Power distribution network operation risk visualization method, system, equipment and medium

By integrating multimodal data fusion and feature extraction with a risk subject identification and positioning model, the accuracy and robustness of safety risk identification in power distribution network operations have been addressed, enabling the visualization and control of risk factors.

CN118940204APending Publication Date: 2024-11-12CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD +3
View PDF 0 Cites 6 Cited by

Patent Information

Application Number
CN202410897526.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Existing technologies for identifying safety risks in power distribution network operations have low accuracy and poor robustness. They fail to effectively consider the interactive changes between multiple risk factors and rely mainly on single-modal data, resulting in insufficient robustness and accuracy in identification and tracking.

Method used

A multimodal data fusion method is adopted to acquire image information, data information and text information. Through feature extraction and data fusion, a trained risk subject identification and localization model is used to identify and locate risk subjects and then visualize the results.

Benefits of technology

It improves the accuracy and robustness of risk identification in power distribution network operations, enables the visualization of multiple risk factors, and enhances the level of risk management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940204A_ABST
    Figure CN118940204A_ABST
Patent Text Reader

Abstract

The invention provides a power distribution network operation risk visualization method, system and device and a medium, and the method comprises the steps: obtaining multi-modal data of a power distribution network in a target operation period, the multi-modal data comprising image information, data information and text information; after feature extraction is carried out on the image information, the data information and the text information, data fusion is carried out; the fused data image data is used as input, and a trained risk subject identification and positioning model is used to identify a risk subject and determine the position of the risk subject in the image; and based on the position of the risk subject in the image and the corresponding risk state type, marking on the image to carry out risk visual display, thereby improving the safety risk management and control level of the power distribution network operation, realizing visual tracking of cross-modal risk factors, improving the operation risk perception capability, and improving the safety and reliability of the power distribution network operation. The occurrence probability of power distribution network operation safety accidents is reduced, and the personal safety of operators is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of distribution network operation safety risk management and control, and in particular to a distribution network operation risk visualization method, system, equipment and medium. Background Art

[0002] The safety risks of distribution network operations involve multiple risk sources such as people, environment, and tools. The various risk factors interact and merge, and the risk subjects change dynamically. Conventional operation safety risk management and control technologies use qualitative descriptions of task risk levels before operations, on-site inspections by supervisors, on-site monitoring of control balls, and identification and monitoring of operation behaviors as means to manage safety risks. This method does not take into account the changes and interactions among the various risk subjects, namely, people, equipment, environment, and tools, and usually only uses single-modal data, namely visible light image data, for risk identification, which makes its risk identification and tracking robustness poor and the identification accuracy low. In addition, the current visualization tracking level of distribution network operation safety risk identification technology needs to be improved. Summary of the invention

[0003] In order to solve the problem of low accuracy and poor robustness in the prior art of risk identification of distribution network operations, the present invention proposes a distribution network operation risk visualization method, comprising:

[0004] Acquire multimodal data of the distribution network within a target operation cycle, wherein the multimodal data includes image information, data information, and text information;

[0005] After extracting features from image information, data information and text information, data fusion is performed;

[0006] Taking the fused image data as input, the risk subject is identified using the trained risk subject identification and positioning model and the position of the risk subject in the image is determined;

[0007] Based on the position of the risk subject in the image and the corresponding risk status type, the image is marked to perform risk visualization.

[0008] Preferably, the image information includes: visible light images and infrared light images; the data information includes: environmental data and equipment operation data; and the text information includes: distribution network operation history text data.

[0009] Preferably, the step of extracting features from the image information, data information and text information and then fusing the data comprises:

[0010] The image features are obtained by cross-modal data fusion and superposition of visible light images and infrared images;

[0011] Extracting features from the environment data and the equipment operation data to obtain environment feature data and equipment feature data;

[0012] Extract features and classify the historical text data of distribution network operations to obtain text features;

[0013] After extracting environmental feature data, equipment feature data and text features, the state features of the distribution network operation are constructed;

[0014] A feature vector is constructed based on the image features and state features, and the feature vector is encoded to obtain fused data.

[0015] Preferably, the state characteristics of the distribution network operation are constructed after extracting the environmental characteristic data, the equipment characteristic data and the text data set, including:

[0016] Perform feature extraction on environmental feature data, equipment feature data and text data sets to obtain environmental features, historical text features and equipment operation features;

[0017] Based on the influence of each equipment operation characteristic on the equipment operation status, the key equipment operation characteristics are selected from the equipment operation characteristics;

[0018] Environmental features, historical text features and key equipment operation features are used as status features.

[0019] Preferably, the step of obtaining image features by cross-modal data fusion and superposition of visible light images and infrared light images includes:

[0020] The infrared light image and the visible light data image are processed in the form of histograms to obtain a gradient histogram feature vector of the infrared light image and a gradient histogram feature vector of the visible light image;

[0021] Convolutional neural networks are used to extract common features from the gradient histogram feature vectors of infrared images and the gradient histogram feature vectors of visible light images respectively;

[0022] A convolutional neural network is used to extract discriminative features from the feature vector of the gradient histogram of the infrared image and the feature vector of the gradient histogram of the visible light image respectively;

[0023] Superimpose the common features and the discriminative features corresponding to the feature vector of the visible light image gradient histogram to obtain the superimposed features of the visible light image;

[0024] Superimpose the common features and the discriminative features corresponding to the feature vector of the infrared image gradient histogram to obtain the superimposed features of the infrared image;

[0025] The visible light image superposition features and the infrared light image superposition features are fused.

[0026] Preferably, the feature extraction of environmental data and equipment operation data includes:

[0027] The environmental characteristics are constructed based on the geographical coordinates of the working environment location and superimposed on the environmental data within the working cycle;

[0028] Collecting equipment operation data of each equipment in the distribution network; the equipment operation data includes equipment operation signals and fault information;

[0029] Feature extraction is performed based on equipment operation signals and fault information to obtain the corresponding relationship between faults and features; and the corresponding equipment operation characteristics are obtained based on the corresponding relationship between each fault and feature.

