A monitoring and warning method and its control system for intelligent substations

Abstract

The present invention relates to the technical field of substation monitoring and early warning, and particularly refers to a monitoring and early warning method and its control system for intelligent substations. By deploying distributed rotatable cameras in multiple areas of the substation, real-time image data of the target is dynamically obtained, and combined with the collection of equipment operation parameters, comprehensive coverage of equipment status and environmental data is achieved. Through time series modeling using the Long Short-Term Memory network (LSTM), the dynamic trends of operation parameters are effectively captured to identify potential anomalies. By combining semantic segmentation technology with equipment type feature extraction, the accuracy of equipment anomaly determination is significantly improved, and the false alarm and missed alarm rates are reduced. Through the combination of rotatable cameras, convolutional neural network (CNN), and multi-modal data fusion, accurate personnel identity recognition and behavior path tracking are achieved, non-staff and abnormal behaviors are intelligently detected, and potential safety threats are timely warned.

Description

Technical Field

[0001] The present invention relates to the technical field of substation monitoring and early warning, and particularly to a monitoring and early warning method and its control system for intelligent substations. Background Art

[0002] With the rapid development of the power system, as a key node for power transmission and distribution, the reliability and safety of substation operation are crucial for the stability of the power system. The monitoring and management of traditional substations mainly rely on manual inspections and single-point equipment monitoring systems. However, these methods have certain limitations. For example, it is difficult to achieve real-time monitoring of the operating status of equipment during manual inspections, and human factors may lead to monitoring omissions; while the single-point equipment monitoring system lacks the ability of overall correlation analysis and is difficult to identify potential equipment failures or safety hazards in a timely manner. The data between the monitoring modules in the existing systems are independent of each other and lack comprehensive analysis capabilities. At the same time, abnormal judgments are made based on fixed thresholds, without considering the timing characteristics of equipment operation, which easily leads to false alarms or missed alarms. The existing substation monitoring systems mainly focus on the operating status of equipment and lack intelligent monitoring of abnormal behaviors of on-site personnel or non-staff, resulting in safety risks. Summary of the Invention

[0003] To solve the above problems, the present invention provides a monitoring and early warning method and its control system for intelligent substations, which realizes intelligent monitoring and early warning by combining the deployment of rotatable cameras, collecting the operating parameters of substation equipment, and combining deep learning.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A monitoring and early warning control system for intelligent substations includes: a real-time monitoring module, an operation abnormality determination module, a behavior determination module, and an early warning control module. The operation abnormality determination module and the behavior determination module are respectively connected to the real-time monitoring module, and the early warning control module is respectively connected to the operation abnormality determination module and the behavior determination module;

[0006] The real-time monitoring module is used to distributively obtain real-time image data of different targets in the substation and is also used to obtain the operating parameters of substation equipment;

[0007] The operation abnormality determination module is used to determine equipment operation abnormalities through timing analysis based on the operating parameters of substation equipment, perform semantic segmentation on real-time image data, and extract features characterized by equipment types to determine equipment characterization abnormalities in the substation;

[0008] The behavior determination module is used to determine the identity of personnel according to real-time image data, match the personnel permissions according to the identity, track the personnel behavior path based on the convolutional neural network according to the real-time image data, and determine the abnormal behavior of personnel;

[0009] The warning control module is used to classify the warnings according to the equipment operation abnormality, equipment characterization abnormality or personnel behavior abnormality and control the execution of warning actions.

[0010] Further, the distributed acquisition of real-time image data of different targets in the substation includes the following steps:

[0011] Set a number of rotatable cameras at different positions within the substation area;

[0012] Preset a plane rectangular coordinate system for the substation area in the cloud platform;

[0013] Match the corresponding coordinates in the rectangular plane of the plane rectangular coordinate system according to the positions of the rotatable cameras;

[0014] Determine the target position based on the orientation of the rotatable camera and the coordinates of the rotatable camera, and obtain real-time images of different targets.

[0015] Further, the operating parameters of the substation equipment include: electrical parameters, environmental parameters, vibration parameters, insulation state parameters and switch action parameters.

[0016] Further, the determination of equipment operation abnormality by time series analysis according to the operating parameters of the substation equipment includes:

[0017] Perform piecewise normalization processing on the real-time collected operating parameters, and extract the change trend of the operating parameters;

[0018] Based on the long short-term memory network model, perform time series modeling on the normalized operating parameters, and predict the parameter changes in a future period of time by learning the characteristics of historical operating data;

[0019] Dynamically compare the predicted parameters with the actual parameters collected in real time, and use the set abnormal threshold to judge the deviation degree;

[0020] Perform clustering analysis on the operating parameters outside the threshold range, match the corresponding abnormal patterns in combination with the abnormal type library, generate the operation abnormality determination result, and associate the equipment components causing the abnormality.

