Intelligent identification and early warning method for electric power operation hidden danger

Abstract

The invention relates to the technical field of power system monitoring, and discloses an intelligent identification and early warning method for power operation hidden dangers, which comprises the following steps: constructing a virtual power distribution system based on three-dimensional space mapping, and establishing association between an equipment entity and a video vision field; executing multi-dimensional data freezing based on millisecond-level clock synchronization, and acquiring electrical parameters and video images which are aligned in time sequence; extracting electrical characteristics such as loop load current and the like and visual physical state characteristics of equipment parts; inferring a conduction state based on the electrical parameters, and performing dual verification with the visual features to identify a state difference; and driving the three-dimensional model to carry out dynamic early warning according to a verification result, and triggering closed-loop control such as physical locking and the like. The actual electrical conduction state is reversely deduced by analyzing the loop load current, the consistency of the actual electrical conduction state and the physical component position characteristics extracted through image recognition is compared, and accurate recognition and active defense of hidden faults such as contact adhesion are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system monitoring, in particular to an intelligent identification and early warning method for power operation hidden dangers. BACKGROUND

[0002] With the advancement of smart grid construction, the operation and maintenance management of distribution rooms and substations gradually changes to unattended and centralized monitoring mode. The current power distribution monitoring system mainly consists of two parts: power monitoring system and video security system. The power monitoring system is responsible for collecting electrical analog and digital quantities such as voltage, current, switch position, etc., and realizes remote measurement and remote signaling through the SCADA system; the video security system is mainly used for real-time monitoring or video storage of the on-site environment and equipment appearance. These two technical means play a fundamental role in ensuring the reliability of power supply and the safety of equipment.

[0003] The existing electrical monitoring and video monitoring usually operate as independent subsystems, lacking a unified time reference and data fusion mechanism. The collection cycle of electrical data is usually millisecond level, while the transmission and coding and decoding of video images have inherent network delays, making it difficult to accurately align the two in time sequence. When transient faults occur or switch opening and closing operations are performed on distribution equipment, due to the inconsistency of time stamps, the operation and maintenance system cannot associate and analyze the electrical waveform and the corresponding video image frame at a specific time. This data fragmentation makes it impossible to restore the real physical state at the moment of failure by mutual verification of multi-dimensional data when an anomaly occurs, limiting the accuracy of fault cause analysis.

[0004] In addition, the traditional monitoring method mainly relies on the electrical feedback signal of the secondary circuit to determine the operating state of the switch device, and there is a monitoring blind area. In actual operation, there may be faults such as contact sticking, transmission mechanism jamming or connecting rod fracture, which cause the handle indication position or auxiliary contact state of the switch to be inconsistent with the actual on-off state of the main circuit. If only relying on electrical signals or single video images, it is difficult to identify such hidden physical faults with inconsistent visual and electrical states. At the same time, the existing monitoring interface is mainly presented in the form of two-dimensional charts or lists, lacking depth mapping with the physical space, resulting in less intuitive fault location. Moreover, the monitoring data often fails to form a closed loop with the physical control system such as the access control system, and cannot automatically trigger regional locking when high-risk hidden dangers are detected, posing a certain risk of personnel entering the safety area. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an intelligent identification and early warning method for power operation hidden dangers, which solves the problem of being difficult to identify hidden physical faults such as contact sticking and mechanical jamming by simply relying on electrical signal feedback.

[0006] To achieve the above object, the present application is implemented by the following technical solutions: An intelligent identification and early warning method for power operation hidden dangers, comprising the following steps: Step S1, constructing a virtual power distribution system based on three-dimensional space mapping, establishing the visual field association relationship between the power equipment entity and the video acquisition equipment in the three-dimensional digital space, and establishing the digital index of the global equipment; Step S2, performing multi-dimensional data freezing based on millisecond-level clock synchronization, sending trigger instructions to heterogeneous devices using a unified clock reference, and realizing time alignment of electrical data acquisition and video frame capture; Step S3, analyzing the data uploaded by the power monitoring gateway, extracting the electrical characteristic data of the target equipment, the electrical characteristic data including the auxiliary contact state of the switch and the loop load current value; Step S4, processing the image frames captured by the video acquisition equipment, and extracting the visual physical state characteristics of the target equipment using image recognition algorithms; Step S5, performing double-checking logic judgment of visual and electrical state, inferring the electrical conduction state according to the electrical characteristic data, and comparing it with the visual physical state characteristics for consistency, generating a device operation state judgment result; Step S6, performing three-dimensional visualization early warning and closed-loop linkage control, updating the three-dimensional model display according to the device operation state judgment result, and triggering hierarchical early warning or physical lockout control when an abnormal state is detected.

[0007] In the process of constructing the virtual power distribution system, establishing the visual field association relationship between the power equipment entity and the video acquisition equipment involves geometric calculation and optical verification. The system traverses the power monitoring object set, calculates the visual line vector from the video acquisition equipment optical center to the center of the target equipment key part, and judges whether the included angle between the visual line vector and the camera optical axis is less than the half field angle of the camera, to determine the field of view coverage range. At the same time, the spatial straight line segment intersection test algorithm is used to construct a virtual detection line segment, to detect whether the line segment intersects with the building structure or equipment model grid in the three-dimensional scene, to determine whether there is physical obstruction. In addition, based on the pinhole imaging principle, the number of projected pixels of the target equipment key part on the imaging plane is calculated, and if the number of pixels is higher than the preset pixel matrix threshold, it is determined that the imaging resolution meets the standard. The system selects the video acquisition equipment that meets the field of view coverage, no physical obstruction and imaging resolution standard at the same time to establish the mapping relationship.

