CN118818222B - Power grid space position analysis method combining GIS service and artificial intelligence technology - Google Patents
Power grid space position analysis method combining GIS service and artificial intelligence technology Download PDFInfo
- Publication number
- CN118818222B CN118818222B CN202411308350.4A CN202411308350A CN118818222B CN 118818222 B CN118818222 B CN 118818222B CN 202411308350 A CN202411308350 A CN 202411308350A CN 118818222 B CN118818222 B CN 118818222B
- Authority
- CN
- China
- Prior art keywords
- power grid
- image
- grid line
- shape
- shape feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 30
- 238000004458 analytical method Methods 0.000 title claims abstract description 25
- 238000005516 engineering process Methods 0.000 title claims abstract description 25
- 238000001514 detection method Methods 0.000 claims abstract description 78
- 230000002159 abnormal effect Effects 0.000 claims abstract description 38
- 238000000034 method Methods 0.000 claims abstract description 34
- 230000005856 abnormality Effects 0.000 claims abstract description 31
- 238000007689 inspection Methods 0.000 claims abstract description 18
- 238000000605 extraction Methods 0.000 claims description 25
- 230000004927 fusion Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 11
- 230000009466 transformation Effects 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 238000002372 labelling Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000012886 linear function Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012876 topography Methods 0.000 description 2
- 239000011800 void material Substances 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 208000025274 Lightning injury Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000004643 material aging Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 230000036561 sun exposure Effects 0.000 description 1
Abstract
The application provides a power grid space position analysis method combining GIS service and artificial intelligence technology, which comprises the following steps: receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection, and a power grid line identifier associated with the to-be-detected image, carrying out anomaly detection on the to-be-detected image through a power grid line anomaly detection model, and acquiring association information of the anomaly line section through a power grid GIS platform based on the anomaly line section; on one hand, the method adopts the RT-DETR model fused with the shape enhancement module to construct a power grid line anomaly detection model, and has higher adaptability, robustness and accuracy for anomaly detection based on line images; on the other hand, by combining GIS service with artificial intelligence technology, accurate positioning of abnormal line sections can be obtained, rapid matching of abnormal line section associated information is realized, and more comprehensive information support is provided for power grid line abnormality analysis and decision.
Description
Technical Field
The application belongs to the technical field of power grid line detection, and particularly relates to a power grid space position analysis method based on an artificial intelligence technology.
Background
The power system is one of the infrastructures in the modern society, but with the continuous expansion of the power grid scale, the power grid area is wide, the operation environment is complex and changeable, and the power grid faces a plurality of potential safety hazards and fault risks in the operation process. The power grid line, especially the overhead line in the field, is in unstable weather for many years, and is subjected to severe weather such as sun exposure, rain, snow, hail and the like, and equipment material aging, corrosion, icing, lightning stroke and the like usually occur, and if the faults of the power transmission line and the equipment are not repaired and treated in time, great potential power supply hazards are brought.
In recent years, the rapid development of artificial intelligence technology provides a new solution for detecting power grid line anomalies. The artificial intelligence technology, in particular to machine learning, deep learning, neural network and the like, can find potential rules and abnormal modes in the power grid through training and analysis of a large amount of historical data, thereby realizing high-efficiency detection of power grid line abnormality. However, the existing detection method based on artificial intelligence still has limitations in terms of operation efficiency and detection accuracy, which limit the effectiveness of the detection method in practical application, and influence the quick response and effective treatment of the power grid line abnormality.
Disclosure of Invention
The application provides a power grid space position analysis method based on an artificial intelligence technology, which is used for improving the power grid line anomaly detection efficiency and accuracy.
In a first aspect, the present application provides a method for analyzing a spatial position of a power grid based on an artificial intelligence technique, including:
Step S1: receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection and a power grid line identifier associated with the to-be-detected image;
step S2: and carrying out anomaly detection on the image to be detected through the power grid line anomaly detection model.
The step S2 includes:
Step S21: acquiring historical image sample data of a power grid line acquired by unmanned aerial vehicle inspection to construct a data set, and marking a line abnormal target contained in an image sample in the data set;
step S22: constructing a power grid line anomaly detection model based on the improved RT-DETR model;
step S23: training the power grid line abnormality detection model based on the data set to obtain a trained power grid line abnormality detection model;
Step S24: inputting a power grid line to-be-detected image shot by unmanned aerial vehicle inspection into a trained power grid line abnormality detection model to obtain a power grid line abnormality detection result; the power grid line abnormality detection result comprises a line abnormality category and a line abnormality positioning frame;
Step S25: if the power grid line is detected to be abnormal, a specific abnormal line section is determined.
