Ground wire positioning method and system based on deep residual network

Abstract

The invention relates to a grounding wire positioning method and system based on a deep residual network. The method comprises the steps of firstly collecting grounding wire signals and performing preprocessing to generate a data set, then extracting multi-scale features by using an improved Teager-Kaiser energy operator, outputting a time-frequency energy spectrogram, then constructing a positioning model containing a deep residual network, training the model by a multi-task learning framework and a loss function based on the time-frequency energy spectrogram, and finally obtaining a positioning model containing a deep residual network. And carrying out lightweight processing on the trained model, deploying the model to an edge terminal, collecting a grounding wire signal in real time, and inputting the grounding wire signal into the lightweight model of the edge terminal, thereby outputting the real-time space coordinate and position state of the grounding wire. Compared with the prior art, the method has the advantages of high positioning precision, strong anti-interference performance, good real-time performance and the like.

Description

Technical Field

[0001] The present invention relates to the field of power transmission technology, and in particular to a ground wire positioning method and system based on a deep residual network. Background Art

[0002] In power systems, transmission line grounding wires are critical protective equipment for maintenance personnel. Their proper installation and condition monitoring are directly related to operational safety. Traditional grounding wire location methods rely primarily on manual inspections or simple sensor monitoring, which suffer from low efficiency, poor real-time performance, and high missed detection rates. This is particularly true in high-voltage / ultra-high-voltage transmission scenarios, where lines are widely distributed and terrain is complex. Manual inspections are costly and difficult to fully cover. In recent years, signal processing-based location methods (such as wavelet transform and Hilbert-Huang transform) have been gradually adopted. However, due to strong electromagnetic interference and multi-source noise coupling, feature extraction accuracy is insufficient, resulting in a high false alarm rate. Meanwhile, deep learning technology has shown great potential in power equipment condition monitoring, but existing models (such as CNN and LSTM) have limited ability to learn features from weak fault signals and fail to fully incorporate the physical characteristics of transmission line grounding wires.

[0003] The existing transmission line grounding wire positioning technology has the following defects: First, traditional manual inspections rely on subjective judgment, making it difficult to achieve all-weather, full-coverage monitoring; second, signal processing methods based on wavelet transform or Hilbert-Huang transform are susceptible to noise pollution in strong electromagnetic interference environments, resulting in the submersion of weak features; third, the existing deep learning model has insufficient ability to extract time-frequency domain coupling features, and has not optimized the correlation between grounding wire vibration and current signals, resulting in low positioning accuracy and high false alarm rate, which makes it difficult to meet the real-time monitoring needs of complex power grid scenarios.

[0004] In summary, the current transmission line ground wire positioning technology has low accuracy, poor anti-interference ability, and cannot achieve real-time monitoring. Summary of the Invention

[0005] The purpose of the present invention is to provide a grounding wire positioning method and system based on deep residual network, so as to achieve accurate positioning and real-time monitoring of the grounding wire status, so as to improve the intelligent level of power grid operation and maintenance.

[0006] The purpose of the present invention can be achieved by the following technical solutions: According to one aspect of the present invention, a ground wire location method based on a deep residual network is provided, and the method steps include: S1, collecting ground wire signals and preprocessing them to generate a signal data set; S2, using the improved Teager-Kaiser energy operator to extract features from the signal data set and output a time-frequency energy spectrum; S3. Build a ground line location model including a deep residual network; S4. Based on the time-frequency energy spectrum, a multi-task learning framework and loss function are used to train the grounding wire location model, and the trained grounding wire location model is output; S5. Perform lightweight processing on the trained ground wire positioning model and deploy the lightweight model to the edge terminal; S6. Collect the ground wire signal in real time and input it into the lightweight model of the edge terminal, thereby outputting the real-time spatial coordinates and position status of the ground wire.

[0007] As a preferred technical solution, the ground wire signal in S1 includes a ground current signal and a mechanical vibration signal. The preprocessing includes: aligning the time of each signal and recording the time stamp information of each signal; and then filtering, denoising and standardizing each signal.