[0008] To address the time alignment issue of multidimensional data, data freezing based on millisecond-level clock synchronization is implemented. The power monitoring gateway performs time calibration with the master clock server via a precision clock synchronization protocol and sends broadcast freeze frames to the underlying sensing devices. The system marks the system time of sending the broadcast freeze frame as an electrical timestamp and reads the operating values ​​locked in the holding registers of each underlying sensing device at the moment the command is received. Simultaneously, a snapshot request is sent to the video management service, extracting the presentation timestamp of the video frame with the smallest deviation from the sampling time as the visual timestamp. The system calculates the difference between the electrical timestamp and the visual timestamp, and only when this difference is less than a preset synchronization tolerance threshold is the data considered valid, thus ensuring that subsequent verification logic is based on the physical state at the same moment.

[0009] During the feature extraction phase, the system parses the data packets uploaded by the power monitoring gateway, extracts the electrical feature data of the target device, and simultaneously extracts the micro-environmental parameters of the environmental auxiliary control sensors to construct an electrical feature vector containing auxiliary contact feedback signals, the maximum value of the three-phase load current, the effective value of the circuit voltage, and the water level value of the cable trench. For video images, the system performs contrast-limited adaptive histogram equalization and Gaussian filtering on the original image frames, and uses a convolutional neural network model to locate and crop the region of interest containing the status indicator component. For the region of interest, the system uses the oriented gradient histogram algorithm to extract edge direction features or statistically analyzes the pixel distribution in the HSV color space, outputting a visual physical state quantization value and a confidence probability value. When the confidence probability value is lower than a preset reliability threshold, the visual physical state quantization value is marked as an anomaly flag.

[0010] During the dual-verification logic judgment phase, the system first infers the physical state based on electrical parameters: it reads the circuit load current value from the electrical characteristic data. When this value is greater than the preset current judgment threshold, the main circuit is determined to be in an electrically conducting state, and the inferred electrical state is defined as conducting. When this value is less than or equal to the preset current judgment threshold, the main circuit is determined to be in an electrically disconnected state, and the inferred electrical state is defined as disconnected. Subsequently, the difference between the visual physical state quantization value corresponding to the visual physical state feature and the inferred electrical state is calculated. If the difference is non-zero, it is determined to be an abnormal situation where the visual state and electrical state are inconsistent, and a hidden danger type classification diagnosis is performed: if the visual physical state quantization value indicates closed and the inferred electrical state indicates open, it is determined whether it is a mechanical fault based on the circuit voltage value; if the visual physical state quantization value indicates open and the inferred electrical state indicates conducting, it is determined to be a level one hidden danger of contact adhesion.

[0011] To avoid false alarms caused by transient interference, the system performs continuous timing verification. A preset verification time window is set, within which data from multiple sampling moments are continuously collected. A fault is confirmed and an alarm signal is generated only when the status judgment results for N consecutive sampling moments are all abnormal and the hazard type classification results are consistent; otherwise, a single abnormality is judged as a transient event and filtered out.

[0012] In the 3D visualization early warning and closed-loop linkage control phase, the system retrieves the corresponding equipment geometric model nodes in the 3D digital space based on the equipment operating status judgment results. When the judgment result is an abnormal state, the model material color is modified to the preset alarm hue, and a self-illuminating texture is loaded. For equipment models containing internal structures, transparency blending rendering is performed to reduce the opacity of the outer shell material while keeping the key internal components opaque and highlighted. Environmental status labels are rendered around the 3D model. When environmental parameters exceed limits, particle effects simulate water immersion or smoke effects. When an abnormal state is detected and an early warning is triggered, the system uses the sampling time that triggered the alarm as the index key to retrieve related data from the time-series database. A comprehensive status information window is generated at the top layer of the 3D view, synchronously displaying the on-site video capture image frames, electrical load current values, and the comparison results between visual and electrical status at that moment. Combined with the event sequence data recorded by the edge computing gateway, an energy consumption load curve analysis graph is generated. In terms of control linkage, when a primary potential hazard of contact adhesion is detected, a locking command is sent to the access controller via industrial fieldbus or LoRa wireless network to lock the physical access control of the relevant area. At the same time, an operation locking signal is sent to the data acquisition and monitoring control system to suppress the remote opening and closing control commands of the faulty equipment through a logical interlocking mechanism. For mechanical faults, a maintenance work order is generated and pushed to the operation and maintenance terminal without triggering physical area locking.

