Method for automatically identifying on-site environment characteristics of distribution network operation without power failure

Abstract

The invention belongs to the technical field of distribution networks, and particularly relates to a distribution network uninterruptible operation site environment feature automatic identification method, which comprises the following specific steps: judging an image acquired by an unmanned aerial vehicle to judge whether fog exists in the image, if so, entering a step 2, if not, entering a step 3, and if not, entering a step 4; the method comprises the following steps: step 1, setting reference points, carrying out defogging operation on an image to ensure that the image is clear, and after the defogging operation is carried out, entering step 2, establishing initial data of a working site, extracting features in the image, and converting lines of the features into a plurality of points. According to the method and the device, the position distance and the position relation between the reference point and the feature are acquired, and the position distance and the position relation are compared with the initial data of the operation site to realize feature identification, so that the problem that personnel need to identify one by one when the environment feature of the operation site in the image is identified at present can be solved; and the feature identification efficiency can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of distribution network, and in particular to a method for automatically identifying on-site environmental characteristics of distribution network uninterrupted power supply operations. Background Art

[0002] The distribution network refers to the power grid that receives electric energy from the transmission network or regional power plants and distributes it locally or step by step according to voltage to various users through distribution facilities. It is composed of overhead lines, cables, towers, distribution transformers, disconnectors, reactive power compensators and some ancillary facilities, and plays an important role in distributing electric energy in the power grid.

[0003] At present, before the non-stop operation of the distribution network, drones are usually used to collect images of the distribution network operation site to facilitate the rapid evaluation and implementation of the non-stop operation plan. However, after the image is collected, it is difficult to automatically identify the environmental features of the operation site in the image, and personnel are required to identify them one by one, which has limitations. Therefore, a method for automatically identifying environmental features of the distribution network non-stop operation site is invented. Summary of the invention

[0004] In view of the above problems and / or the problems existing in the existing methods for automatically identifying the characteristics of the on-site environment of non-stop power distribution network operations, the present invention is proposed.

[0005] Therefore, the purpose of the present invention is to provide a method for automatically identifying the characteristics of the on-site environment of a distribution network without power outages, which can solve the above-mentioned existing problems.

[0006] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0007] The method for automatically identifying the on-site environmental characteristics of a distribution network without power outage operation comprises the following specific steps:

[0008] Step 1: Determine whether there is fog in the image collected by the drone. If there is fog, proceed to step 2; if there is no fog, proceed to step 3;

[0009] Step 2: Defogging the image to ensure the clarity of the image. After defogging, it will proceed to step 3;

[0010] Step 3: Establish initial data of the work site;

[0011] Step 4: Extract features from the image and convert the feature lines into several points;

[0012] Step 5: Find the reference point, and based on this, obtain the position distance and position relationship between the reference point and each point in step 4;

[0013] Step 6: After obtaining the position distances between the reference point and each point and their positional relationships, identify the reference point to analyze the specific position of the reference point in the work site;

[0014] Step 7: Compare the actual position of the reference point and the distance between the reference point and each point and their positional relationship in the established initial data to compare what kind of object the extracted feature is. At this point, the features in the image can be automatically identified.

[0015] As a preferred scheme of the method for automatic identification of on-site environmental features of distribution network non-stop operations described in the present invention, the defogging operation of the image in the step 2 includes a defogging method based on surrounding information, a defogging method based on a deep learning model, a defogging method based on the fusion of multiple different levels of features, a defogging method based on a deep learning neural network, a dehaze method based on a DehazeNet model and a defogging method based on an MSCNN model.

[0016] As a preferred solution of the method for automatically identifying the characteristics of the distribution network non-stop operation site environment of the present invention, the defogging method based on surrounding information is as follows:

[0017] Process 1, haze model estimation: estimate the haze model in the image;

[0018] Process 2, dark channel prior calculation: obtain the global atmospheric light value in the image by calculating the dark channel prior of the image;

[0019] Process 3, fog density estimation: The fog density at each pixel is calculated using the estimated atmospheric light value and haze model;

[0020] Process 4, non-local similarity weight calculation: For each pixel, the non-local similarity weight is calculated based on the similarity between the local block around it and other blocks in the entire image;

[0021] Process 5: Fog removal: The image is dehazed using the calculated non-local similarity weight and fog density. By taking a weighted average of each pixel, the details obscured by the haze can be restored.

