Power transmission channel foreign matter line hanging hidden danger source monitoring method and device

Abstract

The invention discloses a power transmission channel foreign matter line hanging hidden danger source monitoring method and device, and the method comprises the steps: collecting optical satellite remote sensing data of a monitoring region, and constructing a hidden danger source sample library through marking; enhancing the sample library data by using a preset improved GAN network to obtain an enhanced sample library; inputting samples in the enhanced sample library into a semantic segmentation network integrated with an edge information enhancement module for training to obtain a hidden danger source identification model; historical foreign matter hanging line data, meteorological data and electric power information basic data in a monitoring area are collected, and a semi-quantitative evaluation method is adopted to construct a hidden danger source risk evaluation model; and extracting a hidden danger source from the current optical satellite remote sensing data by using a hidden danger source identification model, and carrying out power transmission channel foreign matter line hanging hidden danger dynamic monitoring and early warning by using a risk assessment model according to the extraction result, the current meteorological data and the current power transmission line geographic position information. According to the invention, pertinence and accuracy of foreign matter line hanging hidden danger source monitoring of the power transmission channel can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent external force damage risk monitoring of power transmission channels, and in particular to a method and device for monitoring hidden danger sources of foreign objects hanging on wires in power transmission channels. Background Art

[0002] Ensuring the safe operation of the power system is crucial to the overall national security. Intelligent monitoring and forecasting of external force damage risks to transmission lines is a core requirement and development direction for the company's safe grid operations and maintenance. In windy weather, agricultural covering films such as plastic greenhouse films and ground films, as well as dust screens, can easily become entangled and hung around power equipment and transmission lines, causing electric field distortion, reduced insulation margin, and, in severe cases, phase-to-phase short circuits. Furthermore, once hung, agricultural covering films such as plastic greenhouse films and ground films are difficult to naturally fall off, and the environmental electrical changes after flashovers can easily lead to reclosing failures. In severe cases, secondary flashovers and continuous tripping can lead to the melting of transmission lines, causing irreparable losses.

[0003] Although satellite remote sensing technology has been applied to monitoring the potential dangers of foreign objects hanging on transmission lines, it still faces several challenges. For one thing, the use of satellite remote sensing for monitoring the sources of foreign objects hanging on transmission lines suffers from an imbalanced distribution of data across different categories and a small number of samples in certain categories. This results in a small number of samples and low accuracy in identifying the types of foreign objects hanging on transmission lines. Furthermore, existing risk assessment systems for the sources of foreign objects hanging on transmission lines do not fully consider the impact of dynamic meteorological data on risk. Meteorological factors are a significant contributing factor to the risk of foreign objects hanging on transmission lines. Traditional static assessment methods are unable to reflect changes in meteorological conditions in real time and may miss critical moments when foreign objects hanging on transmission lines occur, resulting in a lack of pertinence and accuracy in the assessment results. Summary of the Invention

[0004] In response to the above technical problems, the present invention proposes a method and device for monitoring hidden danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data, so as to improve the pertinence and accuracy of monitoring hidden danger sources of foreign objects hanging on power transmission channels.

[0005] The present invention adopts the following technical solutions.

[0006] According to a first aspect of the present invention, a method for monitoring the hidden dangers of foreign objects hanging on power transmission channels based on dynamic meteorological data is provided. The method comprises the following steps:

[0007] Collect optical satellite remote sensing data of the monitoring area;

[0008] Marking samples of the optical satellite remote sensing data to construct a sample library of potential hazards of foreign objects hanging on power transmission channels;

[0009] Using a preset improved GAN network to enhance the sample library data to obtain an enhanced sample library;

[0010] Inputting the samples in the enhanced sample library into a semantic segmentation network that incorporates an edge information enhancement module for training, and obtaining a trained semantic segmentation network as a hidden danger source identification model;

[0011] Collect historical data on foreign objects hanging on power lines, meteorological data, and basic power information data in the monitoring area. Using a semi-quantitative assessment method, a risk assessment model for the hidden danger source of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data is constructed based on the collected data.

[0012] The hidden danger source identification model is used to extract the hidden danger sources of foreign objects hanging on the transmission channel from the current optical satellite remote sensing data. Based on the extraction results, current meteorological data and the current geographical location information of the transmission line, the risk assessment model is used to conduct daily dynamic monitoring and early warning of the hidden dangers of foreign objects hanging on the transmission channel.

[0013] Furthermore, optical satellite data of the monitoring area is collected, including:

[0014] Collect original optical satellite remote sensing data of the monitoring area;

[0015] Standardization preprocessing is performed on the original optical satellite remote sensing data to obtain optical satellite remote sensing data; the standardization preprocessing includes: performing radiation correction, geometric correction, atmospheric correction, image fusion, image mosaicking and cropping on the original data.

[0016] Furthermore, the optical satellite remote sensing data is sample-labeled to construct a sample library of hidden danger sources of foreign objects hanging on power transmission channels, including:

[0017] Using a geospatial database as support, we collected real data on potential sources of foreign objects hanging on transmission lines in multiple scenarios. Using remote sensing software, we extracted the characteristic information of these sources from remote sensing images using visual interpretation methods, generating images and marker files.