[0021] Further, the determination of equipment characterization abnormality in the substation by semantic segmentation of real-time image data and feature extraction characterized by equipment type includes:

[0022] Segment the real-time image data based on the DeepLab semantic segmentation model, divide the image into a background area and a device target area, and extract the contour and position features of the device target;

[0023] Extract the shape features, color features, and texture features of the device;

[0024] Use a graph neural network to extract features from the segmented device target area, represent the pixels of the device as graph nodes, and generate a geometric shape feature vector of the device through the aggregation of node features and adjacency relationships;

[0025] Match the generated geometric shape feature vector with a preset feature library of device types, and determine whether there is an abnormality in the device representation based on the similarity measure in the feature embedding space.

[0026] Furthermore, in the graph neural network, the formula for feature update is as follows:

[0027] ;

[0028] Among them, is the representation of each node at the k-th layer; the representation of each node at the k+1-th layer; is the set of neighbor nodes of node ; is a neighbor node of node ; is the normalized weight between node and neighbor node ; represents the weight matrix at the k-th layer; represents the non-linear activation function.

[0029] Furthermore, the determination of the personnel identity based on the real-time image data and the matching of the personnel permissions according to the identity include:

[0030] Extract the feature points of the face and body contour of the personnel through the real-time image data combined with the key point detection algorithm, and generate a feature vector of the personnel based on a preset pose model;

[0031] Use multi-modal data fusion to associate and match the extracted personnel feature vector with the identity feature information obtained based on the device interaction data to generate a comprehensive identity determination;

[0032] Retrieve the behavior rules and device operation permissions corresponding to the identity in the permission management module through the matched identity features, including the accessible areas, the categories of devices to be operated, and the operation time limits.

[0033] Furthermore, the personnel behavior path tracking based on the convolutional neural network according to the real-time image data, and the determination of abnormal personnel behavior include:

[0034] S1. Based on the real-time image data, use the convolutional neural network to detect personnel targets, and generate the unique identity identifier and movement trajectory points of the target personnel;

[0035] S2. Based on the orientation of the rotatable camera and the coordinates of the rotatable camera, determine the target position, and adjust the rotation angle, focal length, and field of view range of the camera in real time, continuously obtain the real-time image of the target personnel, and update the movement trajectory points;

[0036] S3. Based on the updated trajectory points, calculate the movement direction and speed of the target personnel, and when the personnel is about to leave the current camera field of view, predict its expected position in the next rotatable camera field of view based on the coordinates of the rotatable camera, and transmit the prediction parameters and target identity identifier to the next camera;

[0037] S4. After the next camera receives the parameters transmitted in step S3, continue to track the target personnel in combination with the real-time image, generate an interconnected movement trajectory segment, and perform a fusion verification of the time stamp and spatial position of this segment and the trajectory segment recorded by the previous camera to generate a complete behavior path;

[0038] S5. Match and analyze the complete behavior path with the preset activity rules, and identify abnormal personnel behavior based on path morphological features, regional coverage relationships, and the distribution of stop points. The abnormal personnel behavior includes unauthorized area entry, abnormal stay, and abnormal wandering.

[0039] A monitoring and early warning method for an intelligent substation, which is applied to the monitoring and early warning control system for an intelligent substation described in any one of the foregoing items, and includes:

[0040] Distributedly obtain the real-time image data of different targets in the substation and obtain the operating parameters of the substation equipment;

[0041] Judge the abnormal operation of the equipment through time series analysis according to the operating parameters of the substation equipment, perform semantic segmentation through the real-time image data, and extract the features characterized by the equipment type to determine the abnormal representation of the equipment in the substation;

[0042] Determine the identity of personnel according to the real-time image data, match the personnel permissions according to the identity, and perform personnel behavior path tracking based on the convolutional neural network according to the real-time image data to determine abnormal personnel behavior;

[0043] Classify the alarms according to the abnormal operation of the equipment, the abnormal representation of the equipment, or the abnormal personnel behavior, and control the execution of the alarm actions.

[0044] The beneficial effects of the present invention are as follows: By deploying distributed rotatable cameras in multiple areas of the substation to dynamically obtain real-time image data of different targets in the substation and collecting equipment operation parameters, the comprehensive coverage of equipment status and environmental data is ensured. The obtained data is fused through a centralized processing module, breaking the limitation of the independent data of the monitoring modules in the traditional system and realizing comprehensive analysis within the whole substation. By using the time series analysis algorithm and performing time series modeling on the equipment operation parameters through the long short-term memory network (LSTM) model, the dynamic change trend of equipment operation can be effectively captured and potential anomalies can be identified. In addition, by combining semantic segmentation technology with equipment types for feature extraction, the accuracy of equipment anomaly determination is further improved, and the incidence of false alarms and missed alarms is significantly reduced. Aiming at the problem of weak personnel behavior monitoring ability in the traditional system, through the combination of rotatable cameras, convolutional neural network (CNN) and multimodal data fusion technology, not only can the personnel identity be accurately identified, but also the personnel activity trajectory can be dynamically monitored through behavior path tracking. Especially the intelligent identification function for non-staff or abnormal behaviors can timely detect and warn potential security threats. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 FIG. is a schematic structural diagram of a monitoring and early warning control system for an intelligent substation in the present invention.