[0008] As a preferred technical solution, the specific process of S2 includes: S21. Using the improved Teager-Kaiser energy operator to capture the weak features in the grounding line signal and obtain the processed signal; S22, performing time-frequency analysis on the processed signal to generate an initial time-frequency energy spectrum; S23. Use dimensionality reduction technology to process the initial time-frequency energy spectrum to reduce the number of features while retaining preset important positioning information, thereby obtaining a time-frequency energy spectrum.

[0009] As a preferred technical solution, the improved Teager-Kaiser energy operator in S2 introduces an adaptive threshold filtering strategy and a multi-scale energy fusion strategy.

[0010] As a preferred technical solution, the specific formula of the adaptive threshold filtering strategy is: ; in, is the improved Teager-Kaiser energy operator; is the sampling value of the input signal at discrete time point n; is the result of median filtering the input signal; l The noise suppression coefficient is dynamically adjusted by the particle swarm optimization algorithm. The coefficient is automatically adapted according to the local signal-to-noise ratio of the signal. In the preset strong noise section, the l To enhance filtering, in the preset weak noise section, reduce l To preserve signal details.

[0011] As a preferred technical solution, the multi-scale energy fusion strategy includes a sliding window mechanism and a multi-band energy fusion mechanism. The sliding window length covers the typical time-frequency characteristics of a preset ground fault. By calculating the instantaneous energy envelope within the window, a time-varying energy trajectory is generated; and the energy trajectory is further decomposed into 0-5kHz low-frequency band and 5-10kHz high-frequency band components. After assigning weights to each component, they are weighted and fused to enhance the transient impact characteristics of weak grounding events.

[0012] As the preferred technical solution, the deep residual network in S3 uses an improved bottleneck structure as its core module, adopts 1×1 convolution to achieve channel compression and expansion, adopts 3×3 depthwise separable convolution to replace traditional convolution, and embeds a lightweight SE attention mechanism. The deep residual network adopts an 18-layer depth design as a whole, and alleviates the gradient vanishing problem through residual connections.

[0013] As an optimal technical solution, the multi-task learning framework in S4 has a shared feature extraction module at the bottom layer and a dual-branch structure at the top layer. The dual-branch structure includes a position regression branch and a state classification branch. The position regression branch outputs spatial coordinates through the fully connected layer, and the state classification branch predicts two types of states: normal position or abnormal position through the Softmax layer.

[0014] As a preferred technical solution, the Adam algorithm is used to train the grounding wire location model in S4, combined with the cosine annealing learning rate scheduling strategy, and the initial learning rate is set to 1×10 -4 , decaying to 0.9 times the original value every 10 epochs, and using MixUp for data enhancement during training, the input signal is mixed in the time and frequency domains to generate virtual samples to expand the boundary scene data; the loss function is a joint loss function, which is composed of the weighted Smooth L1 loss function responsible for the position regression task and the Focal Loss loss function responsible for the state classification task. The specific formula is: ; in, is the value of the joint loss function, which is used to measure the overall difference between the model prediction results and the true labels and guide the update of model parameters; is the loss weight coefficient of the position regression task; is the Smooth L1 loss function; is the true label of the position regression task; is the model’s predicted value for the position regression task; is the loss weight coefficient of the state classification task; is the Focal Loss loss function; The true label for the state classification task; is the model's predicted value for the state classification task.

[0015] According to another aspect of the present invention, a ground wire positioning system based on a deep residual network is provided, the system comprising a data processing module, a feature extraction module, a ground wire positioning model, a model training module and a model application module; The data processing module collects data and preprocesses the data to generate a data set; The feature extraction module uses the improved Teager-Kaiser energy operator to extract features from the data set and outputs a time-frequency energy spectrum; The ground line location model includes a deep residual network; The model training module trains the grounding wire location model based on the time-frequency energy spectrogram using a multi-task learning framework and loss function, and outputs the trained grounding wire location model. The model application module performs lightweight processing on the trained grounding wire positioning model and deploys the lightweight model to the edge terminal. It collects the grounding wire signal in real time and inputs it into the lightweight model of the edge terminal, thereby outputting the real-time spatial coordinates and position status of the grounding wire to achieve real-time positioning of the grounding wire.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. In the present invention, the method covers the entire process of signal acquisition, feature extraction, model construction and training, lightweight deployment and real-time application, forming a complete technical closed loop. The time-frequency domain feature extraction of weak grounding signals is enhanced by the improved Teager-Kaiser energy operator. By constructing a deep residual network architecture and training it using a multi-task learning framework, combined with edge terminal deployment, the real-time spatial coordinates and position status output of the grounding wire are realized. This method has high positioning accuracy and strong anti-interference ability, meeting the needs of transmission line operation and maintenance for rapid response, promoting the development of transmission line operation and maintenance towards intelligence and less manpower, and ensuring the safe and stable operation of the power grid.