[0018] The images and marker files are cropped into image samples of uniform size and randomly divided into training set, validation set, and test set according to the preset sample ratio.

[0019] Furthermore, the improved GAN network includes a generator and a discriminator;

[0020] A dual-channel attention mechanism is incorporated into the SPADE residual module of the generator, and the convolutional layer of the generator is set as a depth-separable convolutional layer;

[0021] The discriminator incorporates a spectral normalization scheme.

[0022] Furthermore, the dual-channel attention mechanism is configured to include an upper and lower branch, wherein the upper branch introduces a mixed pooling module, and the lower branch introduces a small sample adaptive attention module;

[0023] The hybrid pooling module includes a strip pooling unit and a pyramid pooling unit; the strip pooling unit is used to pool samples in the input sample library in the horizontal and vertical directions to capture the characteristics of long strip objects; the pyramid pooling unit is used to aggregate scene information of different regions of the samples in the input sample library, analyze multi-scale scene feature information, and obtain effective global information of pixel-level scene annotation;

[0024] The small sample adaptive attention module is used to automatically adjust the different levels of attention to the relevant features corresponding to each type of foreign object hanging line hidden danger source through adaptive learning of samples in the input sample library.

[0025] Furthermore, the pooling formulas for the horizontal and vertical directions of the strip pooling unit are respectively:

[0026]

[0027] Where x i,j Indicates the value at position (i, j) in the feature map corresponding to the input sample, y s and y v Represents the pooling results along the horizontal and vertical directions respectively, x i,j ∈R S×V ,y s ∈R S ,y v ∈R V , V and H represent the pooling lengths in the horizontal and vertical directions respectively, R S×V represents a real matrix of dimension S×V, R S represents an S-dimensional real vector, R V Represents a real vector of dimension V.

[0028] Furthermore, the small sample adaptive attention module is further configured as follows:

[0029] Inputting the sample information in the sample library into the SENet structure to convolute the features of the sample information into a multi-channel two-dimensional feature map; wherein the number of channels of the multi-channel two-dimensional feature map is the same as the number of types of hidden danger sources of foreign objects hanging on the wires;

[0030] Compressing the multi-channel two-dimensional feature map into a feature vector by global average pooling;

[0031] constructing, by a fully connected layer, the correlation between the channels in the multi-channel two-dimensional feature map using the feature vector, and automatically learning the feature weight of each channel based on the correlation;

[0032] The learned feature weights are normalized, and then the normalized weights are weighted to the features of each channel, so as to adaptively learn features with different correlations with different types of foreign object hanging line hidden danger sources.

[0033] Furthermore, the convolution operation of the depthwise separable convolution includes depthwise convolution and pointwise convolution.

[0034] Furthermore, the edge information enhancement module is used to:

[0035] Use frequency domain processing to convert remote sensing images into frequency domain through Fourier transform and then perform frequency domain filtering;

[0036] A Butterworth filter transfer function is designed to process the frequency domain features of the filtered remote sensing image in the frequency domain space, weaken the low-frequency components, retain the edge information of the high-frequency components of the hidden danger source of foreign objects hanging on the wire, and suppress noise points with excessively high frequencies, thereby obtaining a processed spectrum graph;

[0037] The processed spectrum is transformed back to the spatial domain by inverse Fourier transform to obtain an edge-enhanced image;

[0038] Among them, the Butterworth bandpass filter formula is as follows:

[0039]

[0040] Among them, H(u,v) is the transfer function, D(u,v) represents the distance from point (u,v) to the center point in the frequency domain, where D O is the cutoff frequency, W is the bandwidth; D O is 20 to 50 pixels, and W is 10 to 40 pixels.

[0041] Furthermore, historical data on foreign objects hanging on power lines, historical meteorological data, geographical location information of transmission lines, and spatial data on potential hazards of foreign objects hanging on power lines in the monitoring area are collected. A semi-quantitative assessment method is used to construct a risk assessment model for potential hazards of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data based on the collected data, including:

[0042] Based on the historical data of foreign objects hanging on the line, determine the risk level value R corresponding to each historical date;

[0043] Based on the spatial data of potential hazards of foreign objects hanging on wires, the relative density factors ReD of various potential hazards of foreign objects hanging on wires corresponding to each historical date are determined;

[0044] Based on the spatial data of the hidden danger sources of foreign objects hanging on the power lines and the geographical location information of the transmission lines, the hidden danger proximity distance factor VeD corresponding to each historical date is determined;

[0045] Based on historical meteorological data, determine the wind speed risk factor Win corresponding to each historical date;

[0046] Based on historical meteorological data and transmission line geographic location information data, the wind direction risk factor WDir corresponding to each historical date is determined;

[0047] Based on R, ReD, VeD, Win, and WDir corresponding to each historical date, the weight factors a, b, c, and d corresponding to ReD, VeD, Win, and WDir are determined respectively, and the risk assessment model shown below is obtained:

[0048] R=ReD×bVeD×cWin×dWDir.