[0030] Preferably, the feature extraction and classification of the distribution network operation history text data to obtain text features includes:

[0031] Vectorization preprocessing of historical text dataset, two sample sentences;

[0032] Perform text overall feature extraction and text feature segmentation on two sample sentences respectively;

[0033] The action text features corresponding to the two sample sentences are fused and classified to obtain the classification result of the action text;

[0034] The relational text features corresponding to the two sample sentences are fused and classified to obtain the relational text classification result;

[0035] The classification results of action text and relationship text are used as text features.

[0036] Preferably, the training process of the risk subject identification and positioning model includes:

[0037] Constructing a sample set based on historical fused data and location information of risk subjects corresponding to the fused data;

[0038] Dividing the sample set into a training set and a test set according to a set ratio;

[0039] Taking the fused data in the training set as input and the location information of the risk subject corresponding to the fused data in the training set as output, the convolutional neural network is trained to obtain a preliminarily trained risk location model;

[0040] The preliminarily trained risk positioning model is tested based on the test set to revise the preliminarily trained risk positioning model to obtain a trained risk subject identification and positioning model.

[0041] Preferably, the method of taking the fused image data as input and using a trained risk subject identification and positioning model to identify the risk subject and determine the position of the risk subject in the image includes:

[0042] The risk location model is used to locate the area where the risk subject in the fused data is located, the area is framed, and the coordinates of the framed area are determined.

[0043] Preferably, the marking on the image based on the position of the risk subject in the image and the corresponding risk status type for risk visualization includes:

[0044] Based on the risk subject in the framed area of ​​the image, determining a corresponding operation risk status type, and determining a risk level according to the operation risk status type;

[0045] The color of the framed area is determined by using the corresponding RGB color according to the risk level and rendered on the image to achieve a visual display of the risk.

[0046] Preferably, determining the risk level according to the operation risk status type includes:

[0047] In the operation scenario, when the risk subject is not in a risk state, the risk level is 0;

[0048] When the risk subject is in a risk state, the risk level index is calculated based on the number of risk categories, severity and duration;

[0049] The risk level is determined according to the risk level indicator.

[0050] Based on the same inventive concept, the present invention also provides a distribution network operation risk visualization system, including: a data acquisition module, a fusion module, a location determination module and a visualization module;

[0051] The data acquisition module is used to: acquire multimodal data of the distribution network within a target operation cycle, wherein the multimodal data includes image information, data information and text information;

[0052] The fusion module is used to: extract features from image information, data information and text information, and then perform data fusion;

[0053] The position determination module is used to: take the fused data image data as input, use the trained risk subject identification and positioning model to identify the risk subject and determine the position of the risk subject in the image;

[0054] The visualization module is used to mark the image based on the position of the risk subject in the image and the corresponding risk status type to perform risk visualization display.

[0055] Preferably, the image information includes: visible light images and infrared light images; the data information includes: environmental data and equipment operation data; and the text information includes: distribution network operation history text data.

[0056] Preferably, the fusion module includes: an image processing submodule, a data processing submodule, a text processing submodule and a fusion processing submodule;

[0057] The image processing submodule is used to: obtain image features by cross-modal data fusion and superposition on visible light images and infrared light images;

[0058] The data processing submodule is used to: extract features from the environment data and the equipment operation data to obtain environment feature data and equipment feature data;

[0059] The text processing submodule is used to: extract features and classify and identify the historical text data of the distribution network operation to obtain text features;

[0060] The fusion processing submodule is used to: construct the state characteristics of the distribution network operation after feature extraction of environmental feature data, equipment feature data and text features; and to construct a feature vector based on image features and state features, and encode the feature vector to obtain fused data.

[0061] Preferably, the fusion processing submodule is specifically used for:

[0062] Perform feature extraction on environmental feature data, equipment feature data and text data sets to obtain environmental features, historical text features and equipment operation features;

[0063] Based on the influence of each equipment operation characteristic on the equipment operation status, the key equipment operation characteristics are selected from the equipment operation characteristics;

[0064] Environmental features, historical text features and key equipment operation features are used as status features.

[0065] Preferably, the image processing submodule is specifically used for:

[0066] The infrared light image and the visible light data image are processed in the form of histograms to obtain a gradient histogram feature vector of the infrared light image and a gradient histogram feature vector of the visible light image;

[0067] Convolutional neural networks are used to extract common features from the gradient histogram feature vectors of infrared images and the gradient histogram feature vectors of visible light images respectively;

[0068] A convolutional neural network is used to extract discriminative features from the feature vector of the gradient histogram of the infrared image and the feature vector of the gradient histogram of the visible light image respectively;

[0069] Superimpose the common features and the discriminative features corresponding to the feature vector of the visible light image gradient histogram to obtain the superimposed features of the visible light image;

[0070] Superimpose the common features and the discriminative features corresponding to the feature vector of the infrared image gradient histogram to obtain the superimposed features of the infrared image;

[0071] The visible light image superposition features and the infrared light image superposition features are fused.

[0072] Preferably, the data processing submodule is specifically used for:

[0073] The environmental characteristics are constructed based on the geographical coordinates of the working environment location and superimposed on the environmental data within the working cycle;

[0074] Collecting equipment operation data of each equipment in the distribution network; the equipment operation data includes equipment operation signals and fault information;

[0075] Feature extraction is performed based on equipment operation signals and fault information to obtain the corresponding relationship between faults and features; and the corresponding equipment operation characteristics are obtained based on the corresponding relationship between each fault and feature.

[0076] Preferably, the text processing submodule is specifically used for:

[0077] Vectorization preprocessing of historical text dataset, two sample sentences;

[0078] Perform text overall feature extraction and text feature segmentation on two sample sentences respectively;

[0079] The action text features corresponding to the two sample sentences are fused and classified to obtain the classification result of the action text;

[0080] The relational text features corresponding to the two sample sentences are fused and classified to obtain the relational text classification result;

[0081] The classification results of action text and relationship text are used as text features.

[0082] Preferably, the system also includes a training process of a risk subject identification and positioning model, specifically including:

[0083] Constructing a sample set based on historical fused data and location information of risk subjects corresponding to the fused data;

[0084] Dividing the sample set into a training set and a test set according to a set ratio;

[0085] Taking the fused data in the training set as input and the location information of the risk subject corresponding to the fused data in the training set as output, the convolutional neural network is trained to obtain a preliminarily trained risk location model;

[0086] The preliminarily trained risk positioning model is tested based on the test set to revise the preliminarily trained risk positioning model to obtain a trained risk subject identification and positioning model.