[0017] 2. This invention uses improved operators to capture subtle features, time-frequency analysis to generate spectrograms, and dimensionality reduction techniques to streamline features. This layered processing preserves the core characteristics of ground wire signals (such as transient impulses and frequency mutations) while removing redundant information. Compared to directly using the original signal, the time-frequency energy spectrogram more intuitively reflects the signal's time-frequency characteristics, and dimensionality reduction further improves model computational efficiency.

[0018] 3. The present invention introduces the Teager-Kaiser energy operator with adaptive threshold filtering and multi-scale energy fusion strategy to improve the feature extraction capability of weak ground signals, suppress strong electromagnetic noise interference, and accurately capture weak signal features in strong interference environments; adaptive filtering dynamically balances noise reduction and detail retention, and multi-scale fusion covers features of different frequency bands, solving the problem of low sensitivity of traditional operators to weak signals and providing richer feature inputs for positioning models. The adaptive filtering strategy is dynamically adjusted by the particle swarm optimization algorithm. l The coefficient is used to adapt the filtering strength to the local signal-to-noise ratio of the signal. In strong noise segments, filtering is strengthened to suppress interference, while in weak noise segments, filtering is weakened to retain details. This avoids feature loss or residual noise caused by traditional fixed threshold filtering, significantly improving the signal-to-noise ratio and feature integrity of the preprocessed signal. In the multi-scale energy fusion strategy, the sliding window covers the typical time-frequency range of the fault, combined with weighted fusion of high and low frequency bands to highlight the transient impact characteristics of weak grounding events. Compared with single-scale analysis, multi-scale fusion can more comprehensively capture the time-frequency distribution of the fault signal and improve the correlation between the characteristics and the grounding line location.

[0019] 4. The deep residual network used in this invention reduces computational complexity through 1×1 convolution and 3×3 depthwise separable convolution, constructing a lightweight residual network. An attention mechanism is embedded to strengthen key feature channels, while the SE attention mechanism enhances key feature weights. 18 layers of depth and residual connections ensure feature expression. This lightweight architecture maintains high-precision positioning, addressing the computational complexity and deployment difficulties of traditional deep networks.

[0020] 5. In the multi-task learning framework of the present invention, a shared feature extraction module is combined with dual branches of position regression and state classification to synchronously output spatial coordinates and states. Compared to single-task models, the dual-branch design can acquire multidimensional information with a single inference, reducing the number of model calls, meeting the integrated multi-task requirements of operation and maintenance, and simultaneously achieving ground wire position regression and state classification. The model integrates energy and spatial features, and processes multi-source signals through a dual-branch network structure, significantly improving positioning accuracy and anti-interference capabilities, meeting the real-time monitoring requirements of complex power grid scenarios and enhancing monitoring robustness in complex environments.

[0021] 6. This invention uses the Adam algorithm combined with cosine annealing learning rate scheduling to achieve rapid convergence of model parameters. MixUp data augmentation expands boundary scene samples to alleviate sample imbalance. A joint loss function balances the weights of regression and classification tasks. This results in a trained model with enhanced generalization capabilities, maintaining stable positioning accuracy and state judgment accuracy even in complex transmission line environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1Schematic diagram of the steps of the ground wire positioning method based on deep residual network in the present invention; Figure 2 Schematic diagram of the steps of feature extraction and outputting time-frequency energy spectrum in the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] This solution aims to address the low accuracy and poor interference resistance of existing grounding wire location technologies by integrating an improved Teager-Kaiser energy operator with a deep residual network. It improves the feature extraction capabilities of weak grounding signals and suppresses strong electromagnetic noise interference; builds a lightweight deep learning model to achieve real-time and accurate positioning of grounding wire status; reduces false alarm rates and improves monitoring robustness in complex environments; and promotes intelligent and less-manned operation and maintenance of transmission lines to ensure safe and stable power grid operation.

