Electric power operation site safety monitoring system based on AI image recognition

Abstract

The invention discloses an electric power operation site safety monitoring system based on AI image recognition, and relates to the technical field of electric power operation safety monitoring, the electric power operation site safety monitoring system comprises a data acquisition assembly, a data processing assembly, a data calculation assembly and an early warning execution assembly, and the data calculation assembly comprises a behavior recognition unit, an environment analysis unit and a safety early warning unit. A data calculation assembly of the system quantifies individual violation risks through multiple parameters, AI image recognition is combined, violation and risk conditions occurring in an electric power operation site are intelligently recognized, the problem that manual monitoring is disjointed in real-time performance and qualification can be solved, the risks are dynamically adapted by integrating environment, equipment and personnel behaviors, and the system has the advantages of being high in safety and high in reliability. And multi-dimensional factors are fused and diffusion of risk conditions is considered, so that real-time quantitative early warning of system risks is realized, objectivity and predictability of power operation safety monitoring are improved, and an excellent intelligent effect is achieved.

Description

Technical Field

[0001] This invention relates to the field of power operation safety monitoring technology, specifically to a power operation on-site safety monitoring system based on AI image recognition. Background Technology

[0002] The power industry is the cornerstone of the national economy. Its work sites, such as substations, distribution rooms, and transmission line maintenance areas, are characterized by high voltage, strong current, dense equipment, and complex environments. Any violation of regulations or abnormal environment during operation may lead to serious accidents such as electric shock, equipment explosion, and large-scale power outages, posing a significant threat to personnel safety and the stable operation of the power grid. Safety accidents in the power industry may stem from personnel violations, such as failure to wear protective equipment according to regulations and accidental contact with live equipment, or uncontrolled environmental risks, such as high temperature and humidity causing equipment insulation failure and strong winds causing falls from heights. Therefore, real-time safety monitoring of work sites is of irreplaceable importance.

[0003] Current power operation site safety monitoring systems lack intelligence. Traditional power operation site monitoring of individual violations often relies on manual inspections combined with paper records. Safety officers need to regularly patrol the work area, visually determine whether personnel are violating regulations (such as not wearing a safety helmet or staying outside the boundary), and manually record the type of violation and personnel information. However, this method may have problems such as long intervals between manual inspections, inability to capture instantaneous violations, and uncertainty in manual supervision, resulting in poor safety monitoring effects at power operation sites.

[0004] Current power operation site safety monitoring systems, when assessing individual violation risks, may only focus on whether violations exist, such as whether a safety helmet is worn or whether boundaries are crossed, while ignoring the degree of matching between the violator's qualifications and the current task, and whether the worker's qualification level matches the task difficulty. These factors directly determine the severity of the potential consequences of violations. Traditional assessment systems may not dynamically link personnel qualification data with violations, resulting in the inability of risk quantification results to distinguish between accidental violations by qualified personnel and intentional violations by unqualified personnel. This makes the assessment lack objective basis and difficult to accurately reflect the essential risk differences of individual violations.

[0005] Current power operation site safety monitoring systems assess environmental risks primarily based on whether environmental parameters exceed safety thresholds (such as humidity exceeding 80% or temperature exceeding equipment allowable values). However, they fail to consider the amplifying effect of the equipment's own health status on environmental risks. For example, after equipment insulation ages, the failure risk under the same environmental conditions may be significantly higher than that of new equipment. Traditional methods separate environmental factors from equipment status, resulting in environmental risk assessments failing to dynamically adapt to the actual health level of the equipment, thus leading to insufficient accuracy in environmental risk warnings throughout the equipment's entire life cycle. Summary of the Invention

[0006] The purpose of this invention is to provide a safety monitoring system for power operation sites based on AI image recognition, which solves the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a power operation site safety monitoring system based on AI image recognition, comprising:

[0008] Data acquisition component: used to acquire data on individual behaviors of power workers, coupled data of the power environment, and data related to safety early warnings, and input the acquired data into the data processing component;

[0009] Data processing component: Cleans the input data on individual behavior of power operations, coupled data of power environment, and data related to safety early warning, and inputs the cleaned data into the data calculation component;

[0010] Data computing components:

[0011] Based on various power operation violations data in individual power operation behavior data, and combined with the risk weight of the i-th type of violation, the detection confidence of the i-th type of violation, the qualification task matching coefficient, the duration of the i-th type of violation, and the maximum duration threshold of a single type of violation, an individual violation risk index is output to accurately locate high-risk violations from the personnel dimension.

[0012] Based on the risk weight of the j-th type of environmental factor, the deviation of the j-th type of environmental factor, the sensitivity coefficient of the work process stage, and the equipment insulation aging correction coefficient in the power environment coupling correlation data, the impact of the environment on the safety of the work site is analyzed. Combined with the individual violation risk index, the environmental coupling risk coefficient is output to reflect the dynamic risk level under the synergistic effect of environmental exceedance and personnel violation.

[0013] Based on the risk synergy amplification coefficient and risk spatiotemporal diffusion coefficient in the safety early warning data, the real-time safety early warning situation at the power site is analyzed. Combined with the individual violation risk index and the environmental coupling risk coefficient, a real-time safety early warning index is output as a comprehensive safety early warning quantitative index for the power operation site.