[0087] Preferably, the location determination module is specifically used for:

[0088] The risk location model is used to locate the area where the risk subject in the fused data is located, the area is framed, and the coordinates of the framed area are determined.

[0089] Preferably, the visualization module is specifically used to include:

[0090] Based on the risk subject in the framed area of ​​the image, determining a corresponding operation risk status type, and determining a risk level according to the operation risk status type;

[0091] The color of the framed area is determined by using the corresponding RGB color according to the risk level and rendered on the image to achieve a visual display of the risk.

[0092] Preferably, the determination of the risk level includes:

[0093] In the operation scenario, when the risk subject is not in a risk state, the risk level is 0;

[0094] When the risk subject is in a risk state, the risk level index is calculated based on the number of risk categories, severity and duration;

[0095] The risk level is determined according to the risk level indicator.

[0096] Based on the same inventive concept, the present invention also provides an electronic device, comprising: at least one processor and a memory;

[0097] The memory and the processor are connected via a bus;

[0098] The memory is used to store one or more programs;

[0099] When the one or more programs are executed by the at least one processor, a distribution network operation risk visualization method provided by the present invention is implemented.

[0100] Based on the same inventive concept, the present invention also provides 24. a readable storage medium having a computer program stored thereon, and when the computer program is executed, a distribution network operation risk visualization method provided by the present invention is implemented.

[0101] Compared with the prior art, the present invention has the following beneficial effects:

[0102] The present invention provides a distribution network operation risk visualization method and system, comprising: acquiring multimodal data of the distribution network within a target operation cycle, the multimodal data comprising image information, data information and text information; performing data fusion after feature extraction on the image information, data information and text information; using the fused data image data as input, using a trained risk subject identification and positioning model to identify the risk subject and determine the position of the risk subject in the image; marking the image based on the position of the risk subject in the image and the corresponding risk status type for risk visualization display. The present invention solves the problem of poor robustness and limited accuracy improvement of single-modal data risk identification and tracking by extracting and fusing features of multimodal data and using the fused data for risk identification and positioning; and through a visualization method, a visualization presentation of multiple risk factors is achieved, thereby improving the visualization management and control level of risk factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 A flow chart of the method for visualizing the risk of power distribution network operation according to the present invention;

[0104] Figure 2 It is a flow chart of the HOG-based image data feature extraction algorithm of the present invention;

[0105] Figure 3 is a flow chart of cross-modal image data feature fusion of the present invention;

[0106] Figure 4 It is a flow chart of feature extraction of historical text data of the present invention;

[0107] Figure 5 and Figure 6 The visual effects diagrams of distribution network operation risks under different states are shown respectively;

[0108] Figure 7 It is a structural schematic diagram of the distribution network operation analysis visualization system of the present invention;

[0109] Figure 8 It is a schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION

[0110] Since the visual risk identification of the current distribution network operation safety risk identification technology requires the fusion and superposition of multi-modal operation risk data, and at the same time, it is necessary to use quantitative evaluation methods to achieve visual tracking of risk subjects, the technical difficulties of the present invention are concentrated in the following two aspects:

[0111] First, cross-modal data fusion and superposition: Considering that the single-modal data of the distribution network operation safety monitoring video image is easily affected by the real environment and acquisition equipment during the collection process, and the description of risk factors is not comprehensive enough, the recognition and classification accuracy of risk factors and subjects is not high. Two-dimensional images can present the overall scene shape, color, texture and other details of the operation site, which are intuitive image features. Environmental meteorological data and equipment operation data can describe the status characteristics of the working environment objects. Using images and status data together to describe the safety risk factors of the distribution network operation scene can obtain more intuitive, comprehensive and reliable information. Feature fusion can retain the features of different data sources, but the extracted features must be represented in the same data format, but the selection of methods for fusion of features at different levels is a technical difficulty.

[0112] Second, in order to achieve the visualization of cross-modal risk data, operational risk factors and subject quantitative indicators are superimposed on image data. The difficulty lies in how to quantify and normalize the risk factors, realize image superposition through target anchor frame and RGB risk visualization technology, visualize the dynamic change characteristics of operational risks, and enhance the perception of operational safety risks.

[0113] In order to better understand the present invention, the content of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0114] Example 1

[0115] The present invention provides a method for visualizing the risks of distribution network operations. Figure 1 As shown, including:

[0116] S1. Acquire multimodal data of a distribution network within a target operation cycle, wherein the multimodal data includes image information, data information, and text information;

[0117] S2, extracting features from image information, data information and text information, and then performing data fusion;

[0118] S3, using the fused data image data as input, using the trained risk subject identification and positioning model to identify the risk subject and determine the position of the risk subject in the image;

[0119] S4. Mark the image based on the position of the risk subject in the image and the corresponding risk status type to perform risk visualization.

[0120] The specific process is as follows:

[0121] The multimodal data in step S1 includes: image information (visible light image, infrared light image), data information (environmental data, equipment operation data) and text information, etc.;

[0122] The visible light image is an image frame obtained from the visible light video data, and the infrared light image is an image frame obtained from the infrared light video data.

[0123] Environmental data includes: the highest and lowest data of ambient temperature, the highest and lowest data of humidity, the highest wind speed, rain, snow, ice, thunderstorm and other meteorological information during the distribution network operation cycle. Equipment operation data includes: equipment operation status information and fault information, such as vibration, partial discharge, voltage and current, etc.

[0124] The text information is the historical text data of distribution network operations, which can be divided into structured, semi-structured and unstructured text data.

[0125] Step S2 specifically includes:

[0126] S2-1, cross-modal data fusion and superposition are used to obtain image features for visible light images and infrared light images;

[0127] Firstly, infrared and visible light data image enhancement methods are used to enhance the image data respectively. Then, through cross-modal unified representation learning, the common feature information of visible light and infrared images is extracted, the feature interference information under each modality is removed, the feature dimension of cross-modal images is reduced, the key feature information of the image is retained, and the influence of single-modal image information on the key features is reduced. Then, cross-modal specific representation learning is used to extract the specific representation information of visible light and infrared. Finally, the common features are aligned through unified representation feature alignment, and the specific representation information is superimposed to realize the fusion and superposition of cross-modal image data.