[0025] Example 1 In this embodiment, a grounding wire positioning method based on a deep residual network is adopted, and an ITKEO operator with adaptive threshold filtering is designed to enhance the time-frequency domain feature extraction of weak grounding signals and suppress strong electromagnetic noise. A lightweight residual network is constructed, and an attention mechanism is embedded to enhance key feature channels, thereby simultaneously achieving grounding wire position regression and state classification. A space-time joint positioning model is proposed, which integrates ITKEO energy features and DRN spatial features, and processes multi-source signals (current harmonics and vibration) through a dual-stream network structure, significantly improving positioning accuracy and anti-interference capabilities to meet the real-time monitoring needs of complex power grid scenarios. The method steps are as follows: Figure 1 As shown, specifically including: S1, collecting ground wire signals and preprocessing them to generate a signal data set; S2, using the improved Teager-Kaiser energy operator to extract features from the signal data set and output a time-frequency energy spectrum; S3. Build a ground line location model including a deep residual network; S4. Based on the time-frequency energy spectrum, a multi-task learning framework and loss function are used to train the grounding wire location model, and the trained grounding wire location model is output; S5. Perform lightweight processing on the trained ground wire positioning model and deploy the lightweight model to the edge terminal; S6. Collect the ground wire signal in real time and input it into the lightweight model of the edge terminal, thereby outputting the real-time spatial coordinates and position status of the ground wire.

[0026] The ground wire signals in S1 include ground current and mechanical vibration signals. Preprocessing involves time-aligning the signals and recording their time stamps. Each signal is then filtered, de-noised, and normalized. Targeted ground current and mechanical vibration signals are collected. Time alignment, filtering, de-noising, and normalization eliminate signal time-scale deviations and noise interference, unifying the data scale. This preprocessed data better meets model input requirements, providing a high-quality foundation for subsequent feature extraction and model training, and reducing the impact of invalid data on positioning results.

[0027] The specific process of S2 is as follows Figure 2 Shown, including: S21. Using the improved Teager-Kaiser energy operator to capture the weak features in the grounding line signal and obtain the processed signal; S22, performing time-frequency analysis on the processed signal to generate an initial time-frequency energy spectrum; S23. Use dimensionality reduction technology to process the initial time-frequency energy spectrum to reduce the number of features while retaining preset important positioning information, thereby obtaining a time-frequency energy spectrum.

[0028] The improved Teager-Kaiser energy operator in S2 introduces an adaptive threshold filtering strategy and a multi-scale energy fusion strategy.

[0029] The specific formula of the adaptive threshold filtering strategy is: ; in, is the improved Teager-Kaiser energy operator; is the sampling value of the input signal at discrete time point n; is the result of median filtering the input signal; l The noise suppression coefficient is dynamically adjusted by the particle swarm optimization algorithm. The coefficient is automatically adapted according to the local signal-to-noise ratio of the signal. In the preset strong noise section, the l To enhance filtering, in the preset weak noise section, reduce l To preserve signal details.

[0030] The multi-scale energy fusion strategy includes a sliding window mechanism and a multi-band energy fusion mechanism. The sliding window length covers the typical time-frequency characteristics of a preset ground fault. By calculating the instantaneous energy envelope within the window, a time-varying energy trajectory is generated; and the energy trajectory is further decomposed into 0-5kHz low-frequency band and 5-10kHz high-frequency band components. After assigning weights to each component, they are weighted and fused to enhance the transient impact characteristics of weak grounding events.

[0031] The deep residual network in S3 uses an improved bottleneck structure as its core module, adopts 1×1 convolution to achieve channel compression and expansion, adopts 3×3 depthwise separable convolution to replace traditional convolution, and embeds a lightweight SE attention mechanism. The deep residual network adopts an 18-layer depth design as a whole, and alleviates the gradient vanishing problem through residual connections.