[0049] According to a second aspect of the present invention, there is provided a device for monitoring the hidden danger of foreign objects hanging on power transmission channels using the method described in the first aspect of the present invention. The device comprises:

[0050] Acquisition module, used to collect optical satellite remote sensing data of the monitoring area;

[0051] A sample construction module is used to perform sample labeling on the optical satellite remote sensing data to construct a sample library of potential hazards of foreign objects hanging on power transmission channels;

[0052] A sample enhancement module is used to enhance the sample library data using a preset improved GAN network to obtain an enhanced sample library;

[0053] An identification training module is used to input samples in the enhanced sample library into a semantic segmentation network integrated with an edge information enhancement module for training, thereby obtaining a trained semantic segmentation network as a hidden danger source identification model;

[0054] The assessment model construction module is used to collect historical data on foreign objects hanging on power lines, meteorological data, and basic power information data in the monitoring area. Based on the collected data, a semi-quantitative assessment method is used to construct a risk assessment model for the hidden danger source of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data;

[0055] The monitoring module uses the hidden danger source identification model to extract the hidden danger sources of foreign objects hanging on the transmission channel from the current optical satellite remote sensing data. Based on the extraction results, current meteorological data and the current geographical location information of the transmission line, the risk assessment model is used to perform daily dynamic monitoring and early warning of hidden dangers of foreign objects hanging on the transmission channel.

[0056] According to a third aspect of the present invention, a terminal is provided, which includes a processor and a storage medium.

[0057] The storage medium is used to store instructions;

[0058] The processor is configured to operate according to the instructions to execute the steps of the method according to the first aspect of the present invention.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. In view of the problem that the existing multi-category extraction technology based on satellite remote sensing has insufficient adaptability to small samples and unbalanced data, which easily leads to missed detection and false detection, the present invention improves the GAN network data enhancement method and constructs a dual-channel attention mechanism. The upper part introduces a hybrid pooling module combining strip pooling unit and pyramid pooling, and the lower part introduces a small sample adaptive attention module. It not only takes into account the contextual information of most long strips such as greenhouses and ground films and non-long strips such as dustproof nets, but also adaptively learns features with low correlation between multi-category labels and features with high correlation within each category, thereby enhancing relevant effective feature channels.

[0061] 2. The generator of the traditional GAN network data enhancement method uses ordinary convolutional layers and incorporates a dual-channel attention mechanism, which increases the computational complexity of the model and easily leads to low efficiency of the model when processing large-scale data. The present invention replaces the ordinary convolutional layers in the generator of the improved GAN network data enhancement method with depthwise separable convolutional layers, which greatly reduces the amount of computation and enables the model to still efficiently generate samples under more lightweight conditions.

[0062] 3. Existing risk assessment systems for the potential danger sources of foreign objects hanging on power transmission channels fail to consider the impact of dynamic meteorological data, making it difficult to capture the dynamic changes in meteorological conditions that affect risk. This paper introduces dynamic meteorological data and employs a semi-quantitative assessment method to construct a risk assessment model for the potential danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data. This model analyzes the impact of daily meteorological conditions on these potential danger sources, significantly improving the accuracy and timeliness of risk assessments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of a method for monitoring hidden danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data of the present invention.

[0064] Figure 2 It is a schematic diagram of a device for monitoring hidden danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data according to the present invention. DETAILED DESCRIPTION

[0065] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without making creative efforts are all within the scope of protection of the present invention.

[0066] Example 1

[0067] This embodiment provides a method for monitoring hidden danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data. Figure 1 In this embodiment, the method for monitoring hidden danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data includes the following steps:

[0068] Step 1: Collect optical satellite remote sensing data.

[0069] Specifically, this step further includes:

[0070] Step 11: Collect raw optical satellite remote sensing data.

[0071] The raw optical satellite remote sensing data is preferably raw optical satellite remote sensing data with a resolution better than 2 meters. Specifically, based on the monitoring needs for potential hazards of foreign objects hanging on transmission line channels, optical satellite remote sensing data with a spatial resolution of no less than 2 meters, such as Gaofen-1, Gaofen-2, Gaofen-6, and the China-Pakistan Resources Satellite 04A, can be used.

[0072] Step 12: Perform standardization preprocessing on the original optical satellite remote sensing data to obtain optical satellite remote sensing data.

[0073] Standardization preprocessing includes radiometric correction, geometric correction, atmospheric correction, image fusion, image mosaicking, and cropping of the raw data. By performing standardization preprocessing on raw optical satellite remote sensing data, radiometric errors and geometric deviations in satellite images can be corrected or eliminated.

[0074] Step 2: Sample labeling is performed on the optical satellite remote sensing data to construct a sample library of potential hazards of foreign objects hanging on power transmission channels.

[0075] Among them, the method of manually marking samples can be used to mark samples of optical satellite remote sensing data.

[0076] Specifically, step 2 may further include:

[0077] Step 21: With the support of geospatial database, collect real data on the hidden danger sources of foreign objects hanging on the transmission channels in multiple scenarios. Use remote sensing software through visual interpretation method to outline and extract the characteristic information of the hidden danger sources of foreign objects hanging on the transmission lines in the remote sensing images to obtain images and marking files.