[0128] The specific process, such as Figure 2 As shown:

[0129] (1) Scale the original image to a fixed size and then grayscale the image to reduce the amount of computation and improve the robustness of the algorithm;

[0130] (2) By calculating the gradient of each pixel, the gradient magnitude and gradient direction of each pixel in the image can be obtained;

[0131] (3) Determine multiple detection windows from the image according to the window sliding step size, select a target detection window from the determined multiple detection windows, and then divide the image into multiple small areas from the target detection window, each area containing a number of pixels.

[0132] (4) For each pixel in each region, the number of gradient directions is counted to obtain a gradient direction histogram. The histogram is divided into multiple units according to the gradient direction, and the size of each unit is 360 degrees divided by the number of units.

[0133] (5) Perform unit-direction projection in the region and calculate the projection sum.

[0134] (6) Normalize the gradient histogram of each region with the gradient histograms of several surrounding regions to reduce the impact of illumination and contrast changes in the image.

[0135] (7) The normalized gradient histograms of all regions are concatenated into a vector to obtain the final Histogram of Oriented Gradient (HOG) feature vector corresponding to the visible light image data and the infrared image data.

[0136] (8) Repeat the above steps (3)-(7) until the image processing is completed.

[0137] After obtaining the HOG feature vectors corresponding to the visible light image and the infrared light image, cross-modal unified representation and specific representation learning are performed. The process is as follows: Figure 3 As shown:

[0138] First, unified representation learning is performed. Convolutional Neural Networks (CNN) are used to extract the common features in the HOG feature vectors of visible light images and the HOG feature vectors of infrared images respectively, so as to eliminate the interference of information that only exists in a specific modality, thereby reducing the risk of the unified representation learning process being affected by specific modality information. Then, the common features corresponding to the HOG feature vectors of visible light images and the common features corresponding to the HOG features of infrared images are aligned.

[0139] On the basis of the above unified representation, cross-modal specific representation learning is performed: a convolutional neural network is used to extract the unique discriminative features (such as gradient center value, distribution shape, and degree of discreteness) in the HOG feature vector of visible light images and the HOG feature vector of infrared light images to distinguish different targets, thereby restoring the discriminative information in the specific modality that is suppressed in the unified representation learning, so as to perform feature compensation of different modal features.

[0140] Then, the discriminative features corresponding to the HOG feature vector of the visible light image are superimposed with the common features corresponding to the HOG feature vector of the visible light image to obtain the superimposed features of the visible light image, and then the discriminative features corresponding to the HOG feature vector of the infrared light image are superimposed with the common features corresponding to the HOG feature vector of the infrared light image to obtain the superimposed features of the infrared light image.

[0141] Then, the superimposed features of the visible light image and the superimposed features of the infrared light image are aligned to fuse the visible light image and the infrared light image.

[0142] A more robust cross-modal image feature representation is achieved by combining cross-modal unified representation and cross-modal specific representation based on the method of specific modality feature compensation.

[0143] Finally, by aligning the unified representation features, the common features are aligned and the specific representation information is superimposed to achieve the fusion and superposition of cross-modal image data.

[0144] Based on the complementary learning framework of specific and unified representations, the similarities between samples of different modalities are first modeled according to their common features, and then the specific modality representation is used to further retrieve potential positive samples, which significantly improves the accuracy of Visible-Infrared ReID (VI-RelD) and the diversity and completeness of cross-modal representations. Based on unified representation and cross-modal representation learning, focusing on the appearance representation of the extracted target takes into account the similarities between the same target and the differences between different targets, which can improve the ability of the image data target recognition model to distinguish different targets with similar appearances.

[0145] S2-2, extracting features from the environment data and the equipment operation data to obtain environment feature data and equipment feature data;

[0146] Normalize and extract features of multimodal data such as environment and equipment operation data, and then scale the multimodal data. Through data feature engineering, multi-dimensional feature expression of environmental data and equipment operation data is achieved, and unified feature-level expression of multimodal data is achieved.

[0147] The environmental data is expressed with characteristics based on the geographic coordinates of the working environment. The highest and lowest values ​​of the ambient temperature, the highest and lowest values ​​of humidity, the highest value of wind speed, and the meteorological information of rain, snow and thunderstorms during the working period are superimposed to construct environmental characteristics.

[0148] For the equipment operation characteristics, it is necessary to sort out and collect the equipment operation signals and fault information of different data systems. Then, the equipment operation status signal is extracted to obtain the signal characteristics, and the signal characteristics are analyzed for uncertainty to obtain the fault characteristics associated with different signal characteristics. Finally, the fault characteristic indicators are arranged according to the uncertainty, and the evidence fusion theory is applied to evaluate the correlation between the indicators, and the fusion rules are selected to perform the characteristic value fusion calculation. The clustering decision model is used to cluster and select the equipment operation characteristics.

[0149] Taking the data acquisition and monitoring control system as an example, the actual information of the data acquisition and monitoring control system is mainly manifested as switch quantity, and the specific manifestation of the switch quantity is protection action and switch state. Similarly, the recording information, current information and voltage information in the data acquisition and monitoring control system are also specific manifestations of electrical quantities. Then, the switch quantity information is preliminarily processed using the reduction function, and the Bayesian network is introduced for refinement mining to obtain the opening and closing state and current direction data. The wavelet change method is used to analyze and calculate the time-frequency characteristics of the electrical quantity changes in detail, and these quantitative data are used as the state signal characteristics of the data acquisition and monitoring control system. The uncertainty analysis of the state signal characteristics of the data acquisition and monitoring control system is carried out to obtain the fault feature indicators associated with the state signal characteristics. The fault feature indicators include Bayesian fault feature indicators, wavelet fault feature indicators, wavelet singular feature indicators and wavelet energy feature indicators. The state signal characteristics are sorted according to the uncertainty, that is, the fault feature indicators, and the clustering decision model is used to classify and select the state signal characteristics. For example, the two features of opening and closing state and current direction are the specific manifestations of the switch quantity, so these two features are classified together, and then the fault characteristic indicators corresponding to the opening and closing state and the current direction are compared, and the feature with the highest correlation with the switch quantity, that is, the feature with the lowest uncertainty, is selected as the representative feature of the switch quantity, and the representative features of different types are summarized to obtain the equipment operation characteristics. S2-3, feature extraction and classification recognition of the distribution network operation history text data to obtain text features;

[0150] Text data can be divided into structured, semi-structured and unstructured text data. Aiming at the processing of historical text data of distribution network operation, a network structure based on Bert pre-training is designed to extract text features. The deep learning method is used to extract features on long text data sets, which can reduce the difficulty of text feature extraction, improve the efficiency of model training, and more accurately express the semantic information of the text.