[0032] The multi-task learning framework in S4 has a shared feature extraction module at the bottom layer and a dual-branch structure at the top layer. The dual-branch structure includes a position regression branch and a state classification branch. The position regression branch outputs spatial coordinates through a fully connected layer, and the state classification branch predicts two types of states: normal position or abnormal position through a Softmax layer.

[0033] When training the grounding line location model in S4, the Adam algorithm is used, combined with the cosine annealing learning rate scheduling strategy, and the initial learning rate is set to 1×10 -4 , decaying to 0.9 times the original value every 10 epochs, and using MixUp for data enhancement during training, the input signal is mixed in the time and frequency domains to generate virtual samples to expand the boundary scene data; the joint loss function is composed of the weighted Smooth L1 loss function responsible for the position regression task and the Focal Loss loss function responsible for the state classification task. Its specific formula is: ; in, is the value of the joint loss function, which is used to measure the overall difference between the model prediction results and the true labels and guide the update of model parameters; is the loss weight coefficient of the position regression task; is the Smooth L1 loss function; is the true label of the position regression task; is the model’s predicted value for the position regression task; is the loss weight coefficient of the state classification task; is the Focal Loss loss function; The true label for the state classification task; is the model's predicted value for the state classification task.

[0034] In summary, this method covers the entire process from signal acquisition, feature extraction, model building and training, lightweight deployment, and real-time application, forming a complete technical closed loop. By using an improved Teager-Kaiser energy operator to enhance the time-frequency domain feature extraction of weak grounding signals, and by constructing a deep residual network architecture and training it using a multi-task learning framework, combined with edge terminal deployment, it can output the real-time spatial coordinates and position status of the grounding wire. This solves the problems of low accuracy and poor anti-interference of existing grounding wire positioning technology, meets the demand for rapid response in transmission line operation and maintenance, promotes the development of transmission line operation and maintenance in the direction of intelligent and less-manned operation, and ensures the safe and stable operation of the power grid.

[0035] Example 2 In this embodiment, a grounding wire positioning system based on a deep residual network is adopted, and the system includes a data processing module, a feature extraction module, a grounding wire positioning model, a model training module and a model application module; The data processing module collects data and preprocesses the data to generate a data set; The feature extraction module uses the improved Teager-Kaiser energy operator to extract features from the data set and outputs a time-frequency energy spectrum; The ground line location model includes a deep residual network; The model training module trains the grounding wire location model based on the time-frequency energy spectrogram using a multi-task learning framework and loss function, and outputs the trained grounding wire location model. The model application module performs lightweight processing on the trained grounding wire positioning model and deploys the lightweight model to the edge terminal. It collects the grounding wire signal in real time and inputs it into the lightweight model of the edge terminal, thereby outputting the real-time spatial coordinates and position status of the grounding wire to achieve real-time positioning of the grounding wire.

[0036] The specific implementation steps of this system are as follows: Step S1: Data acquisition and preprocessing Step S2: Improved Teager-Kaiser energy operator feature extraction Step S3: Deep residual network modeling and training Step S4: Space-time feature fusion and positioning output In step S1, the following steps are included: Step S11: Install high-precision current sensors and vibration sensors at key locations on the ground wire of the transmission line to collect ground current and mechanical vibration data in real time; Step S12: ensuring the time alignment of multi-sensor data through the time synchronization module and recording the time stamp information of the signal; Step S13: Filter, de-noise and standardize the collected signals to generate a high-quality data set for subsequent analysis.

[0037] In step S2, the following steps are included: Step S21: Propose an improved Teager-Kaiser energy operator to more accurately capture the weak features of the ground wire signal through adaptive filtering and multi-scale energy fusion; Step S22: performing time-frequency analysis on the processed signal to generate a time-frequency energy spectrogram to intuitively display the change of signal energy over time and frequency; Step S23: Use dimensionality reduction technology to reduce the number of features while retaining the most important information for the positioning task and improving computational efficiency.