[0014] Early warning execution component: Used to calculate the individual violation risk index, environmental coupling risk coefficient, and real-time safety early warning index output by the input data calculation component. The early warning execution component triggers graded early warnings based on the input data and executes targeted safety management measures.

[0015] Optionally, the data computing component includes: a behavior recognition unit, an environmental analysis unit, and a security early warning unit.

[0016] Optionally, the processing flow of the behavior recognition unit is as follows:

[0017] S1. Use AI image recognition to determine the probability that a target belongs to a violation, so as to reflect the reliability of the detection results, analyze the detection confidence of the i-th type of violation, and combine the risk weight of the i-th type of violation to conduct a comprehensive risk analysis based on the risk situation of multiple individuals.

[0018] S2. By quantifying the degree of matching between the qualifications of the operators and the current task, the corresponding evaluation qualification-task matching coefficient is output.

[0019] S3. By analyzing the cumulative time from the first detection of the violation by AI to the current moment, the time-sensitivity risk of the violation is reflected, as well as the maximum duration allowed for a certain type of violation in the safety procedures. The duration of the i-th type of violation and the maximum duration threshold of a single type of violation are calculated separately. The results are presented in the form of ratio, minimum value function and summation function to finally output the individual violation risk index.

[0020] Optionally, the processing flow of the environmental analysis unit is as follows:

[0021] S1. By analyzing the deviation relationship between the measured values ​​and safety values ​​of multiple environmental factors, the deviation degree of the j-th type of environmental factor is calculated, and combined with the risk weight of the j-th type of environmental factor, the risk situation of environmental factors at the power operation site is analyzed.

[0022] S2. By analyzing the sensitivity of the current power operation process stage to the j-th type of environmental factor, the sensitivity coefficient of the operation process stage is obtained, and combined with the equipment's operating years, The insulation condition of the equipment is used to correct the amplification effect of environmental factors on the risk of equipment failure by measuring the degree of insulation aging, so as to obtain the equipment insulation aging correction coefficient.

[0023] S3. Combine the amplification factor of individual violations on environmental risks with the individual violation risk index, and use a summation function and an activation function to finally output the environmental coupling risk coefficient.

[0024] Optionally, the processing flow of the security early warning unit is as follows:

[0025] S1. By incorporating the individual violation risk index and the environmental coupling risk coefficient into this safety early warning unit, the risk quantification values ​​of personnel violations and environmental coupling are provided respectively, forming the core data source for comprehensive assessment;

[0026] S2. Calculate the risk synergy amplification coefficient by analyzing the synergistic effect between individual violation risk and environmental risk;

[0027] S3. By considering the speed at which the current risk spreads to the surrounding area over time, the physical diffusion effect of the risk is reflected. Based on the duration of the risk, the half-life of the risk diffusion, the width of the safety isolation zone, and the actual distance between the current risk point and the nearest surrounding equipment, the spatiotemporal diffusion coefficient of the risk is comprehensively analyzed, and finally, a real-time safety warning index is output.

[0028] Optionally, the triggering of tiered early warnings in the early warning execution component specifically includes:

[0029] When the real-time safety warning index is in the range of [0, 0.3), it indicates a safety level where there are no obvious violations and environmental risks are controllable.

[0030] When the real-time security warning index is in the range of [0.3, 0.5), it indicates a level of concern. At this point, even minor violations or environmental deviations require enhanced monitoring.

[0031] When the real-time safety warning index is in the range of [0.5, 0.7), it indicates a warning level, which means that there is a moderate level of violation and a relatively high environmental risk, requiring intervention.

[0032] When the real-time safety warning index is in the range of [0.7, 1.0], it indicates an emergency level. At this point, there are serious violations and extremely high environmental risks, and work must be stopped immediately.

[0033] Optionally, the security management measures for the security level and the concern level are as follows:

[0034] When the system is at a safe level, it must continuously monitor without triggering alarms and generate a security status report every hour, which will be pushed to the administrator's terminal.

[0035] When at the attention level;

[0036] Audio-visual prompts are provided, with on-site voice announcements of potential risks, and the specific violations are displayed on the screen in the work area.

[0037] To assist in correction, AI uses cameras to locate violators and sends vibration alerts to the smart safety wearable devices they are wearing;

[0038] Records are archived; the system automatically takes screenshots of violations and stores the associated personnel IDs in the security log.

[0039] Optionally, the safety management measures for the warning level and emergency level are as follows:

[0040] When the alert level is in effect;

[0041] Enhanced alarms: warnings were broadcast repeatedly through loudspeakers on site, and alarm information was simultaneously sent to the mobile phone of the person in charge on site.

[0042] In the event of a localized intervention, if the equipment operation is found to be in violation of regulations, the system will send a command to the equipment's smart lock to temporarily lock the critical operation buttons.

[0043] Video linkage allows managers to access real-time video of violations through the backend and provide remote guidance for correction.

[0044] When the emergency level is in effect;

[0045] In case of emergency shutdown, the system automatically cuts off the power to non-critical equipment in the work area, triggers a full-site audible and visual alarm, and causes red warning lights to flash.

[0046] Personnel evacuation: Voice commands are given to workers to evacuate immediately, and evacuation route navigation is sent to the smart safety wearable devices of all workers.