[0046] Figure 2 FIG. is a flowchart of the steps of a monitoring and early warning method for an intelligent substation in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] Please refer to Figure 1 - Figure 2 As shown in the figure, the present invention relates to a monitoring and early warning control system for an intelligent substation, including: a real-time monitoring module, an operation anomaly determination module, a behavior determination module, and an early warning control module. The operation anomaly determination module and the behavior determination module are respectively connected to the real-time monitoring module, and the early warning control module is respectively connected to the operation anomaly determination module and the behavior determination module;

[0048] The real-time monitoring module is used to distributively obtain real-time image data of different targets in the substation and is also used to obtain the operation parameters of the substation equipment;

[0049] The operation anomaly determination module is used to determine equipment operation anomalies through time series analysis based on the operation parameters of the substation equipment, perform semantic segmentation on the real-time image data, and extract features characterized by equipment types to determine equipment characterization anomalies in the substation;

[0050] The behavior determination module is used to determine the identity of personnel based on real-time image data, match personnel permissions according to the identity, track the behavior path of personnel based on the convolutional neural network according to the real-time image data, and determine abnormal personnel behaviors;

[0051] The early warning control module is used to classify alarms according to equipment operation anomalies, equipment characterization anomalies or personnel behavior anomalies and control the execution of alarm actions.

[0052] In some embodiments, the system realizes the real-time monitoring of different targets in the substation through the hardware composition and integration of the real-time monitoring module. The real-time monitoring module consists of multiple rotatable cameras deployed distributively, and these cameras have high-definition resolution and multi-dimensional adjustable mechanical pan-tilt heads. Through the embedded motion control unit, the cameras can rotate, pitch and adjust the focal length according to the instructions of the central control system to accurately cover the key equipment areas and pedestrian flow channels in the substation. The cameras and multiple environmental sensors (including temperature sensors, humidity sensors, vibration sensors and current-voltage detection modules) form an integrated acquisition network to collect equipment operation parameters and environmental status in real time. The operation anomaly determination module mainly relies on a high-performance edge computing unit for data processing, and its hardware structure includes a multi-core CPU and an efficient GPU running in cooperation. After the operation parameters are collected by the sensors, the edge computing unit completes data cleaning, format conversion and compression, and then inputs them into the time series analysis model. This model runs based on an embedded FPGA acceleration unit and uses a long short-term memory network (LSTM) to perform real-time modeling of the equipment operation parameters. Through this module, the dynamic operation trend of the equipment is captured, and combined with the historical anomaly feature library and the semantic segmentation algorithm, the feature vectors of the equipment characterization are automatically extracted to analyze and determine whether the equipment characterization is abnormal. The hardware configuration of the behavior determination module includes a dedicated AI processing chip (such as the NVIDIA Jetson platform), which is equipped with a deep learning acceleration unit for running a convolutional neural network (CNN) model. The images captured by the cameras in real time are parsed by the neural network in the module to extract the key point features of personnel. Combining multi-modal sensor data, the system uses a behavior path generation algorithm to track the movement trajectory of personnel in real time. In particular, the behavior determination module is built-in with a permission rule memory and an identity comparison unit, and can determine the identity and permissions of personnel through quick matching with the database. For unauthorized personnel or detected abnormal behaviors, the module generates a warning signal and sends it to the early warning control module. The hardware of the early warning control module consists of a central control unit (MCU) based on the ARM architecture and a programmable logic controller (PLC). After receiving the abnormal signal, the early warning control module determines the alarm priority according to the classification algorithm and completes the alarm action through multiple execution units (such as an audible and visual alarm, a remote control switch or a communication interface). At the same time, the warning information will be synchronously uploaded to the cloud platform storage module to provide real-time feedback and post-event analysis basis for the operation and maintenance personnel.

[0053] Furthermore, the distributed acquisition of real-time image data of different targets in the substation includes the following steps:

[0054] Set a number of rotatable cameras at different positions within the substation area;

[0055] Preset a plane rectangular coordinate system for the substation area in the cloud platform;

[0056] Match the corresponding coordinates in the right plane of the plane rectangular coordinate system according to the positions of the rotatable cameras;

[0057] Determine the target positions based on the orientations of the rotatable cameras and the coordinates of the rotatable cameras, and acquire real-time images of different targets.

