Substation equipment top operation protection method and system based on heterogeneous data detection

Abstract

This invention discloses a method and system for protecting substation equipment during top-level operations based on heterogeneous data detection, belonging to the field of heterogeneous data processing technology. The method includes the following steps: constructing a comprehensive protection network based on safety influencing factors of substation equipment and performing status detection on the comprehensive protection network; acquiring multi-source heterogeneous data from the comprehensive protection network, including infrared grating obstruction signals, mechanical data, and image data; performing safety detection on the multi-source heterogeneous data respectively; and providing protection feedback based on a priority-based alarm hierarchical response strategy using a fuse mechanism according to the safety detection results. This invention achieves early warning and proactive protection for the entire process of high-altitude operations on transformers, effectively suppressing safety accidents caused by regulatory negligence.

Description

Technical Field

[0001] This invention relates to the field of heterogeneous data processing technology, and in particular to a method and system for protecting substation equipment tops from heterogeneous data detection. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The safe and stable operation of core equipment in substations, such as large transformers, is directly related to the reliability of the power grid. Transformer inspection and maintenance work usually needs to be carried out at the top of the equipment, which is a typical high-risk operation scenario. Traditional safety protection measures mainly rely on manual monitoring and physical isolation methods such as ordinary fences, but these have many technical defects and safety hazards.

[0004] First, existing physical fences lack proactive early warning capabilities, failing to monitor and alert workers who accidentally enter the area in real time, resulting in significant blind spots. When workers accidentally enter dangerous areas, the situation often relies on manual detection by on-site supervisors, leading to delayed responses and difficulty in promptly preventing dangerous behavior.

[0005] Secondly, the current supervision of violations and inspection of dress codes mainly rely on manual on-site patrols, which is significantly lagging behind. Safety management personnel need to constantly patrol the work site, which is not only labor-intensive but also makes it difficult to achieve 24 / 7, all-round monitoring. Violations such as not wearing safety helmets or anti-static clothing are often only discovered after the fact, making real-time warnings and timely corrections impossible.

[0006] More importantly, the existing monitoring methods operate independently, creating information silos. Video surveillance systems, personnel positioning systems, and environmental monitoring systems operate separately, and data cannot be effectively integrated, making it difficult to provide a comprehensive sense of security situation. This fragmented management approach prevents safety managers from fully grasping the safety status of the work site, affecting the scientific nature and effectiveness of safety decisions.

[0007] In recent years, with the development of technologies such as artificial intelligence, the Internet of Things, and big data, intelligent safety monitoring technology has been widely applied in the industrial sector. In scenarios such as construction sites, AI-based visual analysis-based safety belt wearing recognition systems have been put into use, enabling real-time monitoring and automatic early warning of personnel behavior. Meanwhile, the application of multi-source heterogeneous data fusion technology in hazard source management has also achieved significant results. By integrating multi-dimensional information such as video surveillance, sensor data, and personnel positioning, it has achieved comprehensive perception and intelligent analysis of complex working environments.

[0008] However, current models still suffer from insufficient adaptability to the unique working environment atop substation transformers. Firstly, the limited space and complex equipment layout at the transformer top limit the accuracy and stability of acquiring heterogeneous data, potentially affecting the accuracy of existing machine learning model outputs. Secondly, the inconsistent dimensions among multi-source heterogeneous data result in a lack of reasonable protection and early warning logic in existing systems. Finally, the large computational load caused by processing multi-source data makes it difficult for existing models to balance computational cost and accuracy. Summary of the Invention

[0009] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for protecting substation equipment top operations based on heterogeneous data detection. By combining the mechanical and spatial sensing of the physical layer with the AI ​​vision algorithm of the application layer, it achieves early warning and proactive protection for the entire process of high-altitude operations on transformers, effectively preventing safety accidents caused by regulatory negligence.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solution: The first aspect of this invention provides a method for top-of-station equipment protection based on heterogeneous data detection, comprising the following steps: A comprehensive protection network is constructed based on the safety influencing factors of substation equipment, and the status of the comprehensive protection network is monitored. Acquire multi-source heterogeneous data from the entire protected network, including infrared grating blocking signals, mechanical data, and image data; Security checks are performed on multi-source heterogeneous data. Specifically, the area and duration of infrared grating obstruction signals are detected to determine if there is any object intrusion. Changes in mechanical data and seat belt attachment status are used to determine if there is any impact. Multi-target visual detection of image data is used to determine if there is any violation. Based on the safety test results, a priority-based circuit breaker mechanism is used to implement a graded alarm response strategy for protection feedback.

[0011] A second aspect of the present invention provides a substation equipment top-work protection system based on heterogeneous data detection, comprising: The initialization module is configured to build a comprehensive protection network based on the safety impact factors of substation equipment and to perform status detection on the comprehensive protection network. The data acquisition module is configured to acquire multi-source heterogeneous data in the protected network, including infrared grating blocking signals, mechanical data and image data. The multi-source data processing module is configured to perform security detection on multi-source heterogeneous data. Specifically, it detects the area and duration of infrared grating obstruction signals to determine if there is any object intrusion; it comprehensively judges whether there is any impact by analyzing changes in mechanical data and the seat belt attachment status; and it judges whether there is any violation by performing multi-target visual detection on image data. The alarm module is configured to provide protective feedback based on a priority-based circuit breaker mechanism and alarm tiered response strategy according to the security detection results.

[0012] A third aspect of the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and to execute the steps of the substation equipment top-operation protection method based on heterogeneous data detection as described in the first aspect of the present invention.

[0013] A fourth aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the substation equipment top-operation protection method based on heterogeneous data detection as described in the first aspect of the present invention.

[0014] A fifth aspect of the present invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the substation equipment top-operation protection method based on heterogeneous data detection as described in the first aspect of the present invention.