[0038] In step S21, the following contents are included: To address the problems of the traditional Teager-Kaiser Energy Operator (TKEO) being susceptible to noise contamination and insufficiently capable of extracting weak ground signal features in strong electromagnetic interference environments, this paper proposes an improved ITKEO (Improved Teager-Kaiser Energy Operator). This operator significantly improves the accuracy and robustness of signal feature extraction by incorporating an adaptive threshold filtering mechanism and a multi-scale energy fusion strategy.

[0039] First, based on the classic TKEO formula, the present invention designs a dynamic adjustment term to suppress noise interference. The traditional TKEO operator formula is:

[0040] The ITKEO operator embeds the adaptive threshold filtering module on this basis, and the formula is optimized as follows:

[0041] in, The noise suppression coefficient is dynamically adjusted by the particle swarm optimization algorithm (PSO). This coefficient automatically adapts according to the local signal-to-noise ratio (SNR) of the signal and increases in strong noise sections. To enhance filtering and reduce the noise level in weak noise range To preserve signal details.

[0042] Secondly, to capture the multi-scale characteristics of ground line vibration and current signals, the ITKEO operator employs a sliding window mechanism and a multi-band energy fusion strategy. The sliding window is set to 512 sampling points, covering the typical time-frequency characteristics of ground faults. By calculating the instantaneous energy envelope within the window, a time-varying energy trajectory is generated. The energy trajectory is further decomposed into low-frequency components (0-5 kHz) and high-frequency components (5-10 kHz). These components are assigned weights of 0.6 and 0.4, respectively, and then weighted fusion is performed to enhance the transient impact characteristics of weak ground events (such as virtual connections).

[0043] Step S3 includes the following steps: Step S31: A deep residual network is designed to reduce the number of parameters through the bottleneck structure and attention mechanism, thereby improving the inference speed while maintaining the positioning accuracy; Step S32: adopt a multi-task learning framework, combine position regression and state classification tasks, and achieve efficient model training through joint loss function and optimization algorithm.

[0044] Step S31 includes the following contents: To address the dual requirements of lightweight models and feature expression for the power transmission line ground wire location task, this paper designs a deep separable residual network architecture. This network utilizes an improved bottleneck structure as its core module, achieving channel compression and expansion through 1×1 convolutions. 3×3 depthwise separable convolutions replace traditional convolutions to reduce computational complexity. To enhance the network's sensitivity to key features, a lightweight Squeeze-and-Excitation (SE) attention mechanism is embedded in the residual module. This mechanism generates channel-level statistics through global average pooling, learns nonlinear mapping relationships through two fully connected layers, and ultimately generates channel weight coefficients in the range of 0-1 to dynamically weight the feature maps. The network adopts an 18-layer depth design, and residual connections are used to mitigate the vanishing gradient problem.

[0045] Step S32 includes the following contents: In order to achieve the coordinated optimization of accurate grounding wire position regression and intelligent state classification, the present invention constructs a multi-task learning framework and designs a joint loss function with dynamic weight adjustment. In view of the coupling characteristics of position regression and state classification in the grounding wire positioning task, the bottom layer of the network shares the feature extraction module, and the upper layer adopts a dual-branch structure to process the two types of tasks respectively. The position regression branch outputs the spatial coordinates through the fully connected layer, and the state classification branch outputs the spatial coordinates through the fully connected layer. The layer predicts two types of states: normal / abnormal.

[0046] The joint loss function consists of the Smooth L1 loss for position regression and the Focal Loss weighted for the classification task:

[0047] in, The design solves the sparseness problem of abnormal grounding line state samples through Focal Loss, and its modulation factor is Effectively suppress the weight of easy-to-classify samples, so that the model focuses on difficult examples such as virtual connections.

[0048] The network optimization adopts Adam algorithm, combined with cosine annealing learning rate scheduling strategy, and the initial learning rate is set to , decaying to 0.9 times its original value every 10 epochs. To enhance model generalization, MixUp data augmentation was introduced during training. This method mixes the input signal in the time and frequency domains to generate virtual samples to expand the boundary scene data, significantly outperforming the single-task learning baseline model.