[0022] As a preferred solution of the method for automatically identifying the characteristics of the on-site environment of the distribution network without power outages described in the present invention, the defogging method based on the deep learning model is as follows:

[0023] Process 1, multi-scale processing: operating at multiple scales to capture global and local information of the input image, so as to be able to analyze the image at different granularity levels and obtain more details;

[0024] Process 2, enhanced dehazing: using the enhanced algorithm to combine multiple weak dehazing models to form a more powerful integrated model;

[0025] Process three, dense feature fusion: combining features from different network layers in a dense and interconnected manner to be able to propagate information across different scales and effectively capture low-level and high-level features to achieve dehazing.

[0026] As a preferred solution of the method for automatically identifying the characteristics of the distribution network non-stop operation site environment described in the present invention, the defogging method based on the fusion of multiple different levels of characteristics is as follows:

[0027] Process 1, feature extraction: using convolutional neural networks to extract different levels of image features and capture information of different scales and semantics;

[0028] Process 2, haze perception: haze areas are detected by analyzing the contrast and brightness of feature maps at different levels, and corresponding masks are generated as the basis for subsequent processing;

[0029] Process 3, layer-by-layer fusion: Use the gating mechanism to adaptively select and adjust the weights between different features to achieve better feature fusion effect. Through layer-by-layer fusion, both low-level detail information and high-level semantic information can be considered at the same time, thereby improving the dehazing effect.

[0030] Process 4, dehazing and restoration: By learning a mapping function, the input image together with the estimated haze mask is converted into an image without haze. The whole process is implemented using a convolutional neural network.

[0031] As a preferred solution of the method for automatically identifying the characteristics of the on-site environment of the distribution network without power outages described in the present invention, the defogging method based on the deep learning neural network is as follows:

[0032] Process 1: Input the foggy image into the deep learning neural network;

[0033] Process 2: The deep learning neural network extracts features from foggy images, autonomously learns and extracts fog-related features;

[0034] Process 3: Perform multi-scale mapping on the foggy image, extract the features of the foggy image at different scales, and form a feature map;

[0035] Process 4: Perform local extremum analysis on each pixel on the feature map and keep the resolution unchanged to obtain the processed image;

[0036] Process 5: Perform nonlinear regression on the processed image to obtain the initial transmittance;

[0037] Process 6: Optimizing transmittance using guided filters and performing image smoothing on the processed image;

[0038] Process 7: Calculate atmospheric light parameters;

[0039] Process 8: Restore the fog-free image based on the initial transmittance and atmospheric light parameters.

[0040] As a preferred scheme of the method for automatic identification of on-site environmental characteristics of distribution network non-stop operation described in the present invention, the dehazing method based on the DehazeNet model specifically includes the DehazeNet model, and the DehazeNet model is composed of an input layer, a feature extraction layer, a multi-scale mapping layer, a nonlinear regression layer and an output layer.

[0041] As a preferred solution of the method for automatically identifying on-site environmental features of distribution network non-stop operations described in the present invention, the input layer of the DehazeNet model is a foggy image, the feature extraction layer of the DehazeNet model realizes feature extraction of the input image by adopting a convolution layer + activation layer structure, the multi-scale mapping layer of the DehazeNet model is constructed with convolution layers of different convolution kernel sizes, the nonlinear regression layer of the DehazeNet model is a bilateral rectified linear unit, and the output layer of the DehazeNet model is a transmittance map corresponding to the foggy image.

[0042] As a preferred solution of the method for automatic identification of on-site environmental characteristics of distribution network non-stop operation described in the present invention, the defogging method based on the MSCNN model specifically includes an MSCNN model, and the MSCNN model is composed of a coarse-scale convolutional network and a fine-scale convolutional network.

