Power grid digital twinning technology analysis method and system based on AI image recognition

Abstract

The invention relates to the technical field of power grid data processing, in particular to a power grid digital twinning technology analysis method and system based on AI image recognition, and the method comprises the following steps: collecting a visible light image and an ultraviolet image of an insulator string through a multi-spectral camera carried by an unmanned aerial vehicle, and obtaining line leakage current waveform data at the same time; generating a multi-scale pollution feature matrix; calculating the equivalent resistivity of the polluted conductive path; reconstructing surface electric field distribution of the digital twin based on the equivalent resistivity, and simulating field intensity distortion gradients under different humidity conditions; when the field intensity distortion gradient reaches a preset percentage of the air breakdown field intensity, a graded early warning signal is generated; according to the invention, the electric field distortion of the dirty layer under different humidity conditions is simulated to generate the humidity-field intensity mapping relation, so that whether the insulation performance of the electrical equipment is affected or not can be judged.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power grid data processing, in particular to a power grid digital twin technology analysis method and system based on AI image recognition. BACKGROUND

[0002] In a power grid device, an insulator is a key electrical component, and its performance directly affects the safety and stability of the power system. The main function of the insulator is to prevent current leakage and ensure normal transmission of electricity. However, the surface of the insulator is easily affected by environmental factors, especially the accumulation of dirt. Dirt such as dust, salt, and plant residues can adhere to the surface of the insulator. Over time, these dirt layers gradually thicken, increasing the conductivity of the insulator surface and affecting its electrical performance. In a humid environment, the conductivity of the dirt layer can be significantly exacerbated by increased moisture absorption. This change can lead to dangerous electrical breakdown and flashover.

[0003] Traditional insulator contamination state detection relies on regular patrols and manual visual inspection, which is limited by the frequency and accuracy of manual inspection, making it difficult to accurately grasp the real-time state of the insulator. In addition, although image analysis technology can be used to detect the appearance and thickness of the dirt layer in some cases, it often fails to fully consider the changes in the electrical performance of the dirt layer. Traditional image analysis techniques mainly focus on visible surface features, ignoring the dynamic impact of the dirt layer on the electrical properties of the insulator. SUMMARY

[0004] The application provides a power grid digital twin technology analysis method and system based on AI image recognition.

[0005] The power grid digital twin technology analysis method based on AI image recognition includes the following steps: S1: Collecting visible light images and ultraviolet images of the insulator string through a drone equipped with a multispectral camera, while obtaining line leakage current waveform data; S2: Constructing a dual-channel deep residual network to extract the contamination texture features of the visible light images and the discharge spot distribution features of the ultraviolet images, respectively, and generating a multi-scale contamination feature matrix; S3: Establishing a dynamic contamination-current coupling model to align the multi-scale contamination feature matrix with the leakage current harmonic components in the time domain, and calculating the equivalent resistivity of the contamination conduction path; S4: Reconstructing the surface electric field distribution of the digital twin of the insulator string based on the equivalent resistivity, and simulating the field intensity distortion gradient under different humidity conditions; S5: Generating a graded warning signal when the field intensity distortion gradient reaches a preset percentage of the air breakdown field intensity.

[0006] Optionally, the S1 specifically includes: S11: Integrate multi-spectral imaging equipment on the UAV, including a visible light camera (400-700 nm) and a solar blind ultraviolet camera (240-280 nm), to collect visible light images and ultraviolet images of the insulator string respectively. Set the resolution of the visible light image to 20 million pixels, and the resolution of the ultraviolet image to 1 million pixels; S12: Plan the UAV inspection path to ensure that each insulator is photographed from at least three different angles. The frame rate of the visible light and ultraviolet images is synchronized to 10 fps, and the spatial registration error is less than 0.5 pixels; S13: Install a wireless leakage current sensor at the steel cap of the insulator string. Set the sampling frequency to 10 kHz, and transmit the leakage current waveform data in real time to the UAV through the LoRa communication module.

[0007] Optionally, the S2 specifically includes: S21: Design a visible light channel network using an improved ResNet-34 architecture. Embed a spatial attention module after the third residual block to output a pollution texture feature map, which includes pollution thickness distribution (quantized value of 0-100 pm) and material classification probability (dust / salt crystallization / bird droppings); S22: Construct an ultraviolet channel network based on deformable convolution to build a U-Net structure. Extract the distribution density, area ratio, and morphological complexity features of the discharge light spot through a light spot detection layer to generate a discharge activity index matrix; S23: In the feature fusion stage, perform cross-modal alignment between the fourth residual block feature map (size HxWx512) of the visible light channel and the decoder output (HxWx256) of the ultraviolet channel. Use a bidirectional gating mechanism to realize feature interaction and generate a multi-scale pollution feature matrix; S24: Also includes extracting multi-scale features through a pyramid pooling layer to output feature descriptors including four scales (1x1, 2x2, 4x4, global mean), which constitute a multi-scale pollution feature matrix with a dimension of (HxWx1024).