[0151] For the processing of historical text features, the overall process is roughly the classification and recognition of historical text data sets, so as to convert historical text data into the same representation as image features, which is convenient for text feature fusion during supervision. The feature extraction model can choose to use the Bert model, which uses deep learning methods to extract features on long text data sets, which can reduce the difficulty of text feature extraction, improve the efficiency of model training, and more accurately express the semantic information of the text.

[0152] The feature extraction process of historical text data is as follows: Figure 4 As shown, Figure 5 Taking the above process as an example, the following is explained:

[0153] First, the historical text data is preprocessed to obtain two sample sentences. Sample sentence 1 is: The construction workers are crossing the safety protection fence; sample sentence 2 is: The construction workers have violated the regulations by crossing the fence.

[0154] The overall text feature extraction and text feature segmentation are performed on sample sentences 1 and 2 respectively, and the action text features corresponding to sample sentence 1 are obtained: crossing; and the relationship text features: construction workers → being → crossing → safety protection fence; the action text features corresponding to sample sentence 2 are obtained: crossing; and the relationship text features: construction workers → existence → crossing → safety protection fence → behavior.

[0155] The action text features corresponding to sample sentence 1 and the action text features corresponding to sample sentence 2 are fused and input into the action feature classification module to obtain the classification result of the action text. At the same time, the relationship text features corresponding to sample sentence 1 and the relationship text features corresponding to sample sentence 2 are fused and input into the relationship feature classification module to obtain the relationship text classification result. Finally, the action text classification result and the relationship text classification result are used as historical text features, so the feature extraction fusion is finally obtained: construction personnel → crossing → safety protection fence. .

[0156] It should be noted that in the process of historical text data classification and recognition, the same or similar text data sets can obtain unique information in deep semantics. This generally refers to the short "distance" between two similar text data in vector representation. That is, the smaller the "distance" between two texts, the higher the similarity between the texts; conversely, the larger the "distance", the lower the similarity. The main process of the text feature extraction process for the identification of the subject of operational safety risk lies in the construction of the text data set and the establishment of the text feature extraction model. Combining the idea of ​​​​supervising the corresponding visual features of the environment, equipment operation data, and relational text features, it is necessary to reflect the text data set of operational risk factors in the construction of text data.

[0157] S2-4, extracting features from the environmental feature data, equipment feature data and text data set to construct state features of the distribution network operation;

[0158] Given multiple different input modes of images, data, and text sets, a single distribution network operation risk subject location and model is trained, which can learn to capture any subset of possible modes to achieve the ultimate goal. Different data modes contain information at multiple scales, and a hierarchical latent space with different resolutions is used to directly transfer information from each resolution of the spatial encoder to the corresponding resolution of the latent space, so that high-resolution control features can be better preserved.

[0159] Specifically: normalize the environmental data, historical text data and equipment operation data, extract features from the normalized environmental data, historical text data and equipment operation data to obtain environmental features, historical text features and equipment operation features, screen out key equipment operation features from the equipment operation features based on the influence of each equipment operation feature on the equipment operation status, and use the environmental features, historical text features and key equipment operation features as status features.

[0160] Finally, environmental characteristics, equipment operation characteristics, and historical text characteristics are used as status characteristics of distribution network operations.

[0161] On the basis of obtaining the state characteristics of the distribution network operation, the image characteristics and the state characteristics are encoded into a feature vector, the feature vector is encoded to obtain an output vector, and the fusion data is generated based on the output vector. Specifically:

[0162] The image features and state features are encoded into a feature vector, which is input into a tensor fusion model. The tensor fusion model includes a multimodal encoder and a decoder. When the feature vector is input into the tensor fusion model, the feature vector is encoded by the multimodal encoder to obtain the output vector of the encoder. The output vector is then input into the decoder to generate fusion data to learn high-dimensional intrinsic information between different modalities. The fusion data can be expressed in the form of an image.

[0163] S2-5. Construct a feature vector based on the image features and the state features, and encode the feature vector to obtain fused data.

[0164] Specifically, cross-modal data fusion and superposition are used to obtain image features for visible light images and infrared light images, including:

[0165] The present invention extracts the common features of the two types of images through the above-mentioned means, and superimposes the discriminative features of different types of images on the common features, so as to perform feature extraction and fusion of image data. Through unified representation and specific representation, it pays attention to both the similarities between the same targets and the differences between different targets, which can enhance the ability of the image data target recognition model to distinguish different targets with similar appearances, and further enhance the accuracy of distribution network risk identification.

[0166] Step S3 specifically includes:

[0167] After obtaining the fused data, it is used as input to train the risk subject identification and positioning model to achieve dynamic tracking of the risk subject, and the output is the visible light image data superimposed with the target anchor frame.

[0168] Specifically: Use the risk identification and positioning model to identify the area where the risk subject is located in the fused data, and select the area. Use the positioning module in the risk identification and positioning model to determine the coordinates of the selected area, and display the area corresponding to the coordinates separately from the non-coordinate area.

[0169] Among them, the risk subjects include the operators during distribution network operations, the tools used by the operators, the various equipment in the distribution network operating environment, and the distribution network operating environment.

[0170] According to one embodiment of the present invention, the identification and positioning process of the risk subject of distribution network operation is specifically as follows: based on a heterogeneous multi-source cross-modal data feature set, i.e., fused data, as training data for the risk subject identification and positioning model, so that the risk subject identification and positioning model can realize dynamic tracking of the risk subject, and use the visible light image superimposed on the target anchor frame as output data. After the risk subject identification and positioning model is trained, the fused data is input into the risk subject identification and positioning model to obtain the location information of the risk subject, i.e., the output data of the risk subject identification and positioning model with the visible light data superimposed on the target anchor frame.