[0058] In some embodiments, a number of rotatable cameras are deployed inside the substation area. These cameras have precise positioning capabilities and programmable rotation control functions. The position of each camera is stored in a plane rectangular coordinate system preset on the cloud platform in the form of three-dimensional coordinates. To ensure the accuracy of the coordinate system, the system adopts a camera self-calibration algorithm. By comparing the shooting results of known reference points with the preset reference coordinates, it automatically adjusts the installation angle of the camera and the coordinate record to eliminate the offset caused by installation errors. During the process of real-time target image acquisition, the system first calculates the theoretical position of the target object in the camera's field of view using a coordinate mapping algorithm based on inverse projection according to the position coordinates and orientation information of the rotatable camera. This algorithm converts the global coordinates of the target into the field-of-view coordinates of the camera and calculates the relative position of the target in the imaging plane in combination with the pitch angle and focal length information of the camera. During this process, to reduce the camera rotation control error, the system adjusts the rotation speed and angle of the camera in real time through a proportional-integral-derivative (PID) control algorithm to ensure that the target is always at the center of the best field of view of the camera. When the fields of view of multiple cameras cover the same area, to avoid data redundancy, the system introduces an image feature fusion algorithm. Based on a weighted feature matching method, this algorithm fuses the images from different cameras through geometric correction of feature points and weight fusion of feature vectors to generate a panoramic view. Through this method, not only the problem of the limited perspective of a single camera is solved, but also the integrity and consistency of the image data are enhanced. In addition, to improve the accuracy of target detection, the system combines a deep learning model for target detection and region segmentation during the real-time image acquisition stage. The YOLOv5 (You Only Look Once, Version 5) model is used to classify and locate the targets in the real-time image. The model is optimized using a large amount of actual substation scene data during the training stage to improve the recognition ability for common substation equipment (such as transformers, switchgear) and dynamic targets (such as personnel, vehicles). During real-time operation, the model processes the data streams of multiple cameras through a parallel acceleration algorithm (such as TensorRT optimization) to ensure real-time response capabilities.

[0059] Further, the operating parameters of the substation equipment include: electrical parameters, environmental parameters, vibration parameters, insulation status parameters, and switch operation parameters.

[0060] Specifically, the electrical parameters include core data such as current, voltage, power factor, frequency, and total harmonic distortion. The system uses high-precision current transformers and voltage transformers to collect these signals in real time and performs digital processing through a multi-channel analog-to-digital conversion module (ADC). To ensure the stability and anti-interference ability of the data, the sensors are equipped with low-noise operational amplifiers and combined with differential signal processing technology to eliminate common-mode interference. In addition, the electrical parameters extract harmonic characteristics through an algorithm based on Fourier transform for analyzing power quality and abnormal load conditions. The environmental parameters include the temperature, humidity, and air pressure in the area where the equipment is located. The system obtains these data through a distributed environmental monitoring sensor network. These sensors have the characteristics of wide measurement range and high sensitivity and can operate stably under extreme conditions such as high temperature and high humidity. The environmental parameter data is uploaded to the edge computing node through the LoRa wireless communication module to analyze the impact of environmental changes on the equipment operation in real time and generate trend predictions. The vibration parameters are obtained through high-frequency acceleration sensors, and the sensors are arranged on the mechanical components of key equipment (such as transformers and switchgear). The vibration signals collected by the sensors are processed by a band-pass filter to remove low-frequency noise, and at the same time, key vibration characteristics (such as amplitude and frequency) are extracted through time-domain and frequency-domain analysis algorithms. These characteristics are combined with the fault diagnosis model to determine whether there are potential abnormalities in the mechanical components of the equipment, such as looseness or fatigue. The insulation state parameters involve the insulation resistance value and dielectric loss factor of the equipment. The system uses on-line insulation monitoring equipment to regularly collect the insulation characteristic data of the equipment based on the high-voltage test method and dielectric response analysis method. Through the time-series modeling of the change trend of the insulation parameters, the system can predict the insulation aging process and potential breakdown risks. The switch operation parameters include the opening and closing times, operating current, and current waveform characteristics of the circuit breaker. The system uses a high-frequency sampling unit to record the dynamic current data during the switch operation process and compares it with the reference waveform of the normal switch operation through a waveform matching algorithm to quickly identify the operation abnormalities of the switch mechanical components or electromagnetic mechanisms.

[0061] Further, the determination of equipment operation anomalies through time-series analysis based on the operation parameters of substation equipment includes:

[0062] Perform piecewise normalization processing on the real-time collected operation parameters and extract the change trend of the operation parameters;

[0063] Based on the long short-term memory network model, perform time-series modeling on the normalized operation parameters, and predict the parameter changes in a future period of time by learning the characteristics of historical operation data;

[0064] Dynamically compare the predicted parameters with the actual parameters collected in real time, and use the set anomaly threshold to judge the deviation degree;

[0065] Perform cluster analysis on operating parameters outside the threshold range, match the corresponding abnormal patterns in combination with the abnormal type library, generate an operating abnormality determination result, and associate the equipment components that cause the abnormality.