[0013] This invention provides an intelligent method for identifying and issuing early warnings of potential hazards in power operation. It has the following beneficial effects: 1. This invention establishes a unified clock reference for the entire system by performing multi-dimensional data freezing based on millisecond-level clock synchronization. The system uses broadcast commands to force heterogeneous devices to lock electrical data and video image frames simultaneously, effectively eliminating timing misalignments caused by network transmission delays or differences in device processing speeds. This mechanism ensures that subsequent visual and electrical status comparisons are based on sampled data from the same physical moment, avoiding logical misjudgments caused by data asynchrony and improving the system's ability to capture and analyze transient changes in the distribution network.

[0014] 2. This invention achieves accurate diagnosis of hidden faults in power distribution equipment by constructing a dual verification logic of visual physical state and electrical operating parameters. The system analyzes the circuit load current to infer the actual electrical conduction state and compares it with the physical component position features extracted by image recognition. This method can effectively identify "visual-electrical discrepancy" faults that are difficult to detect by traditional single-signal monitoring methods, such as contact adhesion, mechanical linkage breakage, or status indicator jamming, filling the monitoring gap for inconsistencies between the mechanical and electrical characteristics of equipment.

[0015] 3. This invention enhances the intuitiveness of fault location and the safety of on-site handling by combining three-dimensional spatial mapping technology with a closed-loop linkage control mechanism. The system maps monitoring data to a three-dimensional digital model in real time. When a fault occurs, the model's transparency and high-brightness rendering technology visually display the location of internal hazards, solving the problem of blind spots caused by physical obstructions. Simultaneously, the access control and operation suppression functions, automatically triggered based on the hazard level, prevent maintenance personnel from accidentally entering faulty areas or performing incorrect operations in dangerous conditions, achieving a shift from passive monitoring to proactive safety defense. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the three-dimensional spatial mapping construction process of the virtualized power distribution system of the present invention; Figure 3 This is a timing diagram of the multi-dimensional data synchronous acquisition of the present invention; Figure 4 This is a flowchart of the electrical operating parameters and visual physical state feature extraction process of the present invention; Figure 5 This is a flowchart of the dual visual and electrical status verification logic of the present invention; Figure 6 This is a flowchart of the three-dimensional visualization early warning and closed-loop linkage control of the present invention. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See attached document Figure 1 , Figure 1This is a flowchart of a method for intelligent identification and early warning of potential power operation hazards according to an embodiment of the present invention. The present invention provides a method for intelligent identification and early warning of potential power operation hazards, comprising the following steps: Step S1: Construct a virtualized power distribution system based on three-dimensional spatial mapping, and establish the visual relationship between power equipment entities and video acquisition equipment in three-dimensional digital space; Step S2: Perform multi-dimensional data freezing based on millisecond-level clock synchronization, and use a unified clock reference to trigger electrical data acquisition and video frame capture; Step S3: Analyze the data uploaded by the power monitoring gateway and extract the electrical characteristic data of the target device. The electrical characteristic data includes the status of the switch auxiliary contacts and the circuit load current value. Step S4: Process the image frames captured by the video acquisition device and use image recognition algorithms to extract the visual physical state features of the target device; Step S5: Perform dual verification logic judgment of visual and electrical status, infer the electrical conduction status based on electrical feature data, and compare it with the visual physical status features to generate the equipment operation status judgment result. Step S6: Execute 3D visualization early warning and closed-loop linkage control, update the 3D model display based on the equipment operation status judgment result, and trigger early warning or control interlock when an abnormal state is detected.

[0019] To further clarify the specific implementation details and technical effects of the intelligent identification and early warning method for potential power operation hazards provided by this invention, the implementation methods of each step will be described in detail below in conjunction with the above steps.

[0020] See attached document Figure 2 , Figure 2 This is a schematic diagram illustrating the construction process of a three-dimensional spatial mapping for a virtualized power distribution system according to an embodiment of the present invention. In this embodiment, constructing a virtualized power distribution system based on three-dimensional spatial mapping is the foundation for subsequent visual and electrical data fusion analysis. This process digitizes the physical world's power facilities and their environmental information, and establishes a visual relationship between the physical entities of the power equipment and the video acquisition equipment in a three-dimensional digital space. This step may specifically include the following sub-steps: Step S101: Construct a 3D digital twin model of the power distribution scenario. The system accesses architectural drawings (CAD), geographic information system (GIS) data, and equipment ledger data for the substation or distribution room. Using a 3D modeling engine, the system reconstructs the building structure of the distribution room, including walls, floors, and cable trenches, based on the architectural drawings, and establishes digital geometric models of the power equipment based on the equipment ledger data. The power equipment includes, but is not limited to, high-voltage switchgear, transformers, low-voltage distribution cabinets, capacitor compensation cabinets, and connecting busbars. During the modeling process, the system performs hierarchical component modeling for each power device, clarifying the parent-child coordinate relationships between the main body of the equipment and key active components. Specifically, the system calibrates the bounding box range and geometric center coordinates of the circuit breaker operating handle, the opening / closing status indicator, the energy storage indicator window, and the instrument display panel in the local coordinate system of the equipment model. These local coordinates are transformed and stored as unified world coordinate system data through a model transformation matrix, ensuring that the system can accurately determine the absolute position of the component to be detected within the entire distribution room space. The specific rendering techniques and texture mapping processes for 3D modeling are well-known techniques in the field of computer graphics and will not be elaborated here.