[0043] As a preferred solution of the method for automatic identification of on-site environmental features of distribution network non-stop operation described in the present invention, wherein: the task of the coarse-scale convolutional network is to predict the overall transmittance map of the scene, and the coarse-scale convolutional network is composed of a multi-scale convolutional layer, a pooling layer, an upsampling layer and a linear connection layer; the task of the fine-scale convolutional network is to refine the rough transmittance map output by the coarse-scale convolutional network, and by inputting the foggy image and the rough transmittance map into the fine-scale convolutional network, a more accurate transmittance map can be obtained.

[0044] Compared with existing technologies:

[0045] By setting a reference point and obtaining the position distance and position relationship between the reference point and the feature, and comparing it with the initial data of the work site, the feature can be recognized. This can solve the problem that personnel need to recognize the environmental features of the work site in the image one by one, thereby improving the recognition efficiency of the feature. In addition, by performing a defogging operation on the image, the feature can still be accurately recognized when there is fog in the image. DETAILED DESCRIPTION

[0046] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below.

[0047] The present invention provides a method for automatically identifying on-site environmental characteristics of a distribution network without power outage operation, including the following specific steps:

[0048] Step 1: Determine whether there is fog in the image collected by the drone. If there is fog, proceed to step 2; if there is no fog, proceed to step 3;

[0049] Step 2: Defogging the image to ensure the clarity of the image. After defogging, it will proceed to step 3;

[0050] Step 3: Establish initial data of the work site;

[0051] Step 4: Extract features from the image and convert the feature lines into several points;

[0052] Step 5: Find the reference point, and based on this, obtain the position distance and position relationship between the reference point and each point in step 4;

[0053] Step 6: After obtaining the position distances between the reference point and each point and their positional relationships, identify the reference point to analyze the specific position of the reference point in the work site;

[0054] Step 7: Compare the actual position of the reference point and the distance between the reference point and each point and their positional relationship in the established initial data to compare what kind of object the extracted feature is. At this point, the features in the image can be automatically identified.

[0055] Among them: the dehazing operation of the image includes a dehazing method based on surrounding information, a dehazing method based on a deep learning model, a dehazing method based on the fusion of multiple features at different levels, a dehazing method based on a deep learning neural network, a dehazing method based on a DehazeNet model, and a dehazing method based on an MSCNN model;

[0056] The defogging method based on surrounding information is as follows:

[0057] Process 1, haze model estimation: estimate the haze model in the image;

[0058] Process 2, dark channel prior calculation: obtain the global atmospheric light value in the image by calculating the dark channel prior of the image;

[0059] Process 3, fog density estimation: The fog density at each pixel is calculated using the estimated atmospheric light value and haze model;

[0060] Process 4, non-local similarity weight calculation: For each pixel, the non-local similarity weight is calculated based on the similarity between the local block around it and other blocks in the entire image;

[0061] Process 5: Fog removal: The image is dehazed using the calculated non-local similarity weight and fog density. By taking a weighted average of each pixel, the details obscured by the haze can be restored.

[0062] The dehazing method based on the deep learning model is as follows:

[0063] Process 1, multi-scale processing: operating at multiple scales to capture global and local information of the input image, so as to be able to analyze the image at different granularity levels and obtain more details;

[0064] Process 2, enhanced dehazing: using the enhanced algorithm to combine multiple weak dehazing models to form a more powerful integrated model;

[0065] Process three, dense feature fusion: combining features from different network layers in a dense and interconnected manner to be able to propagate information across different scales and effectively capture low-level and high-level features to achieve dehazing.

[0066] The defogging method based on the fusion of multiple different levels of features is as follows:

[0067] Process 1, feature extraction: using convolutional neural networks to extract different levels of image features and capture information of different scales and semantics;

[0068] Process 2, haze perception: haze areas are detected by analyzing the contrast and brightness of feature maps at different levels, and corresponding masks are generated as the basis for subsequent processing;

[0069] Process 3, layer-by-layer fusion: Use the gating mechanism to adaptively select and adjust the weights between different features to achieve better feature fusion effect. Through layer-by-layer fusion, both low-level detail information and high-level semantic information can be considered at the same time, thereby improving the dehazing effect.

[0070] Process 4, dehazing and restoration: By learning a mapping function, the input image together with the estimated haze mask is converted into an image without haze. The whole process is implemented using a convolutional neural network.