[0008] Optionally, the S3 specifically includes: S31: Obtain the leakage current waveform data and process it using fast Fourier transform to extract each harmonic component in the signal. Extract the 3rd, 5th, and 7th harmonic components respectively; S32: Use a time synchronization module to time-align the multi-scale pollution feature matrix with the harmonic components. The search range is limited to ±10 milliseconds. This ensures accurate alignment within a short time window, thereby reducing the error caused by time synchronization; S33: Constructing a nonlinear relationship between the contamination feature and the harmonic current component, a nonlinear dynamic contamination-current coupling model is established by using the multi-scale contamination features extracted by the pyramid pooling network, the dynamic contamination-current coupling model is weighted and summed after the multi-scale contamination features, and the influence of contamination on current is modeled by combining the exponential function with the corresponding harmonic component, the weight parameter is dynamically learned according to the training process, the influence degree of each scale feature is adjusted, the harmonic attenuation coefficient is introduced to control the nonlinear influence of contamination feature on harmonic, and a noise term is added to consider the error and uncertainty in actual measurement; S34: The influence degree of contamination on current is evaluated by calculating the equivalent resistivity, the equivalent resistivity is obtained by squaring and summing the amplitudes of multiple harmonic components and combining with the effective value of phase voltage, and the value of the equivalent resistivity is used to judge the severity of the contamination on the surface of the insulator.

[0009] Optionally, the dynamic contamination-current coupling model is represented as: , wherein, n represents the harmonic order, represents the weighted sum of the pyramid-pooled feature vectors of the first scale, is a learnable weight parameter; is a harmonic attenuation coefficient, represents the amplitude of the nth harmonic current over time , is the noise term of the nth harmonic, which considers the possible errors or uncontrollable factors in the experiment.

[0010] Optionally, the equivalent resistivity of the contamination conductive path is : , wherein, is the geometric factor of the contamination layer, is the effective value of the phase voltage, is the amplitude of the harmonic component.

[0011] Optionally, S4 specifically includes: S41, constructing a three-dimensional finite element model and mapping the conductivity: a three-dimensional finite element model of the insulator is constructed according to the geometric parameters of the insulator, which is used to describe the spatial form of the insulator and provide a basis for electric field intensity calculation, and in the three-dimensional finite element model, the equivalent resistivity value is converted into the conductivity of the contamination layer, the conductivity is the inverse of the resistivity, which reflects the conductivity of the contamination layer material; ​​S42, introduce humidity correction factor and dynamically adjust conductivity: Considering the effect of humidity on the conductivity of the contamination layer, introduce humidity correction factor and dynamically adjust the conductivity of the contamination layer according to the relative humidity of the environment; S43, solving the electric field distribution using the Poisson equation: The electric field distribution on the insulator surface is solved using the Poisson equation. By solving this equation, the electric potential distribution is obtained, and then the electric field strength is calculated. Boundary conditions are set during the calculation, including applying a phase voltage at the high-voltage end and setting the ground potential to zero. Using this equation, the surface electric field distribution is calculated based on changes in conductivity, and changes in electric field distortion are reflected. S44, calculate the field intensity distortion gradient: calculate the field intensity distortion gradient based on the electric field distribution , field intensity distortion gradient It indicates the severity of the change in electric field intensity and reflects the local non-uniformity of the electric field. The calculation of the field intensity distortion gradient includes taking the partial derivative of the electric field component (change in the X, Y, and Z directions) and taking the maximum value to obtain the gradient of the field intensity distortion; S45, simulating the field intensity distortion gradient under different humidity conditions: The field intensity distortion gradient under different humidity conditions is simulated according to the humidity gradient. By simulating the field intensity distortion gradient under different humidity conditions, a mapping relationship surface between humidity and field intensity is generated to show the impact of humidity changes on field intensity distortion.

[0012] Optionally, the field intensity distortion gradient Expressed as: , represents the electric field strength, that is, the field strength, They are the three components of the electric field strength, representing the components of the electric field in the x-axis, y-axis, and z-axis directions respectively.

[0013] Optionally, the preset percentage in S5 is 45%, and the air breakdown field strength reference value is set to , calculate the critical threshold ; S52: Distort the field intensity gradient Map the spatial coordinates to the digital twin surface of the insulator string to generate a normalized risk coefficient matrix : ; S53: Divide warning levels according to the risk factor matrix.