Optionally, the step S22 includes:
step S22.1: extracting the characteristics of an input image by using a backbone network, and outputting three characteristic graphs with different dimensions { S3, S4, S5}, wherein the characteristic graphs are used as the input of a hybrid encoder;
Step S22.2: the internal scale feature interaction module based on the attention mechanism in the AIFI of the hybrid encoder extracts attention features of the S5 feature map to obtain an output F5, and performs feature fusion of different scales on { S3, S4 and F5} through the CCFM-based cross-scale feature fusion module to output a multi-scale feature sequence;
Step S22.3: a shape enhancement module is constructed, gray level processing is carried out on an input image to obtain a gray level image, the gray level image is input into the shape enhancement module to extract shape features of the image to obtain a target shape feature sequence, and the target shape feature sequence is spliced with a multi-scale feature sequence output by a CCFM (code division multiple frequency modulation) to obtain a fusion feature sequence, wherein the fusion feature sequence is used as an image feature sequence output by a hybrid encoder;
step S22.4: and adopting IoU perception query selection module, selecting the characteristics with high classification score and high IoU score to initialize the object query of the decoder based on the image characteristic sequence output by the hybrid encoder during model training, adopting a transducer decoder structure, generating the final object query based on the image characteristic sequence from the hybrid encoder and the initialized object query through iterative optimization, and sending the output of the decoder into a prediction head, wherein the prediction head comprises a classification prediction head and a bounding box regression prediction head which are respectively used for target classification prediction and bounding box coordinate prediction.
In some embodiments, the shape enhancement module includes an edge feature extraction unit and a shape feature extraction unit, the shape feature extraction unit including a first convolution layer, a second void convolution layer, and a third void convolution layer in parallel;
The edge feature extraction unit is used for extracting edge features of the gray level image by adopting a gradient transformation algorithm to obtain an edge feature image F e;
The shape feature extraction unit is used for extracting shape features of the edge feature map F e through the first convolution layer, the second cavity convolution layer and the third cavity convolution layer which are parallel to each other to obtain a first shape feature map, a second shape feature map and a third shape feature map, and performing concat splicing of space dimensions after the first shape feature map, the second shape feature map and the third shape feature map are flattened respectively to obtain a target shape feature sequence.
In the first convolution layer, usingEach convolution kernelRespectively convolving the edge feature map F e to obtainA first shape feature matrix to beStacking the first shape feature matrixes to obtain a first shape feature diagram; wherein the convolution step size stride=I.e. the first convolution layer convolution step and the convolution kernelThe sizes are the same;
In the second hole convolution layer, using Each convolution kernelRespectively carrying out cavity convolution on the edge feature map F e to obtainA second shape feature matrix to beStacking the second shape feature matrixes to obtain a second shape feature graph, wherein the expansion rate of the second cavity convolution layer is as follows；
In the third hole convolution layer, usingEach convolution kernelRespectively carrying out cavity convolution on the edge feature map F e to obtainA third shape feature matrix to beStacking the third shape feature matrixes to obtain a third shape feature diagram, wherein the expansion rate of the third cavity convolution layer is as follows，＞；
Flattening the first shape feature map, the second shape feature map and the third shape feature map respectively, and then performing concat splicing of space dimension to obtain a target shape feature sequence;
wherein the convolution kernel The calculation formula is as follows:
Wherein 1 < > ≤，Is an integer of the number of the times,Represent the firstEach convolution kernelMiddle (f)Line 1The elements of the column are arranged such that,In (a)The range of the values of (a) is all，Representing the convolution kernel size.
The edge feature extraction unit extracts the gray scale image edge feature by adopting a gradient transformation algorithm, wherein the gradient transformation algorithm is a sobel operator.
Optionally, in the first, second, or third hole convolution layers, edge filling is performed on the edge feature map prior to convolution.
In a second aspect, the present application provides a method for analyzing a spatial position of a power grid by combining GIS service and artificial intelligence technology, and the present application further includes the following steps on the basis of the first embodiment:
Step S3: and acquiring the associated information of the abnormal line section through a power grid GIS platform based on the abnormal line section.
In addition, the application also provides a power grid space position analysis system based on the artificial intelligence technology, which comprises the following steps: the system comprises a data receiving module, an image detection module and a spatial position analysis module;
the data receiving module is used for receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection and a power grid line identifier associated with the to-be-detected image;
The image detection module is used for carrying out anomaly identification on the image to be detected through the power grid line anomaly detection model;
Further optionally, the system further comprises a spatial position analysis module, which is used for acquiring the association information of the abnormal line section through a power grid GIS platform based on the abnormal line section.
Compared with the prior art, the application has the following advantages:
The application provides a power grid space position analysis method based on an artificial intelligence technology, which is used for constructing a power grid line abnormity detection model based on an improved RT-DETR model, on one hand, the adoption of the RT-DETR model can capture multi-scale characteristics from low level to high level while maintaining the calculation efficiency, and the detection performance is improved; on the other hand, in the actual scene of the abnormal detection of the power grid line, the special shape characteristics of the power grid line are considered, in order to improve the accuracy and the efficiency of target detection, the shape characteristics of the power grid line detection image are extracted through the shape enhancement module and are fused with the multi-scale characteristics extracted by the RT-DETR, the shape information of the power grid line detection image is enhanced, and the detection and identification capability of the model on the abnormal power grid line is enhanced. Therefore, the method adopts the RT-DETR model fused with the shape enhancement module to construct the power grid line anomaly detection model, and has higher adaptability, robustness and accuracy for line image-based anomaly detection compared with a general RT-DETR model.