[0047] In an emergency response, the system automatically dials the power emergency command center and sends on-site video and personnel location data to initiate the emergency repair process.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] I. This invention utilizes AI image recognition technology to accurately identify individual violations in real time and dynamically link it to a personnel qualification database. It transforms multi-dimensional information such as whether a violation occurred, who committed the violation, the duration of the violation, and whether qualifications match into quantitative parameters. This solves the problems of poor real-time performance, disconnect from qualifications, and lack of quantification associated with traditional manual monitoring. It achieves an objective quantitative assessment of individual violation risks. Based on AI image recognition technology, it acquires personnel behavior data in real time, combining it with environmental parameters collected by sensors and equipment log data. It dynamically couples the relationship between environmental parameters, equipment status, and personnel behavior, resolving the problems of disconnect from traditional isolated sensors and equipment status, and subjective coupling relationships. This enables dynamic adaptation and accurate quantification of environmental risks.

[0050] Second, the behavior recognition unit outputs an individual violation risk index. By integrating multiple parameters such as violation risk weight, detection confidence, qualification-task matching coefficient, violation duration, maximum allowable duration, and total number of violations, a multi-dimensional collaborative individual violation risk quantification model is constructed. By quantifying the risk level of different violations, differentiated weight allocation for violations is achieved. Based on the confidence of AI image recognition, low-confidence detection results are filtered to ensure the reliability of input data. Personnel qualifications are bound to the current task, accurately distinguishing the severity of the consequences of violations by personnel with different qualifications. Thus, the individual violation risk index is upgraded from judging whether a behavior is a violation to a multi-dimensional quantitative assessment, comprehensively improving the objectivity, accuracy, and scenario adaptability of individual violation risk assessment.

[0051] Second, this invention outputs an environmental coupling risk coefficient through an environmental analysis unit. By integrating multiple parameters such as environmental factor risk weights, environmental deviation, work process stage sensitivity coefficients, equipment insulation aging coefficients, individual violation amplification coefficients, and individual violation risk indices, it constructs a risk assessment framework that couples the environment, equipment, and personnel. Based on the degree of impact of environmental factors on power safety, it achieves differentiated weighting of environmental risks. By quantifying the degree of deviation of environmental parameters from safety benchmarks, it avoids static assessments that only judge whether standards are exceeded. It can also adapt to the risk sensitivity of different work stages, dynamically adjust the environmental risk assessment benchmark, and incorporate equipment health status into environmental risk assessment by combining equipment operating years and insulation resistance values, reflecting the higher sensitivity of older equipment to the environment. This achieves dynamic coupling between personnel violations and environmental risks, upgrading the environmental coupling risk system from independent assessment of environmental parameters to multi-dimensional coupling quantification of environmental factors, equipment status, and personnel violations, significantly improving the dynamism, relevance, and scenario adaptability of environmental risk assessment.

[0052] Third, this invention outputs a real-time safety early warning index through a safety early warning unit. By integrating parameters such as individual violation risk index, environmental coupling risk coefficient, risk synergy amplification coefficient, and risk spatiotemporal diffusion coefficient, a progressive early warning model from local risk to systemic risk is constructed, realizing the integration of personnel and environmental risks. Based on the introduction of the risk spatiotemporal diffusion coefficient, the degree of diffusion of local risks to surrounding areas is quantified, enabling the prediction of risk evolution trends. This upgrades the real-time safety early warning index from the current risk value superimposed to a system-level early warning of basic risks, synergistic effects, and spatiotemporal diffusion, significantly improving the predictability, comprehensiveness, and accuracy of safety early warnings, and providing a scientific quantitative basis for pre-event prevention and graded intervention at power operation sites. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0054] Figure 2 This is a schematic diagram of the data acquisition component of the present invention;

[0055] Figure 3 This is a schematic diagram of the structure of the data computing component of the present invention;

[0056] Figure 4 This is a schematic diagram illustrating the operation of the data calculation component of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figures 1 to 4 This implementation provides an AI-based image recognition-based power operation site safety monitoring system, including:

[0059] Data acquisition component: used to acquire data on individual behaviors of power workers, coupled data of the power environment, and data related to safety early warnings, and input the acquired data into the data processing component;

[0060] It is worth noting that further explanation is needed regarding data collection;

[0061] I. Image Data Acquisition;

[0062] Data collection methods:

[0063] Deploy high-definition AI cameras and support nighttime infrared mode;

[0064] Installation location:

[0065] In critical work areas, such as transformer areas, switchgear areas, and aerial work platforms, 360° panoramic cameras can be installed to cover the details of personnel operations.

[0066] Safety boundaries, such as safety isolation zones for live equipment and entrances to work areas, can be equipped with directional cameras to monitor personnel crossing the boundaries.

[0067] High-risk operation points, such as live wiring and equipment maintenance stations, can be equipped with close-up cameras to capture details of personnel attire, such as safety helmets, insulated clothing, and tool usage.

[0068] Data type:

[0069] Personnel behavior type (e.g., climbing and touching equipment), clothing status type (e.g., wearing of safety helmets, insulated clothing and reflective vests), and location coordinates type (distance relative to the safe area).