[0014] The power grid digital twin technology analysis system based on AI image recognition is used to implement the above-mentioned power grid digital twin technology analysis method based on AI image recognition, and includes the following modules: Image acquisition and data fusion: Using a multispectral camera mounted on a drone, multi-dimensional image data, including visible light and ultraviolet images, is collected, and line leakage current waveform data is simultaneously acquired; Dirty feature extraction and analysis: using a dual-channel deep residual network, the multi-dimensional image data of the dirty texture feature and the discharge light spot distribution feature are extracted respectively, a multi-scale dirty feature matrix is generated to reflect the distribution and state of the dirty layer; Dynamic contamination current coupling model: by establishing a dynamic contamination-current coupling model, the multi-scale contamination features and the leakage current harmonic components are time-domain aligned, the equivalent resistivity of the contamination conduction path is calculated, and the electrical characteristics of the contamination layer are quantified; Digital twin reconstruction and field strength simulation: based on the equivalent resistivity, the digital twin of the insulator string is reconstructed, and the field strength distortion gradient under different humidity conditions is simulated; Hierarchical early warning mechanism: when the field strength distortion gradient reaches the preset percentage of the set air breakdown field strength, a hierarchical early warning signal is automatically generated.

[0015] The beneficial effects of the present application are: The present application, by comprehensively utilizing the harmonic analysis of leakage current and multi-scale feature extraction technology, combined with the environmental humidity correction factor, can dynamically evaluate the conductivity change of the contamination layer, break through the limitation of traditional single measurement technology, not only can real-time monitor the electrical characteristics of the contamination layer, but also can adjust the evaluation results according to the actual environmental conditions (such as humidity change), improve the accuracy and real-time of the contamination layer state evaluation, so as to provide more reliable data support for the operation safety of power system.

[0016] The present application, by using the finite element model-based electric field distribution solution, combined with the calculated contamination layer conductivity and the influence of environmental humidity, can realize the simulation of field strength distortion gradient, simulate the electric field distortion under different humidity conditions of the contamination layer, generate the humidity-field strength mapping relationship, and then judge whether the insulation performance of electrical equipment is affected, when the field strength distortion gradient reaches the set risk threshold, trigger the hierarchical early warning, effectively predict the potential flashover risk in advance, and improve the safety protection ability of power equipment. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0018] Fig. 1 The analysis method flowchart of the embodiment of the present application; Fig. 2 The field strength distortion gradient calculation and simulation schematic diagram of the embodiment of the present application. DETAILED DESCRIPTION

[0019] The application will be described in greater detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments below are the best, preferred embodiments, and other alternative embodiments can also be implemented by those skilled in the art for some known technologies; and the accompanying drawings are only used to more specifically describe the embodiments and are not intended to specifically limit the application.

[0020] It should be noted that in the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiment can include a specific feature, structure or property, but not necessarily every embodiment includes the specific feature, structure or property. In addition, when a specific feature, structure or property is described in combination with an embodiment, it should be within the knowledge of those skilled in the related art to realize such a feature, structure or property in combination with other embodiments (whether or not explicitly described).

[0021] Generally, the terms can be understood at least in part from the context in which they are used. For example, depending on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular or can be used to describe combinations of features, structures, or characteristics, whether large or small, whether used alone or in combination. In addition, the term "based on" can be understood as not necessarily intending to convey a set of exclusive factors, but can instead, depending at least in part on the context, allow the presence of other factors not necessarily explicitly described.

[0022] As shown in Figs. 1-2 The AI image recognition-based power grid digital twin technology analysis method includes the following steps: S1: Collecting visible light images and ultraviolet images of insulator strings by unmanned aerial vehicles carrying multispectral cameras, and simultaneously acquiring line leakage current waveform data; S2: Constructing a dual-channel deep residual network to extract the contamination texture features of the visible light images and the discharge spot distribution features of the ultraviolet images, respectively, and generating a multi-scale contamination feature matrix; S3: Establishing a dynamic contamination-current coupling model, time-domain aligning the multi-scale contamination feature matrix with the leakage current harmonic components, and calculating the equivalent resistivity of the contamination conduction path; S4: Reconstructing the surface electric field distribution of the digital twin of the insulator string based on the equivalent resistivity, and simulating the field intensity distortion gradient under different humidity conditions; S5: When the field intensity distortion gradient reaches a preset percentage of the air breakdown field intensity, a graded warning signal is generated.

[0023] S1 specifically includes: S11: Integrate multi-spectral imaging equipment on the UAV, including a visible light camera (400-700 nm) and a solar blind ultraviolet camera (240-280 nm), to capture visible light images and ultraviolet images of insulator strings respectively. Set the visible light image resolution to 20 million pixels and the ultraviolet image resolution to 1 million pixels; S12: Plan the UAV inspection path to ensure that each insulator is photographed from at least three different angles. Synchronize the frame rate of visible light and ultraviolet images to 10 fps, and the spatial registration error is less than 0.5 pixels. S13: Install a wireless leakage current sensor at the steel cap of the insulator string, set the sampling frequency to 10 kHz, and transmit the leakage current waveform data to the UAV in real time through the LoRa communication module. Embed a time synchronization module in the UAV flight control system to ensure that the timestamps of multi-spectral images and leakage current waveforms are aligned. Collect multi-spectral images and leakage current waveform data and transmit them back to the ground station in real time through the 5G network to build a spatiotemporally aligned multi-modal data set.