[0171] Afterwards, the on-site locations are divided and located according to the safety type and safety level. The safety precautions and hazard prevention measures corresponding to each location are associated with the safety level, precautions, operating standards and other information corresponding to the equipment at that location. The safety content is decomposed, and the location of the operators performing distribution network operations is located based on the location information of the risk subject. Information such as ongoing operations, maintenance work, current location safety information, risk prevention measures, and historical accident records is presented.

[0172] Step S4 specifically includes:

[0173] The risk factors in the operation scenario are expressed in two dimensions, including: constructing the correspondence between the quantitative evaluation index of risk factors and the visualized RGB color; and performing risk data analysis based on the risk factor feature set, i.e., each fused data, and quantitative fuzzy evaluation. The risk level of the risk subject area location is visualized.

[0174] Specifically, in the operation scenario, when the risk subject is not in a risk state, its risk level is 0; when the risk subject is in a risk state, the risk level index is calculated based on the number of risk categories N, the severity M (value 1 or 2, 1 for general violations and 2 for serious violations, determined according to the operating procedures) and the duration T (calculated in seconds).

[0175] Calculation rules for risk level indicator S score: S = N·M·T.

[0176] According to the score of the risk level indicator, the corresponding RGB color is used to mark the location of the main risk of the operation. The specific color can be selected according to the actual occasion and public habits. The following is only an exemplary introduction to the colors corresponding to each level for reference.

[0177] When S=0, blue is selected; when 1≤S<2, yellow is selected; when 2≤S<4, orange is selected; when S≥4, red is selected and rendered on the image.

[0178] The visual presentation effect of distribution network operation risks under different conditions is as follows: Figure 5 and Figure 6 As shown, in Figure 5 The red anchor box in the middle selects the operator performing distribution network operations, the blue anchor box selects the fence set up at the distribution network operation site, that is, the equipment in the distribution network operation, and the operator's ongoing behavior, that is, the behavior of crossing the fence, is marked above the red anchor box.

[0179] exist Figure 6 In the figure, the green anchor box selects the operator who performs distribution network operations, and the status of the current operator is described above the green anchor box. The blue anchor box selects the tool used at the distribution network operation site, namely the ladder, and the status bar above the blue anchor box describes the status of the ladder.

[0180] Example 2

[0181] The present invention based on the same inventive concept also provides a distribution network operation risk visualization system, such as Figure 7 As shown, it includes: a data acquisition module, a fusion module, a position determination module and a visualization module;

[0182] The data acquisition module is used to: acquire multimodal data of the distribution network within a target operation cycle, wherein the multimodal data includes image information, data information and text information;

[0183] The fusion module is used to: extract features from image information, data information and text information, and then perform data fusion;

[0184] The location determination module is used to: take the fused data image data as input, use the trained risk subject identification and positioning model to identify the risk subject and determine the location of the risk subject in the image;

[0185] The visualization module is used to mark the image based on the position of the risk subject in the image and the corresponding risk status type to perform risk visualization display.

[0186] Specifically, the image information includes: visible light images and infrared light images; the data information includes: environmental data and equipment operation data; and the text information includes: distribution network operation history text data.

[0187] Here, the fusion module may include: an image processing submodule, a data processing submodule, a text processing submodule, and a fusion processing submodule, etc.;

[0188] The image processing submodule is used to: obtain image features by cross-modal data fusion and superposition of visible light images and infrared light images;

[0189] Furthermore, the image processing submodule is specifically used for:

[0190] The infrared light image and the visible light data image are processed in the form of histograms to obtain a gradient histogram feature vector of the infrared light image and a gradient histogram feature vector of the visible light image;

[0191] Convolutional neural networks are used to extract common features from the gradient histogram feature vectors of infrared images and the gradient histogram feature vectors of visible light images respectively;

[0192] A convolutional neural network is used to extract discriminative features from the feature vector of the gradient histogram of the infrared image and the feature vector of the gradient histogram of the visible light image respectively;

[0193] Superimpose the common features and the discriminative features corresponding to the feature vector of the visible light image gradient histogram to obtain the superimposed features of the visible light image;

[0194] Superimpose the common features and the discriminative features corresponding to the feature vector of the infrared image gradient histogram to obtain the superimposed features of the infrared image;

[0195] The visible light image superposition features and the infrared light image superposition features are fused.

[0196] The data processing submodule is used to: extract features from the environmental data and the equipment operation data to obtain environmental feature data and equipment feature data;

[0197] Furthermore, the data processing submodule is specifically used for:

[0198] The environmental characteristics are constructed based on the geographical coordinates of the working environment location and superimposed on the environmental data within the working cycle;

[0199] Collecting equipment operation data of each equipment in the distribution network; the equipment operation data includes equipment operation signals and fault information;

[0200] Feature extraction is performed based on equipment operation signals and fault information to obtain the corresponding relationship between faults and features; and the corresponding equipment operation characteristics are obtained based on the corresponding relationship between each fault and feature.

[0201] The text processing submodule is used to: extract features and classify and identify the historical text data of distribution network operations to obtain text features;

[0202] Furthermore, the text processing submodule is specifically used for:

[0203] Vectorization preprocessing of historical text dataset, two sample sentences;

[0204] Perform text overall feature extraction and text feature segmentation on two sample sentences respectively;

[0205] The action text features corresponding to the two sample sentences are fused and classified to obtain the classification result of the action text;

[0206] The relational text features corresponding to the two sample sentences are fused and classified to obtain the relational text classification result;

[0207] The classification results of action text and relationship text are used as text features.

[0208] The fusion processing submodule is used to: construct the state characteristics of the distribution network operation after feature extraction of environmental feature data, equipment feature data and text data sets; and to construct feature vectors based on image features and state features, and encode the feature vectors to obtain fused data.

[0209] Furthermore, the fusion processing submodule is specifically used for:

[0210] Extract features from environmental feature data, equipment feature data, and text data sets to obtain environmental features, historical text features, and equipment operation features;

[0211] Based on the influence of each equipment operation characteristic on the equipment operation status, the key equipment operation characteristics are selected from the equipment operation characteristics;

[0212] Environmental features, historical text features and key equipment operation features are used as status features.