[0078] Among them, the hidden danger sources of foreign objects hanging on the transmission channel may include, for example, dust screens, greenhouses, ground films, etc. The marking file may include label information for marking the hidden danger sources of foreign objects hanging on the transmission channel.

[0079] Step 22: Crop the image and marker files into image samples of uniform size, and randomly divide them into training set, validation set, and test set according to the preset sample ratio.

[0080] The cropped image sample is preferably a PNG image sample with a size of 384*384, and the preset sample ratio is preferably 7:1:2.

[0081] Step 3: Use the preset improved GAN network to enhance the sample library data to obtain an enhanced sample library.

[0082] The improved GAN network includes a generator and a discriminator. The generator's SPADE residual module incorporates a dual-channel attention mechanism, and the generator's convolutional layers are set to depthwise separable convolutional layers. The discriminator incorporates a spectral normalization model.

[0083] By building a dual-channel attention mechanism and integrating it into the SPADE residual module of the GAN network generator, the residual module's ability to learn features can be optimized. The dual-channel attention mechanism is divided into two branches: the upper branch introduces a hybrid pooling module that combines strip pooling units and pyramid pooling units, and the lower branch introduces a small sample adaptive attention module.

[0084] The strip pooling unit is used to pool samples in the input sample library horizontally and vertically to capture the characteristics of long strip objects. Specifically, through strip pooling in the vertical and horizontal directions, position relationships are established in these directions. For most long strip foreign objects hanging on the lines, such as greenhouses and ground films, the use of the strip pooling kernel can eliminate information interference from other irrelevant areas and capture isolated position information over longer distances. The horizontal and vertical pooling formulas are:

[0085]

[0086] Where x i,j Indicates the value at position (i, j) in the feature map corresponding to the input sample, y s and y v Represents the pooling results along the horizontal and vertical directions respectively, x i,j ∈RS×V ,y s ∈R S ,y v ∈R V , V and H represent the pooling lengths in the horizontal and vertical directions respectively, R S×V represents a real matrix of dimension S×V, R S represents an S-dimensional real vector, R V Represents a real vector of dimension V.

[0087] The pyramid pooling unit is used to aggregate scene information of different regions of samples in the input sample library, analyze multi-scale scene feature information, and obtain effective global information of pixel-level scene annotation.

[0088] The small sample adaptive attention module is used to automatically adjust the different levels of attention to the relevant features corresponding to different types of foreign objects hanging on the line hidden danger sources through adaptive learning of samples in the input sample library. The small sample adaptive attention module is configured as follows:

[0089] Inputting the sample information in the sample library into a Squeeze-and-Excitation Networks (SENet) structure to convolute the features of the sample information into a multi-channel two-dimensional feature map; wherein the number of channels in the multi-channel two-dimensional feature map is the same as the number of types of foreign body hanging wire hazard sources;

[0090] Compressing the multi-channel two-dimensional feature map into a feature vector through global average pooling;

[0091] constructing, by a fully connected layer, the correlation between the channels in the multi-channel two-dimensional feature map using the feature vector, and automatically learning the feature weight of each channel based on the correlation;

[0092] The learned feature weights are normalized, and then the normalized weights are weighted to the features of each channel, so as to adaptively learn the low-correlation features and high-correlation features corresponding to each type of foreign object hanging line hazard source.

[0093] This small-sample adaptive attention module can enhance relevant and effective feature channels and improve the ability to reuse labeled data features. For example, when the source of the extracted foreign object hanging on wires is a greenhouse, the small-sample adaptive attention module adaptively learns the weights of each channel and determines that white is a highly relevant feature and red is a less relevant feature.

[0094] The outputs of the mixed pooling module and the small sample adaptive attention module are used as the input of subsequent layers in the generator. The overall output of the generator of the improved GAN network is jointly determined by the features extracted by the mixed pooling module and the small sample adaptive attention module.

[0095] Modifying the normalization mode of the discriminator to spectral normalization and integrating it into the discriminator can improve the discriminator's ability to capture image features and promote stable model training.

[0096] The convolution operation of depth-wise separable convolution includes scene information of the relationship between depth-wise convolution and point-wise convolution, which brings better feature representation capability than global pooling. The core idea of depth-wise separable convolution is to decompose the complete convolution operation into two steps, consisting of depth-wise convolution and point-wise convolution. In depth-wise separable convolution, the input feature map size is L a ×L a ×W, the convolution kernel size is L b ×L b ×W, the number of convolution kernels is N, and the total amount of computation for depthwise separable convolution is F DSConv =L a ×L a ×L b ×L b ×W+W×N×L a ×L a Among them, the total amount of calculation of depth convolution is F DConv =L a ×L a ×L b ×L b ×W, the total amount of calculation for point convolution is F PConv =W×N×L a ×L a The computational complexity of depth-wise separable convolution and ordinary convolution is shown below:

[0097]

[0098] Among them, r F is the computational ratio of depthwise separable convolution to ordinary convolution, N is the number of convolution kernels, L b is the size of the convolution kernel in height and width. Therefore, by replacing the ordinary convolution layer in the generator of the improved GAN network data enhancement method with a depth-separable convolution layer, the number of parameters and the amount of calculation can be effectively reduced.