[0021] Step S102: Establish a digital index and parameterized definition for all devices. The system defines a set of power monitoring objects and assigns a unique topology identifier to each physical power circuit that needs to be monitored. This topology identifier is associated with the circuit's communication address, rated voltage, rated current, and three-dimensional spatial coordinates of its physical installation location within the electrical monitoring system. Simultaneously, the system defines a set of video acquisition nodes, incorporating all fixed cameras, PTZ cameras, and inspection robot cameras deployed on-site into unified management. For each video acquisition node, the system configures inherent optical parameters and spatial pose parameters in its attribute file. The inherent optical parameters include the camera's focal length, sensor size (length and width of the sensor target surface), and image acquisition resolution; the spatial pose parameters include the three-dimensional coordinates of the camera's optical center in the world coordinate system, and the pitch, yaw, and roll angles of the optical axis direction vector in the world coordinate system.

[0022] Step S103: Perform the field-of-view spatial mapping calculation between heterogeneous devices. The system traverses each target device in the power monitoring object set and searches for cameras that can effectively cover the target device in the video acquisition node set, thereby establishing a "device-camera" mapping relationship table. This mapping calculation process is automatically generated based on the calculation logic of three-dimensional geometric space, without the need for manual calibration. The specific calculation logic is as follows: In terms of field-of-view coverage calculation, the system obtains the three-dimensional coordinates of the center point of a key part of the target device (such as an operating handle) and constructs a line-of-view vector pointing from the camera's optical center to this center point. The system projects this line-of-view vector onto the horizontal and vertical planes of the camera coordinate system and calculates the horizontal and vertical angles between the projected vector and the camera's optical axis, respectively. The target device is determined to be within the camera's geometric field of view only when the horizontal angle is less than the camera's horizontal half-field-of-view angle and the vertical angle is less than the camera's vertical half-field-of-view angle.

[0023] For line-of-sight occlusion detection, a spatial line segment intersection test algorithm is used to construct a virtual detection line segment with the camera's optical center as the starting point and the target device's center point as the ending point. The system detects whether this line segment intersects with the geometric mesh surface of pillars, walls, open cabinet doors, or other device models in the scene. If an intersection point is detected and the distance from the intersection point to the camera is less than the distance from the target device to the camera, then physical occlusion is determined to exist, and the camera cannot serve as a valid observation source for the target device.

[0024] In terms of validating imaging resolution, the system performs pixel occupancy estimation based on the pinhole imaging principle. The system calculates the projected size of a key component (e.g., a handle) on the camera's imaging plane based on its physical dimensions, camera focal length, and the straight-line distance from the target device to the camera. Then, combining the sensor size and image acquisition resolution, the projected size is converted into the number of pixels. The system compares the calculated number of pixels with the minimum pixel matrix threshold (e.g., 50×50 pixels) required by the preset image recognition algorithm. If the number of pixels is lower than the threshold, the camera is deemed to have insufficient clarity despite covering the field of view, failing to meet feature extraction requirements, and is not associated with the device. Only cameras that simultaneously meet the three conditions of being within the field of view, having no physical obstructions, and meeting the imaging resolution standard are written into the mapping table as associated video sources for the power equipment. In cases where multiple valid video sources exist, the system selects the primary video source based on either the optimal viewing angle principle (i.e., the smallest angle between the line-of-sight vector and the device panel normal vector) or the maximum pixel density principle. Through the above steps, the system completes the logical binding between physical devices and their observation perspectives in the three-dimensional digital space, providing data support with spatial accessibility and visual resolvability for subsequent retrieval of video streams from specific devices for status analysis.

[0025] See attached document Figure 3 , Figure 3This is a timing diagram for multi-dimensional data synchronization acquisition according to an embodiment of the present invention. In this embodiment, multi-dimensional data freezing based on millisecond-level clock synchronization is performed to eliminate the deviation between power operation data and visual image data at the acquisition time point, ensuring strict alignment of heterogeneous data sources in the time dimension, thereby providing a reliable spatiotemporal reference for subsequent dual verification. This step may specifically include the following sub-steps: Step S201: Establish a unified high-precision time synchronization and clock calibration mechanism for the entire system. The system deploys a power monitoring gateway with edge computing capabilities as the field data aggregation center. This gateway integrates a high-stability crystal oscillator and a time synchronization module. Regarding clock source acquisition, the gateway prioritizes receiving standard time synchronization signals from the BeiDou or GPS satellite navigation systems as the primary time reference. In physical environments where satellite signals are blocked, the gateway runs the Precision Clock Synchronization Protocol (IEEE 1588PTP) through an industrial Ethernet interface to request time synchronization messages from the clock node to the upper-level master clock server. The gateway dynamically compensates for the local system clock by calculating the round-trip path delay and crystal oscillator drift rate, ensuring that the absolute synchronization error between the local time and Coordinated Universal Time (UTC) remains within the millisecond range. The specific PTP protocol message interaction and crystal oscillator calibration algorithms are well-known technologies in the field of communications and will not be elaborated upon here.