In addition, the application also provides a power grid space position analysis method combining the GIS service and the artificial intelligence technology, by combining the GIS service and the artificial intelligence technology, the accurate positioning of the abnormal line section can be obtained, the rapid matching of the related information of the abnormal line section can be realized, and more comprehensive information support is provided for the power grid line abnormality analysis and decision.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.
Fig. 1 is a schematic flow chart of a power grid spatial position analysis method based on an artificial intelligence technology according to an embodiment of the present application.
FIG. 2 is a schematic diagram of an improved RT-DETR model provided by an embodiment of the application.
Fig. 3 is a schematic flow chart of a method for detecting an abnormality of an image to be detected through a power grid line abnormality detection model according to an embodiment of the present application.
Fig. 4 is a schematic flow chart of a method for constructing a power grid line anomaly detection model based on an improved RT-DETR model according to an embodiment of the present application.
Fig. 5 is a schematic flow chart of a method for analyzing a spatial position of a power grid by combining a GIS service and an artificial intelligence technology according to an embodiment of the present application.
Specific embodiments of the present application have been shown by way of the above drawings and will be described in more detail below. The drawings and the written description are not intended to limit the scope of the inventive concepts in any way, but rather to illustrate the inventive concepts to those skilled in the art by reference to the specific embodiments.
Detailed Description
The application is further described below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present application, and are not intended to limit the scope of the present application.
Term interpretation:
grid GIS platform: the power grid GIS system integrates and manages various resource information of power enterprises, including power equipment, towers, power transmission and distribution networks and the like, and natural environment information of mountainous, topography, towns, roads, weather, hydrology, geology, resources and the like through a Geographic Information System (GIS) technology to form a unified informationized production management platform.
RT-DETR (Real-Time DEtection TRansformer) model: in 2023, the hundred degree research and development team in The paper DETRs Beat YOLOs on Real-time Object Detection published by The international top-level conference has proposed an RT-DETR model that, as a real-time end-to-end detector based on DETR architecture, achieves SOTA (State-Of-The-Art) performance in terms Of speed and accuracy.
The architecture of the RT-DETR model includes ResNet or HGNetv as a backbone network, a hybrid encoder, and a transducer decoder with auxiliary pre-header. The model processes the multi-scale features through the hybrid encoder, and optimizes the initial object query through IoU-aware query selection technology, thereby improving the performance of the model.
As shown in fig. 1, a first embodiment of the present application provides a method for analyzing a spatial location of a power grid based on an artificial intelligence technique, including:
Step S1: receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection and a power grid line identifier associated with the to-be-detected image;
It can be appreciated that the unmanned aerial vehicle usually performs flight inspection according to a planned path and hovers over a pre-planned line section to be detected, so that each image to be detected acquired by the unmanned aerial vehicle is associated with a corresponding line section to be detected identifier.
Preferably, in the unmanned aerial vehicle inspection process, the unmanned aerial vehicle is positioned above a power grid line to perform nodding, and the inclination angle can be set in the range of 0-30 degrees.
Step S2: and carrying out anomaly detection on the image to be detected through the power grid line anomaly detection model.
As shown in fig. 3, the step S2 includes:
step S21: and acquiring historical image sample data of the power grid lines acquired by unmanned aerial vehicle inspection to construct a data set, and marking a line abnormality target contained in an image sample in the data set.
The power grid line historical image sample comprises power grid line images acquired for different line sections during unmanned aerial vehicle inspection.
And labeling the targets with abnormal lines contained in the image samples in the data set to generate labeling files corresponding to the image samples. The line anomaly category may be set according to a type of fault common to the grid line, for example, the grid line anomaly category includes line breakage, line cracking, line slush, tree barriers, and the like.
Labeling a line abnormality target contained in an image sample in the dataset through LabelImg labeling tools to generate a labeling file corresponding to the image sample; the annotation file comprises a line anomaly class label and fault point boundary frame coordinate information. Specifically, the marking tool LabelImg is used for marking fault point examples of each image in the dataset, the positions of all fault point examples in each image are accurately marked, and marking information of the corresponding image is generated in a txt file form, wherein the marking information comprises line anomaly class labels and fault point boundary frame coordinates.
Further alternatively, to improve the robustness of the model, to prevent overfitting of the results, the image samples may be processed to augment the data set using data enhancement techniques including random clipping, random flipping, random dimensional changes, and the like.
Step S22: and constructing a power grid line anomaly detection model based on the improved RT-DETR model.