[0066] In some embodiments, first, for the real-time collected operating parameter data (including electrical parameters, environmental parameters, vibration parameters, etc.), the system uses a segmented normalization processing algorithm to standardize the original data. The normalization process scales the data to a unified numerical range (such as [0, 1]) through a linear transformation formula to eliminate the model processing deviation caused by different dimensions or different numerical ranges of equipment parameters. At the same time, the system performs segmented analysis on the operating parameters according to the time period characteristics, extracts the change trend of each segment of data, and uses a sliding window mechanism to generate a smoothed trend sequence for subsequent modeling. In the time series modeling stage, the system uses a long short-term memory network (LSTM) model to construct a time series prediction framework. The LSTM model can effectively process the long-term dependence relationship and short-term change characteristics of equipment operating parameters through its gated unit structure. The input of the model is the normalized parameter sequence, and the output is the parameter prediction value for a specified future time period. During the training process, the system uses a large amount of historical operating data to perform supervised learning on the LSTM model. The objective function is to minimize the mean square error (MSE) between the predicted value and the actual value, and the prediction accuracy of the model is improved by iteratively optimizing the weights. Dynamically comparing the predicted parameters with the actual parameters collected in real time is a key step in anomaly detection. The system uses a dynamic threshold setting method in the comparison process. The threshold is set based on the historical distribution characteristics of the operating parameters and the current working condition requirements, and is dynamically adjusted through statistical methods (such as standard deviation calculation) to adapt to the diversity of equipment operation. When the comparison result shows that the actual parameter deviates from the predicted parameter and exceeds the threshold range, the system marks the parameter as an abnormal candidate value. For the marked abnormal operating parameters, the system further identifies the abnormal pattern through a clustering analysis algorithm. Specifically, the system uses a density-based clustering method (such as DBSCAN) to cluster the abnormal parameter data, separating isolated abnormal points and dense abnormal clusters. The clustering result is matched with the pattern features in the abnormal type library to determine the abnormal type. Combining the analysis of the abnormal pattern, the system associates the abnormal data with the equipment components that may cause the abnormality through a decision tree model to generate the final operating abnormality determination result.

[0067] Further, the determination of abnormal equipment representation in the substation by performing semantic segmentation on real-time image data and extracting features characterized by equipment type includes:

[0068] Perform segmentation processing on the real-time image data based on the DeepLab semantic segmentation model, divide the image into a background area and an equipment target area, and extract the contour and position features of the equipment target;

[0069] Extract the shape features, color features, and texture features of the extraction device;

[0070] Use a graph neural network to extract features from the segmented device target area. Represent the pixel points of the device as graph nodes, and generate a geometric shape feature vector of the device through the aggregation of node features and adjacency relationships;

[0071] Match the generated geometric shape feature vector with a preset feature library of device types, and determine whether there is an abnormality in the device representation based on the similarity measure in the feature embedding space.

[0072] In some embodiments, first, the system uses the DeepLab semantic segmentation model to process real-time image data, and segments the image into a background area and a device target area. The DeepLab model uses the Atrous Convolution technique to enhance the receptive field of feature extraction while ensuring high resolution. Through pre-training of the deep learning model, the system can identify common devices (such as transformers, switchgear) and background elements (such as the ground, walls) in a substation scene. When the model runs, it receives images collected by a real-time camera, extracts the contour and position features of the device target area, and generates a mask for subsequent feature processing. After segmentation, the system further extracts the shape features, color features, and texture features of the device target. The shape features are extracted by an algorithm based on shape descriptors, such as calculating parameters such as the area, perimeter, and aspect ratio of the device contour. The color feature extraction is based on a statistical method of color histograms to describe the color distribution on the device surface; the texture features are modeled by algorithms such as Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG) to obtain the texture information on the device surface. These features together constitute a multi-dimensional description of the device representation. Subsequently, the system uses a graph neural network (GNN) to perform advanced feature extraction on the segmented device target area. The pixel points in the device area are represented as graph nodes, and the connection relationship between the nodes is established according to the pixel position proximity. The GNN aggregates and updates the node features through the Message Passing Mechanism to generate a geometric shape feature vector of the device. The node features include the color value of the pixel, the texture feature vector, and local geometric information, and the aggregation process uses a weighted graph convolution operation. Through multiple layers of graph neural networks, the system can capture the global structure information of the device area. The generated geometric shape feature vector is used to match a preset feature library of device types. The feature library is constructed based on standard images of specific devices and contains geometric shape descriptions of different device types under ideal conditions. During the matching process, the system uses a similarity measure method in the embedding space based on cosine similarity to calculate the similarity between the real-time features and the preset features. If the similarity is lower than the set threshold, the system determines that there is an abnormality in the device representation.

[0073] Furthermore, in the graph neural network, the formula for feature update is as follows:

[0074] ;

[0075] where is the representation of each node at the k-th layer; the representation of each node at the (k + 1)-th layer; is the set of neighbor nodes of node ; is a neighbor node of node ; is the normalized weight between node and neighbor node ; represents the weight matrix at the k-th layer; represents the non-linear activation function.