[0070] Data collection method:

[0071] Real-time streaming can be transmitted to edge computing nodes via 5G or industrial Ethernet with a latency of ≤200ms;

[0072] II. Environmental Data Collection;

[0073] Data collection methods: Deploy multi-parameter sensors, such as temperature and humidity sensors, anemometers, light meters, and infrared thermometers;

[0074] Installation location:

[0075] Temperature and humidity sensors and infrared thermometers are installed around the equipment (such as the transformer body and cable joints);

[0076] An anemometer and a light meter are installed in the center of the work area to monitor the impact of the environment on personnel operations.

[0077] Data type:

[0078] Real-time temperature, humidity, wind speed, light intensity, and equipment surface temperature information;

[0079] Data collection method:

[0080] The sensor samples at a frequency of 1Hz and transmits the data to the data center via LoRa / Wi-Fi, where it is stored as time-series data.

[0081] III. Personnel Qualifications and Task Data Collection;

[0082] Data collection methods:

[0083] Connect to the enterprise's operational personnel management system database;

[0084] Data source:

[0085] The enterprise intranet database is synchronized in real time via API interface;

[0086] Data type:

[0087] Basic personnel information, including name, employee number, qualification type (such as high-voltage electrician certificate and high-altitude operation certificate), qualification level (primary / intermediate / advanced), and qualification validity period;

[0088] Task Information:

[0089] Current task ID, task difficulty level (1-5, with level 1 being the lowest), and required qualification type;

[0090] Data collection method:

[0091] Ten minutes before the start of the task, the system automatically matches the employee's qualifications with the current task based on their employee ID, and generates a qualification and task matching table.

[0092] IV. Equipment status data acquisition;

[0093] Data collection methods:

[0094] It can be integrated with power equipment ledger management system and insulation resistance testing system;

[0095] Data source:

[0096] Equipment ledgers are used to store equipment models, design lifespan, commissioning dates (operating years), and historical fault records;

[0097] Insulation resistance test report, which contains the insulation resistance values ​​of the equipment collected periodically using an insulation resistance tester;

[0098] Data type:

[0099] Equipment operating years, design life, current insulation resistance value, and standard insulation resistance value, etc.;

[0100] Data collection method:

[0101] One hour before the operation, the system automatically synchronizes the basic equipment data from the ledger system and the latest insulation resistance value from the testing system.

[0102] Data processing component: Cleans the input data on individual behavior of power operations, coupled data of power environment, and data related to safety early warning, and inputs the cleaned data into the data calculation component;

[0103] Data computing components:

[0104] The data computing component includes: a behavior recognition unit, an environmental analysis unit, and a security early warning unit;

[0105] Based on various power operation violations data in individual power operation behavior data, and combined with the risk weight of the i-th type of violation, the detection confidence of the i-th type of violation, the qualification task matching coefficient, the duration of the i-th type of violation, and the maximum duration threshold of a single type of violation, an individual violation risk index is output to accurately locate high-risk violations from the personnel dimension.

[0106] Based on the risk weight of the j-th type of environmental factor, the deviation of the j-th type of environmental factor, the sensitivity coefficient of the work process stage, and the equipment insulation aging correction coefficient in the power environment coupling correlation data, the impact of the environment on the safety of the work site is analyzed. Combined with the individual violation risk index, the environmental coupling risk coefficient is output to reflect the dynamic risk level under the synergistic effect of environmental exceedance and personnel violation.

[0107] Based on the risk synergy amplification coefficient and risk spatiotemporal diffusion coefficient in the safety early warning data, the real-time safety early warning situation at the power site is analyzed. Combined with the individual violation risk index and the environmental coupling risk coefficient, a real-time safety early warning index is output as a comprehensive safety early warning quantitative index for the power operation site.

[0108] Early warning execution component: Used to calculate the individual violation risk index, environmental coupling risk coefficient, and real-time safety early warning index output by the input data calculation component. The early warning execution component triggers graded early warnings based on the input data and executes targeted safety management measures.

[0109] When the real-time safety warning index is in the range of [0, 0.3), it indicates a safety level where there are no obvious violations and environmental risks are controllable.

[0110] When at the security level;

[0111] The system must continuously monitor without triggering alarms and generate a security status report every hour, which will be pushed to the administrator's terminal.

[0112] When the real-time security warning index is in the range of [0.3, 0.5), it indicates a level of concern. At this point, even minor violations or environmental deviations require enhanced monitoring.

[0113] When at the attention level;

[0114] Audio-visual prompts are provided, with on-site voice announcements of potential risks, and the specific violations are displayed on the screen in the work area.

[0115] To assist in correction, AI uses cameras to locate violators and sends vibration alerts to the smart safety wearable devices they are wearing;

[0116] Records are archived; the system automatically takes screenshots of violations and stores the associated user IDs in the security log.

[0117] When the real-time safety warning index is in the range of [0.5, 0.7), it indicates a warning level, which means that there is a moderate level of violation and a relatively high environmental risk, requiring intervention.

[0118] When the alert level is in effect;

[0119] Enhanced alarms: warnings were broadcast repeatedly through loudspeakers on site, and alarm information was simultaneously sent to the mobile phone of the person in charge on site.

[0120] In the event of a localized intervention, if the equipment operation is found to be in violation of regulations, the system will send a command to the equipment's smart lock to temporarily lock the critical operation buttons.