As an alternative implementation mode, an RT-DETR model of a fused shape enhancement module is adopted to construct a power grid line anomaly detection model. On the one hand, the RT-DETR model is used as an end-to-end target detection model, and is superior to the existing target detection model in calculation speed and detection accuracy, so the embodiment selects the RT-DETR model as a basic model. On the other hand, in the actual scene of the abnormal detection of the power grid line, the special shape characteristics of the power grid line are considered, in order to improve the accuracy and the efficiency of target detection, the shape characteristics of the power grid line detection image are extracted through the shape enhancement module and are fused with the multi-scale characteristics extracted by the RT-DETR, the shape information of the power grid line detection image is enhanced, and the detection and identification capability of the model on the abnormal power grid line is enhanced. Therefore, the method optimizes the RT-DETR model by introducing the shape enhancement module, and builds the power grid line anomaly detection model by adopting the RT-DETR model fused with the shape enhancement module, so that compared with the general RT-DETR model, the method has higher adaptability, robustness and accuracy for anomaly detection based on power grid line detection images.
Specifically, as shown in fig. 2 and 4, the step S22 includes:
Step S22.1: a backbone network is used for extracting the characteristics of the input image, and three characteristic diagrams with different dimensions { S3, S4, S5} are output as the input of the hybrid encoder.
Where the input image is an RGB profile, the backbone network may employ ResNet or HGNetv networks, for example.
The backbone network is used for receiving an input image and carrying out feature extraction to generate feature graphs with different scales, wherein { S3, S4 and S5} are sequentially carried out from low dimension to high dimension, the highest dimension feature graph S5 contains the most abundant global semantic information and is suitable for large target detection, and the other dimension feature graphs S3 and S4 contain detail information, and the high resolution characteristic of the feature graph S5 enables the feature graph to be suitable for small target detection;
Step S22.2: carrying out attention feature extraction on the S5 feature map by AIFI (an internal scale feature interaction module based on an attention mechanism) in the hybrid encoder to obtain an output F5 feature map, and carrying out feature fusion on { S3, S4 and F5} on different scales by using a CCFM (a trans-scale feature fusion module of CNN) to output a multi-scale feature sequence;
Step S22.3: the method comprises the steps of constructing a shape enhancement module, carrying out gray processing on an input image to obtain a gray image, carrying out shape feature extraction on the image in the gray image input shape enhancement module to obtain a target shape feature sequence, carrying out concat splicing of channel dimensions on the target shape feature sequence and a multi-scale feature sequence output by a CCFM (cross-scale feature fusion module of CNN) to obtain a fusion feature sequence, and taking the fusion feature sequence as an image feature sequence output by an encoder.
In some embodiments, the shape enhancement module comprises: the edge feature extraction unit and the shape feature extraction unit comprise a first convolution layer, a second cavity convolution layer and a third cavity convolution layer which are parallel; the edge feature extraction unit is used for extracting edge features of the gray level image by adopting a gradient transformation algorithm to obtain an edge feature image F e. The shape feature extraction unit is used for extracting shape features of the edge feature map F e through the first convolution layer, the second cavity convolution layer and the third cavity convolution layer which are parallel to each other to obtain a first shape feature map, a second shape feature map and a third shape feature map, and performing concat splicing of space dimensions after the first shape feature map, the second shape feature map and the third shape feature map are flattened respectively to obtain a target shape feature sequence.
Firstly, when the edge feature extraction unit extracts the edge feature of the gray level image, a gradient transformation algorithm is adopted to obtain an initial edge feature matrixThe gradient transformation algorithm is a Sobel operator, and a convolution kernel constructed by the Sobel operator is as follows:
Where a is the convolution kernel that computes G (x) and b is the convolution kernel that computes G (y).
Initial edge feature matrixGray value of each element in (a)The calculation formula is as follows:
wherein, G (x) is a gray value of the gray image subjected to gradient transformation in the x direction, that is, a result value of the convolution of the gray image by the convolution kernel a, and G (y) is a gray value of the gray image subjected to gradient transformation in the y direction, that is, a result value of the convolution of the gray image by the convolution kernel b.
For initial edge feature matrixNormalization and binarization are performed to obtain a target edge feature matrix G, which is used to characterize the edge feature map F e as can be appreciated.
Secondly, in order to extract multi-scale shape feature information, the shape feature extraction unit is used for extracting shape features of the edge feature map Fe through the first convolution layer, the second cavity convolution layer and the third cavity convolution layer which are parallel;
In the first convolution layer, using Each convolution kernelRespectively convolving the edge feature map F e to obtainA first shape feature matrix to beStacking the first shape feature matrixes to obtain a first shape feature diagram; wherein the convolution step size stride=；
In the second hole convolution layer, usingEach convolution kernelRespectively carrying out cavity convolution on the edge feature map F e to obtainA second shape feature matrix to beStacking the second shape feature matrixes to obtain a second shape feature graph, wherein the expansion rate of the second cavity convolution layer is as follows；
In the third hole convolution layer, usingEach convolution kernelRespectively carrying out cavity convolution on the edge feature map F e to obtainA third shape feature matrix to beStacking the third shape feature matrixes to obtain a third shape feature diagram, wherein the expansion rate of the third cavity convolution layer is as follows，＞；
And flattening the first shape feature map, the second shape feature map and the third shape feature map respectively, and then performing concat splicing of space dimension to obtain a target shape feature sequence.