[0076] Specifically, the core process of feature update is completed through the linear transformation of each node's own features and the weighted aggregation of its neighbor nodes' features. In each update, the feature representation of a node consists of two parts: one is the information contribution of the node itself, which reflects the role of the node's original features in the overall graph structure; the other is the information contribution of the neighbor nodes, which reflects the relationship strength and structural influence between this node and its surrounding nodes. This process is repeated in each layer of the graph neural network to gradually expand the receptive field of each node, evolving from the description of local relationships to the representation of the global structure. To ensure the effectiveness of feature aggregation and the stability of graph structure information, the connection relationship between each node and its neighbor nodes is assigned a normalized weight to adjust the intensity of information propagation between nodes. The normalized weight is calculated based on the number of connections of the node (i.e., the degree of the node) to balance the deviation of the relationships between nodes and prevent the excessive influence of high-degree nodes on feature update. In the initial stage of the graph neural network, the node features are jointly composed of the color values, texture features, and geometric position information of the pixel points in the device area. These features are updated layer by layer, continuously integrating local and global topological structure information, and finally generating the high-level feature representation of the nodes. In the last layer of the network, through global feature aggregation operations (such as average or max pooling), the feature vectors of all nodes in the graph are combined into the geometric shape feature representation of the overall device. The finally generated geometric shape feature vector is used for similarity matching with a preset device feature library. By measuring the similarity between the real-time extracted features and the standard features, the system can accurately determine whether there are abnormalities in the representation of the device. This feature extraction method based on the graph neural network greatly enhances the ability to describe the shape and structural features of the device, providing a reliable foundation for device anomaly detection and fault warning.

[0077] Further, determining the identity of a person based on real-time image data and matching the identity with the person's permissions includes:

[0078] Extracting feature points of the person's face and body contour through real-time image data combined with a key point detection algorithm, and generating a feature vector of the person based on a preset pose model;

[0079] Using multi-modal data fusion to associate and match the extracted feature vector of the person with the identity feature information obtained based on device interaction data, and generating a comprehensive identity determination;

[0080] Through the matched identity features, retrieve the behavior rules and device operation permissions corresponding to this identity in the permission management module, including the accessible areas, the categories of devices to be operated, and the operation time limits.

[0081] In some embodiments, first, facial and body contour feature points of a person are extracted by using real-time image data in combination with a key point detection algorithm. Specifically, the system separates human targets from images and identifies key points, including facial feature points (such as eyes, nose, and corners of the mouth) and body feature points (such as shoulders, elbows, and knees), by deploying a high-performance object detection model (such as OpenPose or Mediapipe). These key points are geometrically normalized to remove the scale and position differences caused by different camera angles or distances, generating a set of standardized feature points. Based on these key points, the system models the pose information of the person by combining a preset pose model, generating a set of high-dimensional feature vectors to represent the facial and body characteristics of the individual. On this basis, the system further associates and matches the person feature vectors extracted from the image with the identity feature information generated based on device interaction data through multimodal data fusion technology. The device interaction data includes identity data generated when the person passes through access control, device control terminals, or other identity verification devices in the substation. These data and the image feature vectors are mapped and compared through a unified embedding space. Specifically, the system constructs a feature embedding model using a deep neural network to map multimodal data to the same high-dimensional space, and performs feature matching through cosine similarity or Euclidean distance, thereby achieving comprehensive identity determination. During the matching process, the system assigns a confidence score to the matching result. If the confidence is lower than the set threshold, the system will trigger a secondary verification process to ensure the accuracy of identity determination. When the system successfully matches an identity, it enters the permission verification stage. The permission verification relies on the permission management module to dynamically load the behavior rules and operation permissions of the person by retrieving the permission rules associated with the identity in the database. The permission rules include the area range where the person is allowed to enter, the categories of devices that can be operated, and the permitted operation time periods. For example, for a maintenance engineer, the permission rules may include the maintenance permission for specific devices and a limited time; while for a visitor, the permission may be limited to a short stay in the public area. The system determines whether the current person's activities comply with their permission rules through logical matching. If a violation is detected, such as entering an unauthorized area or operating a restricted device, the system generates a warning signal and notifies the relevant management personnel.

[0082] Further, the determining of abnormal human behavior by tracking the human behavior path based on convolutional neural network according to real-time image data includes:

[0083] S1. Based on real-time image data, a convolutional neural network is used to detect a human target, generating a unique identity identifier and motion trajectory points of the target person;

[0084] S2. Based on the orientation of the rotatable camera and the coordinates of the rotatable camera, the target position is determined and the rotation angle, focal length, and field of view range of the camera are adjusted in real time, continuously obtaining real-time images of the target person and updating the motion trajectory points;

[0085] S3. Calculate the movement direction and speed of the target person based on the updated trajectory points, and when the person is about to leave the current camera's field of view, predict their expected position in the next rotatable camera's field of view based on the coordinates of the rotatable camera, and transmit the prediction parameters and the target identity identifier to the next camera.

[0086] S4. After the next camera receives the parameters transmitted in step S3, continue to track the target person in combination with the real-time image, generate an adjacent movement trajectory segment, and perform a fusion verification of the time stamp and spatial position of this segment with the trajectory segment recorded by the previous camera to generate a complete behavior path.

[0087] S5. Match and analyze the complete behavior path with the preset activity rules, and identify abnormal human behaviors based on path morphological features, regional coverage relationships, and stop point distributions. The abnormal human behaviors include unauthorized area entry, abnormal stay, and abnormal wandering.