[0099] It can be seen that in this step, an improved GAN network data enhancement method is adopted, in which a dual-channel attention mechanism is constructed. The upper part introduces a hybrid pooling module that combines strip pooling units and pyramid pooling, taking into account the contextual information of most long strips such as greenhouses and ground films and non-long strips such as dust screens. The lower part introduces a small sample adaptive attention module to adaptively learn features with low correlation between multi-category labels and features with high correlation within each category, thereby enhancing relevant effective feature channels. In this way, the problems of unbalanced distribution of various types and small sample size in the constructed sample library of transmission channel foreign body hanging line hidden danger sources such as dust screens, greenhouses, and ground films can be overcome. At the same time, the ordinary convolutional layer in the generator of the improved GAN network data enhancement method is replaced with a depth-separable convolutional layer, which increases the number of samples more stably with a lower amount of computation.

[0100] Step 4: Input the samples in the enhanced sample library into the semantic segmentation network that incorporates the edge information enhancement module for training, and obtain the trained semantic segmentation network as a hidden danger source identification model.

[0101] In this step, the edge information enhancement module is used to:

[0102] Using frequency domain processing, the remote sensing image is converted to the frequency domain through Fourier transform and then frequency domain filtering is performed. Among them, frequency domain filtering is to perform Fourier transform on the image, converting the image from image space to frequency domain space.

[0103] The transfer function of the Butterworth filter is designed to process the frequency domain features of the filtered remote sensing image in the frequency domain space, weakening the low-frequency components, retaining the edge information of the high-frequency components (the source of the hidden danger of foreign objects hanging on the wires), and suppressing the noise points with excessively high frequencies to obtain the processed spectrum.

[0104] The processed spectrum is transformed back to the spatial domain by inverse Fourier transform to obtain an edge-enhanced image;

[0105] Among them, the Butterworth bandpass filter formula is as follows:

[0106]

[0107] Among them, H(u,v) is the transfer function, D(u,v) represents the distance from point (u,v) to the center point in the frequency domain, where D O is the cutoff frequency, W is the bandwidth; D O is 20 to 50 pixels, and W is 10 to 40 pixels.

[0108] In a specific application example, the edge information enhancement module can be integrated into the semantic segmentation network to extract potential hazards such as dust screens, greenhouses, and ground films hanging on power transmission channels. For example, the Adam optimizer can be used with an initial learning rate of 10 -5 The batch size is 16, the maximum number of iterations is 30,000, and overfitting of the data is avoided by setting dropout to 0.2 and adding random Gaussian noise to the data.

[0109] Step 5: Collect historical data on foreign objects hanging on power lines, meteorological data, and basic power information data in the monitoring area, and use a semi-quantitative assessment method to build a risk assessment model for the hidden danger source of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data based on the collected data.

[0110] The collected historical data on foreign objects hanging on the wires include the time when the foreign objects hanging on the wires occurred, the type of foreign objects hanging on the wires, the meteorological data on the day when the foreign objects hanging on the wires occurred, the impact of the foreign objects hanging on the wires, etc.

[0111] The collected meteorological data include the daily maximum wind speed value and its corresponding wind direction.

[0112] The transmission line geographic location information data includes the length information, location information, etc. of each transmission line.

[0113] The spatial data of hidden danger sources caused by foreign objects hanging on wires include information such as the type of hidden danger source, the geographical location of the hidden danger source, and the area of the hidden danger source.

[0114] The step 5 specifically includes:

[0115] Step 51: Based on the historical foreign body hanging line data, determine the risk level value R corresponding to each historical date.

[0116] Step 52: Based on the spatial data of potential hazards caused by foreign objects hanging on wiring, determine the relative density factors ReD of various potential hazards caused by foreign objects hanging on wiring corresponding to each historical date.

[0117] As an example, a 100-meter buffer zone can be established near the potential source of foreign body hanging on the line. The relative density factor ReD of each type of potential source of foreign body hanging on the line can be calculated using the following formula:

[0118]

[0119] Among them, A yh A is the total area of various hidden danger sources of foreign objects hanging on wires within the buffer zone, hcq is the buffer zone area.

[0120] Step 53: Based on the spatial data of the hidden danger source of foreign objects hanging on the power lines and the geographical location information data of the power transmission lines, determine the hidden danger proximity distance factor VeD corresponding to each historical date.

[0121] Specifically, the hazard proximity distance factor, VeD, is derived based on the vertical distance between the source of the foreign object hanging on the power line and the transmission line. As an example, based on the collected geographic location information of the transmission line and the location of the hazard source, spatial neighbor analysis is performed to calculate the vertical distance between the source of the foreign object hanging on the power line and the transmission line. Combined with semi-quantitative assessment theory, the hazard proximity distance factor, VeD, is graded as shown in Table 1 below:

[0122] Table I

[0123] Vertical distance (m) ≤100 100-500 500-1000 1000-1500 ≥1500 Hidden danger proximity distance factor value VeD 1 0.8 0.5 0.3 0.1

[0124] Step 54: Based on the historical meteorological data, determine the wind speed risk factor Win corresponding to each historical date.

