uses a Softmax classifier to correspond to the normal state and various typical fault types. The network weight parameters are obtained through offline training and are fixed in the fault identification module, supporting online forward inference calculation. The identification results are output in the form of fault type codes and probability values.

[0057] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for detecting transmission line faults based on electromagnetic detection by unmanned aerial vehicles (UAVs), characterized in that, include: Establish a transmission line model to obtain the three-phase current variation characteristics of the transmission line under normal conditions and various fault conditions; A three-phase transmission line spatial magnetic field calculation model is established, which uses the three-phase current variation characteristics and structural parameters of the transmission line as input variables and the horizontal and vertical components of the magnetic induction intensity at the observation point coordinates as output variables. The calculation process includes: for any observation point in space, calculating the magnetic induction intensity generated by each phase conductor in the transmission line at the observation point. The magnetic induction intensity generated by the three-phase conductors is vector-superimposed in the horizontal and vertical directions to obtain the horizontal component and vertical component of the total magnetic induction intensity at the observation point. The three-phase current variation characteristics are combined with the structural parameters of the transmission line and substituted into the spatial magnetic field calculation model of the three-phase transmission line to calculate the theoretical magnetic field data of the space around the transmission line under different faults at the UAV observation point. The actual magnetic field data of power transmission lines under normal operating conditions are collected by drones, and the theoretical magnetic field data is corrected by the actual magnetic field data. The corrected theoretical magnetic field data and the actual magnetic field data are fused together to form a training magnetic field dataset, including: Calculate the deviation between the measured magnetic field data of the UAV under normal conditions and the theoretical magnetic field data under normal conditions; Based on the distribution pattern of the deviation, the type of deviation is determined, including systematic deviation and random deviation; To address the aforementioned system bias, a correction function is established to correct the theoretical magnetic field data. This correction function includes one or more of the following methods: Deviation compensation method: Calculate the average deviation between the measured magnetic field data under normal conditions and the theoretical magnetic field data under normal conditions at the same observation point, and add the average deviation as a compensation amount to the theoretical magnetic field data; Parameter calibration method: The structural parameters of the transmission line are used as the variables to be calibrated. The goal is to minimize the deviation between the measured magnetic field data under normal conditions and the theoretical magnetic field data under normal conditions. The structural parameters of the transmission line are corrected by inversion through optimization algorithm. Environmental compensation method: Collect environmental magnetic field data through auxiliary sensors carried by UAVs, establish an environmental magnetic field background model, and deduct the environmental background magnetic field from the theoretical magnetic field data or add it as an independent compensation item; To address the aforementioned random deviations, a filtering algorithm is used to smooth the theoretical magnetic field data. A deep learning network is constructed and trained using a training magnetic field dataset. The deep learning network is used to extract multi-scale local features and temporal features of magnetic field sequence data. The magnetic field sequence data is constructed according to the observation points along the flight path of the UAV. The data of each observation point includes the horizontal component of magnetic induction intensity, the vertical component of magnetic induction intensity and the corresponding spatial coordinate information at the observation point. The spatial coordinate information includes horizontal position coordinates and vertical height coordinates. A trained deep learning network is used to identify the magnetic field data of power transmission lines collected by drones during actual flight, thereby obtaining the fault type of the power transmission lines.

2. The method for detecting transmission line faults based on UAV electromagnetic detection according to claim 1, characterized in that, The transmission line model is built using a simulation platform based on the structural parameters of the actual transmission line. A fault simulation module is set up in the simulation platform to simulate the operation status of the transmission line under different fault types, different fault locations, and different line current magnitudes. The structural parameters of the transmission line include tower type, conductor arrangement, phase-to-phase distance, conductor type, height above ground, and total line length. By running the fault simulation module, the characteristic data of three-phase current variation of the transmission line under different fault types, different fault locations, and different combinations of line current magnitudes are obtained.

3. The method for detecting transmission line faults based on UAV electromagnetic detection according to claim 1, characterized in that, The deep learning network includes an input layer, a multi-scale convolutional feature extraction layer, a bidirectional long short-term memory network layer, and a classification output layer. The input layer is used to receive magnetic field sequence data, which is preprocessed training magnetic field data. The multi-scale convolutional feature extraction layer includes multiple parallel one-dimensional convolutional structures. Different one-dimensional convolutional structures use convolutional kernels of different sizes. The convolutional kernels slide along the flight path direction to extract local magnetic field features at different spatial scales. Small-sized convolutional kernels are used to capture abrupt changes in the magnetic field near the fault point, while large-sized convolutional kernels are used to capture the overall distribution trend of the magnetic field along the transmission line. The bidirectional long short-term memory network layer includes a forward long short-term memory network and a backward long short-term memory network, which are used to simultaneously learn the forward and backward temporal features of magnetic field sequence data; The classification output layer maps the feature vectors output by the bidirectional long short-term memory network layer to the fault category space through a fully connected layer, and outputs the probability values ​​corresponding to each fault type through a classification function.

4. The method for detecting transmission line faults based on UAV electromagnetic detection according to claim 3, characterized in that, The forward temporal features characterize the magnetic field change pattern from the flight start point to the current observation point, and the backward temporal features characterize the magnetic field change pattern from the current observation point to the flight end point. Through bidirectional learning, the temporal dependency of magnetic field changes before and after a transmission line fault is captured.

5. A transmission line fault detection system based on UAV electromagnetic detection, used to implement the transmission line fault detection method based on UAV electromagnetic detection as described in any one of claims 1-4, characterized in that, include: Unmanned aerial vehicle (UAV) flight platform; magnetic field acquisition module, mounted on the UAV flight platform, for collecting magnetic field data of the space surrounding the power transmission line along a preset flight path; The positioning module, mounted on the UAV flight platform, is used to acquire the spatial coordinates of the observation point; The data preprocessing module is used to synchronize the magnetic field data with the spatial coordinates in time and space, and convert them into sequential data; The fault identification module has a deep learning network deployed inside it to identify the fault type of the transmission line after preprocessing the sequence data.

Images