[0121] Video linkage allows managers to access real-time video of violations through the backend and provide remote guidance for correction.

[0122] When the real-time safety warning index is in the range of [0.7, 1.0], it indicates an emergency level, at which point serious violations and environmental risks are extremely high, and work must be stopped immediately.

[0123] When the emergency level is in effect;

[0124] In case of emergency shutdown, the system automatically cuts off the power to non-critical equipment in the work area, triggers a full-site audible and visual alarm, and causes red warning lights to flash.

[0125] Personnel evacuation: Voice commands are given to workers to evacuate immediately, and evacuation route navigation is sent to the smart safety wearable devices of all workers.

[0126] In an emergency response, the system automatically dials the power emergency command center and sends on-site video and personnel location data to initiate the emergency repair process.

[0127] Based on the above, this system constructs a comprehensive assessment system from local behavioral risks to system security status through a progressive logical combination of behavior recognition units, environmental analysis units, and security early warning units. The overall effect is reflected in three levels:

[0128] (1) A systematic closed loop for risk assessment;

[0129] The behavior recognition unit focuses on individual violations, the environmental analysis unit integrates the coupled risks of individual violations, environmental factors, and equipment status, and the safety early warning unit integrates individual, environmental, and diffuse system risks to form a closed loop of local, interactive, and systemic risk assessment. This avoids the defects of fragmented assessment and neglect of risk correlation in traditional monitoring and achieves a holistic portrayal of the safety status of power operation sites.

[0130] (2) The dynamic deepening of risk perception;

[0131] All three units in the data computing component incorporate dynamic parameters, such as the duration of the behavior recognition unit, the operation stage of the environmental analysis unit, and the risk duration of the safety warning unit. This enables risk assessment to be dynamically updated over time, operation stage, and equipment status, rather than a static snapshot assessment. This better reflects the dynamic characteristics of the continuous changes in personnel behavior, environmental parameters, and equipment status at power operation sites, and improves the real-time nature and depth of risk perception.

[0132] (3) Precise support for security decision-making;

[0133] The real-time safety warning index output by the safety warning unit directly corresponds to the three-level response standards of safety, warning, and emergency by limiting the quantitative value within a certain range. This provides clear decision-making basis for on-site safety management personnel, solves the problems of ambiguous risk levels and unclear response measures in traditional warnings, and improves the accuracy and efficiency of safety decision-making.

[0134] Detection confidence level IVB for type i violation i It relies on AI image recognition models to infer real-time from images of the work site, outputting the detection confidence of violations such as not wearing a safety helmet and staying in dangerous areas, providing a basis for the quantification of risk weights;

[0135] Duration of Type i violation (IVD) i By using AI target tracking algorithms to correlate consecutive frames of violating targets (such as people not wearing safety helmets), the cumulative time from the first detection to the present is calculated, thereby achieving dynamic duration monitoring of violations.

[0136] AI image recognition provides the location coordinates of violators, providing spatial data for calculating the distance between the risk point and surrounding equipment in the risk spatiotemporal diffusion coefficient of the subsequent safety early warning unit;

[0137] AI image recognition can be used to locate the relative positions of people and equipment in real time (such as whether people have entered the equipment's safety isolation zone), providing spatial correlation evidence for quantifying the environmental risk aggravated by personnel violations in the environmental coupling risk coefficient ED.

[0138] AI image recognition continuously tracks the location and duration of risk points (such as overheated equipment and unauthorized personnel), providing real-time temporal and spatial data for the risk spatiotemporal diffusion coefficient of the safety early warning unit, so that risk diffusion assessment is based on actual on-site dynamics rather than preset parameters;

[0139] The output of the behavior recognition unit serves as the individual violation risk index IV, which is directly input into the environmental analysis unit through the E1×IV term. This quantifies the active amplification effect of individual violations on environmental risks. In other words, the higher the degree of violation by personnel (the larger the IV), the greater the threat coefficient of environmental factors (such as humidity and temperature) to personnel. This forms a positive feedback loop of individual violation → increased environmental risk, solving the problem of isolated calculation of personnel and environmental risks in traditional assessments.

[0140] The product of the outputs of the behavior recognition unit and the environmental analysis unit constitutes the core risk basis of the safety early warning unit, reflecting the synergistic consequences of individual violation risk and environmental coupled risk. That is, when both individual violation and environmental risk are high, the product effect of the two significantly increases the overall risk, rather than simply adding them together. This is more in line with the actual law that accidents are caused by the superposition of personnel violations and adverse environments.

[0141] Please see Figure 1 , Figure 2 , Figure 3 and Figure 4 The processing flow of the behavior recognition unit is as follows:

[0142] 1. Use AI image recognition to determine the probability that a target constitutes a violation, so as to reflect the reliability of the detection results, analyze the detection confidence of the i-th type of violation, and combine the risk weight of the i-th type of violation to conduct a comprehensive risk analysis by considering the risk situation of multiple individuals;

[0143] Second, by quantifying the degree of matching between the qualifications of the operators and the current tasks, the corresponding output results of the qualification-task matching coefficient are evaluated.