[0015] The above one or more technical solutions have the following beneficial effects: This invention proposes a method and system for protecting substation equipment top-level operations based on heterogeneous data detection. It innovatively designs an adaptive environmental stability judgment mechanism. By constructing a global sensor state matrix and calculating physical steady-state indicators in real time during the initialization phase, it automatically switches from debugging to deployment mode only when multimodal data converges, thus achieving "plug and play" functionality and eliminating traditional manual zeroing operations. In terms of core protection logic, this invention constructs a three-track parallel monitoring system: Logical track A employs a "spatiotemporal feature-physical state coupling" strategy, effectively filtering out environmental noise such as birds and wind loads and accurately judging high-risk human behaviors through orthogonal verification of infrared gratings and mechanical states, and dual-channel gating technology; Logical track B utilizes mechanical nodes with built-in edge computing to execute an "edge reflection" mechanism, achieving millisecond-level blocking of fall impacts and real-time monitoring of mechanical states to ensure static protection against long-term safety belt disengagement; Logical track C deploys cascaded visual algorithms, using geometric space topological constraints and temporal logic filters to perform high-precision compliance monitoring of missing safety helmets, smoking, and improper clothing. Ultimately, all monitored events converge to the central arbitrator, which maps multi-source heterogeneous data to a three-level audible and visual alarm response matrix according to a strict "priority pyramid" scheduling rule. This ensures a circuit-breaking preemptive response to critical disruption-level emergencies under resource competition, reducing the risk of multi-dimensional alarm confusion and response delays.

[0016] This invention combines the mechanical and spatial sensing of the physical layer with the AI ​​vision algorithm of the application layer to achieve early warning and proactive protection for the entire process of high-altitude transformer operations, effectively preventing safety accidents caused by regulatory negligence.

[0017] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the substation equipment top operation protection method based on heterogeneous data detection in Embodiment 1 of the present invention. Detailed Implementation

[0020] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] Example 1: Embodiment 1 of the present invention provides a method for top-of-station equipment protection based on heterogeneous data detection, such as... Figure 1 As shown, it includes the following steps: S1: Construct a comprehensive protection network based on the safety impact factors of substation equipment, and perform status monitoring on the comprehensive protection network.

[0023] In one specific implementation, this embodiment determines that the safety influencing factors of substation equipment are human safety violations and sudden emergencies. Therefore, a full-domain protection network is constructed by using multi-dimensional indicators such as object intrusion volume, object intrusion time, changes in mechanical data, and image data detection.

[0024] To address the pain points of traditional power fences, such as the need for manual reset, cumbersome installation and debugging, and high false alarm rates, this embodiment designs an Adaptive Environment Stability Decision Mechanism (AESDM). This mechanism immediately takes over the main control logic after the system is powered on, constructing a global sensor state matrix to calculate the physical steady-state indicators of the environment in real time. The system can seamlessly switch from Debug Mode to Armed Mode only if the state vectors of all subsystems meet the convergence condition. Debug Mode refers to the initialization state where the system only performs self-tests, calibrations, and silent data acquisition; Armed Mode refers to the operating state where the system formally activates the logic arbitration function and allows the execution of audible and visual alarms based on sensor trigger signals.

[0025] S1.1: Build a protective network covering the entire area and perform handshake and high-precision clock synchronization initialization operations.

[0026] In one specific implementation, upon system startup, the edge computing host immediately initiates a global node scan and handshake process. The host prioritizes broadcasting a wake-up frame to verify the integrity of the node topology, ensuring the aggregation... All nodes All are online and ready; among them, Indicates an infrared grating node. Represents a mechanical sensing node. This represents a visual monitoring node. If any node is detected to be missing (such as a mechanical node),... (Not yet online) will trigger an L1-level blocking alarm until the entire protected network topology is closed. Based on this, to ensure the timing accuracy of multimodal data arbitration, a lightweight broadcast-based time synchronization protocol is executed in parallel to dynamically calibrate the network's... Local clocks of each node With the main control benchmark The deviation makes it satisfy the convergence condition. .in This represents the maximum permissible synchronization error threshold, which is set to [value] in this embodiment. This effectively suppresses clock jitter and ensures that the fusion error of infrared and mechanical data is controllable.

[0027] S1.2: Perform closure and environmental steady-state scanning detection on the infrared grating link.

[0028] In one specific implementation, to address the optical path alignment deviations caused by uneven installation on substation equipment surfaces, and the dynamic interference from frequent personnel crossing fences during installation, this embodiment introduces a time-domain integral steady-state criterion using the occlusion signal of an infrared grating. During the initialization phase, the system first executes the optical path alignment auxiliary logic, monitoring the first... Signal quality of segment infrared grating , here Characterized as the ratio of the light intensity received at the receiver to the light intensity emitted at the transmitter; if arbitrary The main controller will drive the audio-visual unit to issue an L1-level prompt, guiding the operator to fine-tune the column angle until the link is closed. A preset threshold is set for signal quality. Subsequently, the system enters a static stability scan (5-second rule) to define the global grating state function. : .

[0029] in, This is an indicator function. express Time of the first Real-time logic state of the infrared grating segment. This represents the effective state constant when the optical path is closed and unobstructed. The necessary and sufficient condition for the system to enter arming mode is... In the observation window (set as) The integral value within a range reaches full rank, i.e., satisfies: .

[0030] in, This indicates the current system timestamp. The physical meaning of this criterion is to confirm that the personnel have completely evacuated the monitoring section and that the physical environment has reached an absolute steady state in terms of thermodynamics and structural mechanics.

[0031] S1.3: Perform benchmark self-test and state space definition detection on the force sensor.

[0032] In one specific implementation, to eliminate sensor zero drift and establish the decision boundaries of the runtime state machine, the system performs mechanical self-checks and state space initialization in parallel, abandoning the traditional "power-on zeroing" logic and instead adopting a state space partitioning strategy based on effective preload for mechanical effectiveness verification. First, within the infrared steady-state observation window... Inside, high-frequency sampling is performed on the sensor, frequency And calculate the force value sampled by the force sensor. mean : .

[0033] in, This indicates the total number of sampling points within the observation window. Indicates the sampling point index.

[0034] Then, validity verification is performed: If in, A unified preset threshold for system power-on ( If the connection is deemed "invalid" and an L1-level warning is triggered, otherwise it is deemed valid and the average value is locked as the runtime baseline. Based on this benchmark, the system automatically generates three-state decision boundaries. The initialization of the mapping from physical quantities to logical states has been completed: .

[0035] in, Indicates a state of being out of insurance coverage. Indicates the anchoring status. Indicates an impact state. Represents a mapping function. The waiver determination coefficient (taken as follows) ), and These are the lethal force threshold and the impact jerk threshold, respectively.

[0036] S1.4: Perform mode switching and arming feedback based on security detection results.

[0037] In one specific implementation, the main control unit immediately terminates the initialization process and executes a chain-like deployment action only after the aforementioned network synchronization, infrared steady-state scanning, and mechanical effectiveness verification have all converged successfully. First, the system state machine transitions from debug mode to armed mode, officially activating the multi-track monitoring threads of infrared spatial arbitration, mechanical edge reflection, and visual compliance monitoring; simultaneously, the human-machine interface terminal is driven to turn the global indicator light to a solid green; at the same time, the system closes the evidence chain, starts writing the ring buffer pointer, and begins cyclically recording global sensor data based on a unified clock reference, marking the entire system entering a "plug-and-play" non-intrusive protection state.

[0038] S2: Acquire multi-source heterogeneous data in the protected network, including infrared grating blocking signals, mechanical data, and image data.

[0039] In this embodiment, infrared grating occlusion signals are acquired through a grating equipped with infrared devices within the protected global network. Mechanical data is obtained by deploying multiple mechanical sensors, and image data of the target area is acquired through a high-definition camera. In this embodiment, the target area can be defined as a region within a preset radius centered on the center of the grating's internal area.

[0040] S3: Perform security checks on multi-source heterogeneous data respectively.

[0041] This embodiment uses three logic tracks to perform security detection on three types of heterogeneous data: logic track A based on infrared-triggered spatial arbitration, logic track B based on mechanically triggered edge reflection, and logic track C based on visual compliance monitoring (Vision Logic). Specifically, it includes the following steps: S3.1: Determine whether there is an intrusion by detecting the area and duration of the infrared grating blocking signal.

[0042] In one specific implementation, logical track A serves as the decision-making hub for the entire protected network. The edge computing host executes a heterogeneous data fusion strategy based on "spatiotemporal feature-physical state coupling" on logical track A. Unlike traditional alarms that rely on a single signal source, this embodiment constructs an orthogonal verification plane: while capturing the spatiotemporal occlusion signal (spatial dimension) of the infrared grating in real time, the host synchronously indexes the current load state (physical dimension) of the mechanical nodes via a low-latency bus at millisecond levels. By running a multimodal cross-validation algorithm, single-dimensional logical artifacts can be effectively eliminated, accurately determining the true intentions of the operators (whether it is a controlled maintenance action or an uncontrolled accidental intrusion), ultimately driving the hierarchical response matrix to output deterministic alarm commands, fundamentally solving the false alarm dilemma. The specific steps are as follows: S3.1.1: A multi-level filtering mechanism based on morphological features is used to reduce noise in infrared grating occlusion signals.

[0043] Specifically, the electromagnetic environment at substation sites is complex and subject to uncontrollable interference from birds, falling leaves, and wind loads. Therefore, this embodiment employs multi-dimensional feature extraction to remove environmental noise, allowing only valid signals with human torso characteristics to pass through.

[0044] First, a low-density morphological erosion algorithm is employed to address high-frequency transient noise such as birds flying at high speeds or small tools falling. This logic sets a minimum effective occlusion area and a time threshold; if a certain number of occluded beams are detected... Root and duration The algorithm identifies it as discrete, non-connected noise and performs hardware-level filtering directly at the underlying level.

[0045] Secondly, to address the swaying of flexible support poles caused by strong winds or transformer vibrations, common-mode rejection logic, similar to differential signal processing, was introduced. When the entire area is monitored... The gratings located on opposite sides of the topology (e.g., east and west sides) are in When a synchronous interruption occurs within the window, based on the kinematic axiom that a human body cannot simultaneously touch the opposite side fence within a microsecond time, it is identified as structural resonance or wind load disturbance, and thus such synchronous signals are shielded.

[0046] Finally, spatial granularity screening is performed for single-beam obstruction caused by accidental touch by small animals or slight hand contact. If only a single beam is detected at any time, the system classifies it as a non-invasive accidental touch and only drives the audio-visual unit to issue an L1-level soft warning to avoid overreacting and disturbing the operator.

[0047] S3.1.2: Dual-channel gating based on spatiotemporal integral weights is used to determine spatial intrusion of the denoised infrared grating occlusion signal.

[0048] Specifically, this embodiment only processes suspected human targets after physical layer noise reduction screening. To resolve the contradiction between missing fatal threats and false alarms of normal actions, a dual-channel spatiotemporal gating logic is introduced, which diverts intrusion behavior to processing channels of different priorities based on the temporal integral weight of the occlusion signal.

[0049] On the one hand, a slow integration channel is established for routine scenarios such as workers climbing, crawling through, or accidentally tripping and slowly impacting the fence with their torso. This channel employs a time-domain cumulative confirmation mechanism: if the number of obstructed beams is detected... The root (exhibiting large-area occlusion characteristics) and the duration of occlusion Exceeding the confirmation threshold (In this example, we assume...) This is considered a physical intrusion. The delay in confirmation here is intended to filter out glitches caused by rapid human movement and generate an activation signal to trigger the regular arbitration process.

[0050] In parallel, to prevent the slow channel from filtering out human bodies falling at excessively high speeds (passing through the grating in milliseconds) as noise, a fast interruption channel is specifically designed. This logic is based on free fall acceleration (…). The model has a very short capture window: when a detection is made... The beams of light were blocked simultaneously and instantaneously, even if the duration fell within The area remains classified as a suspected fall zone. The activation signal generated by this channel has the highest interrupt privileges, allowing it to immediately preempt system bus resources and ensure a zero-delay response to catastrophic emergencies.

[0051] S3.1.3: Make a final decision on the object intrusion situation based on the space intrusion determination results.

[0052] Specifically, when an activation signal is generated Then, the main control unit immediately executes scene freeze, locking all sensor data within the current time window, and simultaneously queries the real-time status of the wireless mechanical monitoring nodes. It applies a strict binary decision-making logic based on different situations.

[0053] The first scenario is controlled overstepping of boundaries, which inhibits compliance actions.

[0054] If the query returns This indicates that the worker is currently under the effective protection of their safety belt. The system logic engine infers that the infrared trigger may have originated from "leaning forward after the safety belt is fastened" or "interference from a large floating object (such as a rainproof tarpaulin)." Based on the principle of "prioritizing work continuity," a high-risk alarm suppression strategy is implemented, recording only the log without triggering audible and visual blocking, thereby avoiding unnecessary work stoppages.

[0055] The second scenario involves unprotected violations, leading to the blocking of potentially fatal risks.

[0056] Conversely, if the query results show (i.e., unprotected or impacted state), which constitutes a fatal combination of "space intrusion + lack of protection." The current action is immediately determined to be either an unprotected violation of boundaries or a genuine fall. At this point, the central arbitrator issues an L3-level blocking command, driving the on-site buzzer to emit a high-decibel long blast and the indicator light to flash red, using the highest priority physical means to prevent the escalation of the danger.

[0057] S3.2: Determine whether there is an impact by comprehensively analyzing changes in mechanical data and the seat belt engagement status.

[0058] In one specific implementation, logical track B operates independently in a closed loop by mechanical sensing nodes deployed at the high-altitude work site. Its design is intended to handle extreme conditions such as sudden falls or mechanical accidents within the fence. Unlike conventional logic that relies on master control scheduling, logical track B directly invokes the parameters established during the initialization phase. State boundaries have the highest interrupt privileges and execution priority on the bus.

[0059] S3.2.1: Determine the status of the Static Connection Guardian.

[0060] Specifically, this step is dedicated to real-time monitoring of the safety belt attachment status of workers, aiming to mitigate the risks of false fastening or habitual violations caused by human negligence during operations. During system operation, the state machine within the node continuously performs windowed analysis on the real-time mechanical waveform. When the algorithm detects a change in state from... (Anchored state, i.e., effective preload detected) A transition occurs, switching to... When the system is in a state of no protection (i.e., the force is zero), it does not immediately trigger an alarm, but instead starts a configurable countdown window. (This embodiment is set as) ).

[0061] This window is designed based on statistical analysis of the stepping frequency in high-altitude power operations, aiming to provide a legitimate relocation buffer period for workers transferring between different attachment points. If in The sensor detected again before it ran out of power. If the signal is received and the countdown is automatically reset, the system will determine that the current behavior constitutes an unnecessary long-term disconnection from protection and will then activate the audible and visual unit to issue an L2-level warning. This mechanism not only prevents blind alarms but also enables precise and flexible correction of violations.

[0062] S3.2.2: Millisecond edge reflection blocking based on actual mechanical curves.

[0063] Specifically, this step simulates the spinal reflex arc mechanism in biology at the electronic engineering level, constructing a decentralized distributed stress response link and realizing local rapid interruption response based on edge computing nodes. Traditional centralized alarm systems require a long link of "sensor sampling - wireless transmission - main control computing - command feedback," often with a delay of hundreds of milliseconds. In this embodiment, the mechanical node has an independent edge computing unit built in.

[0064] Once the sensor detects locally that the mechanical curve satisfies The triggering condition, namely instantaneous tension. or urgency The node will unconditionally circuit break all currently low-priority background tasks, among which, For the set tensile threshold, The threshold for agitation is set. Instead of waiting for polling commands from the main control unit, it directly sends L3 blocking frames to the audio-visual terminal via the ESP-NOW proprietary low-latency protocol. Actual measurements show that this mechanism compresses the end-to-end response latency to [value missing]. Within this range, it ensures that the last line of defense for life takes effect before the main control is activated, thus buying precious physical time for the fall rescue, before the infrared logic completes the complex spatial arbitration.

[0065] Step 3.2.3: Set up a data locking mechanism based on a circular buffer.

[0066] Specifically, to provide immutable legal evidence and technical analysis data after an incident, this embodiment integrates a data locking mechanism based on a ring buffer. In normal monitoring mode, the mechanical nodes continuously overwrite their internal storage queues at a high-frequency sampling rate.

[0067] Whether it's the long-term write protection trigger in step 3.2.1 or the impact event trigger in step 3.2.2, once the alarm flag is set, the node immediately performs a "write protection" operation, freezing the current buffer pointer. This completely locks the steady-state data up to the trigger moment. (e.g., 5 seconds) and dynamic response data after the trigger moment (e.g., 5S), combined to form the total length The high-frequency full waveform recording was then used. This data packet was subsequently marked as high-priority evidence and asynchronously uploaded to the main control unit via an encrypted channel, thereby accurately reconstructing the mechanical impact curve and the behavioral trajectory of the workers at the moment of the accident.

[0068] S3.3: Determine whether there is any violation by performing multi-target visual detection on image data.

[0069] In one specific implementation, the logic track C adopts a cascaded processing architecture: the first stage is based on a single-stage target detection network, which outputs a high-precision structured Boolean state through spatial topology and geometric constraints; the second stage runs a logic filter to perform time-domain smoothing and frequency control on the Boolean state, and finally generates a violation judgment result.

[0070] S3.3.1: First stage: Front-end algorithm and geometric space verification (Algorithm Stage).

[0071] S3.3.1.1: Use a single-stage object detection network to perform feature extraction and preliminary detection to obtain the target bounding box.

[0072] Specifically, considering the characteristics of factory environments such as large field of view, cluttered backgrounds, and significant differences in the scale of illegal targets (e.g., human bodies are large targets, while cigarettes are extremely small targets), this embodiment adopts an improved single-stage target detection network based on the YOLOv8 architecture to extract image features.

[0073] First, image data augmentation preprocessing is performed. Considering the characteristics of large changes in factory lighting and severe equipment occlusion, Mosaic data augmentation technology is introduced during the model training stage. By randomly scaling, cropping, and arranging four images as input, not only is the background semantic information of the detected target enriched, but the frequency of small targets (such as cigarettes and unworn helmets) in the training samples is also significantly increased, improving the robustness of the model.

[0074] Then, a multi-scale feature fusion strategy is used to analyze and fuse the preprocessed image data. The single-stage object detection network employs... As the backbone network In feature fusion neck Phase, adopt Structure. To address the issue of missed detection of small targets, this embodiment focuses on enhancing the utilization of shallow feature maps (80*80 resolution). Through bottom-up path aggregation, the rich geometric information (contours, edges) contained in the shallow layer is losslessly transferred to the deep network, ensuring that the features of tiny cigarette targets are not diluted by the downsampling process.

[0075] Finally, the fusion results are decoupled and output. The detection head uses a decoupled structure to handle the classification and regression tasks separately. The model ultimately outputs two sets of targets: The main target set P: contains all detected personnel, denoted as , where b is the total number of main targets.

[0076] Attribute target set: contains all violations, categorized as: Smoke, No-Helmet, Short-Sleeve, Short-Pants.

[0077] S3.3.1.2: Constructing topological relationships between subjects and attributes in target boxes Since the bounding boxes output by a single-stage object detection network are independent and lack logical connections, this step constructs spatial topological relationships to associate attribute targets with specific individuals.

[0078] Distance matrix calculation: Let the first... Individual objectives The coordinates of the center point c are , No. Attribute Target The coordinates of the center point are Calculate the Euclidean distance matrix. , where w represents the number of rows in the matrix and v represents the number of columns in the matrix. The elements Represents attributes With personnel Distance: .

[0079] Greedy matching strategy: Set a maximum association threshold (1 / 5 of the image diagonal). For each attribute target In order to satisfy Under the given conditions, select the person with the shortest distance. As its candidate parent node. If the distance of all people is greater than the threshold, the attribute target is marked as an isolated target and directly rejected as a false positive.

[0080] S3.3.1.3: False detection filtering operation based on dual filtering funnel mechanism for geometric constraints.

[0081] Specifically, this step aims to address the high false positive rate caused by background interference. A dual filtering funnel mechanism is designed for attribute targets at different scales, including a first branch for false positives of large-scale attribute items and a second branch for verification of small-scale attribute items.

[0082] The first branch performs coverage rate checks on large-scale attributes (short-sleeved shirts, shorts). This logic filters false positives for clothing hanging on walls or placed on chairs. It defines a bounding box for each candidate person. Attribute detection box (e.g., short sleeves) .

[0083] Calculate the intersection: First, calculate the intersection of the two rectangles ( , )and( , Coordinates of the overlapping region: , .

[0084] like and Then the intersection area ;otherwise .

[0085] Containment Ratio Calculation and Determination: The containment ratio is defined as the area of ​​intersection. Area of ​​the target attribute itself ratio : .

[0086] Judgment rule: Set the coverage rate .

[0087] like Determine if the clothing is "worn" by the subject, retain the result, and output Is. Short_Sleeves =True.

[0088] like The clothing item is determined to be background interference (such as clothes hanging on the wall) and is therefore removed. Short_Sleeves =False.

[0089] Second branch: This logic is used to use ROI region mapping to verify and filter false alarms such as light tubes and reflective points that resemble cigarette butts.

[0090] Define the functional ROI active area: based on people detection boxes The geometric parameters are used to dynamically divide functional areas and set up personnel detection frames. The width is The height is : Head ROI ( ): Used for helmet verification. Defined as the top 1 / 3 area of ​​the personnel frame.

[0091] .

[0092] Hand-mouth ROI( ): Used for cigarette calibration. Considering that the hand may extend beyond the body contour when smoking, the ROI area is appropriately expanded horizontally, with an expansion coefficient. .

[0093] .

[0094] Center point mapping determination: Obtain the center point of the suspected cigarette target. Judgment rules: like If the center point coordinates fall within the rectangular area of ​​the active hand-mouth region, it is considered a valid smoking behavior, and Is_Smoking=True is output; otherwise... (For example, light tubes located overhead, or reflective litter at one's feet) are marked as static environmental noise and removed, outputting Is. Smoking =False.

[0095] Obtain the center point of the suspected helmet target Judgment rules: like If the center point coordinates fall within the rectangular area of ​​the active head region, the helmet detection result is considered to have spatial consistency, and Is is output. Helmet =True; if (For example, a circular highlight reflection appearing on the side, behind, or in the background area of ​​a person's frame) is marked as a false detection, and Is is output.Helmet =False.

[0096] S3.3.2: Second Stage: Back-end Logic Filtering and Alarm Generation (Logic Stage).

[0097] Specifically, given the inherent probabilistic fluctuations of deep learning algorithms in dynamic scenes, and the fact that visual nodes do not directly respond to the original inference results, this embodiment deploys a temporal logic filter. Its core function is to perform temporal verification and frequency suppression on the structured Boolean data from the algorithm module, aiming to smooth out the algorithm's random jitter and minimize the interference of false alarms on operators while ensuring the detection rate. Specifically, it includes the following steps: S3.3.2.1: Helmet Detection based on a time-domain integration mechanism using continuous negative samples.

[0098] Specifically, to address the potential fluctuations in algorithm confidence caused by changes in the body posture of personnel at the work site, this logic module abandons the "single-frame judgment" mode and instead adopts a "continuous frame loss confirmation mechanism." The system collects input data Is_Helmet in real time and maintains a sliding time window. (This embodiment is set as follows) Only when the inference results of all consecutive frames within that time window remain in the False (i.e., not wearing) state is the violation considered statistically real, thus generating a valid alarm signal.

[0099] Once a confirmed violation is triggered, the main controller will drive the audio-visual unit to issue an L2 level warning. Simultaneously, to prevent intermittent alarm signals due to personnel being at the edge of the monitoring area, an anti-interference cooling strategy has been introduced: automatically entering [a state of alert] after the alarm is triggered. The silent refractory period (Cooling Period) ignores repeated triggering requests, thus providing timely reminders while avoiding auditory fatigue.

[0100] S3.3.2.2: Smoking Behavior Monitoring Based on Transient Feature-Based Edge Latching Mechanism.

[0101] Specifically, unlike helmet monitoring, smoking is highly concealed and transient (e.g., rapid hand covering, flickering cigarette flame). Therefore, this embodiment employs a transient event latching mechanism. This logic does not require stability confirmation across multiple consecutive frames; instead, it uses a highly sensitive positive edge triggering strategy: as long as the system receives a single frame Is at any given time... Smoking A signal that is True will immediately trigger an internal interrupt and perform an event latch, regardless of how short its duration.

[0102] This mechanism ensures that even extremely brief smoking actions can be accurately captured. Once latched successfully, the main controller immediately outputs an L2-level alarm and simultaneously captures a high-definition image of the current frame and the preceding and following frames. The video streams are packaged and uploaded to the cloud-based violation log database as a basis for subsequent investigation.

[0103] S3.3.2.3: Attire Compliance Monitoring Based on Keyframe Low-Frequency Polling Mechanism.

[0104] Specifically, regarding clothing violations such as wearing short sleeves and shorts, given that these are continuous physical states rather than sudden actions and do not change frequently over time, this embodiment designs a resource-saving static entry check mechanism. This mechanism does not occupy the NPU's real-time inference channel but instead employs a keyframe polling strategy: it only checks when a person first enters the screen area (ROIEntry Event) or at preset long intervals (e.g., ...). When ), the wake-up algorithm model responds to Is. Short_Sleeves A snapshot check is performed on the status. Once a violation is detected, an L2-level alarm is triggered. This strategy effectively reduces the continuous load on edge computing units and optimizes computing power allocation.

[0105] S4: Provide protection feedback based on the alarm grading response strategy of the priority circuit breaker mechanism according to the security detection results.

[0106] In one specific implementation, this embodiment sets up an alarm grading strategy based on a priority circuit breaker mechanism, with three levels (L1, L2, and L3) set according to importance from low to high, and prioritizing the implementation of alarm mechanisms with higher importance levels: Level L1: As a warning level, it is suitable for gentle reminders of incomplete initialization, accidental touches by a single infrared beam, and minor violations. It is characterized by a slow flashing yellow light. ), single short whistle ( ).

[0107] Level L2: As a warning level, it is used for common violations such as missing helmets, smoking, improper clothing, and prolonged periods without safety insurance. It is indicated by a rapidly flashing orange light. ), intermittent rapid beeping ( Ming / Stop, loop.

[0108] Level L3: As a blocking level, it is used in emergency scenarios such as unprotected boundary crossings, real falls, and fatal impacts. This level has the highest interrupt privileges. It is characterized by a flashing red light. (Still lit), continuous long beep / high frequency alarm (continuous output).

[0109] When multiple events occur simultaneously, scheduling is based on the following priority circuit breaker mechanism: Highest priority L3 blocking. Triggered by unprotected violation or fall in infrared logic, or fatal impact or edge reflection in mechanical logic.

[0110] Scheduling employs a non-maskable preemption and deadlock approach. This level grants absolute system control. Upon triggering, a hard interrupt signal is immediately sent to the underlying driver layer, unconditionally terminating all currently executing L1 prompt or L2 warning processes. The audible and visual outputs are physically locked to an L3 blocking state (red light flashing + high-frequency continuous beep), and the system enters a software deadlock mode. In this mode, except for hardware reset or authorized manual deactivation commands, the system refuses to respond to any new sensor inputs, ensuring that evidence is fully preserved and the alarm remains effective.

[0111] Medium-priority L2 alert. Triggered by compliance monitoring (helmet / smoking / clothing) in visual logic and static connection protection (long-term warranty breach) in mechanical logic. Scheduling employs escalation and peer masking. If the system is currently in an L1 idle or alert state, the event will immediately suspend lower-level tasks and take over the output peripherals. If the system is already in an L2 alarm state, and a new L2 event is triggered (such as detected smoking), the arbitrator executes a first-come, first-served strategy, maintaining the currently playing alarm stream and temporarily suppressing (masking) subsequent peer requests. This logic aims to prevent auditory confusion caused by multiple concurrent alarm tones, ensuring clear communication of each instruction.

[0112] Low-priority L1 level alert. Triggered by initialization process, single-beam infrared accidental touch, or network handshake feedback.

[0113] Scheduling is performed using the Idle Slot Execution method. Tasks at this level have only the lowest CPU time slice allocation rights. L1 logic is only allowed to execute when the system is completely idle, i.e., no L2 or L3 level events have occurred. During execution, any high-priority signal input can interrupt the current L1 state at any time, instantly overwriting it without waiting for the current prompt tone to finish playing.

[0114] To more clearly illustrate the scheduling mechanism in scenarios with concurrent multi-source heterogeneous data, this embodiment, combined with actual working conditions, elaborates on two typical alarm state transition processes.

[0115] Actual operating scenario 1: Preemptive interruption of compliance warnings in high-risk situations. Assuming that the worker does not wear a safety helmet when entering the monitoring area, the visual monitoring node triggers the logic first, and the system is currently in a Level 2 warning state (the central indicator light flashes orange rapidly, and the buzzer sounds intermittently and rapidly), and the warning will be displayed repeatedly during the cooling cycle.

[0116] In this state, if the infrared grating suddenly detects a high-density crossing signal (fast channel trigger), and simultaneously queries that the mechanical node is in a "disabled state" ( The system determines that an "unprotected high-risk fall" event has occurred, belonging to the highest priority level (L1). At this point, the central arbitrator immediately executes the circuit breaker mechanism: The main control unit immediately sends a hard interrupt command, forcibly terminating the currently executing L2-level safety helmet warning thread and cutting off the output control of the orange light and intermittent alarm. The system state machine transitions to the L3-level blocking state within milliseconds. The buzzer seamlessly switches to a continuous, high-decibel siren, and the indicator light turns to a red strobe mode. Regardless of whether the previous L2-level alarm has finished playing, the system prioritizes ensuring the unobstructed flow of life-saving channels until manually reset to release the L3 lock.

[0117] Actual operating scenario 2: A tiered upgrade of proactive warnings to passive alerts. Assume that a small animal passing by or an obstruction by a foreign object triggers the lowest-level single infrared grating. The system determines this as a single-beam mis-touch, placing it in L1 level alert status (central indicator light flashes yellow slowly, buzzer sounds briefly once), a low-priority task.

[0118] During this period, if the visual node detects a worker smoking and generates a smoking confirmation signal, it falls under Level 2. Since Level 2 has higher priority than Level 1, an alarm escalation strategy is executed: The main control unit suspends the current L1 level silent alert logic and releases the audio-visual channel resources. The alarm level is then upgraded from L1 to L2. The indicator light changes from yellow to orange, and the buzzer activates a three-short warning rhythm. This process achieves a smooth transition from "passive feedback" to "active violation removal," ensuring that violations are corrected promptly.

[0119] Example 2: Embodiment 2 of the present invention provides a substation equipment top operation protection system based on heterogeneous data detection, comprising: The initialization module is configured to build a comprehensive protection network based on the safety impact factors of substation equipment and to perform status detection on the comprehensive protection network. The data acquisition module is configured to acquire multi-source heterogeneous data in the protected network, including infrared grating blocking signals, mechanical data and image data. The multi-source data processing module is configured to perform security detection on multi-source heterogeneous data. Specifically, it detects the area and duration of infrared grating obstruction signals to determine if there is any object intrusion; it comprehensively judges whether there is any impact by analyzing changes in mechanical data and the seat belt attachment status; and it judges whether there is any violation by performing multi-target visual detection on image data. The alarm module is configured to provide protective feedback based on a priority-based circuit breaker mechanism and alarm tiered response strategy according to the security detection results.

[0120] Example 3: Embodiment 3 of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing the steps of the substation equipment top operation protection method based on heterogeneous data detection as described in Embodiment 1 of the present invention.

[0121] Example 4: Embodiment 4 of the present invention provides a computer device, the device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps of the substation equipment top operation protection method based on heterogeneous data detection as described in Embodiment 1 of the present invention.

[0122] Example 5: Embodiment 5 of the present invention provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the substation equipment top-operation protection method based on heterogeneous data detection as described in Embodiment 1 of the present invention.

[0123] The steps and methods involved in Examples 2, 3, 4 and 5 above correspond to those in Example 1. For specific implementation methods, please refer to the relevant description section of Example 1.

[0124] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc. The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for top-of-station equipment protection based on heterogeneous data detection, characterized in that, Includes the following steps: A comprehensive protection network is constructed based on the safety influencing factors of substation equipment, and the status of the comprehensive protection network is monitored. Acquire multi-source heterogeneous data from the entire protected network, including infrared grating blocking signals, mechanical data, and image data; Security checks are performed on multi-source heterogeneous data. Specifically, the area and duration of infrared grating obstruction signals are detected to determine if there is any object intrusion. Changes in mechanical data and seat belt attachment status are used to determine if there is any impact. Multi-target visual detection of image data is used to determine if there is any violation. The specific steps for determining whether there is an intrusion by detecting the area and duration of the infrared grating blocking signal are as follows: A multi-level filtering mechanism based on morphological features is used to reduce noise in infrared grating-blocked signals. A dual-channel gating system based on spatiotemporal integral weights is used to determine spatial intrusion of the denoised infrared grating occlusion signal. The dual-channel spacetime gating logic includes a slow integration channel and a fast interrupt channel; The slow integration channel employs a time-domain accumulation confirmation mechanism: if the number of blocked beams is detected... Root and occlusion duration Exceeding the confirmation threshold This is determined to be an entity intrusion; the rapid interruption channel is based on a free-fall acceleration model and sets an extremely short capture window: when detected... The beams of light were blocked simultaneously and instantaneously, even if the duration fell within The area is still considered a suspected fall zone; The final decision on object intrusion is made based on the spatial intrusion determination result; specifically, when the activation signal is generated, the main control unit immediately executes scene freeze, locks all sensor data within the current time window, and simultaneously queries the real-time status of the wireless mechanical monitoring node, and executes strict binary decision logic according to different situations. Based on the safety test results, a priority-based circuit breaker mechanism is used to implement a graded alarm response strategy for protection feedback.

2. The substation equipment top job protection method based on heterogeneous data detection of claim 1, wherein, The specific steps for constructing a comprehensive protection network based on the safety influencing factors of substation equipment and performing status monitoring on the comprehensive protection network are as follows: Build a protective network covering the entire area and perform handshake and high-precision clock synchronization initialization operations; Closed-loop and steady-state environmental scanning detection of infrared grating links; Perform baseline self-testing and state space definition detection on the force sensor; Mode switching and deployment feedback are performed based on security test results.

3. The substation equipment top job protection method based on heterogeneous data detection of claim 1, wherein, The specific steps for determining whether an impact has occurred by comprehensively considering changes in mechanical data and the seatbelt's engagement status are as follows: Determine the status of static connection protection; Millisecond-level edge reflection blocking is performed based on actual mechanical curves; edge reflection blocking simulates the spinal reflex arc mechanism in biology at the electronic engineering level, constructing a decentralized distributed stress response link; The mechanical node has a built-in independent edge computing unit, triggered by instantaneous tension. or urgency The node will unconditionally circuit break all currently low-priority background tasks, among which, For the set tensile threshold, The set jerk threshold is used; the end-to-end response latency is compressed to a minimum directly through the ESP-NOW proprietary low-latency protocol. within; Configure a data locking mechanism based on a circular buffer.

4. The substation equipment top job protection method based on heterogeneous data detection of claim 1, wherein, Determining whether there are violations by performing multi-object visual detection on image data includes: The first stage is based on a single-stage target detection network, which outputs high-precision structured Boolean states through spatial topology and geometric constraints; the dual filtering funnel mechanism includes a first branch for false positives of large-scale attribute items and a second branch for verification of small-scale attribute items. First branch: Coverage rate verification for large-scale attributes; Second branch: This logic is used to use ROI region mapping to verify and filter false alarms such as light tubes and reflective points that resemble cigarette butts. The second stage runs a logic filter to smooth the Boolean state in the time domain and control its frequency, ultimately generating a result for judging violations. Three completely different logical filtering mechanisms are used for different violations: the safety helmet detection uses a time-domain integration mechanism with continuous negative samples, which requires confirmation of multiple consecutive frames; the smoking detection uses an edge latching mechanism with transient features, which adopts a highly sensitive positive edge triggering strategy; and the clothing detection uses a low-frequency polling mechanism with key frames, which only detects when a specific event is triggered.

5. The substation equipment top operation protection method based on heterogeneous data detection as described in claim 1, characterized in that, The alarm tiered response strategy based on the priority circuit breaker mechanism sets three levels—L1, L2, and L3—according to their importance from low to high, prioritizing the implementation of alarm mechanisms with higher importance levels. Level L1: As a prompt level, it is suitable for flexible reminders of incomplete initialization, accidental touch of a single infrared beam, and minor violations; Level L2: As a warning level, it is used for common violation scenarios such as missing helmets, smoking, improper clothing, and prolonged periods without safety protection. Level L3: As a blocking level, it is used for emergency scenarios such as unprotected crossings, real falls, and fatal impacts.

6. A substation equipment top-operation protection system based on heterogeneous data detection, characterized in that, include: The initialization module is configured to build a comprehensive protection network based on the safety impact factors of substation equipment and to perform status detection on the comprehensive protection network. The data acquisition module is configured to acquire multi-source heterogeneous data in the protected network, including infrared grating blocking signals, mechanical data and image data. The specific steps for determining whether there is an intrusion by detecting the area and duration of the infrared grating blocking signal are as follows: A multi-level filtering mechanism based on morphological features is used to reduce noise in infrared grating-blocked signals. A dual-channel gating system based on spatiotemporal integral weights is used to determine spatial intrusion of the denoised infrared grating occlusion signal. The dual-channel spacetime gating logic includes a slow integration channel and a fast interrupt channel; The slow integration channel employs a time-domain accumulation confirmation mechanism: if the number of blocked beams is detected... Root and occlusion duration Exceeding the confirmation threshold This is determined to be an entity intrusion; the rapid interruption channel is based on a free-fall acceleration model and sets an extremely short capture window: when detected... The beams of light were blocked simultaneously and instantaneously, even if the duration fell within The area is still considered a suspected fall zone; The final decision on object intrusion is made based on the spatial intrusion determination result; specifically, when the activation signal is generated, the main control unit immediately executes scene freeze, locks all sensor data within the current time window, and simultaneously queries the real-time status of the wireless mechanical monitoring node, and executes strict binary decision logic according to different situations. The multi-source data processing module is configured to perform security detection on multi-source heterogeneous data. Specifically, it detects the area and duration of infrared grating obstruction signals to determine if there is any object intrusion; it comprehensively judges whether there is any impact by analyzing changes in mechanical data and the seat belt attachment status; and it judges whether there is any violation by performing multi-target visual detection on image data. The alarm module is configured to provide protective feedback based on a priority-based circuit breaker mechanism and alarm tiered response strategy according to the security detection results.

7. A computer program product, characterised in that, The computer program product includes a computer program that, when executed by a processor, implements the substation equipment top operation protection method based on heterogeneous data detection as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-5: a method for protecting substation equipment top operations based on heterogeneous data detection.

9. A computer device, comprising: include: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the substation equipment top-operation protection method based on heterogeneous data detection as described in any one of claims 1-5.

Images