[0125] Specifically, the wind speed risk factor Win is determined based on the risk of damage to the transmission line caused by the daily maximum wind speed value. As an example, based on the daily maximum wind speed value and combined with the semi-quantitative assessment theory, the daily maximum wind speed value and the wind speed influence coefficient can be set as shown in the following Table II:

[0126] Table II

[0127] The maximum daily wind speed w 0 1 2 3 4 5 6 7+ <![CDATA[Wind speed influence coefficient α w > 0 0 0 0.1 0.15 0.2 0.25 0.5

[0128] The calculation formula of wind speed risk factor Win is as follows:

[0129]

[0130] Among them, w is the maximum wind speed of the day, α w is the wind speed influence coefficient.

[0131] Step 55: Based on the historical meteorological data and the geographical location information data of the transmission line, determine the wind direction risk factor WDir corresponding to each historical date.

[0132] In this step, the wind speed risk factor WDir is obtained based on the angle between the wind direction and the transmission line. As an example, combining the wind direction corresponding to the maximum daily wind speed value and the semi-quantitative evaluation theory, the wind direction and wind direction influence coefficient are set as shown in Table III below:

[0133] Table III

[0134] Angle ≤90° 90° 90°-180° 180°-270° 270° 270°-360° <![CDATA[Included angle level θ y > 2 3 1 2 3 1 <![CDATA[Wind direction influence coefficient β y > 0.25 0.5 0.1 0.25 0.5 0.1

[0135] The calculation formula of wind direction risk factor WDir is as follows:

[0136] WDir=θ y *β y

[0137] Among them, θ y is the angle between wind direction and transmission line, β yis the wind direction influence coefficient.

[0138] Step 56: Based on R, ReD, VeD, Win, and WDir corresponding to each historical date, weight factors a, b, c, and d corresponding to ReD, VeD, Win, and WDir are determined respectively, and the following risk assessment model is obtained:

[0139] R=ReD×bVeD×cWin×dWDir

[0140] The weight factors a, b, c, and d can be determined by expert scoring or by training in combination with a neural network.

[0141] Step 6: Use the hidden danger source identification model to extract the hidden danger source of foreign objects hanging on the transmission channel from the current optical satellite remote sensing data. Based on the extraction results, current meteorological data and the current geographical location information of the transmission line, use the risk assessment model to dynamically monitor and provide early warning for the hidden danger of foreign objects hanging on the transmission channel.

[0142] Specifically, based on the extraction results of potential foreign object hanging on power lines, combined with current meteorological data and the current transmission line location, we can further calculate the relative density factor ReD, the proximity factor VeD, and the wind direction risk factor WDir for each type of foreign object hanging on power lines. The wind speed risk factor Win can be calculated from the current meteorological data. This allows us to calculate the corresponding risk level using the risk assessment model developed in step 5.

[0143] When conducting assessments and early warnings, the standard deviation classification method can be used to categorize risk levels into four levels, from smallest to largest: Level 4 (low), Level 3 (high), Level 2 (high), and Level 1 (very high). This allows us to use optical satellite remote sensing imagery to extract data on potential sources of foreign objects hanging on power transmission lines, such as dust screens, greenhouses, and ground film, and conduct risk assessments based on daily meteorological data, thereby providing early warnings.

[0144] In steps 5-6, by introducing dynamic meteorological data and adopting a semi-quantitative assessment method, a risk assessment model for the hidden danger sources of foreign objects hanging on the power transmission channels based on dynamic meteorological data was constructed. The impact of daily meteorological conditions on the hidden danger sources of foreign objects hanging on the power transmission channels was analyzed, which significantly improved the accuracy and timeliness of the risk assessment.

[0145] Example 2

[0146] This embodiment provides a device for monitoring the hidden danger source of foreign objects hanging on the power transmission channel. The device adopts the method described in Example 1. Specifically, Figure 2 , the device comprises:

[0147] Acquisition module, used to collect optical satellite remote sensing data of the monitoring area;

[0148] A sample construction module is used to perform sample labeling on the optical satellite remote sensing data to construct a sample library of potential hazards of foreign objects hanging on power transmission channels;

[0149] A sample enhancement module is used to enhance the sample library data using a preset improved GAN network to obtain an enhanced sample library;

[0150] An identification training module is used to input samples in the enhanced sample library into a semantic segmentation network integrated with an edge information enhancement module for training, thereby obtaining a trained semantic segmentation network as a hidden danger source identification model;

[0151] The assessment model construction module is used to collect historical data on foreign objects hanging on power lines, meteorological data, and basic power information data in the monitoring area. Based on the collected data, a semi-quantitative assessment method is used to construct a risk assessment model for the hidden danger source of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data;

[0152] The monitoring module uses the hidden danger source identification model to extract the hidden danger sources of foreign objects hanging on the transmission channel from the current optical satellite remote sensing data. Based on the extraction results, current meteorological data and the current geographical location information of the transmission line, the risk assessment model is used to perform daily dynamic monitoring and early warning of hidden dangers of foreign objects hanging on the transmission channel.