[0213] Furthermore, the risk subject identification and positioning model in the location determination module can be trained in the following manner:

[0214] Building a sample set based on historical fused data and location information of risk subjects corresponding to the fused data;

[0215] Dividing the sample set into a training set and a test set according to a set ratio;

[0216] Taking the fused data in the training set as input and the location information of the risk subject corresponding to the fused data in the training set as output, the convolutional neural network is trained to obtain a preliminarily trained risk location model;

[0217] The preliminarily trained risk positioning model is tested based on the test set to revise the preliminarily trained risk positioning model to obtain a trained risk subject identification and positioning model.

[0218] Furthermore, the location determination module is specifically used for:

[0219] The risk location model is used to locate the area where the risk subject in the fused data is located, the area is framed, and the coordinates of the framed area are determined.

[0220] Furthermore, the visualization module is specifically used to include:

[0221] Based on the risk subject in the framed area of ​​the image, determining a corresponding operation risk status type, and determining a risk level according to the operation risk status type;

[0222] The color of the framed area is determined by using the corresponding RGB color according to the risk level and rendered on the image to achieve a visual display of the risk.

[0223] The determination of risk levels here includes:

[0224] In the operation scenario, when the risk subject is not in a risk state, the risk level is 0;

[0225] When the risk subject is in a risk state, the risk level index is calculated based on the number of risk categories, severity and duration;

[0226] The risk level is determined according to the risk level indicator.

[0227] Example 3

[0228] like Figure 8 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store a computer execution program, and an exemplary computer execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0229] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding functions, so as to realize the steps of a distribution network operation risk visualization method in the above-mentioned embodiment.

[0230] Example 4

[0231] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in electronic devices and, of course, extended storage media supported by electronic devices. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a distribution network operation risk visualization method in the above embodiment.

[0232] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0233] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0234] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0235] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0236] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A method for visualizing distribution network operation risks. It is characterized in that The method comprises: Acquire multimodal data of the distribution network within a target operation cycle, wherein the multimodal data includes image information, data information, and text information; After extracting features from image information, data information and text information, data fusion is performed; Taking the fused image data as input, the risk subject is identified using the trained risk subject identification and positioning model and the position of the risk subject in the image is determined; Based on the position of the risk subject in the image and the corresponding risk status type, the image is marked to perform risk visualization.

2. The method according to claim 1, It is characterized in that The image information includes: visible light images and infrared light images; the data information includes: environmental data and equipment operation data; the text information includes: distribution network operation history text data.

3. The method according to claim 1, It is characterized in that After extracting features from the image information, data information and text information, data fusion is performed, including: The image features are obtained by cross-modal data fusion and superposition of visible light images and infrared images; Extracting features from the environment data and the equipment operation data to obtain environment feature data and equipment feature data; Extract features and classify the historical text data of distribution network operations to obtain text features; After extracting environmental feature data, equipment feature data and text features, the state features of the distribution network operation are constructed; A feature vector is constructed based on the image features and state features, and the feature vector is encoded to obtain fused data.

4. The method according to claim 3, It is characterized in that The state characteristics of the distribution network operation are constructed after extracting the characteristics of the environment characteristic data, the equipment characteristic data and the text data set, including: Perform feature extraction on environmental feature data, equipment feature data and text data sets to obtain environmental features, historical text features and equipment operation features; Based on the influence of each equipment operation characteristic on the equipment operation status, the key equipment operation characteristics are selected from the equipment operation characteristics; Environmental features, historical text features and key equipment operation features are used as status features.

5. The method according to claim 3, It is characterized in that The method of obtaining image features by cross-modal data fusion and superposition of visible light images and infrared light images includes: The infrared light image and the visible light data image are processed in the form of histograms to obtain a gradient histogram feature vector of the infrared light image and a gradient histogram feature vector of the visible light image; Convolutional neural networks are used to extract common features from the gradient histogram feature vectors of infrared images and the gradient histogram feature vectors of visible light images respectively; A convolutional neural network is used to extract discriminative features from the feature vector of the gradient histogram of the infrared image and the feature vector of the gradient histogram of the visible light image respectively; Superimpose the common features and the discriminative features corresponding to the feature vector of the visible light image gradient histogram to obtain the superimposed features of the visible light image; Superimpose the common features and the discriminative features corresponding to the feature vector of the infrared image gradient histogram to obtain the superimposed features of the infrared image; The visible light image superposition features and the infrared light image superposition features are fused.

6. The method according to claim 3, It is characterized in that The feature extraction of environmental data and equipment operation data includes: The environmental characteristics are constructed based on the geographical coordinates of the working environment location and superimposed on the environmental data within the working cycle; Collecting equipment operation data of each equipment in the distribution network; the equipment operation data includes equipment operation signals and fault information; Feature extraction is performed based on equipment operation signals and fault information to obtain the corresponding relationship between faults and features; and the corresponding equipment operation characteristics are obtained based on the corresponding relationship between each fault and feature.

7. The method according to claim 3, It is characterized in that The feature extraction and classification of the distribution network operation history text data to obtain text features includes: Vectorization preprocessing of historical text dataset, two sample sentences; Perform text overall feature extraction and text feature segmentation on two sample sentences respectively; The action text features corresponding to the two sample sentences are fused and classified to obtain the classification result of the action text; The relational text features corresponding to the two sample sentences are fused and classified to obtain the relational text classification result; The classification results of action text and relationship text are used as text features.

8. The method according to claim 3, It is characterized in that The training process of the risk subject identification and positioning model includes: Constructing a sample set based on historical fused data and location information of risk subjects corresponding to the fused data; Dividing the sample set into a training set and a test set according to a set ratio; Taking the fused data in the training set as input and the location information of the risk subject corresponding to the fused data in the training set as output, the convolutional neural network is trained to obtain a preliminarily trained risk location model; The preliminarily trained risk positioning model is tested based on the test set to revise the preliminarily trained risk positioning model to obtain a trained risk subject identification and positioning model.

9. The method according to claim 1, It is characterized in that The method uses the fused image data as input, uses the trained risk subject identification and positioning model to identify the risk subject and determine the position of the risk subject in the image, including: The risk location model is used to locate the area where the risk subject in the fused data is located, the area is framed, and the coordinates of the framed area are determined.

10. The method according to claim 9, It is characterized in that The step of marking the image based on the position of the risk subject in the image and the corresponding risk status type for risk visualization includes: Based on the risk subject in the framed area of ​​the image, determining a corresponding operation risk status type, and determining a risk level according to the operation risk status type; The color of the framed area is determined by using the corresponding RGB color according to the risk level and rendered on the image to achieve a visual display of the risk.