In order to maintain the consistency of the features of different dimensions and facilitate the subsequent channel dimension concat splicing operation, the width-height dimensions of the first shape feature map, the second shape feature map and the third shape feature map respectively correspond to the dimensions of the feature maps S3, S4 and F5.
Wherein the convolution kernelThe calculation formula is as follows:
Wherein 1 < > ≤，Is an integer of the number of the times,Represent the firstEach convolution kernelMiddle (f)Line 1The elements of the column are arranged such that,In (a)The range of the values of (a) is all，Representing the convolution kernel size.
Optionally, in the first, second or third hole convolution layers, edge filling may be performed on the edge feature map prior to convolution, typically with zero-valued padding, i.e., adding a circle of zero-valued pixels around the boundary of the image, in order to facilitate subsequent concat stitching operations.
Wherein the size of the convolution kernelAnd expansion ratio、Can be adjusted according to different application scenes. For example, convolution kernel sizeCan be set to, for example, 3x3 or 8 x 8, hole convolution rateLess than，、The values of 2,3 or 6 can be taken to adapt to different image processing requirements.
It will be appreciated that the number of components,ComprisesThe linear functions are as follows:
thus, the first and second substrates are bonded together, Essentially, the linear function isThe parameters are equally spaced to be takenAnd each. Wherein, The range of the values is，≥16。
For the first convolution layer, calculating a first shape feature matrix by convolutionThe following formula is shown:
wherein, Representing a first shape feature matrixMiddle (f)Line 1The elements of the column are arranged such that,The first edge feature matrix G of the edge feature map F e Line 1The elements of the column are arranged such that,Is the convolution kernel size.
As an example, in step S22.1, the size of the input image is h×w×c which is 576×576×3, and three feature maps S3, S4, S5 of different dimensions are output after feature extraction of the input image through the backbone network, and the feature map sizes h×w×c are 72×72×256, 36×36×256, and 18×18×256, respectively; in step S22.2, feature fusion of different scales is carried out through CCFM to output a multi-scale feature sequence, wherein the size is 6804 multiplied by 256;
In step S22.3, the input image is sequentially subjected to gradation processing and edge feature extraction, and the dimension h×w×c of the edge feature map is 576×576×1, and the convolution kernel is obtained The size of (c) is set to 8 x 8,Taking 32, in the first convolution layer, the convolution step size stride=8, and the size h×w×c of the obtained first feature map is 72×72×32;
In the second hole convolution layer, the expansion ratio is 3, the convolution step size stride=16, in order to make the second feature map size correspond to the width-height size of the feature map S4, firstly, edge filling padding=3 is performed on the edge feature map, that is, 3 circles of pixels with zero values are added around the boundary of the edge feature map, and after the edge feature map after edge filling is subjected to hole convolution, the size of the second feature map is 36×36×32;
In the third hole convolution layer, the expansion ratio is 6, the convolution step size stride=32, in order to make the second feature image size correspond to the width-height size of the feature image S4, firstly, edge filling is performed on the edge feature image, 5 columns of pixels with zero values and 6 columns of pixels with zero values are respectively added on the left boundary and the right boundary of the edge feature image, 5 rows of pixels with zero values and 6 rows of pixels with zero values are respectively added on the left boundary and the right boundary, and the size of the second feature image is 18×18×32 after the hole convolution is performed on the edge feature image after edge filling;
And performing concat splicing of space dimension after flattening the first shape feature map, the second shape feature map and the third shape feature map respectively to obtain a target shape feature sequence, wherein the size is 6804 multiplied by 32, and performing concat splicing of channel dimension with a multi-scale feature sequence output by CCFM to obtain a fusion feature sequence, and the size is 6804 multiplied by 288.
It will be appreciated that the sequence of target shape features essentially represents the shape features consisting of different straight lines present in the greyscale map. In the step, a plurality of convolution kernels are generated through linear functions of various different inclination angles, shape features composed of different straight lines in the edge feature map are identified and extracted according to the different convolution kernels, and feature information of images on a plurality of scales is captured through adding a cavity convolution layer, so that the model can process objects of different sizes, and the receptive field of the model is widened. Obviously, the number of convolution kernels in the shape feature extraction processThe larger the number of (c) the more informative the shape feature captured, but the slower the training time.
Step S22.4: and adopting IoU a perception query selection module, selecting characteristics with high classification scores and high IoU scores to initialize object queries of a decoder based on an image characteristic sequence output by the encoder during model training, adopting a transducer decoder structure, generating a final object query based on the image characteristic sequence from the encoder and the initialized object queries through iterative optimization, and sending the output of the decoder into a prediction head, wherein the prediction head comprises a classification prediction head and a bounding box regression prediction head which are respectively used for target classification prediction and bounding box coordinate prediction.