[0024] The visible light camera uses a global shutter CMOS sensor with a resolution of 20 million pixels and a dynamic range of ≥70 dB, supporting HDR mode to handle strong light / backlight scenarios.

[0025] The solar blind ultraviolet camera uses an AlGaN focal plane array with a spectral response of 240-280 nm and a quantum efficiency of ≥25%, equipped with a narrowband filter to suppress background light interference.

[0026] Achieve dual-camera hardware synchronization through FPGA.

[0027] UAV inspection path: flying speed 2 m / s, maintaining a safe distance of 2-3 m from the insulator string.

[0028] Shooting angle: each insulator piece is photographed from 0° (straight ahead), 45° (oblique side), and 90° (straight side) to ensure full coverage.

[0029] The wireless leakage current sensor uses a Rogowski coil type current sensor with a range of 0-100 mA and an accuracy of ±0.5%.

[0030] Wireless transmission uses LoRa modulation with a bandwidth of 125 kHz, a transmission distance of ≥2 km, and a packet loss rate of <0.1%.

[0031] S2 specifically includes: S21: Design a visible light channel network using an improved ResNet-34 architecture, embed a spatial attention module after the third residual block, and output a pollution texture feature map containing pollution thickness distribution (quantized value 0-100 μm) and material classification probability (dust / salt crystallization / bird droppings). 1. The spatial attention module in the visible light channel design is calculated as follows: Input: The feature map output for the third residual block, where is the spatial size, is the number of channels, is the spatial height, is the spatial width; average pooling and maximum pooling : ; Concatenate the pooling results: , denotes concatenation; Convolution operation: where is a convolution kernel, outputting a single-channel feature map; Sigmoid activation: where, is the Sigmoid function, outputting a spatial attention weight matrix ; Output the weighted feature map: where denotes element-wise multiplication to apply the spatial attention weight matrix to the original input feature map, resulting in a weighted feature map; The output layer adopts a multi-task head design: Regression branch: predicts the thickness of contamination (L1 Loss); Classification branch: Softmax outputs the material category (cross-entropy loss); Contamination thickness regression branch: where is the predicted value of the thickness of contamination, and the loss function is L1Loss: is the true thickness label; Material classification branch: where, is the material classification probability (dust, salt crystals, birds), and the loss function is cross-entropy loss, is the true material label.

[0032] S22: Construct the ultraviolet channel network, based on deformable convolution to construct the U-Net structure, through the light spot detection layer to extract the distribution density, area proportion and morphological complexity features of the discharge light spot, and generate the discharge activity index matrix; The core structure design of the ultraviolet channel is: Encoder: 4 layers of deformable convolution (Deformable Conv), each with a stride of 2; Decoder: Transposed convolution + skip connection, restored to the original resolution; Spot detection layer: Morphology-inspired annular convolution kernel (radius 3 pixels, thickness 1 pixel), enhanced spot edge detection capability; The spot detection layer uses a morphology-inspired annular convolution kernel (radius 3 pixels, thickness 1 pixel) to enhance the spot edge detection capability, and the output matrix contains spot density, area ratio, and morphological parameters.

[0033] In deformable convolution calculation, the input , is the feature map of the ultraviolet image, and the deformable convolution is represented as: where is the output position coordinate, is the convolution kernel weight, is the conventional convolution kernel offset, is the learnable offset.

[0034] S23: In the feature fusion stage, the fourth residual block feature map of the visible light channel (size HxWx512) is aligned with the decoder output of the ultraviolet channel (HxWx256) across modalities, and a bidirectional gating mechanism is used to realize feature interaction, generating a multi-scale contamination feature matrix; The bidirectional gating mechanism for feature interaction is as follows: 1. Visible light gating calculation: where represents that the visible light feature and the ultraviolet feature are spliced together as input into convolution, then the Sigmoid activation function is used to obtain the gating weight ; 2. Ultraviolet gating calculation: Similarly, the ultraviolet feature and the visible light feature are spliced, passed through convolution and Sigmoid activation function, and the gating weight is obtained; 3. Feature fusion: Finally, the gating weights and are respectively multiplied by the visible light feature and the ultraviolet feature , and the results are added to obtain the final fusion feature ; S24: Also includes extracting multi-scale features through a pyramid pooling layer, outputting feature descriptors including four scales (1x1, 2x2, 4x4, global mean), forming a multi-scale contamination feature matrix with dimensions (HxWx1024).