[0026] Step S202: Perform instantaneous freezing and acquisition of electrical data based on broadcast commands. The system sets a global discrete sampling time sequence. At each preset sampling time tk, the power monitoring gateway does not use the traditional polling method, but instead sends a broadcast freeze frame to all connected smart meters, circuit breaker protection devices, and switch quantity acquisition modules via RS-485 bus, LoRa wireless self-organizing network, or fieldbus. Upon receiving the broadcast command, each underlying sensing device immediately locks the operating values ​​in its current register, including three-phase current, voltage, active power, and switch position status, and transfers these instantaneous values ​​to a specific holding register. The power monitoring gateway marks the system time of sending the broadcast freeze frame as the unified electrical timestamp for this batch of electrical data. Subsequently, the gateway reads the data from the holding registers of each device in sequence. Since the data has been frozen, subsequent read delays no longer affect the time-domain properties of the data, thus ensuring the consistency of the sampling time of all electrical nodes at the physical level.

[0027] Step S203: Perform precise extraction of video image frames based on time index. Since video surveillance systems typically record continuously at a fixed frame rate (e.g., 25fps or 30fps), the system employs a frame matching mechanism based on timestamp index to acquire synchronized images. At the aforementioned sampling time... The system sends a snapshot request with a specific timestamp to the video management service. The video management service then searches the video stream buffer queue for a snapshot that matches the sampling time. The video frame with the smallest time deviation between the given frames is selected. The system extracts the Presentation Timestamp (PTS) from the header information of this video frame and marks it as the visual timestamp. .

[0028] Step S204: Perform time synchronization validity verification on heterogeneous data. To prevent data time alignment deviations caused by network congestion, video encoding / decoding delays, or device response timeouts, the system verifies the validity of data pairs using a synchronization verification inequality before subsequent analysis. The synchronization verification inequality is defined as follows: ; in, For timestamps of electrical data, The timestamp of the video image frame. This is a preset synchronization tolerance threshold. The settings are based on the power grid frequency cycle and the video frame interval, for example, set to 100 milliseconds. Only when the above inequality is true will the system determine that the set of electrical data and the visual image belong to the same physical moment as valid samples, allowing them to enter the subsequent feature extraction process; otherwise, the system will mark the set of data as invalid data to prevent correlation analysis of state data from different times.

[0029] See attached document Figure 4 , Figure 4 This is a flowchart illustrating the extraction of electrical operating parameters and visual physical state features according to an embodiment of the present invention. In this embodiment, the extraction of electrical operating parameters and visual physical state features is a processing step connecting data acquisition and hazard identification. This step transforms the original binary communication messages and unstructured video images into mathematical feature vectors that can be understood by a computer. This step may specifically include the following sub-steps: Step S301: Parse the data packets uploaded by the power monitoring gateway and construct the electrical feature vector. The system receives data frames containing timestamps from the power monitoring gateway and performs unpacking and engineering quantity conversion according to the frame format definition of the Modbus-RTU, DL / T645, or IEC60870-5-104 communication protocol. The system locates the data fields in the data frames and extracts the target device's data at the sampling time. Key operating parameters include the auxiliary contact feedback signal, the effective value of the three-phase load current (A / B / C), and the loop voltage. For the auxiliary contact feedback signal, the system reads the state of a specific bit in the digital input register and converts it into a binary variable. The value 1 represents the closed position, and the value 0 represents the open position. For the load current, the system selects the maximum value among the three-phase currents as a scalar to characterize the circuit load level. For loop voltage, the system extracts the effective value of the line voltage. Based on the extracted scalars, the system constructs the device's time... electrical feature vector The vector is defined as: ; in, This represents the vector transpose. This eigenvector provides standardized input data for subsequent electrical conduction state inference.

[0030] Step S302: Perform image enhancement preprocessing on the synchronously acquired source image. System retrieval and sampling time. The video image frame with the smallest time deviation is used as the source image. To address potential low illumination or uneven lighting conditions in the power distribution room environment, the system first converts the source image to grayscale, then applies the Limiting Contrast Adaptive Histogram Equalization (CLAHE) algorithm. This algorithm divides the image into non-overlapping sub-regions, calculates the pixel brightness histogram within each sub-region, and remaps it, thereby limiting excessive noise amplification and enhancing local image contrast. Next, the system uses a Gaussian filtering algorithm to smooth the image, removing thermal noise generated by the image sensor to highlight the edge contours and texture features of the equipment surface, providing high-quality image data for subsequent target detection.

[0031] Step S303: Perform region of interest (ROI) localization and cropping based on a deep neural network. The system loads a pre-trained convolutional neural network (CNN) object detection model, which adopts a YOLO (YouOnly-Look-Once) or Faster-R-CNN architecture. The pre-processed full-frame image is input into the model, which has been trained on a large number of samples containing different types of circuit breaker panels and has the ability to extract deep features of the image. The model searches for regions in the full-frame image that match the appearance features of the target device and outputs the bounding box coordinates containing key status indicators. The key status indicators include the circuit breaker's operating handle, knob, or mechanical status indicator window. Based on the output bounding box coordinates, the system crops a sub-image from the original image containing only the status indicators, defining it as the ROI. If the model fails to detect a valid target or the output detection confidence is lower than a preset detection threshold, the system marks the frame image as invalid and terminates subsequent visual analysis.