Step S23: training the power grid line abnormality detection model based on the data set to obtain a trained power grid line abnormality detection model.
The input image is adjusted to a uniform resolution H x W x C to ensure consistency of model inputs. For example, the input image resolution is unified to 576×576×3, the training round number (epochs) can be set to 200 during training, the learning rate is set to 0.001 with Adam optimizer, and the training size per batch (batch) is 8.
Step S24: and inputting the image to be detected of the power grid line shot by the unmanned aerial vehicle inspection into a trained power grid line abnormality detection model to obtain a power grid line abnormality detection result.
The power grid line abnormality detection result comprises a line abnormality category and a line abnormality positioning frame.
Step S25: if the power grid line is detected to be abnormal, a specific abnormal line section is determined.
Specifically, when the existence of the abnormal power grid line is detected, a specific abnormal line section is determined according to the association relation between the power grid line to-be-detected image and the power grid line identifier, which are shot by the unmanned aerial vehicle inspection.
In the embodiment of the application, the RT-DETR model of the fusion shape enhancement module is adopted to construct the power grid line abnormity detection model, on one hand, the RT-DETR model can be adopted to capture multi-scale characteristics from low level to high level while maintaining the calculation efficiency, and the detection performance is improved; on the other hand, in the actual scene of the abnormal detection of the power grid line, the special shape characteristics of the power grid line are considered, in order to improve the accuracy and the efficiency of target detection, the shape characteristics of the power grid line detection image are extracted through the shape enhancement module and are fused with the multi-scale characteristics extracted by the RT-DETR, the shape information of the power grid line detection image is enhanced, and the detection and identification capability of the model on the abnormal power grid line is enhanced. Therefore, the method adopts the RT-DETR model fused with the shape enhancement module to construct the power grid line anomaly detection model, and has higher adaptability, robustness and accuracy for line image-based anomaly detection compared with a general RT-DETR model.
The second embodiment of the present application provides a method for analyzing a spatial position of a power grid by combining GIS service and artificial intelligence technology, as shown in fig. 5, and in the second embodiment, the present application further includes the following steps based on the first embodiment:
Step S3: and acquiring the associated information of the abnormal line section through a power grid GIS platform based on the abnormal line section.
The association information at least comprises space positioning information, ledger information and environment information of the abnormal line section.
Typically, the grid GIS platform data includes the information of the power equipment, towers, the station account of the power transmission and distribution network, the information of space location, and the environmental information of mountains, terrains, towns, roads, weather, hydrology, geology, resources and the like.
Therefore, when the existence of the power grid line abnormality is detected through the unmanned aerial vehicle inspection image, the related information of the abnormal line section, such as the space positioning information of the abnormal line section, the environmental information of nearby weather, trees, topography and the like, and the account information of the line tower model, voltage level, historical maintenance record and the like, can be obtained through the power grid GIS platform, so that more comprehensive information support is provided for power grid line abnormality analysis and decision.
A third embodiment of the present application provides an artificial intelligence technology-based power grid spatial location analysis system for operating the artificial intelligence technology-based power grid spatial location analysis method as provided in the first embodiment of the present application, the system including: the device comprises a data receiving module and an image detecting module;
the data receiving module is used for receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection and a power grid line identifier associated with the to-be-detected image;
the image detection module is used for carrying out anomaly detection on the image to be detected through the power grid line anomaly detection model;
Further optionally, the system further comprises a spatial position analysis module, which is used for acquiring the association information of the abnormal line section through a power grid GIS platform based on the abnormal line section.
Embodiments of the present application also provide an electronic device that may include one or more processors of a processing core, one or more memories of a computer-readable storage medium, a communication component, and the like. Wherein the processor, the memory and the communication means are connected by a bus.
In a particular implementation, at least one processor executes computer-executable instructions stored in a memory, causing the at least one processor to perform a predictive method as described above.
The Processor may be a central processing unit (english: central Processing Unit, abbreviated as CPU), other general purpose processors, a digital signal Processor (english: DIGITAL SIGNAL Processor, abbreviated as DSP), an Application specific integrated Circuit (english: application SPECIFIC INTEGRATED Circuit, abbreviated as ASIC), or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the present invention may be embodied directly in a hardware processor for execution, or in a combination of hardware and software modules in a processor for execution.
The Memory may include high-speed Memory (Random Access Memory, RAM) or may further include Non-volatile Memory (NVM), such as at least one disk Memory.
The bus may be an industry standard architecture (Industry Standard Architecture, ISA) bus, an external device interconnect (PERIPHERAL COMPONENT, PCI) bus, or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, among others. The buses may be divided into address buses, data buses, control buses, etc.
The embodiment of the application also provides a computer readable storage medium, and the computer readable storage medium is stored with computer program instructions which execute the grid space position analysis method provided by the embodiment of the application when being read and run by a processor.