[0035] 1x1 scale: Provides a global perspective for assessing the overall condition of the insulator string.

[0036] 2x2 scale: Provides a regional perspective for identifying areas of contamination accumulation and discharge hotspots.

[0037] 4x4 scale: Provides a local perspective for analyzing contamination details of individual insulator segments.

[0038] Global mean scale: Provides a statistical perspective for quantifying the statistical properties of overall contamination and discharge.

[0039] S3 specifically includes: S31: Obtain leakage current waveform data, which contains information about the changes in the time domain. Process the leakage current waveform data using fast Fourier transform to extract the harmonic components in the signal, including the 3rd, 5th, and 7th harmonics. These harmonic components represent different frequency components of the current waveform, which can reflect the degree of distortion of the current and reveal the impact of contamination characteristics on the current. S32: To effectively match the multi-scale contamination feature matrix and harmonic component data in time, we need to perform time domain alignment. Specifically, use the time synchronization module to align the multi-scale contamination feature matrix with the harmonic components within a range of ±10 milliseconds. This ensures accurate alignment within a short time window, reducing errors caused by time synchronization. S33: Establish a nonlinear relationship between contamination features and harmonic current components. Use the multi-scale contamination features extracted by the pyramid pooling network to establish a nonlinear dynamic contamination-current coupling model. After weighting and summing the multi-scale contamination features, combine them with the corresponding harmonic components using an exponential function to model the impact of contamination on current. The weight parameters are dynamically learned during the training process to adjust the influence of each scale feature. Introduce a harmonic attenuation coefficient to control the nonlinear impact of contamination features on harmonics, and add a noise term to consider errors and uncertainties in actual measurements. S34: Evaluate the impact of contamination on current by calculating the equivalent resistivity. The equivalent resistivity is obtained by squaring and summing the amplitudes of multiple harmonic components and combining them with the effective value of the phase voltage. The value of the equivalent resistivity is used to judge the severity of the contamination on the insulator surface. If the value is lower than the normal level, it may indicate that the conductivity of the contamination layer has increased, posing potential electrical risks.

[0040] Perform fast Fourier transform on the acquired leakage current waveform data to extract the amplitude of the 3rd, 5th and 7th harmonic components and total harmonic distortion THD; The dynamic pollution-current coupling model is expressed as: ,in, , represents the harmonic order, Represents the pyramid pooling The weighted sum of the scale eigenvectors, is the learnable weight parameter; is the harmonic attenuation coefficient, Indicates time On the Subharmonic current amplitude, For the The noise term of subharmonics takes into account possible errors or uncontrollable factors in the experiment; =3, that is, the third harmonic, Take 0.8, Take 1.1; =5, that is, the 5th harmonic, Take 0.6, Take 0.9; =7, that is, the 7th harmonic, Take 0.4, Take 0.7.

[0041] Equivalent resistivity of dirty conductive path : ,in, It is the pollution layer geometry factor, which reflects the influence of the pollution layer geometry on the equivalent resistivity. It is related to the insulator shed structure and links the macroscopic geometric characteristics of the pollution layer with the microscopic resistivity. The larger its value, the more significant the influence of the pollution layer on the electrical performance of the insulator. is the effective value of the phase voltage, is the amplitude of the harmonic component (the amplitude of the 3rd, 5th, and 7th harmonics respectively), The effective value of is calculated as follows: ,in, is the function of voltage changing with time, is the voltage cycle time, Represents small increments of time.

[0042] Comparison table of values ​​corresponding to insulator models: The higher the THD value, the more harmonic components in the leakage current, which means that the insulator surface is more severely contaminated or the partial discharge activity is more active. During the model training stage, sample data with higher contamination levels can be screened out based on the THD value to ensure the model's generalization ability under extreme working conditions.