[0088] Specifically, first, a convolutional neural network (CNN) is used to perform object detection on real-time images. The model adopts the YOLOv5 (You Only Look Once Version 5) architecture. Through its single-step detection ability, it quickly locates target personnel in each frame of the image and generates bounding boxes and class labels. The input to the model is the normalized and resized camera image, and the output is the spatial coordinates and confidence of the target personnel. For each detected target personnel, the system assigns a unique identification number (ID), and at the same time generates the corresponding motion trajectory points for this frame, including the coordinate position and timestamp information of the target. Then, the system dynamically adjusts its orientation and focal length through the built-in rotation control module of the camera to ensure that the target personnel are always at the center of the field of view. This process is achieved through the proportional-integral-derivative (PID) control algorithm. The PID control algorithm receives the offset between the real-time position of the target personnel and the center of the field of view, and dynamically adjusts the rotation angle and pitch angle of the camera based on the error feedback. At the same time, to improve the accuracy of target position calculation, the system combines the built-in geometric correction model of the camera to convert the bounding box position into the position in the global coordinate system, ensuring the consistency of the trajectory points between different cameras. When the target personnel are about to leave the field of view of the current camera, the system uses a motion prediction model to calculate their next position. The motion prediction is based on the Kalman Filter, and the motion trajectory of the target personnel is predicted by updating the position and velocity estimates in real time. The system combines the fixed layout and field of view coverage of the cameras in the substation scenario to calculate the predicted position of the target in the field of view of the next camera. The prediction parameters (including the target identification number, predicted position coordinates, and velocity vector) are transmitted to the next camera through the local area network (LAN) to initiate the real-time tracking of the next camera. After the next camera receives the prediction parameters, it continues to track the target personnel in combination with the real-time image data in its field of view and generates a new motion trajectory segment. To ensure the continuity of the multi-camera trajectory, the system uses timestamp and spatial verification algorithms to fuse the trajectory segments generated by different cameras. Specifically, the timestamp verification ensures the temporal continuity of the trajectory segments by comparing the time intervals of adjacent camera data transmissions; the spatial verification ensures the accuracy of position matching by comparing the positions of the start and end points of the trajectory segments in the global coordinate system. During the fusion process, the system interpolates and smooths the trajectory points to eliminate the jump errors that may be introduced by camera switching and generates a complete behavior path. After the complete behavior path is generated, the system matches and analyzes the path with the preset activity rules. The activity rules are defined by the safety management strategy of the substation, including the boundaries of the authorized area, the allowed stay time, and the reasonable motion pattern. The path analysis is carried out through the following three dimensions: Path morphological characteristics: By analyzing the trajectory curvature, angle change, and consistency of the motion direction, it is identified whether there are abnormal behaviors of the target personnel (such as frequent turning back or non-straight paths).Area coverage relationship: Determine whether a person enters an unauthorized area by the intersection of the path and the boundary of the authorized area. If the path covers the area around the sensitive device and does not conform to the permission rules, the system marks it as abnormal. Distribution of stay points: Detect whether a person stays in a specific area for a long time by clustering and analyzing the stay points. Alarms are triggered for unauthorized stays or behaviors exceeding the set time.

[0089] The present invention further includes a monitoring and warning method for an intelligent substation, which is applied to a monitoring and warning control system for an intelligent substation described in any one of the foregoing, and includes:

[0090] Distributively acquire real-time image data of different targets in the substation and obtain the operating parameters of the substation equipment;

[0091] Determine the abnormal operation of the equipment through time series analysis based on the operating parameters of the substation equipment, perform semantic segmentation on the real-time image data, extract features characterized by the equipment type, and determine the abnormal representation of the equipment in the substation;

[0092] Determine the identity of the person based on the real-time image data, match the person's permissions according to the identity, and track the person's behavior path based on the convolutional neural network according to the real-time image data to determine the abnormal behavior of the person;

[0093] Classify the alarms according to the abnormal operation of the equipment, the abnormal representation of the equipment, or the abnormal behavior of the person, and control the execution of the alarm actions.