[0144] Third, by analyzing the cumulative time from the first detection of the violation by AI to the current moment, the time-sensitivity risk of the violation is reflected, as well as the maximum duration allowed for a certain type of violation in the safety procedures. The duration of the i-th type of violation and the maximum duration threshold of a single type of violation are calculated separately, and the ratio, minimum value function and summation function are used to finally output the individual violation risk index.

[0145] More specifically, the calculation process of the behavior recognition unit is as follows:

[0146] ;

[0147] in:

[0148] IV refers to the Individual Violation Risk Index, which represents the comprehensive quantitative value of an individual's violation risk and reflects the cumulative risk level of personnel's violations (type, duration, and qualification matching).

[0149] By calculating the Individual Violation Risk Index (IV), high-risk violations can be accurately identified from the personnel perspective, distinguishing between minor and serious violations. This avoids a one-size-fits-all approach to assessing all violations and provides basic risk data from the personnel perspective for subsequent environmental risk coupling, thus achieving the first closed loop from single behavior detection to risk quantification assessment.

[0150] n refers to the total number of violations detected, which is the number of different types of violations detected by AI at the current moment. For example, not wearing a helmet and staying in a dangerous area are counted as two categories. The violation type labels in the AI ​​detection results can be used to deduplicate the count. If the same person is both not wearing a helmet and not wearing a seat belt, then n=2. This parameter is used to normalize the difference in the number of violations, eliminate the evaluation bias of more minor violations and fewer serious violations, and achieve horizontal comparability in different scenarios.

[0151] i refers to the index of detected violations;

[0152] IVA i The risk weight refers to the type i violation, which is used to quantify the degree of threat to safety of different violations. The higher the weight, the greater the probability that the violation will lead to an accident. The value ranges from 0 to 1, and the sum of the weights of all violation types is 1. For example, not wearing a safety helmet IVA=0.4, not wearing insulated clothing IVA=0.3, and staying in a dangerous area IVA=0.3.

[0153] Based on power industry safety regulations and historical accident statistics, such as the percentage of head injuries caused by not wearing a safety helmet and the percentage of electric shocks caused by staying in dangerous areas in the past 5 years, the weights can be adjusted in conjunction with expert experience to ensure that high-consequence violations (such as misoperation of equipment) have a higher weight than low-consequence violations (such as not wearing a reflective vest). This parameter is introduced to distinguish the threat level of different violations, and high-consequence violations (such as staying in dangerous areas) are given a higher weight to ensure that the risk assessment is clearly prioritized.

[0154] IVB i The detection confidence level refers to the probability that the AI ​​image recognition model judges a target to be a violation, reflecting the reliability of the detection result. The value range is 0-1 (0 = completely impossible, 1 = completely certain).

[0155] To elaborate further:

[0156] A pre-labeled dataset of power operation scenarios (including violation types such as not wearing a safety helmet and entering dangerous areas) can be used to train a target detection model, with real-time images from on-site cameras as input.

[0157] Dehazing of the input image is performed for outdoor substations; illumination normalization is performed, and contrast can be enhanced using the CLAHE algorithm; target area cropping is performed to focus on the area where workers are active.

[0158] The model outputs bounding boxes and confidence scores (IVB) for detected non-compliant targets (such as the head region not matching helmet features). i For example, if the model detects that a person is not wearing a helmet, it outputs IVB. i =0.92 (meaning there is a 92% probability that it was determined that no helmet was being worn);

[0159] By using AI to detect and filter low-reliability results, the risk of misjudgment caused by false detections can be avoided, thereby improving the accuracy of the assessment.

[0160] IVC i The qualification-task matching coefficient refers to the matching coefficient between the qualifications of the personnel committing the i-th type of violation and the current task. It is used to quantify the degree of matching between the qualifications of the operators and the current task. The greater the mismatch between qualifications, the higher the risk of consequences of the violation. The data collection process can be based on the enterprise's operator qualification database (including personnel ID, qualification type, qualification level, qualification validity period, and historical task assessment level). Then, according to the power safety work regulations, the work tasks are divided into 5 levels (Level 1 = low difficulty, such as equipment inspection; Level 5 = extremely high difficulty, such as 220kV busbar live-line work).

[0161] If the personnel do not have the qualifications required for the current task (e.g., performing live-line work without a live-line work permit), then IVC i=2.0;

[0162] If the personnel qualification level is less than the task difficulty level (e.g., an intermediate qualification level performing a level 4 task), then IVC i =1.5;

[0163] If the personnel qualification level equals the task difficulty level and is within the validity period, then IVC i =1.0;

[0164] If the personnel qualification level is greater than the task difficulty level (e.g., a high-level qualification person performing a level 2 task), then IVC i =0.8;

[0165] This qualification task matching coefficient IVC i The introduction of the correlation between personnel qualifications and task matching degree means that the greater the mismatch between qualifications, such as unlicensed operation, the more significant the risk amplification, thus addressing the difference in the impact of who violates the rules on the consequences.