[0049] Step S4 further includes the following steps: Step S41: Deploy the lightweight model to the edge terminal, adapt the hardware and design electromagnetic shielding to ensure detection accuracy under strong electric fields; Step S42: Verify the stability and accuracy of the system under extreme conditions through laboratory and field multi-scenario testing.

[0050] Step S41 also includes the following contents: To meet the real-time requirements of ground wire status monitoring for power transmission lines, an edge computing terminal deployment architecture based on the NVIDIA Jetson TX2 development board was designed. This platform features a 256-core Pascal-based GPU and a dual-core Denver 2 64-bit CPU, along with 8GB of LPDDR4 memory and 32GB of eMMC storage, providing 15W of low-power computing power. Using the TensorRT optimization engine, the lightweight residual network model is quantized using FP16 / INT8 hybrid quantization, reducing the model size to 12.7MB and increasing inference speed by 2.3 times, achieving a single-frame response of 12ms. To address complex electromagnetic environments, a two-layer electromagnetic protection design is employed: an outer aluminum alloy shielding housing, based on the Faraday cage principle, achieves -80dB electromagnetic shielding effectiveness at 1GHz; an inner three-stage filtering circuit, consisting of common-mode inductors, X / Y capacitors, and feed-through filters, implements differential / common-mode hybrid filtering for power and signal lines, ensuring 97.6% detection accuracy even in the presence of strong electric field interference of 10kV / m. The hardware interface uses a military-grade aviation plug, has an operating temperature range of -40°C to 85°C, and is IP67 certified. System integration utilizes Docker containerization to decouple model inference, data preprocessing, and communication protocols, supporting OTA remote updates.

[0051] Step S42 also includes the following contents: In order to fully verify the performance of the system, multi-dimensional testing and verification were carried out. In the laboratory stage, this solution constructed a test set of typical grounding faults such as false connection and broken line to simulate different fault characteristics, focusing on testing the model's recognition ability and response speed for weak signals; the field test selected ultra-high voltage DC and AC lines, and the equipment was operated for a long time under extreme temperature and humidity conditions to verify the hardware weather resistance and software stability; for the micro-meteorological areas in mountainous areas, severe climates such as freezing rain and strong winds were simulated to test the reliability of the equipment under complex stresses such as mechanical vibration and low-temperature condensation; at the same time, electromagnetic compatibility testing was carried out to ensure that the equipment maintains communication stability and data acquisition accuracy under strong electromagnetic interference. Comprehensive tests show that the system has achieved design goals in key indicators such as positioning accuracy, response speed and environmental adaptability. In particular, the fault recognition rate under complex working conditions has been significantly improved compared with traditional methods, providing reliable technical support for the intelligent operation and maintenance of transmission lines.

[0052] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A ground wire location method based on deep residual network, characterized in that: The method steps include: S1. Collect grounding wire signals and preprocess them to generate a signal data set; S2. Use the improved Teager-Kaiser energy operator to extract features from the signal data set and output a time-frequency energy spectrum; S3. Construct a grounding wire positioning model containing a deep residual network; S4. Based on the time-frequency energy spectrum, use a multi-task learning framework and loss function to train the grounding wire positioning model, and output the trained grounding wire positioning model; S5. Perform lightweight processing on the trained grounding wire positioning model and deploy the lightweight model to the edge terminal; S6. Collect grounding wire signals in real time and input them into the lightweight model of the edge terminal, thereby outputting the real-time spatial coordinates and position status of the grounding wire.

2. A ground wire location method based on deep residual network according to claim 1, characterized in that: The ground wire signal in S1 includes a ground current signal and a mechanical vibration signal. The preprocessing includes: aligning the time of each signal and recording the time stamp information of each signal; and then filtering, denoising and standardizing each signal.

3. The ground wire location method based on deep residual network according to claim 1 is characterized in that: The specific process of S2 includes: S21. Using the improved Teager-Kaiser energy operator to capture the weak features in the grounding line signal and obtain the processed signal; S22, performing time-frequency analysis on the processed signal to generate an initial time-frequency energy spectrum; S23. Use dimensionality reduction technology to process the initial time-frequency energy spectrum to reduce the number of features while retaining preset important positioning information, thereby obtaining a time-frequency energy spectrum.