[0043] S4 specifically includes: S41, Construct a 3D finite element model and map conductivity: A 3D finite element model of the insulator is constructed based on its geometric parameters. This model is used to describe the insulator's spatial morphology and provide a basis for calculating electric field strength. In the 3D finite element model, the equivalent resistivity value is converted into the conductivity of the contamination layer. Conductivity is the reciprocal of resistivity and reflects the conductive properties of the contamination layer material. This conversion process lays the physical foundation for subsequent simulation of the electric field distribution. The 3D finite element model is a discretized mathematical representation of the insulator and is used for numerical calculations (such as electric field distribution and stress analysis). S42, introduces a humidity correction factor and dynamically adjusts conductivity: Considering the effect of humidity on the conductivity of the contamination layer, a humidity correction factor is introduced to dynamically adjust the conductivity of the contamination layer according to the relative humidity of the environment. This correction factor can perform nonlinear corrections to the conductivity based on humidity changes, thereby simulating the conductivity changes of the contamination layer under different humidity conditions. This step reflects the significant impact of humidity on the electric field distribution on the insulator surface, making the model more realistic. S43, solving the electric field distribution using the Poisson equation: The electric field distribution on the insulator surface is solved using the Poisson equation. By solving this equation, the electric potential distribution is obtained, and then the electric field strength is calculated. Boundary conditions are set during the calculation, including applying a phase voltage at the high-voltage end and setting the ground potential to zero. Using this equation, the surface electric field distribution is calculated based on changes in conductivity, and changes in electric field distortion are reflected. S44, calculate the field intensity distortion gradient: calculate the field intensity distortion gradient based on the electric field distribution , field intensity distortion gradient It indicates the severity of the change in electric field strength, reflects the local non-uniformity of the electric field, and helps us understand the electric field strength distribution at different locations on the insulator surface, thus providing a basis for analyzing the insulation performance of the contamination layer. The specific calculation method is to calculate the field strength distortion gradient by taking the partial derivative of the electric field components (changes in the X, Y, and Z directions), taking the maximum value, and obtaining the gradient of the field strength distortion. S45, simulating the field intensity distortion gradient under different humidity conditions: The field intensity distortion gradient under different humidity conditions is simulated according to the humidity gradient. By simulating the field intensity distortion gradient under different humidity conditions, a mapping relationship surface between humidity and field intensity is generated to show the impact of humidity changes on field intensity distortion.

[0044] After building the three-dimensional finite element model according to the insulator geometric parameters (shed angle, spacing, diameter), the equivalent resistivity is mapped to the conductivity of the pollution layer . Building a three-dimensional finite element model is a geometric modeling process, which is as follows: Use CAD software or finite element software to create a three-dimensional geometric model of the insulator; Input the key geometric parameters of the insulator: Shed angle (e.g. 20°); Shed spacing (e.g. 50mm); Shed diameter (e.g. 300mm); Meshing: divide the geometric model into finite element meshes, which can use tetrahedral elements; Set the grid density: increase the grid density in the pollution layer and the electric field concentration area (shed edge), and appropriately reduce the grid density in other areas.

[0045] Material property definition: Define the dielectric constant of the insulator material; Define the conductivity of the pollution layer.

[0046] The humidity correction factor is represented as , which adjusts the conductivity , to obtain the adjusted conductivity : where is the relative humidity of the environment, is the reference humidity, , are the material moisture absorption characteristic coefficients, , represents the degree of influence of humidity on conductivity change, , determines the degree of nonlinearity of humidity influence; The surface electric field distribution is solved by the Poisson equation, which is represented as: where represents the surface electric field gradient, and the boundary conditions are: the high-voltage end is applied with phase voltage , and the ground end potential is 0; The field strength distortion gradient is represented as: , represents the electric field strength, i.e. the field strength, are the three components of the field strength, respectively representing the components of the electric field in the x-axis, y-axis and z-axis directions.

[0047] Simulate the field strength distortion gradient according to the humidity gradient of 50%, 70% and 90% respectively to obtain the humidity-field strength mapping relationship surface, The 50%, 70%, and 90% are respectively brought into the humidity correction factor , and then the pollution layer conductivity is adjusted according to the humidity correction factor , and finally the field strength gradient in three directions is calculated by solving the Poisson equation based on the adjusted pollution layer conductivity The 50%, 70%, and 90% are selected as three typical humidity points, covering dry, humid, and saturated states, and the humidity-field strength mapping relationship surface is a three-dimensional surface graph, with the horizontal axis representing humidity, the vertical axis representing position coordinates, and the vertical axis representing field strength distortion gradient values.

[0048] The preset percentage in S5 is 45%, and in the power system, the design and testing of power equipment such as insulators usually take 30 kV / cm as the reference value of air breakdown field strength, which helps to ensure the safety of equipment in operation. Set the air breakdown field strength reference value , and calculate the critical threshold S52: Map the field strength distortion gradient to the digital twin surface of the insulator string according to the spatial coordinates (map ∇E from the finite element grid to the digital twin surface vertices, use nearest neighbor interpolation for mapping, display on the digital twin, and render a heat map on the digital twin surface), and generate a normalized risk coefficient matrix S53: According to the risk coefficient matrix, divide the warning levels into three levels: Yellow warning Mark the risk area coordinates and increase the monitoring frequency to once every minute; Orange warning Generate a cleaning path planning graph and push it to the operation and maintenance terminal; Red warning : Trigger an emergency power-off command and start the insulator self-explosion protection mechanism; Use HSV color space to render the heat map, with hue gradually changing from green to red , and saturation positively correlated with risk coefficient. The hierarchical warning signal and heat map are transmitted in real time to the dispatch center through the LoRaWAN network, with a response delay of less than 200ms.