[0166] IVD i This refers to the duration of the i-th type of violation, the cumulative time from when the AI ​​first detected the violation to the current moment, reflecting the time-sensitive risk of the violation (the longer the duration, the higher the probability of an incident), and is measured in seconds. The specific acquisition process is as follows:

[0167] For violations detected by AI (such as people not wearing safety helmets), multi-target tracking is performed using the DeepSORT algorithm, assigning a unique ID to each target [based on appearance features (such as clothing color and body shape)] and matching motion trajectories (such as position and speed);

[0168] If k consecutive frames of the same target ID are detected as violations, then the duration IVD i = k × (1 / frame rate) (frame rate = 25fps, i.e., 0.04 seconds per frame), for example, if a person with ID=123 is detected not wearing a safety helmet for 50 consecutive frames, IVDi = 50 × 0.04 = 2 seconds;

[0169] By introducing a time dimension, the risk of continuous accumulation of violations can be quantified, avoiding the over-evaluation of momentary violations;

[0170] IVE refers to the maximum duration threshold for a single type of violation, indicating the maximum time that a certain type of violation is allowed to continue in the safety procedure. After exceeding this threshold, the risk no longer increases linearly with time (to avoid interference from extreme values). The unit is seconds, and the value range is 5-30 seconds (set according to the type of violation, for example, IVE=10 seconds for staying in a dangerous area, and IVE=30 seconds for not wearing insulated clothing).

[0171] The introduction of the maximum duration threshold (IVE) for single-type violations limits the risk ceiling of a single violation, prevents the risk value from being amplified indefinitely due to excessive duration, and ensures the stability of the assessment.

[0172] Based on the above, the core function of this behavior recognition unit is to quantify the cumulative risk of individual violations from the perspective of personnel behavior, breaking through the limitations of traditional safety monitoring that only judges whether a violation has occurred and ignores the differences in the consequences of violations. Its contribution lies in:

[0173] Risk refinement involves integrating the risk weight of violations, AI detection confidence, duration, and qualification task matching coefficient to transform isolated violations such as not wearing a safety helmet or staying in dangerous areas into comparable and cumulative risk values, thereby avoiding misjudgments of risk caused by single violation types or momentary false detections.

[0174] By linking responsibility, a qualification-task matching coefficient is introduced to bind personnel qualifications with the risk of violations, clarifying the logic that the consequences of violations by personnel with mismatched qualifications are more serious, and providing data support for subsequent accountability and personnel management;

[0175] Dynamic and real-time, through the cumulative calculation of duration, transforms the static state of violation into a dynamic risk process, reflecting the risk evolution trend of violation behavior over time, and providing a time window for early intervention;

[0176] Traditional individual violation risk assessments often rely on the type of violation and the confidence level of the test, ignoring the impact of differences in the qualifications of the violators on the consequences, which leads to the underestimation of the violation risk of unqualified individuals.

[0177] By linking the personnel qualification database (which is objective data, not subjective assessment), the risk calculation includes who is violating the rules, which significantly amplifies the violation risk of personnel with mismatched qualifications (such as unlicensed operation or low qualifications performing high-difficulty tasks), and reasonably reduces the risk of personnel with matching qualifications, thus solving the unreasonableness of traditional assessments where the consequences of the same violation are the same for different personnel.

[0178] The parameter data comes from the enterprise qualification database (including qualification type, level, validity period, etc.), avoiding the ambiguity caused by relying on subjective data such as facial features and movement characteristics, making the risk assessment results traceable and verifiable, and meeting the power industry's requirements for data rigor.

[0179] Please see Figure 1 , Figure 2 , Figure 3 and Figure 4 The processing flow of the environmental analysis unit is as follows:

[0180] 1. By analyzing the deviation relationship between the measured values ​​and safe values ​​of multiple environmental factors, the deviation degree of the j-th type of environmental factor is calculated, and combined with the risk weight of the j-th type of environmental factor, the risk situation of environmental factors at the power operation site is analyzed.

[0181] Second, by analyzing the sensitivity of the current power operation process stage to the j-th type of environmental factor, the sensitivity coefficient of the operation process stage is obtained, and this is combined with the equipment's operating years, The insulation condition of the equipment is used to correct the amplification effect of environmental factors on the risk of equipment failure by measuring the degree of insulation aging, so as to obtain the equipment insulation aging correction coefficient.

[0182] Third, the environmental risk amplification coefficient of individual violations is combined with the individual violation risk index, and the environmental coupling risk coefficient is finally output through the form of a summation function and an activation function.

[0183] More specifically, the calculation process for the environmental analysis unit is as follows:

[0184] ;

[0185] in:

[0186] ED stands for Environmental Coupling Risk Coefficient, which is the quantitative value of risk after environmental factors and personnel violations are coupled together, reflecting the dynamic risk level under the synergistic effect of environmental exceedances and personnel violations.

[0187] By calculating the environmental coupling risk coefficient ED, environmental risks can be assessed from a system perspective, avoiding focusing only on a single environmental factor, such as temperature, while ignoring the amplifying effect of human behavior. This improves the comprehensiveness of risk assessment and provides risk data on the coupling of environment and human factors for the final comprehensive early warning, thus making up for the one-sidedness of traditional monitoring that emphasizes either the environment or human factors.

[0188] m refers to the total number of environmental factors;

[0189] j refers to the index of environmental factors;

[0190] EDA j The risk weight refers to the type j environmental factor, used to quantify the degree of impact of different environmental factors on safety (e.g., humidity has a greater impact on insulation than wind speed), with a value range of 0-1 and a sum of 1 (e.g., EDA). 温度 =0.3, EDA 湿度 =0.33, EDA 风速 =0.2k=0.2, EDA 光照 =0.2);

[0191] Risk weights of environmental factors of type j (EDA) j The introduction of this feature is used to differentiate the impact weight of different environmental factors (such as humidity having a greater impact on insulation than wind speed), ensuring that key environmental factors (such as temperature) are prioritized for evaluation.