[0032] Step S304: Extract the visual physical state features of the region of interest and quantize the output. For the cropped region of interest, the system performs specific state classification. For handle-type components, the system uses the Histogram of Oriented Gradients (HOG) algorithm to extract the edge direction features of the handle, or uses Principal Component Analysis (PCA) to calculate the principal axis direction angle of the handle region's binarized mask. The system determines the physical orientation based on the angle between the handle's principal axis and the vertical line. When the handle features are detected as vertically upward or biased towards the preset closing indicator side, it is determined to be in a closed state; when the handle features are detected as vertically downward or biased towards the preset opening indicator side, it is determined to be in an open state. For indicator window-type components, the system converts the image to the HSV color space and calculates the pixel distribution histogram of the H (hue) channel. If the proportion of red pixels exceeds a preset color ratio threshold, it is determined to be in a closed state; if the proportion of green pixels exceeds a preset color ratio threshold, it is determined to be in an open state.

[0033] Step S305: Generate visual physical state quantization values. The system defines visual physical state quantities. The above identification results are encoded. When the determination result is a closed state, a value is assigned. When the determination result is a tripped state, assign a value. Simultaneously, the system outputs the confidence probability value of the classification algorithm. To prevent misidentification due to obstruction by foreign objects, deliberate alteration, or an open cabinet door, the system incorporates a confidence threshold mechanism. When When the reliability falls below a preset threshold, the system will forcibly... The value is assigned to -1. This value of -1 serves as an anomaly flag, indicating that the visual data at this moment is unreliable and will not be included in the hazard determination by the subsequent logic verification module, thus ensuring the rigor of the dual verification mechanism.

[0034] See attached document Figure 5 , Figure 5 This is a flowchart illustrating the dual verification logic of visual and electrical status according to an embodiment of the present invention. In this embodiment, the dual verification logic judgment of visual and electrical status is a processing step of the system. It achieves the identification of hidden faults that are difficult to detect by a single monitoring method through the mutual verification of heterogeneous data of electrical features and visual physical features. This step may specifically include the following sub-steps: Step S401: Infer the actual conduction state of the electrical circuit based on the electrical load current. The system reads the electrical feature vector constructed in step S301. Load current value in Since the auxiliary contact signal only represents the position of the auxiliary mechanism inside the switch and is not directly equivalent to the on / off state of the main circuit, the system introduces inference logic based on current amplitude. The system sets a preset current judgment threshold close to zero. .when When the system determines that the main circuit is in an electrically conductive state, the inferred electrical state is defined. ;when When the system determines that the main circuit is in an electrically disconnected state, the inferred electrical state is defined. .

[0035] Step S402: Construct a state difference verification function and perform a consistency comparison. The system retrieves data from the same sampling time. Visual physical state quantities Inferring electrical state Before performing the comparison, the system checks... The validity indicator. If A value of -1 indicates that the visual recognition result is unreliable, and the system terminates the current verification logic and records the log. If... Valid, the system calculates the state difference value. The calculation formula is as follows: ; when When this occurs, it indicates that the visually observed physical handle position matches the actual on / off state of the electrical circuit. The system determines that the equipment is operating normally and outputs this status result to the status database. Step S403: Perform hazard type classification diagnosis based on differential characteristics. When When this occurs, it indicates that the system has encountered an abnormal situation where the visual state and electrical state are inconsistent. The system further distinguishes the nature of the fault based on the characteristics of the state combination.

[0036] like and The system identifies this as an abnormal state of physical closing but electrical disconnection. This state indicates that although the circuit breaker handle is in the closed position, the main circuit is not connected. The system queries the synchronous circuit voltage value. Perform auxiliary judgment. If If normal, the fault is likely caused by the circuit breaker's internal operating mechanism tripping but the handle not resetting, poor contact of the internal main contacts, or a fault in the auxiliary signal line. The system generates a mechanical fault warning code. and The system determines this to be an abnormal state of physical tripping but electrical continuity. This state indicates that the operator has performed the tripping operation or the handle is in the tripping indication position, but there is still load current in the main circuit. This situation is usually caused by welding between the moving and stationary contacts of the circuit breaker, preventing the contacts from separating. After recognizing this mode, the system generates a Level 1 hazard alarm code and triggers the safety interlocking logic.

[0037] Step S404: Perform continuous timing verification to eliminate transient interference. To avoid false alarms caused by transient processes during switching operations, the system introduces a timing smoothing filtering mechanism. The system sets a preset verification time window. Verification is only performed when continuous... State difference value at each sampling time When all values ​​remain at 1 and the hazard type classification results are consistent, the system confirms the fault and issues an alarm signal. For single anomalies that do not meet the continuity condition, the system records them as transient events.