Wherein the storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disk, and the like.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to related descriptions of other embodiments. The technical features of the foregoing embodiments may be arbitrarily combined, and for brevity, all of the possible combinations of the technical features of the foregoing embodiments are not described, however, all of the combinations of the technical features should be considered as being within the scope of the disclosure.
Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
It is to be understood that the application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.
Claims (10)
1. A method for analyzing a spatial position of a power grid by combining a GIS service and an artificial intelligence technology, the method comprising:
Step S1: receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection and a power grid line identifier associated with the to-be-detected image;
Step S2: performing anomaly detection on the image to be detected through a power grid line anomaly detection model;
the step S2 includes:
Step S21: acquiring historical image sample data of a power grid line acquired by unmanned aerial vehicle inspection to construct a data set, and marking a line abnormal target contained in an image sample in the data set;
step S22: constructing a power grid line anomaly detection model based on the improved RT-DETR model;
step S23: training the power grid line abnormality detection model based on the data set to obtain a trained power grid line abnormality detection model;
Step S24: inputting a power grid line to-be-detected image shot by unmanned aerial vehicle inspection into a trained power grid line abnormality detection model to obtain a power grid line abnormality detection result; the power grid line abnormality detection result comprises a line abnormality category and a line abnormality positioning frame;
Step S25: if the power grid line is detected to be abnormal, determining a specific abnormal line section;
wherein, the step S22 includes:
step S22.1: extracting the characteristics of an input image by using a backbone network, and outputting three characteristic graphs with different dimensions { S3, S4, S5}, wherein the characteristic graphs are used as the input of a hybrid encoder;
step S22.2: the internal scale feature interaction module of AIFI in the hybrid encoder is used for carrying out attention feature extraction on the S5 feature map to obtain an output F5 feature map, and carrying out feature fusion on { S3, S4 and F5} on the basis of the CCFM cross-scale feature fusion module to output a multi-scale feature sequence;
Step S22.3: a shape enhancement module is constructed, gray level processing is carried out on an input image to obtain a gray level image, the gray level image is input into the shape enhancement module to extract shape features of the image to obtain a target shape feature sequence, and the target shape feature sequence is spliced with a multi-scale feature sequence output by a CCFM (code division multiple frequency modulation) to obtain a fusion feature sequence, wherein the fusion feature sequence is used as an image feature sequence output by a hybrid encoder;
step S22.4: and adopting IoU perception query selection module, selecting the characteristics with high classification score and high IoU score to initialize the object query of the decoder based on the image characteristic sequence output by the hybrid encoder during model training, adopting a transducer decoder structure, generating the final object query based on the image characteristic sequence from the hybrid encoder and the initialized object query through iterative optimization, and sending the output of the decoder into a prediction head, wherein the prediction head comprises a classification prediction head and a bounding box regression prediction head which are respectively used for target classification prediction and bounding box coordinate prediction.
2. The method for analyzing the spatial position of the power grid by combining the GIS service and the artificial intelligence technology according to claim 1, wherein the shape enhancement module comprises an edge feature extraction unit and a shape feature extraction unit, and the shape feature extraction unit comprises a first convolution layer, a second cavity convolution layer and a third cavity convolution layer which are parallel;
The edge feature extraction unit is used for extracting edge features of the gray level image by adopting a gradient transformation algorithm to obtain an edge feature image F e;
The shape feature extraction unit is used for extracting shape features of the edge feature map F e through the first convolution layer, the second cavity convolution layer and the third cavity convolution layer which are parallel to each other to obtain a first shape feature map, a second shape feature map and a third shape feature map, and performing concat splicing of space dimensions after the first shape feature map, the second shape feature map and the third shape feature map are flattened respectively to obtain a target shape feature sequence.
3. The method for analyzing the spatial position of a power grid combining GIS service and artificial intelligence technology according to claim 2, wherein in said first convolution layer, usingEach convolution kernelRespectively convolving the edge feature map F e to obtainA first shape feature matrix to beStacking the first shape feature matrixes to obtain a first shape feature diagram;
In the second hole convolution layer, using Each convolution kernelRespectively carrying out cavity convolution on the edge feature map F e to obtainA second shape feature matrix to beStacking the second shape feature matrixes to obtain a second shape feature graph, wherein the expansion rate of the second cavity convolution layer is as follows；
In the third hole convolution layer, usingEach convolution kernelRespectively carrying out cavity convolution on the edge feature map F e to obtainA third shape feature matrix to beStacking the third shape feature matrixes to obtain a third shape feature diagram, wherein the expansion rate of the third cavity convolution layer is as follows，＞；
Flattening the first shape feature map, the second shape feature map and the third shape feature map respectively, and then performing concat splicing of space dimension to obtain a target shape feature sequence;
wherein the convolution kernel The calculation formula is as follows:
Wherein 1 < > ≤，Is an integer of the number of the times,Represent the firstEach convolution kernelMiddle (f)Line 1The elements of the column are arranged such that,In (a)The range of the values of (a) is all，Representing the convolution kernel size.