4. The ground wire location method based on deep residual network according to claim 1 is characterized in that: The improved Teager-Kaiser energy operator in S2 introduces an adaptive threshold filtering strategy and a multi-scale energy fusion strategy.

5. A ground wire location method based on deep residual network according to claim 4, characterized in that: The specific formula of the adaptive threshold filtering strategy is: ； in, is the improved Teager-Kaiser energy operator; is the sampling value of the input signal at discrete time point n; is the result of median filtering the input signal; λ The noise suppression coefficient is dynamically adjusted by the particle swarm optimization algorithm. The coefficient is automatically adapted according to the local signal-to-noise ratio of the signal. In the preset strong noise section, the λ To enhance filtering, in the preset weak noise section, reduce λ To preserve signal details.

6. A ground wire location method based on deep residual network according to claim 4, characterized in that: The multi-scale energy fusion strategy includes a sliding window mechanism and a multi-band energy fusion mechanism. The sliding window length covers the typical time-frequency characteristic range of a preset ground fault. By calculating the instantaneous energy envelope within the window, a time-varying energy trajectory is generated. The energy trajectory is further decomposed into 0-5kHz low-frequency band and 5-10kHz high-frequency band components, which are weighted and then fused to enhance the transient impact characteristics of weak grounding events.

7. The ground wire location method based on deep residual network according to claim 1 is characterized in that: The deep residual network in S3 uses an improved bottleneck structure as its core module, adopts 1×1 convolution to achieve channel compression and expansion, adopts 3×3 depthwise separable convolution to replace traditional convolution, and embeds a lightweight SE attention mechanism. The deep residual network adopts an 18-layer depth design as a whole, and alleviates the gradient vanishing problem through residual connections.

8. The ground wire location method based on deep residual network according to claim 1, characterized in that: The multi-task learning framework in S4 has a shared feature extraction module at the bottom layer and a dual-branch structure at the top layer. The dual-branch structure includes a position regression branch and a state classification branch. The position regression branch outputs spatial coordinates through a fully connected layer, and the state classification branch predicts two types of states: normal position or abnormal position through a Softmax layer.

9. A ground wire location method based on deep residual network according to claim 8, characterized in that: When training the ground wire location model in S4, the Adam algorithm is used, combined with the cosine annealing learning rate scheduling strategy, and the initial learning rate is set to 1×10 -4 , decaying to 0.9 times the original value every 10 epochs, and using MixUp for data enhancement during training, the input signal is mixed in the time and frequency domains to generate virtual samples to expand the boundary scene data; the loss function is a joint loss function, which is composed of the weighted Smooth L1 loss function responsible for the position regression task and the Focal Loss loss function responsible for the state classification task. Its specific formula is: ； in, is the value of the joint loss function, which is used to measure the overall difference between the model prediction results and the true labels and guide the update of model parameters; is the loss weight coefficient of the position regression task; is the Smooth L1 loss function; is the true label of the position regression task; is the model’s predicted value for the position regression task; is the loss weight coefficient of the state classification task; is the Focal Loss loss function; The true label for the state classification task; is the model's predicted value for the state classification task.

10. A ground wire positioning system based on deep residual network, characterized in that: The system applies a grounding wire positioning method based on a deep residual network as described in any one of claims 1 to 9, and the system includes a data processing module, a feature extraction module, a grounding wire positioning model, a model training module and a model application module; The data processing module collects data and pre-processes the data to generate a data set; The feature extraction module uses the improved Teager-Kaiser energy operator to extract features from the data set and outputs a time-frequency energy spectrogram; The ground line location model includes a deep residual network; The model training module trains the grounding wire location model based on the time-frequency energy spectrogram using a multi-task learning framework and a loss function, and outputs the trained grounding wire location model; The model application module performs lightweight processing on the trained grounding wire positioning model and deploys the lightweight model to the edge terminal, collects the grounding wire signal in real time and inputs it into the lightweight model of the edge terminal, thereby outputting the real-time spatial coordinates and position status of the grounding wire to achieve real-time positioning of the grounding wire.

Images