[0153] Example 3

[0154] Embodiment 3 of the present invention provides a computer-readable storage medium.

[0155] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of the method for monitoring the hidden danger source of foreign objects hanging on the power transmission channel as described in Example 1 of the present invention.

[0156] The detailed steps are the same as those of the method for monitoring hidden danger sources of foreign objects hanging on power transmission channels provided in Example 1, and will not be repeated here.

[0157] The beneficial effects of the present invention are as follows:

[0158] 1. In view of the problem that the existing multi-category extraction technology based on satellite remote sensing has insufficient adaptability to small samples and unbalanced data, which easily leads to missed detection and false detection, the present invention improves the GAN network data enhancement method and constructs a dual-channel attention mechanism. The upper part introduces a hybrid pooling module combining strip pooling unit and pyramid pooling, and the lower part introduces a small sample adaptive attention module. It not only takes into account the contextual information of most long strips such as greenhouses and ground films and non-long strips such as dustproof nets, but also adaptively learns features with low correlation between multi-category labels and features with high correlation within each category, thereby enhancing relevant effective feature channels.

[0159] 2. The generator of the traditional GAN network data enhancement method uses ordinary convolutional layers and incorporates a dual-channel attention mechanism, which increases the computational complexity of the model and easily leads to low efficiency of the model when processing large-scale data. The present invention replaces the ordinary convolutional layers in the generator of the improved GAN network data enhancement method with depthwise separable convolutional layers, which greatly reduces the amount of computation and enables the model to still efficiently generate samples under more lightweight conditions.

[0160] 3. Existing risk assessment systems for the potential danger sources of foreign objects hanging on power transmission channels fail to consider the impact of dynamic meteorological data, making it difficult to capture the dynamic changes in meteorological conditions that affect risk. This paper introduces dynamic meteorological data and employs a semi-quantitative assessment method to construct a risk assessment model for the potential danger sources of foreign objects hanging on power transmission channels based on dynamic meteorological data. This model analyzes the impact of daily meteorological conditions on these potential danger sources, significantly improving the accuracy and timeliness of risk assessments.

[0161] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0162] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0163] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0164] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, and conventional procedural programming languages such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., utilizing an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions. The electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring hidden danger sources of foreign objects hanging on power transmission channels, characterized in that: The following steps are involved: Collect optical satellite remote sensing data of the monitoring area; Marking samples of the optical satellite remote sensing data to construct a sample library of potential hazards of foreign objects hanging on power transmission channels; Performing data enhancement on the sample library using a preset improved GAN network to obtain an enhanced sample library; Inputting the samples in the enhanced sample library into a semantic segmentation network that incorporates an edge information enhancement module for training, and obtaining a trained semantic segmentation network as a hidden danger source identification model; Collect historical data on foreign objects hanging on power lines, meteorological data, and geographical location information of transmission lines in the monitoring area. Using a semi-quantitative assessment method, a risk assessment model for the hidden danger source of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data is constructed based on the collected data. The hidden danger source identification model is used to extract the hidden danger sources of foreign objects hanging on the transmission channel from the current optical satellite remote sensing data. Based on the extraction results, current meteorological data and the current geographical location information of the transmission line, the risk assessment model is used to dynamically monitor and provide early warning for the hidden dangers of foreign objects hanging on the transmission channel.

2. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 1 is characterized in that: The collecting of optical satellite data of the monitoring area includes: Collect original optical satellite remote sensing data of the monitoring area; Standardization preprocessing is performed on the original optical satellite remote sensing data to obtain optical satellite remote sensing data; the standardization preprocessing includes: performing radiation correction, geometric correction, atmospheric correction, image fusion, image mosaicking and cropping on the original data.

3. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 1 is characterized in that: The step of marking samples of the optical satellite remote sensing data and constructing a sample library of potential hazards of foreign objects hanging on power transmission channels includes: Using a geospatial database as support, we collected real data on potential sources of foreign objects hanging on transmission lines in multiple scenarios. Using remote sensing software, we extracted the characteristic information of these sources from remote sensing images using visual interpretation methods, generating images and marker files. The images and marker files are cropped into image samples of uniform size and randomly divided into training set, validation set, and test set according to the preset sample ratio.

4. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 1 is characterized in that: The improved GAN network includes a generator and a discriminator; A dual-channel attention mechanism is integrated into the SPADE residual module of the generator, and the convolution layer of the generator is set as a depth-separable convolution layer; The discriminator incorporates a spectral normalization scheme.

5. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 4 is characterized in that: The dual-channel attention mechanism is configured to include an upper and lower branch, wherein the upper branch introduces a mixed pooling module, and the lower branch introduces a small sample adaptive attention module; The hybrid pooling module includes a strip pooling unit and a pyramid pooling unit; the strip pooling unit is used to pool samples in the input sample library in the horizontal and vertical directions to capture the characteristics of long strip objects; the pyramid pooling unit is used to aggregate scene information of different regions of the samples in the input sample library, analyze multi-scale scene feature information, and obtain effective global information of pixel-level scene annotation; The small sample adaptive attention module is used to automatically adjust the different levels of attention to the relevant features corresponding to each type of foreign object hanging line hidden danger source through adaptive learning of samples in the input sample library.

6. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 5 is characterized in that: The pooling formulas for the horizontal and vertical directions of the strip pooling unit are: Where x i,j Indicates the value at position (i, j) in the feature map corresponding to the input sample, y s and y v Represents the pooling results along the horizontal and vertical directions respectively, x i,j ∈R S×V ,y s ∈R S ,y v ∈R V , V and H represent the pooling lengths in the horizontal and vertical directions respectively, R S×V represents a real matrix of dimension S×V, R S represents an S-dimensional real vector, R V Represents a real vector of dimension V.

7. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 5, characterized in that: The small sample adaptive attention module is configured as follows: Inputting the sample information in the sample library into the SENet structure to convolute the features of the sample information into a multi-channel two-dimensional feature map; wherein the number of channels of the multi-channel two-dimensional feature map is the same as the number of types of hidden danger sources of foreign objects hanging on the wires; Compressing the multi-channel two-dimensional feature map into a feature vector by global average pooling; constructing, by a fully connected layer, the correlation between the channels in the multi-channel two-dimensional feature map using the feature vector, and automatically learning the feature weight of each channel based on the correlation; The learned feature weights are normalized, and then the normalized weights are weighted to the features of each channel, so as to adaptively learn the low-correlation features and high-correlation features corresponding to each type of foreign object hanging line hazard source.

8. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 4 is characterized in that: The convolution operation of the depth-wise separable convolution includes depth-wise convolution and point-wise convolution.

9. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 1, characterized in that: The edge information enhancement module is used to: Use frequency domain processing to convert remote sensing images into frequency domain through Fourier transform and then perform frequency domain filtering; A Butterworth filter transfer function is designed to process the frequency domain features of the filtered remote sensing image in the frequency domain space, weaken the low-frequency components, retain the edge information of the high-frequency components of the hidden danger source of foreign objects hanging on the wire, and suppress noise points with excessively high frequencies, thereby obtaining a processed spectrum graph; The processed spectrum is transformed back to the spatial domain by inverse Fourier transform to obtain an edge-enhanced image; Among them, the Butterworth bandpass filter formula is as follows: Among them, H(u,v) is the transfer function, D(u,v) represents the distance from point (u,v) to the center point in the frequency domain, where D O is the cutoff frequency, W is the bandwidth; D O is 20 to 50 pixels, and W is 10 to 40 pixels.

10. The method for monitoring hidden danger sources of foreign objects hanging on power transmission channels according to claim 1, characterized in that: Collect historical data on foreign objects hanging on power lines in the monitoring area, historical meteorological data, geographical location information of transmission lines, and spatial data on potential hazards of foreign objects hanging on power lines. A semi-quantitative assessment method is used to construct a risk assessment model for potential hazards of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data based on the collected data, including: Based on the historical data of foreign objects hanging on the line, determine the risk level value R corresponding to each historical date; Based on the spatial data of potential hazards of foreign objects hanging on wires, the relative density factors ReD of various potential hazards of foreign objects hanging on wires corresponding to each historical date are determined; Based on the spatial data of the hidden danger sources of foreign objects hanging on the power lines and the geographical location information of the transmission lines, the hidden danger proximity distance factor VeD corresponding to each historical date is determined; Based on historical meteorological data, determine the wind speed risk factor Win corresponding to each historical date; Based on historical meteorological data and transmission line geographic location information data, the wind direction risk factor WDir corresponding to each historical date is determined; Based on R, ReD, VeD, Win, and WDir corresponding to each historical date, the weight factors a, b, c, and d corresponding to ReD, VeD, Win, and WDir are determined respectively, and the risk assessment model shown below is obtained: R=ReD×bVeD×cWin×dWDir.

11. A device for monitoring hidden danger sources of foreign objects hanging on power transmission channels using the method according to any one of claims 1 to 10, characterized in that: include: Acquisition module, used to collect optical satellite remote sensing data of the monitoring area; A sample construction module is used to perform sample labeling on the optical satellite remote sensing data to construct a sample library of potential hazards of foreign objects hanging on power transmission channels; A sample enhancement module is used to enhance the sample library data using a preset improved GAN network to obtain an enhanced sample library; An identification training module is used to input samples in the enhanced sample library into a semantic segmentation network integrated with an edge information enhancement module for training, thereby obtaining a trained semantic segmentation network as a hidden danger source identification model; The assessment model construction module is used to collect historical data on foreign objects hanging on power lines, meteorological data, and basic power information data in the monitoring area. Based on the collected data, a semi-quantitative assessment method is used to construct a risk assessment model for the hidden danger source of foreign objects hanging on power lines in transmission channels based on dynamic meteorological data; The monitoring module uses the hidden danger source identification model to extract the hidden danger sources of foreign objects hanging on the transmission channel from the current optical satellite remote sensing data. Based on the extraction results, current meteorological data and the current geographical location information of the transmission line, the risk assessment model is used to perform daily dynamic monitoring and early warning of hidden dangers of foreign objects hanging on the transmission channel.

12. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 10.

Images