[0049] The power grid digital twin technology analysis system based on AI image recognition is used to realize the above analysis method, which includes the following modules:​​​​​​​​​ Image acquisition and data fusion: Multi-spectral cameras are carried by the UAV to collect multi-dimensional image data including visible light images and ultraviolet images, and to synchronously obtain line leakage current waveform data; Contamination feature extraction and analysis: A dual-channel deep residual network is used to extract contamination texture features and discharge spot distribution features from multi-dimensional image data, respectively, to generate a multi-scale contamination feature matrix reflecting the distribution and state of the contamination layer; Dynamic contamination current coupling model: A dynamic contamination-current coupling model is established to time-domain align the multi-scale contamination features and leakage current harmonic components, calculate the equivalent resistivity of the contamination conductive path, and quantify the electrical characteristics of the contamination layer; Digital twin reconstruction and field strength simulation: Based on the equivalent resistivity, the digital twin of the insulator string is reconstructed, and the field strength distortion gradient under different humidity conditions is simulated; Hierarchical early warning mechanism: When the field strength distortion gradient reaches a preset percentage of the set air breakdown field strength, a hierarchical early warning signal is automatically generated.

[0050] The present application encompasses any substitutions, modifications, equivalent methods and solutions made to the essence and scope of the present application. In order for the public to have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can be fully understood without these details by those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0051] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should be considered as the protection scope of the present application.

Claims

1. The power grid digital twin technology analysis method based on AI image recognition is characterized by: The following steps are involved: S1: A multispectral camera mounted on a drone is used to capture visible light and ultraviolet images of the insulator string, while also acquiring line leakage current waveform data. S2: Construct a dual-channel deep residual network to extract the pollution texture features of the visible light image and the discharge spot distribution features of the ultraviolet image, respectively, and generate a multi-scale pollution feature matrix; S3: Establish a dynamic pollution-current coupling model, align the multi-scale pollution characteristic matrix with the leakage current harmonic components in the time domain, and calculate the equivalent resistivity of the pollution conductive path; S4: Reconstruct the surface electric field distribution of the digital twin based on equivalent resistivity and simulate the field intensity distortion gradient under different humidity conditions; S5: When the field intensity distortion gradient reaches a preset percentage of the air breakdown field intensity, a graded warning signal is generated.

2. The power grid digital twin technology analysis method based on AI image recognition according to claim 1 is characterized in that: Said S1 specifically includes: S11: A multispectral imaging device is integrated on the drone, including a visible light camera and a solar-blind ultraviolet camera, to collect visible light images and ultraviolet images of the insulator string respectively; S12: Plan the drone inspection route to ensure that each insulator is photographed from at least three different angles; S13: A wireless leakage current sensor is installed on the steel cap of the insulator string. The sampling frequency is set to 10 kHz, and the leakage current waveform data is transmitted to the drone in real time through the LoRa communication module.

3. The power grid digital twin technology analysis method based on AI image recognition according to claim 1 is characterized in that: The S2 specifically includes: S21: Design a visible light channel network using an improved ResNet-34 architecture. Embed a spatial attention module after the third residual block to output a dirt texture feature map, including dirt thickness distribution and material classification probability. S22: Construct a UV channel network and a U-Net structure based on deformable convolution. Use the spot detection layer to extract the distribution density, area ratio, and morphological complexity characteristics of the discharge spot and generate a discharge activity index matrix. S23: In the feature fusion stage, the feature map of the fourth residual block of the visible light channel is cross-modally aligned with the decoder output of the ultraviolet channel, and a bidirectional gating mechanism is used to achieve feature interaction to generate a multi-scale contamination feature matrix; S24: It also includes extracting multi-scale features through the pyramid pooling layer, and outputting feature descriptors including four scales.

4. The power grid digital twin technology analysis method based on AI image recognition according to claim 3 is characterized in that: The S3 specifically includes: S31: Acquire leakage current waveform data, process the leakage current waveform data using fast Fourier transform, extract each harmonic component in the signal, and respectively extract the third, fifth, and seventh harmonic components; S32: Using the time synchronization module, the multi-scale pollution feature matrix is ​​time-aligned with the harmonic components, and the search range is limited to ±10 milliseconds; S33: Construct a nonlinear relationship between pollution features and harmonic current components. A nonlinear dynamic pollution-current coupling model is established by using multi-scale pollution features extracted using a pyramid pooling network. This dynamic pollution-current coupling model weights and sums the multi-scale pollution features, then combines them with the corresponding harmonic components using an exponential function to model the impact of pollution on current. The weight parameters are dynamically learned during the training process to adjust the influence of each scale feature. A harmonic attenuation coefficient is introduced to control the nonlinear impact of pollution features on harmonics. S34: The degree of contamination's impact on current is assessed by calculating the equivalent resistivity, which is obtained by summing the squares of the amplitudes of multiple harmonic components and combining them with the effective value of the phase voltage. The equivalent resistivity value is used to determine the severity of contamination on the insulator surface.