11. The method according to claim 10, It is characterized in that Determining the risk level according to the operation risk status type includes: In the operation scenario, when the risk subject is not in a risk state, the risk level is 0; When the risk subject is in a risk state, the risk level index is calculated based on the number of risk categories, severity and duration; The risk level is determined according to the risk level indicator.

12. A distribution network operation risk visualization system, It is characterized in that include: Data acquisition module, fusion module, location determination module and visualization module; The data acquisition module is used to: acquire multimodal data of the distribution network within a target operation cycle, wherein the multimodal data includes image information, data information and text information; The fusion module is used to: extract features from image information, data information and text information, and then perform data fusion; The position determination module is used to: take the fused data image data as input, use the trained risk subject identification and positioning model to identify the risk subject and determine the position of the risk subject in the image; The visualization module is used to mark the image based on the position of the risk subject in the image and the corresponding risk status type to perform risk visualization display.

13. The system of claim 12, It is characterized in that The image information includes: visible light images and infrared light images; the data information includes: environmental data and equipment operation data; the text information includes: distribution network operation history text data.

14. The system of claim 12, It is characterized in that The fusion module includes: an image processing submodule, a data processing submodule, a text processing submodule and a fusion processing submodule; The image processing submodule is used to: obtain image features by cross-modal data fusion and superposition on visible light images and infrared light images; The data processing submodule is used to: extract features from the environment data and the equipment operation data to obtain environment feature data and equipment feature data; The text processing submodule is used to: extract features and classify and identify the historical text data of the distribution network operation to obtain text features; The fusion processing submodule is used to: construct the state characteristics of the distribution network operation after feature extraction of environmental feature data, equipment feature data and text features; and to construct a feature vector based on image features and state features, and encode the feature vector to obtain fused data.

15. The system of claim 14, It is characterized in that The fusion processing submodule is specifically used for: Perform feature extraction on environmental feature data, equipment feature data and text data sets to obtain environmental features, historical text features and equipment operation features; Based on the influence of each equipment operation characteristic on the equipment operation status, the key equipment operation characteristics are selected from the equipment operation characteristics; Environmental features, historical text features and key equipment operation features are used as status features.

16. The system of claim 14, It is characterized in that The image processing submodule is specifically used for: The infrared light image and the visible light data image are processed in the form of histograms to obtain a gradient histogram feature vector of the infrared light image and a gradient histogram feature vector of the visible light image; Convolutional neural networks are used to extract common features from the gradient histogram feature vectors of infrared images and the gradient histogram feature vectors of visible light images respectively; A convolutional neural network is used to extract discriminative features from the feature vector of the gradient histogram of the infrared image and the feature vector of the gradient histogram of the visible light image respectively; Superimpose the common features and the discriminative features corresponding to the feature vector of the visible light image gradient histogram to obtain the superimposed features of the visible light image; Superimpose the common features and the discriminative features corresponding to the feature vector of the infrared image gradient histogram to obtain the superimposed features of the infrared image; The visible light image superposition features and the infrared light image superposition features are fused.

17. The system of claim 14, It is characterized in that The data processing submodule is specifically used for: The environmental characteristics are constructed based on the geographical coordinates of the working environment location and superimposed on the environmental data within the working cycle; Collecting equipment operation data of each equipment in the distribution network; the equipment operation data includes equipment operation signals and fault information; Feature extraction is performed based on equipment operation signals and fault information to obtain the corresponding relationship between faults and features; Based on the corresponding relationship between each fault and feature, the corresponding equipment operation characteristics are obtained.

18. The system of claim 14, It is characterized in that The text processing submodule is specifically used for: Vectorization preprocessing of historical text dataset, two sample sentences; Perform text overall feature extraction and text feature segmentation on two sample sentences respectively; The action text features corresponding to the two sample sentences are fused and classified to obtain the classification result of the action text; The relational text features corresponding to the two sample sentences are fused and classified to obtain the relational text classification result; The classification results of action text and relationship text are used as text features.

19. The system of claim 14, It is characterized in that The system also includes a training process of a risk subject identification and positioning model, specifically including: Constructing a sample set based on historical fused data and location information of risk subjects corresponding to the fused data; Dividing the sample set into a training set and a test set according to a set ratio; Taking the fused data in the training set as input and the location information of the risk subject corresponding to the fused data in the training set as output, the convolutional neural network is trained to obtain a preliminarily trained risk location model; The preliminarily trained risk positioning model is tested based on the test set to revise the preliminarily trained risk positioning model to obtain a trained risk subject identification and positioning model.

20. The system of claim 12, It is characterized in that The location determination module is specifically used for: The risk location model is used to locate the area where the risk subject in the fused data is located, the area is framed, and the coordinates of the framed area are determined.

21. The system of claim 12, It is characterized in that The visualization module is specifically used to include: Based on the risk subject in the framed area of ​​the image, determining a corresponding operation risk status type, and determining a risk level according to the operation risk status type; The color of the framed area is determined by using the corresponding RGB color according to the risk level and rendered on the image to achieve a visual display of the risk.

22. The system of claim 21, It is characterized in that The determination of the risk level includes: In the operation scenario, when the risk subject is not in a risk state, the risk level is 0; When the risk subject is in a risk state, the risk level index is calculated based on the number of risk categories, severity and duration; The risk level is determined according to the risk level indicator.

23. An electronic device, It is characterized in that include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the distribution network operation risk visualization method according to any one of claims 1 to 11 is implemented.

24. A readable storage medium, It is characterized in that A computer program is stored thereon, and when the computer program is executed, the distribution network operation risk visualization method as described in any one of claims 1 to 11 is implemented.

Citation Information

Cited By

  • Automatic intelligent operation and maintenance method and system for power grid dispatching based on multi-modal large model

    CN119379263A

  • Grid Dispatching Automation Intelligent Operation and Maintenance Method and System Based on Multimodal Large Model

    CN119379263B

  • Power generation enterprise operation risk identification method and system based on multi-modal large model

    CN119539494A

  • Risk assessment method, device and equipment for operation state of power grid and medium

    CN120087748A

  • Safety tool intelligent management full-life-cycle tracing system based on Internet of Things

    CN120634277A