[0071] The dehazing method based on deep learning neural network is as follows:

[0072] Process 1: Input the foggy image into the deep learning neural network;

[0073] Process 2: The deep learning neural network extracts features from foggy images, autonomously learns and extracts fog-related features;

[0074] Process 3: Perform multi-scale mapping on the foggy image, extract the features of the foggy image at different scales, and form a feature map;

[0075] Process 4: Perform local extremum analysis on each pixel on the feature map and keep the resolution unchanged to obtain the processed image;

[0076] Process 5: Perform nonlinear regression on the processed image to obtain the initial transmittance;

[0077] Process 6: Optimizing transmittance using guided filters and performing image smoothing on the processed image;

[0078] Process 7: Calculate atmospheric light parameters;

[0079] Process 8: Restore the fog-free image based on the initial transmittance and atmospheric light parameters.

[0080] The dehazing method based on the DehazeNet model specifically includes a DehazeNet model, which is composed of an input layer, a feature extraction layer, a multi-scale mapping layer, a nonlinear regression layer and an output layer. The input layer of the DehazeNet model is a foggy image, the feature extraction layer of the DehazeNet model realizes feature extraction of the input image by adopting a convolution layer + activation layer structure, the multi-scale mapping layer of the DehazeNet model is constructed with convolution layers of different convolution kernel sizes, the nonlinear regression layer of the DehazeNet model is a bilateral rectified linear unit, and the output layer of the DehazeNet model is a transmittance map corresponding to the foggy image.

[0081] The dehazing method based on the MSCNN model specifically includes an MSCNN model, wherein the MSCNN model is composed of a coarse-scale convolutional network and a fine-scale convolutional network. The task of the coarse-scale convolutional network is to predict the overall transmittance map of the scene, and the coarse-scale convolutional network is composed of a multi-scale convolutional layer, a pooling layer, an upsampling layer, and a linear connection layer; the task of the fine-scale convolutional network is to refine the rough transmittance map output by the coarse-scale convolutional network. By inputting the foggy image and the rough transmittance map into the fine-scale convolutional network, a more accurate transmittance map can be obtained.

[0082] Although the present invention has been described above with reference to the embodiments, various modifications may be made thereto and parts thereof may be replaced by equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the various features in the embodiments disclosed in the present invention may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for automatically identifying on-site environmental characteristics of distribution network non-stop operation, characterized in that: The specific steps are as follows: Step 1: Determine whether there is fog in the image collected by the drone. If there is fog, proceed to step 2; if there is no fog, proceed to step 3; Step 2: Defogging the image to ensure the clarity of the image. After defogging, it will proceed to step 3; Step 3: Establish initial data of the work site; Step 4: Extract features from the image and convert the feature lines into several points; Step 5: Find the reference point, and based on this, obtain the position distance and position relationship between the reference point and each point in step 4; Step 6: After obtaining the position distances between the reference point and each point and their positional relationships, identify the reference point to analyze the specific position of the reference point in the work site; Step 7: Compare the actual position of the reference point and the distance between the reference point and each point and their positional relationship in the established initial data to compare what kind of object the extracted feature is. At this point, the features in the image can be automatically identified.

2. The method for automatically identifying on-site environmental characteristics of distribution network non-stop operation according to claim 1 is characterized in that: The dehazing operation of the image in the step 2 includes a dehazing method based on surrounding information, a dehazing method based on a deep learning model, a dehazing method based on the fusion of multiple different levels of features, a dehazing method based on a deep learning neural network, a dehazing method based on a DehazeNet model, and a dehazing method based on an MSCNN model.

3. The method for automatically identifying the on-site environmental characteristics of a distribution network without power outage according to claim 2 is characterized in that: The defogging method based on surrounding information is specifically as follows: Process 1, haze model estimation: estimate the haze model in the image; Process 2, dark channel prior calculation: obtain the global atmospheric light value in the image by calculating the dark channel prior of the image; Process 3, fog density estimation: The fog density at each pixel is calculated using the estimated atmospheric light value and haze model; Process 4, non-local similarity weight calculation: For each pixel, the non-local similarity weight is calculated based on the similarity between the local block around it and other blocks in the entire image; Process 5: Fog removal: The image is dehazed using the calculated non-local similarity weight and fog density. By taking a weighted average of each pixel, the details obscured by the haze can be restored.