4. The grid space position analysis method combining GIS service and artificial intelligence technology according to claim 2, wherein the edge feature extraction unit extracts gray scale image edge features by using a gradient transformation algorithm, and the gradient transformation algorithm adopts a sobel algorithm.
5. A grid space position analysis method combining GIS services and artificial intelligence techniques according to claim 3, wherein edge filling is performed on edge feature maps before convolution in the first, second or third convolution layers.
6. The method for analyzing the spatial position of a power grid by combining GIS service and artificial intelligence technology according to claim 1, wherein the method further comprises the step of S3: and acquiring the associated information of the abnormal line section through a power grid GIS platform based on the abnormal line section, wherein the associated information at least comprises space positioning information, standing book information and environment information of the abnormal line section.
7. A grid space position analysis system combining GIS services and artificial intelligence techniques for operating a grid space position analysis method combining GIS services and artificial intelligence techniques as claimed in claim 1, comprising: the device comprises a data receiving module and an image detecting module;
the data receiving module is used for receiving a power grid line to-be-detected image shot by unmanned aerial vehicle inspection and a power grid line identifier associated with the to-be-detected image;
The image detection module is used for carrying out anomaly detection on the image to be detected through the power grid line anomaly detection model.
8. The grid space location analysis system combining GIS services and artificial intelligence techniques of claim 7, further comprising: and the space position analysis module is used for acquiring the associated information of the abnormal line section through the power grid GIS platform based on the abnormal line section.
9. An electronic device, comprising: a memory and a processor, the memory having stored therein computer program instructions that, when read and executed by the processor, perform the method of any of claims 1-6.
10. A computer readable storage medium, having stored thereon computer program instructions which, when read and executed by a processor, perform the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411308350.4A CN118818222B (en) | 2024-09-19 | Power grid space position analysis method combining GIS service and artificial intelligence technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411308350.4A CN118818222B (en) | 2024-09-19 | Power grid space position analysis method combining GIS service and artificial intelligence technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118818222A CN118818222A (en) | 2024-10-22 |
| CN118818222B true CN118818222B (en) | 2024-11-15 |
Family
ID=
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109765462A (en) * | 2019-03-05 | 2019-05-17 | 国家电网有限公司 | Fault detection method, device and the terminal device of transmission line of electricity |
| CN115984222A (en) * | 2023-01-05 | 2023-04-18 | 北京国网富达科技发展有限责任公司 | Power distribution network overhead line defect detection method and system |
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109765462A (en) * | 2019-03-05 | 2019-05-17 | 国家电网有限公司 | Fault detection method, device and the terminal device of transmission line of electricity |
| CN115984222A (en) * | 2023-01-05 | 2023-04-18 | 北京国网富达科技发展有限责任公司 | Power distribution network overhead line defect detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN113378686B (en) | Two-stage remote sensing target detection method based on target center point estimation | |
| Zhou et al. | Octr: Octree-based transformer for 3d object detection | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN111680556B (en) | Method, device, equipment and storage medium for identifying traffic gate vehicle type | |
| CN116994140A (en) | Cultivated land extraction method, device, equipment and medium based on remote sensing image | |
| CN111738165A (en) | Method for extracting individual plant canopy from high-resolution unmanned aerial vehicle visible light remote sensing image | |
| Nordeng et al. | DEBC detection with deep learning | |
| CN113761999A (en) | Target detection method and device, electronic equipment and storage medium | |
| CN114519819B (en) | Remote sensing image target detection method based on global context awareness | |
| CN111611933B (en) | Information extraction method and system for document image | |
| CN117557775B (en) | Substation power equipment detection method and system based on infrared and visible light fusion | |
| CN116168246A (en) | Method, device, equipment and medium for identifying waste slag field for railway engineering | |
| Li et al. | Pixel‐Level Recognition of Pavement Distresses Based on U‐Net | |
| CN117496223A (en) | Light insulator defect detection method and device based on deep learning | |
| CN116385911A (en) | Lightweight target detection method for unmanned aerial vehicle inspection insulator | |
| CN115908786A (en) | Electrical cabinet grounding cable abnormity identification method and system based on deep learning | |
| CN113591614B (en) | Remote sensing image road extraction method based on close-proximity spatial feature learning | |
| CN118818222B (en) | Power grid space position analysis method combining GIS service and artificial intelligence technology | |
| CN113569600A (en) | Method and device for identifying weight of object, electronic equipment and storage medium | |
| CN118818222A (en) | Power grid space position analysis method combining GIS service and artificial intelligence technology | |
| Mao et al. | City object detection from airborne Lidar data with OpenStreetMap‐tagged superpixels | |
| Li et al. | Focus on local: transmission line defect detection via feature refinement | |
| CN114882292B (en) | Remote sensing image ocean target identification method based on cross-sample attention mechanism graph neural network | |
| Ibrahim et al. | Challenging Weather for License Plate Detection: Impaired visibility |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |