Power plant violation behavior real-time identification method, system, device and storage medium

Abstract

The present application relates to the technical field of machine learning, and specifically provides a power plant violation behavior real-time identification method, system, device and storage medium, comprising: collecting multi-source monitoring data of a power plant operation area in real time, wherein the multi-source monitoring data comprises video data, personnel position data and environment state data; extracting features from the multi-source monitoring data, and fusing the extracted features into multi-dimensional feature information for representing personnel behavior state; inputting the multi-dimensional feature information into a pre-trained risk level evaluation model to obtain a multi-level risk evaluation result corresponding to current personnel behavior; the multi-level risk evaluation result at least comprises a high risk level and a low risk level; according to the multi-level risk evaluation result, starting a graded early warning response matched with the risk level. The present application can identify different levels of risk before an accident occurs and trigger real-time early warning, changing passive response to active prevention.

Description

Technical Field

[0001] This invention belongs to the field of machine learning technology, specifically relating to a method, system, device, and storage medium for real-time identification of violations in power plants. Background Technology

[0002] Traditional power plant safety monitoring primarily relies on video recordings for post-incident verification and simple real-time alarms (such as those for electronic fence intrusion). The former is not real-time and cannot proactively intervene when an accident occurs; the latter uses a binary alarm mode of "violation / non-violation," frequently triggering alarms for numerous minor or false alarms, easily leading to "alarm fatigue" among staff, causing them to ignore truly high-risk signals. Existing technologies lack the ability to perform real-time, accurate, and quantitative assessments of the risk level of violations, making it difficult to achieve "risk-level control" and optimized allocation of early warning resources, and failing to meet the urgent needs of modern power plants for proactive prevention of safety accidents. Summary of the Invention

[0003] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method, system, device and storage medium for real-time identification of power plant violations, so as to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides a method for real-time identification of violations in power plants, comprising: Real-time acquisition of multi-source monitoring data from the power plant's operating area, including video data, personnel location data, and environmental status data; Features are extracted from the multi-source monitoring data, and the extracted features are fused into multi-dimensional feature information to characterize the state of human behavior. The multi-dimensional feature information is input into a pre-trained risk level assessment model to obtain a multi-level risk assessment result corresponding to the current personnel behavior; the multi-level risk assessment result includes at least a high-risk level and a low-risk level. Based on the results of the multi-level risk assessment, a graded early warning response matching the risk level is initiated.

[0005] In one optional implementation, multi-source monitoring data of the power plant's operating area are collected in real time, including: High-definition cameras deployed in the work area are used to collect video data containing personnel behavior. The precise real-time location and movement trajectory data of the personnel are collected by using positioning base stations deployed in the work area and positioning tags carried by the personnel. Data on the operating status of equipment is collected by status sensors installed on hazardous equipment. Environmental status data is collected by deploying environmental sensors in the work area.

[0006] In an optional implementation, features are extracted from the multi-source monitoring data, and the extracted features are fused into multi-dimensional feature information for characterizing the state of human behavior, including: Extract behavioral features, spatial relationship features, and global scene features related to the current person from the video data; Based on the real-time location and movement trajectory data of the personnel, combined with the spatial relationship features, the spatiotemporal interaction trajectory features of the personnel are generated. The behavioral features, spatial relationship features, global scene features, and spatiotemporal interaction trajectory features are fused to generate the multi-dimensional feature information.

[0007] In an optional implementation, behavioral features, spatial relationship features, and global scene features related to the current person are extracted from the video data, including: Target detection is performed on the current frame of the video data to identify and locate multiple targets, including at least the current personnel and one or more dangerous devices. Extract the skeletal key point sequence of the current person; Based on the location information of the current personnel and the one or more hazardous devices, calculate one or more measures that characterize the spatial relationship between them; The dynamic descriptor determined based on the skeletal keypoint sequence, the one or more metrics, and the global scene descriptor extracted from the current frame are fused to form the behavioral features, spatial relationship features, and global scene features. The behavioral features are mainly generated by dynamic descriptors determined based on the skeletal keypoint sequence. The dynamic descriptors include at least one of motion speed and posture change rate calculated based on skeletal keypoints within a temporal segment. The spatial relationship features are mainly generated by one or more of the metrics, which include at least one of the following: normalized distance between the current personnel and the hazardous equipment, relative direction, and area overlap. The global scene features are mainly generated from the global scene descriptor extracted from the current frame.

[0008] In an optional implementation, based on the real-time location and movement trajectory data of the personnel, and combined with the spatial relationship features, spatiotemporal interaction trajectory features of the personnel are generated, including: Obtain the continuous location sequence of the personnel within a time window of a predetermined duration, with the current time as the cutoff point; Based on the continuous position sequence, calculate at least one of the following movement indicators of the person within the time window: movement speed, movement direction, and acceleration. Based on the location of at least one hazardous device associated with the continuous location sequence and the spatial relationship features, calculate the rate of change or proximity trend of the distance between the personnel and the hazardous device; The spatiotemporal interaction trajectory feature is generated by combining the at least one movement indicator with the rate of change or the approximate trend.

[0009] In one optional implementation, the pre-trained risk level assessment model is a classification model based on the random forest algorithm; Obtain multi-level risk assessment results, including: The multi-dimensional feature information is input into the random forest classification model, and the multiple decision trees in the model output the classification result corresponding to the current behavior; The number or proportion of trees belonging to each predetermined risk level in the classification results output by the multiple decision trees is statistically analyzed. Based on the preset threshold rules and the statistical results, the final multi-level risk assessment result is determined; wherein the predetermined risk level includes no risk, low risk, medium risk and high risk.

[0010] In an optional implementation, based on the results of the multi-level risk assessment, a tiered early warning response matching the risk level is initiated, including: When the multi-level risk assessment result is low risk, a first early warning response is triggered, which includes activating the on-site audible and visual alarm equipment to issue a first-level warning. When the multi-level risk assessment result is medium risk, a second early warning response is triggered. The second early warning response includes sending early warning information to the terminal device of a preset first-category responsible person. When the multi-level risk assessment result is high risk, a third early warning response is triggered. The third early warning response includes performing at least two of the following operations simultaneously: sending alarm information to the terminal devices of the preset second-category responsible personnel, activating the plant area emergency broadcast, and automatically executing linkage control commands associated with the hazardous area or equipment.

[0011] Secondly, the present invention provides a real-time identification system for power plant violations, comprising: The data acquisition module is used to collect multi-source monitoring data of the power plant's operating area in real time. The multi-source monitoring data includes video data, personnel location data, and environmental status data. The feature extraction module is used to extract features from the multi-source monitoring data and fuse the extracted features into multi-dimensional feature information for characterizing the state of human behavior. The risk identification module is used to input the multi-dimensional feature information into a pre-trained risk level assessment model to obtain a multi-level risk assessment result corresponding to the current personnel behavior; the multi-level risk assessment result includes at least a high-risk level and a low-risk level. The early warning response module is used to initiate a graded early warning response that matches the risk level based on the results of the multi-level risk assessment.

[0012] Thirdly, a device is provided, comprising: Memory, used to store the real-time identification program for power plant violations; A processor is configured to implement the steps of the real-time identification method for power plant violations as provided in the first aspect when executing the real-time identification program for power plant violations.

[0013] Fourthly, a computer-readable storage medium is provided, on which a real-time identification program for power plant violations is stored, wherein when the real-time identification program for power plant violations is executed by a processor, the program implements the steps of the real-time identification method for power plant violations provided in the first aspect.

[0014] The beneficial effects of this invention are as follows: The real-time identification method, system, equipment, and storage medium for power plant violations provided by this invention can achieve accurate risk quantification and real-time intervention. Through multi-source data fusion and a multi-level risk assessment model, it transforms traditional binary alarms into continuous risk rating, enabling the identification of different risk levels and real-time triggering of early warnings before accidents occur, changing passive response to proactive prevention. It effectively avoids alarm fatigue by using a differentiated graded early warning mechanism to push personnel intervention or emergency response only to medium- and high-risk behaviors, reducing false alarms and low-value alarms, and ensuring that important alarms are addressed promptly. It conforms to the principle of risk-level control, matching early warning responses with risk levels, optimizing the allocation of safety management resources, and improving the systematicness and effectiveness of overall safety management. It possesses continuous optimization capabilities; the model can be continuously iterated with new cases, enhancing its adaptability to complex scenarios and improving identification accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic flowchart of a method according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic block diagram of a system according to an embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0020] Unless otherwise defined, 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. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0021] The real-time identification method for power plant violations provided in this embodiment of the invention is executed by a computer device, and correspondingly, the real-time identification system for power plant violations runs on the computer device.

[0022] Figure 1 This is a schematic flowchart illustrating a method according to an embodiment of the present invention. Wherein, Figure 1 The implementing entity could be a real-time identification system for violations in power plants. Depending on different needs, the order of the steps in this flowchart can be changed, and some steps can be omitted.

[0023] like Figure 1 As shown, the method includes: S1. Real-time acquisition of multi-source monitoring data in the power plant's operating area, including video data, personnel location data, and environmental status data; S2. Extract features from the multi-source monitoring data and fuse the extracted features into multi-dimensional feature information for characterizing the state of human behavior; S3. Input the multi-dimensional feature information into a pre-trained risk level assessment model to obtain a multi-level risk assessment result corresponding to the current personnel behavior; the multi-level risk assessment result includes at least a high-risk level and a low-risk level. S4. Based on the results of the multi-level risk assessment, initiate a graded early warning response that matches the risk level.

[0024] In one embodiment of the present invention, based on step S1, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0025] S101. Video Data Acquisition System: Several network-connected high-definition infrared cameras are deployed in key operational areas of the plant, such as the high-voltage power distribution room, turbine platform, boiler area, chemical storage area, and main inspection passages. These cameras use H.264 / H.265 encoding format and output real-time video streams at a resolution of 1080P or higher at a rate of no less than 25 frames per second. The streams are transmitted to the data processing server via industrial Ethernet to ensure clear capture of personnel operation behavior, the wearing of personal protective equipment, and the status of surrounding equipment.

[0026] S102. Personnel Precision Positioning System: Employs Ultra-Wideband (UWB) positioning technology. UWB positioning base stations are deployed in a grid pattern within the factory, covering all work areas. All on-site personnel wear safety helmets or work badges with built-in UWB tags. The system measures the time difference between the tag signals arriving at each base station, calculates and outputs in real-time the three-dimensional coordinates (X, Y, Z), movement speed, direction of movement, and historical trajectory sequence for each person, achieving centimeter-level positioning accuracy.

[0027] S103. Equipment Condition Monitoring System: For hazardous equipment such as rotating machinery, high-voltage switchgear, and energized busbars, appropriate condition sensors are installed. For example, vibration sensors are installed on the equipment casing to monitor abnormal vibrations; current / voltage sensors are installed through electrical contacts to monitor unplanned energization; and "Do Not Close" smart locks are installed on equipment under maintenance, and their locking status is monitored. These sensors upload real-time operating and safety status signals of the equipment via industrial buses (such as Modbus and Profinet) or IoT gateways.

[0028] S104. Environmental Status Monitoring System: In critical environments such as confined spaces, flammable and explosive areas, and cable tunnels, combustible gas detectors, oxygen concentration sensors, temperature and humidity sensors, and smoke detectors are deployed. These environmental sensors collect real-time data on gas concentration, temperature, humidity, and fire alarm signs in their respective areas and transmit this data to the central processing unit via the same Internet of Things (IoT) network.

[0029] In one embodiment of the present invention, based on step S2, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0030] S201. Extract behavioral features, spatial relationship features, and global scene features related to the current person from the video data.

[0031] The input video stream is analyzed frame by frame. First, the pre-trained YOLOv5 object detection model is used to process the current frame, identify and select all key targets in the scene, including various types of people, dangerous equipment (such as High-Voltage Cabinet, Turbine), and safety facilities (such as Safety Helmet, Insulating Mat), and obtain their bounding box coordinates and category confidence scores.

[0032] Spatial Relationship Feature Extraction: Centered on the tracked "current person," the spatial relationship between them and each surrounding hazardous device is constructed. The system calculates three core metrics: 1) Normalized Distance Score, based on the pixel distance between the centers of the bounding boxes of both parties, normalized by the scene size; 2) Relative Direction, calculated as the angle between the main orientation of the person's skeleton (defined by torso keypoints) and the vector pointing to the device's center; 3) Area Overlap (IOU), determining whether the person's bounding box intersects with the preset electronic fence area of ​​the device. For each pair (person, device), these three values ​​are combined into a sub-vector. After summing the sub-vectors of all devices, the spatial relationship feature vector of the person is formed.

[0033] Behavioral (Temporal Dynamic) Feature Extraction: The system caches the current frame and the previous 14 frames (15 frames in total, approximately 0.5 seconds of data) as an analysis segment. For each frame in this segment, the OpenPose pose estimation algorithm is used to extract 25 skeletal keypoints of the person. Based on this keypoint sequence, a dynamic descriptor is calculated, including: the average movement velocity of the left and right wrist keypoints (representing the rate of hand movement), and the rate of change of the torso bending angle between adjacent frames (representing abrupt changes in body posture). These calculated values ​​constitute a temporal dynamic feature vector describing short-term movement patterns.

[0034] Global scene feature extraction: At the same time, a high-dimensional feature vector is extracted from the end output of the YOLOv5 model backbone network. This vector encodes the overall scene semantic information of the current frame (such as scene type, lighting, and global layout) and serves as a global scene descriptor.

[0035] S202. Based on the real-time location and movement trajectory data of the personnel, and combined with the spatial relationship features, generate the spatiotemporal interaction trajectory features of the personnel.

[0036] Define a sliding time window of length T seconds (e.g., T=2), and use the current time as the cutoff point to query and extract the continuous location sequence of the target personnel within the window from the personnel positioning system database. Each of them Include information.

[0037] Motion index calculation: Based on sequence P, the system calculates the individual's core kinematic indicators within the time window: Average moving speed: Calculated by dividing the displacement difference between all adjacent locations by the average of the time intervals.

[0038] Direction of movement: Calculate from arrive The displacement vector is obtained and normalized to obtain the main direction of motion.

[0039] Acceleration: Calculated by measuring the rate of change of velocity over consecutive time intervals.

[0040] Interactive dynamic analysis: From the spatial relationship features generated in step S201, identify one or more hazardous devices (such as device D) that are currently most closely related to the person in space (e.g., have the highest distance score). Obtain the preset coordinates or real-time bounding box center coordinates loc of device D. D Subsequently, the location sequence P of each point p is calculated. i to loc D Euclidean distance d i , forming a distance sequence {d t-T ,...,d t}

[0041] Analyze the distance series and calculate its linear trend within the time window (the slope obtained through linear fitting; a negative slope indicates continued approach, and a positive slope indicates moving away) or calculate the average rate of change of distance (e.g., (d)). t -d t-T ) / T).

[0042] Feature Combination Generation: Finally, the calculated movement indicators (such as speed and acceleration) are concatenated with the distance change trend (or rate of change) characterizing the risk interaction to form a new spatiotemporal interaction trajectory feature vector. For example, a feature vector could be [average speed, acceleration, distance change trend]. This feature directly quantifies the key risk behavior pattern of "whether a person is rapidly approaching a dangerous device".

[0043] S203. The behavioral features, spatial relationship features, global scene features, and spatiotemporal interaction trajectory features are fused to generate the multi-dimensional feature information.

[0044] Receive four feature vectors from the preceding steps: Behavioral characteristics (F) behavior ): Originating from S201, it is a dynamic descriptor vector calculated based on the sequence of skeletal key points, such as [average hand velocity, rate of change of torso angle].

[0045] Spatial relationship characteristics (F) spatial): Originating from S201, it is a spliced ​​vector describing the static spatial relationship between personnel and various hazardous equipment, such as [distance fraction with equipment A, orientation angle, overlap, distance fraction with equipment B, ...].

[0046] Global scene features (F global ): Originating from S201, it is a global scene descriptor vector extracted from the object detection backbone network.

[0047] Spatiotemporal interaction trajectory features (F trajectory ): Derived from S202, it is a feature vector that integrates personnel movement indicators and equipment proximity trends, such as [average speed, acceleration, and distance change trend to the nearest hazardous equipment].

[0048] Feature standardization: First, Z-score standardization is performed on the four feature vectors to eliminate the influence of differences in feature dimensions and numerical ranges on the model.

[0049] Feature fusion: Subsequently, a vector concatenation method is used for fusion. The four standardized feature vectors are concatenated end-to-end in a predetermined order to form a unified multi-dimensional feature information vector F. final :

[0050] Here, Concat represents vector concatenation, and Norm represents normalization.

[0051] Output: The final generated F final It is a dense vector with a fixed dimension that simultaneously encodes a person's micro-actions, macro-spatial relationships, overall scene context, and dynamic interaction intentions.

[0052] In one embodiment of the present invention, based on step S3, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0053] The core model used in this method is a classification model pre-trained based on the random forest algorithm. During the training phase, this model uses manually labeled and augmented historical behavioral feature data to learn how to map multi-dimensional features to four predetermined risk levels: no risk, low risk, medium risk, and high risk.

[0054] During the real-time operation phase, the system will generate the standardized multi-dimensional feature information vector F in step S203. final The input is fed into this random forest model. This model consists of N (e.g., N=100) decision trees with distinct structures. Each decision tree is based on F... final Different feature subsets are used to perform independent reasoning and output a preliminary classification result of the current person's behavior, that is, to determine it as one of the four risk levels mentioned above.

[0055] Subsequently, the system performs collective voting statistics. It collects the outputs of all N decision trees, counts the number of decision trees classified as "no risk", "low risk", "medium risk" and "high risk", and calculates the proportion of each level (number of votes / N).

[0056] Finally, the system adjudicates the statistical results based on preset threshold rules, determining a unified multi-level risk assessment result. An example of the threshold rules is as follows: If the proportion of high-risk votes is ≥75%, it will be ultimately determined as high-risk.

[0057] Otherwise, if the proportion of medium-risk votes is ≥50%, it will be ultimately determined as medium-risk.

[0058] Otherwise, if the proportion of low-risk votes is ≥25%, it will be ultimately determined as low-risk.

[0059] Otherwise, it is determined to be risk-free.

[0060] This decision-making mechanism based on voting ratios and threshold rules ensures that the risk assessment results are both robust and interpretable, effectively avoiding overfitting or misjudgment that may occur with a single tree, thereby achieving accurate and reliable classification of personnel behavior risks.

[0061] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0062] S401. Low Risk (First Early Warning Response) When the risk assessment is deemed low (e.g., not wearing a safety helmet properly, or speeding in an unrestricted area), the system sends a command to the intelligent audible and visual alarm in the area where the person in question is located via the industrial control network. The alarm activates a first-level warning, characterized by a slowly flashing yellow warning light and a pre-recorded, gentle voice prompt (e.g., "Please wear protective equipment properly"). Simultaneously, the violation (including time, location, person ID, description of the risky behavior, and assessment result) is recorded in the safety log database for subsequent statistical analysis, but is not pushed to management personnel to avoid interference.

[0063] S402. Medium Risk (Second Early Warning Response) When the assessment result is medium risk (e.g., entering an area with important equipment but failing to maintain a safe distance, or using a mobile phone in a critical area), the system triggers a Level 2 response. In addition to activating an escalating alarm with red lights and an urgent sound on-site, the core operation is to generate a structured warning message, which is pushed in real-time via the company intranet or dedicated wireless network to the mobile terminals or workstations of pre-defined personnel in the first category of responsibility. These personnel are typically the team leader or on-duty safety officer of the violator's work group. The warning message is sent via a dedicated app or SMS and includes: the person's name / ID, precise location (linked to a factory map), a description of the risky behavior, a real-time video screenshot, and suggested corrective measures, enabling management to remotely communicate or immediately respond on-site.

[0064] S403. High Risk (Third Warning Response) When the assessment indicates a high risk (e.g., intrusion into a high-voltage energized compartment, unauthorized hot work in a flammable or explosive area, or body parts entering the hazardous area of ​​rotating machinery), the system immediately triggers the highest-level integrated emergency response. This response requires the parallel execution of at least two core operations, typically including: Information reporting and alarms: The highest priority alarms are sent synchronously to the terminals of the second category of responsible personnel (including the central control room of the plant, the head of the safety department, and the emergency command center). The information includes emergency handling guidelines and continuously updates the on-site video stream.

[0065] Emergency Broadcast: Automatically activates the factory's public broadcasting system or directional broadcasting in the affected area, playing preset emergency evacuation or warning messages.

[0066] Linkage control: Through the safety interface of industrial control systems (such as DCS and SCADA), linkage control commands associated with hazardous areas or equipment are automatically executed. For example, a "forced lockout" command is sent to an intruder in a live compartment, an "emergency stop" command is sent to dangerous rotating machinery, or an "emergency lockout" command is sent to the access control system of a relevant area to physically block the danger.

[0067] In some embodiments, the real-time identification system for power plant violations may include multiple functional modules composed of computer program segments. The computer programs for each program segment in the real-time identification system for power plant violations may be stored in the memory of a computer device and executed by at least one processor to perform (see details). Figure 1 (Description) Function for real-time identification of power plant violations.

[0068] In this embodiment, the real-time identification system for power plant violations can be divided into multiple functional modules based on its functions, such as... Figure 2As shown. The module referred to in this invention is a series of computer program segments that can be executed by at least one processor and perform a fixed function, and is stored in memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0069] The data acquisition module is used to collect multi-source monitoring data of the power plant's operating area in real time. The multi-source monitoring data includes video data, personnel location data, and environmental status data. The feature extraction module is used to extract features from the multi-source monitoring data and fuse the extracted features into multi-dimensional feature information for characterizing the state of human behavior. The risk identification module is used to input the multi-dimensional feature information into a pre-trained risk level assessment model to obtain a multi-level risk assessment result corresponding to the current personnel behavior; the multi-level risk assessment result includes at least a high-risk level and a low-risk level. The early warning response module is used to initiate a graded early warning response that matches the risk level based on the results of the multi-level risk assessment.

[0070] Figure 3 The real-time identification method for power plant violations provided in this application embodiment can be applied to equipment. Those skilled in the art will understand that the equipment structure involved in the embodiments of this invention does not constitute a limitation on the equipment. The equipment may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the equipment includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The equipment may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0071] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will understand that the server structure shown in the figure does not constitute a limitation of the present invention. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0072] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 is able to perform some or all of the steps in the above method embodiments.

[0073] The processor 310 serves as the control center of the storage device, connecting various parts of the electronic device via various interfaces and lines. It executes software programs and / or modules stored in the memory 320, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 310 may consist only of a central processing unit (CPU). In this embodiment of the invention, the CPU may have a single processing core or include multiple processing cores.

[0074] The communication unit 330 is used to establish a communication channel, enabling the storage device to communicate with other devices. It can receive user data sent by other devices or send user data to other devices.

[0075] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program, which, when executed, may include some or all of the steps provided in the embodiments of the present invention. The storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0076] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other medium capable of storing program code. It includes several instructions to cause a computer device (which may be a personal computer, a server, or a second device, network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0077] The same or similar parts between the various embodiments in this specification can be referred to mutually. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0078] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or modules may be electrical, mechanical, or other forms.

[0079] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0081] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. A method for real-time identification of violations in power plants, characterized in that, include: Real-time acquisition of multi-source monitoring data from the power plant's operating area, including video data, personnel location data, and environmental status data; Features are extracted from the multi-source monitoring data, and the extracted features are fused into multi-dimensional feature information to characterize the state of human behavior. The multi-dimensional feature information is input into a pre-trained risk level assessment model to obtain a multi-level risk assessment result corresponding to the current personnel behavior; the multi-level risk assessment result includes at least a high-risk level and a low-risk level. Based on the results of the multi-level risk assessment, a graded early warning response matching the risk level is initiated.

2. The method according to claim 1, characterized in that, Real-time acquisition of multi-source monitoring data from the power plant's operating area, including: High-definition cameras deployed in the work area are used to collect video data containing personnel behavior. The precise real-time location and movement trajectory data of the personnel are collected by using positioning base stations deployed in the work area and positioning tags carried by the personnel. Data on the operating status of equipment is collected by status sensors installed on hazardous equipment. Environmental status data is collected by deploying environmental sensors in the work area.

3. The method according to claim 1, characterized in that, Features are extracted from the multi-source monitoring data, and the extracted features are fused into multi-dimensional feature information to characterize the state of human behavior, including: Extract behavioral features, spatial relationship features, and global scene features related to the current person from the video data; Based on the real-time location and movement trajectory data of the personnel, combined with the spatial relationship features, the spatiotemporal interaction trajectory features of the personnel are generated. The behavioral features, spatial relationship features, global scene features, and spatiotemporal interaction trajectory features are fused to generate the multi-dimensional feature information.

4. The method according to claim 3, characterized in that, From the video data, behavioral features, spatial relationship features, and global scene features related to the current person are extracted, including: Target detection is performed on the current frame of the video data to identify and locate multiple targets, including at least the current personnel and one or more dangerous devices. Extract the skeletal key point sequence of the current person; Based on the location information of the current personnel and the one or more hazardous devices, calculate one or more measures that characterize the spatial relationship between them; The dynamic descriptor determined based on the skeletal keypoint sequence, the one or more metrics, and the global scene descriptor extracted from the current frame are fused to form the behavioral features, spatial relationship features, and global scene features. The behavioral features are mainly generated by dynamic descriptors determined based on the skeletal keypoint sequence. The dynamic descriptors include at least one of motion speed and posture change rate calculated based on skeletal keypoints within a temporal segment. The spatial relationship features are mainly generated by one or more of the metrics, which include at least one of the following: normalized distance between the current personnel and the hazardous equipment, relative direction, and area overlap. The global scene features are mainly generated from the global scene descriptor extracted from the current frame.

5. The method according to claim 3, characterized in that, Based on the real-time location and movement trajectory data of the personnel, and combined with the spatial relationship features, spatiotemporal interaction trajectory features of the personnel are generated, including: Obtain the continuous location sequence of the personnel within a time window of a predetermined duration, with the current time as the cutoff point; Based on the continuous position sequence, calculate at least one of the following movement indicators of the person within the time window: movement speed, movement direction, and acceleration. Based on the location of at least one hazardous device associated with the continuous location sequence and the spatial relationship features, calculate the rate of change or proximity trend of the distance between the personnel and the hazardous device; The spatiotemporal interaction trajectory feature is generated by combining the at least one movement indicator with the rate of change or the approximate trend.

6. The method according to claim 1, characterized in that, The pre-trained risk level assessment model is a classification model based on the random forest algorithm; Obtain multi-level risk assessment results, including: The multi-dimensional feature information is input into the random forest classification model, and the multiple decision trees in the model output the classification result corresponding to the current behavior; The number or proportion of trees belonging to each predetermined risk level in the classification results output by the multiple decision trees is statistically analyzed. Based on the preset threshold rules and the statistical results, the final multi-level risk assessment result is determined; wherein the predetermined risk level includes no risk, low risk, medium risk and high risk.

7. The method according to claim 1, characterized in that, Based on the results of the multi-level risk assessment, a tiered early warning response matching the risk level is initiated, including: When the multi-level risk assessment result is low risk, a first early warning response is triggered, which includes activating the on-site audible and visual alarm equipment to issue a first-level warning. When the multi-level risk assessment result is medium risk, a second early warning response is triggered. The second early warning response includes sending early warning information to the terminal device of a preset first-category responsible person. When the multi-level risk assessment result is high risk, a third early warning response is triggered. The third early warning response includes performing at least two of the following operations simultaneously: sending alarm information to the terminal devices of the preset second-category responsible personnel, activating the plant area emergency broadcast, and automatically executing linkage control commands associated with the hazardous area or equipment.

8. A real-time identification system for violations in power plants, characterized in that, include: The data acquisition module is used to collect multi-source monitoring data of the power plant's operating area in real time. The multi-source monitoring data includes video data, personnel location data, and environmental status data. The feature extraction module is used to extract features from the multi-source monitoring data and fuse the extracted features into multi-dimensional feature information for characterizing the state of human behavior. The risk identification module is used to input the multi-dimensional feature information into a pre-trained risk level assessment model to obtain a multi-level risk assessment result corresponding to the current personnel behavior; The results of the multi-level risk assessment include at least a high-risk level and a low-risk level; The early warning response module is used to initiate a graded early warning response that matches the risk level based on the results of the multi-level risk assessment.

9. A real-time identification device for violations in power plants, characterized in that, include: Memory, used to store the real-time identification program for power plant violations; A processor is configured to implement the steps of the real-time identification method for power plant violations as described in any one of claims 1-7 when executing the real-time identification program for power plant violations.

10. A computer-readable storage medium storing a computer program, characterized in that, The readable storage medium stores a real-time identification program for power plant violations, which, when executed by a processor, implements the steps of the real-time identification method for power plant violations as described in any one of claims 1-7.

Images