5. The power grid digital twin technology analysis method based on AI image recognition according to claim 4 is characterized in that: The dynamic pollution-current coupling model is expressed as: ,in, , represents the harmonic order, Represents the pyramid pooling The weighted sum of the scale eigenvectors, is the learnable weight parameter; is the harmonic attenuation coefficient, Indicates time On the Subharmonic current amplitude, For the The noise term of subharmonics takes into account possible errors or uncontrollable factors in the experiment.

6. The power grid digital twin technology analysis method based on AI image recognition according to claim 5 is characterized in that: The equivalent resistivity : ,in, is the pollution layer geometry factor, is the effective value of the phase voltage, is the harmonic component amplitude.

7. The power grid digital twin technology analysis method based on AI image recognition according to claim 1 is characterized in that: The S4 specifically includes: S41, construct a 3D finite element model and map the conductivity: A 3D finite element model of the insulator is constructed based on the geometric parameters of the insulator to describe the spatial shape of the insulator and provide a basis for the calculation of the electric field strength. In the 3D finite element model, the equivalent resistivity value is converted into the conductivity of the contamination layer. The conductivity is the inverse of the resistivity and reflects the conductive properties of the contamination layer material. S42, introduce humidity correction factor and dynamically adjust conductivity: Considering the effect of humidity on the conductivity of the contamination layer, introduce humidity correction factor and dynamically adjust the conductivity of the contamination layer according to the relative humidity of the environment; S43, solving the electric field distribution using the Poisson equation: The electric field distribution on the surface of the insulator is solved using the Poisson equation. By solving the equation, the electric potential distribution is obtained, and then the electric field strength is calculated. Boundary conditions are set during the calculation. The boundary conditions include applying a phase voltage at the high-voltage end and setting the ground potential to zero. The surface electric field distribution is calculated based on the change in conductivity. S44, calculate the field intensity distortion gradient: calculate the field intensity distortion gradient based on the electric field distribution , field intensity distortion gradient It indicates the severity of the change in electric field intensity and reflects the local non-uniformity of the electric field. The calculation of the field intensity distortion gradient includes taking the maximum value by taking the partial derivative of the electric field component to obtain the gradient of the field intensity distortion; S45, simulating the field intensity distortion gradient under different humidity conditions: The field intensity distortion gradient under different humidity conditions is simulated according to the humidity gradient. By simulating the field intensity distortion gradient under different humidity conditions, a mapping relationship surface between humidity and field intensity is generated to show the impact of humidity changes on field intensity distortion.

8. The power grid digital twin technology analysis method based on AI image recognition according to claim 7 is characterized in that: The field intensity distortion gradient Expressed as: , represents the electric field strength, that is, the field strength, They are the three components of the electric field strength, representing the components of the electric field in the x-axis, y-axis, and z-axis directions respectively.

9. The power grid digital twin technology analysis method based on AI image recognition according to claim 8 is characterized in that: The preset percentage in S5 is 45%, setting the air breakdown field strength reference value , calculate the critical threshold ; S52: Distort the field intensity gradient Map the spatial coordinates to the surface of the digital twin to generate a normalized risk coefficient matrix : ; S53: Divide warning levels according to the risk factor matrix.

10. A power grid digital twin technology analysis system based on AI image recognition, used to implement the power grid digital twin technology analysis method based on AI image recognition as described in any one of claims 1 to 9, characterized in that: Includes the following modules: Image acquisition and data fusion: Using a multispectral camera mounted on a drone, multi-dimensional image data, including visible light and ultraviolet images, is collected, and line leakage current waveform data is simultaneously acquired; Pollution feature extraction and analysis: A dual-channel deep residual network is used to extract pollution texture features and discharge spot distribution features from multi-dimensional image data, generating a multi-scale pollution feature matrix that reflects the distribution and status of the pollution layer. Dynamic pollution current coupling model: By establishing a dynamic pollution-current coupling model, the multi-scale pollution characteristics and the harmonic components of the leakage current are aligned in the time domain, the equivalent resistivity of the pollution conductive path is calculated, and the electrical characteristics of the pollution layer are quantified; Digital twin reconstruction and field strength simulation: Reconstruct the digital twin of the insulator string based on equivalent resistivity and simulate the field strength distortion gradient under different humidity conditions; Grading warning mechanism: When the field intensity distortion gradient reaches a preset percentage of the set air breakdown field intensity, a graded warning signal is automatically generated.

Images