[0038] See attached document Figure 6 , Figure 6 This is a flowchart of a three-dimensional visualization early warning and closed-loop linkage control system according to an embodiment of the present invention. In this embodiment, the three-dimensional visualization early warning and closed-loop linkage control step is the process of converting the front-end recognition results into intuitive human-computer interaction information and active safety control commands. This step may specifically include the following sub-steps: Step S501: Drive the 3D digital twin scene to perform state mapping and dynamic rendering. The system receives the updated device operating status determination results from the status database, and retrieves the corresponding device geometric model node in the 3D scene graph based on the device's unique topology identifier. The rendering engine modifies the rendering attribute parameters of the model material according to the determination results. When the device is in normal operating condition, the system sets the main material color of the device model to a preset first state color (e.g., green); when the device is determined to be in an abnormal state where the visual state and electrical state are inconsistent, the system sets the main material color of the device model to a preset second state color (e.g., red) and loads a periodically changing self-illuminating texture to produce a flashing prompt effect. For switch cabinet models that include internal structures, the system performs transparency blending rendering, reducing the opacity of the cabinet shell material while keeping key internal components (such as circuit breaker contacts) completely opaque and highlighted. This allows for a perspective view of potential internal hazards while maintaining the overall outline of the equipment. Additionally, for nodes that access environmental data, the system renders environmental status labels around the 3D model. If water levels exceed limits or temperatures are too high, particle effects are used to simulate water immersion or smoke effects, providing a more intuitive display of multi-dimensional potential hazards.

[0039] Step S502: Generate a comprehensive status information window and display the associated evidence data. The system monitors interactive events in the 3D scene in real time. When a selection command is received for a highlighted alarm device model, the information display logic is triggered. The system uses the sampling time when the alarm is triggered. Using the index key, retrieve related multidimensional data from the time-series database. The system displays a floating window overlaid on top of the view, simultaneously presenting the following data: time. The system captures on-site video frames (including bounding boxes generated by the target recognition algorithm and handle position markers), the electrical load current value at that moment, and a comparison of the system's inferred electrical continuity state with the visual physical state. By displaying the original video images and electrical parameters side-by-side in the same time slice, it provides maintenance personnel with reference data for fault diagnosis.

[0040] Step S503 executes closed-loop safety linkage control based on hazard level. The system has a built-in safety strategy control module that performs hierarchical control according to the hazard type output in step S403. When a level 1 hazard of "physically open but electrically connected" (i.e., contact adhesion) is detected, the system executes forced interlocking logic. The system sends a write register command or interlocking level signal to the access controller deployed in the power distribution room via industrial fieldbus or LoRa wireless network, controlling the access control electromagnetic lock to be locked, restricting personnel from entering the energized area. At the same time, the system sends an operation interlocking signal to the upper-level SCADA (Data Acquisition and Monitoring Control) system. This signal, through a logic AND gate circuit or software logic interlocking mechanism, forcibly suppresses the remote opening and closing control commands of the faulty circuit breaker, preventing safety accidents caused by secondary misoperation. For mechanical faults of "physically closed but electrically disconnected", the system generates a maintenance work order and pushes it to the handheld terminal of the maintenance team, without triggering physical area interlocking. This process realizes deep linkage between remote control and remote viewing in the "five remote" functions (remote measurement, remote signaling, remote control, remote adjustment, and remote viewing).

[0041] Step S504: Perform anomaly marking and cleaning of energy consumption data. For periods identified as potential hazards, the system post-processes historical energy consumption data. If the judgment result shows that the equipment is in an abnormal period where the visual state and electrical state are inconsistent, the system marks the electricity data generated during that period as a fault loss status in the energy consumption database. For electricity generated by unexpected power supply due to contact adhesion, the system removes it from the normal load electricity consumption statistics and classifies it separately as line loss statistics, thereby correcting the calculation benchmark of the distribution network line loss rate and ensuring the accuracy of energy efficiency analysis data.

Claims

1. A method for intelligent identification and early warning of potential hazards in power operation, characterized in that, Includes the following steps: Step S1: Construct a virtualized power distribution system based on three-dimensional spatial mapping, establish the visual domain association between power equipment entities and video acquisition equipment in three-dimensional digital space, and establish a digital index of all devices. Step S2: Perform multi-dimensional data freezing based on millisecond-level clock synchronization, and send trigger commands to heterogeneous devices using a unified clock reference to achieve time alignment between electrical data acquisition and video frame capture. Step S3: Analyze the data uploaded by the power monitoring gateway and extract the electrical characteristic data of the target device. The electrical characteristic data includes the status of the switch auxiliary contacts and the circuit load current value. Step S4: Process the image frames captured by the video acquisition device and use image recognition algorithms to extract the visual physical state features of the target device; Step S5: Perform dual verification logic judgment of visual and electrical status, infer the electrical conduction status based on the electrical feature data, and compare it with the visual physical status features to generate the equipment operation status judgment result; Step S6: Execute 3D visualization early warning and closed-loop linkage control. Update the 3D model display based on the equipment operation status determination result, and trigger graded early warning or physical interlock control when an abnormal state is detected.

2. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, The step S1, which establishes the field-of-view association between the power equipment entity and the video acquisition device, specifically includes: Traverse the set of power monitoring objects and calculate the line-of-sight vector from the optical center of the video acquisition device to the center of the key part of the target device; Determine whether the angle between the line-of-sight vector and the camera's optical axis is less than the camera's half field of view to determine the field of view coverage. Using a spatial line segment intersection test algorithm, a virtual detection line segment is constructed to detect whether the virtual detection line segment intersects with the mesh of building structure or equipment model in the 3D scene, so as to determine whether there is physical occlusion. The number of projected pixels of key parts of the target device on the imaging plane is calculated based on the pinhole imaging principle. If the number of pixels is higher than the preset pixel matrix threshold, the imaging resolution is determined to meet the standard. A mapping relationship is established by selecting video acquisition devices that simultaneously meet the requirements of the field of view coverage, have no physical obstructions, and have qualified imaging resolution.

3. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, Step S2 specifically includes: The power monitoring gateway performs time calibration with the master clock server through a precision clock synchronization protocol and sends broadcast freeze frames to the underlying sensing devices. The system time that sends the broadcast freeze frame is marked as an electrical timestamp, and the operating values ​​locked in the holding registers of each underlying sensing device at the moment the instruction is received are read. Send a snapshot request to the video management service and extract the presentation timestamp of the video frame with the smallest deviation from the sampling time as the visual timestamp; The difference between the electrical timestamp and the visual timestamp is calculated, and the data is deemed valid only if the difference is less than a preset synchronization tolerance threshold.

4. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, The step S3 involves extracting electrical characteristic data of the target device, and also includes simultaneously extracting micro-environmental parameters from environmental auxiliary control sensors. The system constructs an electrical feature vector, which includes the auxiliary contact feedback signal, the maximum value of the three-phase load current, the effective value of the circuit voltage, and the water level in the cable trench. Step S4 involves extracting the visual physical state features of the target device, specifically including: Image frames are subjected to contrast-limited adaptive histogram equalization and Gaussian filtering. The region of interest containing the state indicator component is located and cropped using a convolutional neural network model; For the region of interest, the edge direction features are extracted using the Histogram of Oriented Gradients algorithm or the pixel distribution in the HSV color space is statistically analyzed, and the visual physical state quantization value and confidence probability value are output. When the confidence probability value is lower than the preset reliability threshold, the visual physical state quantization value is marked as an anomaly flag.

5. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, Step S5, which involves inferring the electrical conduction state based on electrical characteristic data, specifically includes: Read the loop load current value from the electrical characteristic data; A preset current judgment threshold is set. When the load current value of the circuit is greater than the preset current judgment threshold, the main circuit is determined to be in an electrically conductive state, and the inferred electrical state is defined as conductive. When the load current value of the circuit is less than or equal to the preset current determination threshold, the main circuit is determined to be in an electrically disconnected state, and the inferred electrical state is defined as disconnected.

6. The intelligent identification and early warning method for potential power operation hazards according to claim 5, characterized in that, Step S5 involves performing a dual verification logic judgment of visual and electrical states, specifically including: Calculate the difference between the visual physical state quantization value corresponding to the visual physical state feature and the inferred electrical state; If the difference is non-zero, it is determined to be an abnormal situation where the visual state and electrical state are inconsistent, and a hazard type classification diagnosis is performed: If the visual physical state quantification value indicates that the circuit is closed and the inferred electrical state indicator is open, the circuit voltage value is used to determine whether it is a mechanical fault. If the visual physical state quantification value indicates that the circuit is open and the inferred electrical state indicates that the circuit is closed, it is determined to be a level one hidden danger of contact adhesion.

7. The intelligent identification and early warning method for potential power operation hazards according to claim 6, characterized in that, Step S5 further includes a continuous timing verification step: Set a preset verification time window, and continuously collect data at multiple sampling times within the verification time window; The fault is confirmed and an alarm signal is generated only when the status judgment results of N consecutive sampling times are all abnormal and the hazard type classification results are consistent. Otherwise, single anomalies will be classified as transient events and filtered out.

8. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, The step S6, updating the 3D model display, specifically includes: Based on the equipment operating status determination result, the corresponding equipment geometric model node is retrieved in the three-dimensional digital space; When the judgment result is an abnormal state, the model material color is changed to the preset alarm color tone, and a self-illuminating texture is loaded; For device models containing internal structures, perform transparency blending rendering, reducing the opacity of the outer shell material while keeping key internal components opaque and highlighted; Render environmental status labels around the 3D model. When environmental parameters exceed limits, simulate water immersion or smoke effects using particle effects.

9. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, The specific methods for triggering an early warning when an abnormal state is detected in step S6 include: Using the sampling time that triggered the alarm as the index key, retrieve related data from the time series database; A comprehensive status information window is generated at the top of the 3D view, which simultaneously displays the on-site video capture image frames, electrical load current values, and the comparison results between the visual and electrical status at that moment. By combining the event sequence data recorded by the edge computing gateway, an energy consumption load curve analysis map is generated.

10. The intelligent identification and early warning method for potential power operation hazards according to claim 1, characterized in that, The specific steps of triggering the physical interlocking control in step S6 include: When a primary hazard of contact adhesion is detected, a locking command is sent to the access controller via industrial fieldbus or LoRa wireless network to lock the physical access control of the relevant area; At the same time, it sends an operation interlock signal to the data acquisition and monitoring control system to suppress the remote opening and closing control commands of faulty equipment through a logic interlock mechanism; For the aforementioned mechanical faults, a maintenance work order is generated and pushed to the operation and maintenance terminal without triggering physical area locking.

Images