[0094] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A monitoring and early warning control system for an intelligent substation, characterized in that, Including: A real-time monitoring module, which is used to distributively acquire real-time image data of different targets in the substation and is also used to acquire the operating parameters of substation equipment; An operating anomaly determination module, which is used to determine equipment operating anomalies through time series analysis based on the operating parameters of substation equipment, perform semantic segmentation on real-time image data, extract features characterized by equipment types, generate a geometric shape feature vector of the equipment, match the generated geometric shape feature vector with a preset feature library of equipment types, and determine whether there are anomalies in equipment characterization based on the similarity measurement in the feature embedding space; A behavior determination module, which is used to determine the identity of personnel based on real-time image data, match personnel permissions according to the identity, track the personnel behavior path based on the convolutional neural network according to the real-time image data, and determine personnel behavior anomalies; An early warning control module, which is used to classify alarms according to equipment operating anomalies, equipment characterization anomalies or personnel behavior anomalies and control the execution of alarm actions; The determination of the identity of personnel based on real-time image data and the matching of personnel permissions according to the identity include: Extracting feature points of the face and body contour of the personnel through real-time image data combined with a key point detection algorithm, and generating a feature vector of the personnel based on a preset pose model; Using multi-modal data fusion to associate and match the extracted personnel feature vector with the identity feature information obtained based on equipment interaction data to generate a comprehensive identity determination; Retrieving the behavior rules and equipment operation permissions corresponding to this identity in the permission management module through the matched identity features, including the areas that can be entered, the categories of equipment that can be operated, and the operation time limit; The tracking of the personnel behavior path based on the convolutional neural network according to the real-time image data and the determination of personnel behavior anomalies include: S1. Based on real-time image data, using a convolutional neural network to detect personnel targets, generating a unique identity identifier and movement trajectory points of the target personnel; S2. Based on the orientation of the rotatable camera and the coordinates of the rotatable camera, determining the target position and adjusting the rotation angle, focal length and field of view range of the camera in real time, continuously acquiring real-time images of the target personnel and updating the movement trajectory points; S3. Based on the updated trajectory points, calculating the movement direction and speed of the target personnel, and predicting its expected position in the next rotatable camera field of view based on the coordinates of the rotatable camera when the personnel is about to leave the current camera field of view, and transmitting the prediction parameters and target identity identifier to the next camera; S4. After the next camera receives the parameters transmitted in step S3, continuing to track the target personnel in combination with real-time images, generating an interconnected movement trajectory segment, and performing timestamp and spatial position fusion verification on this segment and the trajectory segment recorded by the previous camera to generate a complete behavior path; S5. Matching and analyzing the complete behavior path with preset activity rules, and identifying personnel behavior anomalies based on path shape features, regional coverage relationships and stop point distributions. The personnel behavior anomalies include unauthorized area entry, abnormal stay and abnormal wandering.

2. The monitoring and early warning control system for an intelligent substation according to claim 1, characterized in that The distributive acquisition of real-time image data of different targets in the substation includes the following steps: Setting a number of rotatable cameras at different positions in the substation area; Preset a plane rectangular coordinate system in the substation area in the cloud platform; Match the corresponding coordinates in the rectangular plane of the plane rectangular coordinate system according to the position of the rotatable camera; Determine the target position based on the orientation of the rotatable camera and the coordinates of the rotatable camera, and obtain real-time images of different targets.

3. The monitoring and early warning control system for an intelligent substation according to claim 1, characterized in that, The operating parameters of the substation equipment include: electrical parameters, environmental parameters, vibration parameters, insulation status parameters, and switch action parameters.

4. The monitoring and early warning control system for an intelligent substation according to claim 1, characterized in that, The determination of equipment operation anomalies through time series analysis based on the operating parameters of the substation equipment includes: Perform segmented normalization processing on the real-time collected operating parameters, and extract the change trend of the operating parameters; Based on the long short-term memory network model, perform time series modeling on the normalized operating parameters, and predict the parameter changes in a future period of time by learning the characteristics of historical operating data; Dynamically compare the predicted parameters with the actual parameters collected in real time, and use the set anomaly threshold to judge the deviation degree; Perform clustering analysis on the operating parameters that exceed the threshold range, match the corresponding anomaly patterns in combination with the anomaly type library, generate the operation anomaly determination result, and associate the equipment components that cause the anomaly.

5. The monitoring and early warning control system for an intelligent substation according to claim 1, wherein The semantic segmentation through the real-time image data, and the feature extraction characterized by the equipment type to generate the geometric morphological feature vector of the equipment; specifically includes: Based on the DeepLab semantic segmentation model, perform segmentation processing on the real-time image data, divide the image into the background area and the equipment target area, and extract the contour and position features of the equipment target; Extract the external shape features, color features, and texture features of the equipment; Use the graph neural network to extract features from the segmented equipment target area, represent the pixel points of the equipment as graph nodes, and generate the geometric morphological feature vector of the equipment through the aggregation of node features and adjacency relationships.

6. The monitoring and early warning control system for an intelligent substation according to claim 5, characterized in that, In the graph neural network, the formula for feature update is as follows: ； wherein, is the representation of each node at the k-th layer; is the representation of each node at the (k + 1)-th layer; is the set of neighbor nodes of node ; is the neighbor node of node ; is the normalized weight between node and neighbor node ; represents the weight matrix at the k-th layer; represents the non-linear activation function.

7. A monitoring and early warning method for an intelligent substation, which is applied to a monitoring and early warning control system for an intelligent substation according to any one of claims 1-6, characterized in that, Includes: Distributedly obtain the real-time image data of different targets in the substation and obtain the operating parameters of the substation equipment; Determine the equipment operation anomalies through time series analysis based on the operating parameters of the substation equipment, perform semantic segmentation through the real-time image data, and perform feature extraction characterized by the equipment type to determine the equipment characterization anomalies in the substation; Determine the identity of the personnel according to the real-time image data, match the personnel permissions according to the identity, and track the personnel behavior path based on the convolutional neural network according to the real-time image data to determine the personnel behavior anomalies; Classify the alarms according to the equipment operation anomalies, equipment characterization anomalies, or personnel behavior anomalies, and control the execution of the alarm actions.

Images