4. The method for automatically identifying on-site environmental characteristics of distribution network non-stop operation according to claim 2 is characterized in that: The defogging method based on the deep learning model is as follows: Process 1, multi-scale processing: operating at multiple scales to capture global and local information of the input image, so as to be able to analyze the image at different granularity levels and obtain more details; Process 2, enhanced dehazing: using the enhanced algorithm to combine multiple weak dehazing models to form a more powerful integrated model; Process three, dense feature fusion: combining features from different network layers in a dense and interconnected manner to be able to propagate information across different scales and effectively capture low-level and high-level features to achieve dehazing.

5. The method for automatically identifying the on-site environmental characteristics of a distribution network without power outage according to claim 2 is characterized in that: The defogging method based on fusing multiple features at different levels is specifically as follows: Process 1, feature extraction: using convolutional neural networks to extract different levels of image features and capture information of different scales and semantics; Process 2, haze perception: haze areas are detected by analyzing the contrast and brightness of feature maps at different levels, and corresponding masks are generated as the basis for subsequent processing; Process 3, layer-by-layer fusion: Use the gating mechanism to adaptively select and adjust the weights between different features to achieve better feature fusion effect. Through layer-by-layer fusion, both low-level detail information and high-level semantic information can be considered at the same time, thereby improving the dehazing effect. Process 4, dehazing and restoration: By learning a mapping function, the input image together with the estimated haze mask is converted into an image without haze. The whole process is implemented using a convolutional neural network.

6. The method for automatically identifying the on-site environmental characteristics of a distribution network without power outage according to claim 2 is characterized in that: The defogging method based on deep learning neural network is as follows: Process 1: Input the foggy image into the deep learning neural network; Process 2: The deep learning neural network extracts features from foggy images, autonomously learns and extracts fog-related features; Process 3: Perform multi-scale mapping on the foggy image, extract the features of the foggy image at different scales, and form a feature map; Process 4: Perform local extremum analysis on each pixel on the feature map and keep the resolution unchanged to obtain the processed image; Process 5: Perform nonlinear regression on the processed image to obtain the initial transmittance; Process 6: Optimizing transmittance using guided filters and performing image smoothing on the processed image; Process 7: Calculate atmospheric light parameters; Process 8: Restore the fog-free image based on the initial transmittance and atmospheric light parameters.

7. The method for automatically identifying on-site environmental characteristics of distribution network non-stop operation according to claim 2 is characterized in that: The dehazing method based on the DehazeNet model specifically includes a DehazeNet model, and the DehazeNet model is composed of an input layer, a feature extraction layer, a multi-scale mapping layer, a nonlinear regression layer and an output layer.

8. The method for automatically identifying on-site environmental characteristics of distribution network non-stop operation according to claim 7 is characterized in that: The input layer of the DehazeNet model is a foggy image, the feature extraction layer of the DehazeNet model realizes feature extraction of the input image by adopting a convolution layer + activation layer structure, the multi-scale mapping layer of the DehazeNet model is constructed with convolution layers of different convolution kernel sizes, the nonlinear regression layer of the DehazeNet model is a bilateral rectified linear unit, and the output layer of the DehazeNet model is a transmittance map corresponding to the foggy image.

9. The method for automatically identifying on-site environmental characteristics of distribution network non-stop operation according to claim 2 is characterized in that: The defogging method based on the MSCNN model specifically includes an MSCNN model, and the MSCNN model is composed of a coarse-scale convolutional network and a fine-scale convolutional network.

10. The method for automatically identifying on-site environmental characteristics of distribution network non-stop operation according to claim 9, characterized in that: The task of the coarse-scale convolutional network is to predict the overall transmittance map of the scene. The coarse-scale convolutional network consists of a multi-scale convolutional layer, a pooling layer, an upsampling layer, and a linear connection layer. The task of the fine-scale convolutional network is to refine the rough transmittance map output by the coarse-scale convolutional network. By inputting the foggy image and the rough transmittance map into the fine-scale convolutional network, a more accurate transmittance map can be obtained.