[0192] EDB jThis refers to the deviation of the j-th type of environmental factor, that is, the degree to which the measured value of the environmental factor deviates from the safety threshold. The greater the deviation, the higher the risk. The data source can be obtained from sensors deployed at the work site, such as temperature and humidity sensors, anemometers, and light meters.

[0193] The specific formula for calculating the deviation, taking temperature as an example, is as follows:

[0194] EDB j温度 =max[(T_actual_measurement-T_safety) / T_safety,0];

[0195] Where T is the measured real-time temperature of the sensor;

[0196] Where T_safety is the maximum allowable operating temperature of the equipment (e.g., T_safety = 105℃ for transformer windings).

[0197] For example, if the measured temperature T is 110℃, then EDB j温度 =(110-105) / 105≈0.048;

[0198] Deviation of type j environmental factor (EDB) j The introduction of this method is used to quantify the degree to which measured environmental values ​​deviate from safety thresholds. The greater the deviation, the higher the risk contribution, which intuitively reflects the degree of environmental exceedance.

[0199] EDC j This refers to the sensitivity coefficient of a work process stage, used to quantify the sensitivity of the current work process stage to environmental factors of type j. It dynamically adjusts the actual impact weight of environmental factors based on the current work process stage (high-risk stages are more sensitive to environmental factors). Different process stages have different tolerances for environmental risks; the actual impact of the same environmental factor will be amplified in a high-sensitivity stage and reduced in a low-sensitivity stage. The value range is 0.7-1.8 (EDC). j 0.7 = Final stage, EDC j For stage 1.0, EDC j For the 1.6 = live operation stage, EDC j (1.8 = Emergency Repair Phase)

[0200] The specific acquisition and collection process is as follows:

[0201] The power operation process is divided into 5 stages, including the preparation stage (which may be the arrangement of safety measures), the power verification stage (which may be the confirmation that the equipment is de-energized), the operation stage (which may be the core operation, such as wiring / maintenance), the testing stage (which may be the functional verification), and the closing stage (which may be the cleanup of the site). Then, as shown above, the values ​​for each stage are assigned.

[0202] The introduction of the sensitivity coefficient EDCj for different work stages dynamically adjusts the sensitivity of different work stages to environmental risks (e.g., the live-line operation stage is more sensitive to humidity), which is in line with the stage-specific risk differences in the power work process.

[0203] EDD j This refers to the equipment insulation aging correction factor, which is the effect of the degree of equipment insulation aging on the j-th type of environmental factor. It is based on the degree of equipment insulation aging and corrects the amplification effect of environmental factors on the risk of equipment failure (after insulation aging, the impact of environmental factors is more significant). The value range is ≥1.0 (1.0 = new equipment, 1.5 = moderate aging, 2.0 = severe aging). It can be obtained from the equipment ledger database, the equipment operating years and equipment design life, and the current insulation resistance value and standard insulation resistance value from the insulation resistance test report.

[0204] ;

[0205] For example, if a cable has a design life of 20 years and has been in operation for 15 years, and its current insulation resistance is 60% of the standard value (i.e., current / standard) = 0.6, then EDDj = 1 + 0.5 × (15 / 20 + 1 - 0.6) = 1.575;

[0206] The introduction of the equipment insulation aging correction factor EDDj links the equipment insulation aging state with environmental risk. The more severe the aging (such as the decrease in insulation resistance), the more significant the amplification of environmental risk, thus solving the problem of the difference in environmental risk between new and old equipment.

[0207] E1 refers to the amplification factor of individual violations on environmental risks. It is used to quantify the active amplification effect of personnel violations on environmental risks (such as the increased threat factor of humidity to personnel when not wearing insulated clothing). The value range can be 0.3-0.7 (adjusted according to the type of violation, such as E1=0.6 when not wearing insulated clothing and E1=0.3 when not wearing a safety helmet).

[0208] Through accident simulation experiments (e.g., in an environment with 80% humidity, the probability of electric shock for a person not wearing insulating clothing when contacting a live conductor is 3 times that of a person wearing insulating clothing, so E1=0.6 is set);

[0209] The introduction of the amplification factor of individual violations on environmental risks reflects the active amplification effect of personnel violations on environmental risks (such as the risk of humidity doubling when not wearing insulated clothing), realizing the coupling relationship between personnel and environmental risks;

[0210] Sigmoid is a non-linear activation function whose core function is to compress any real-valued input into the range of 0-1, preventing risk assessment results from overflowing due to excessively large input values. It also simulates the smooth transition of risk from low to high. In environmental analysis units, Sigmoid is used to transform the linear combination risk of environmental factors and personnel violations into a 0-1 environmental coupling risk coefficient ED, facilitating subsequent calculation in conjunction with the individual violation risk index I. The Sigmoid function can typically be transformed into Sigmoid(x) = 1 / (1+e^(-1 / 2)) -x );

[0211] Where e is the natural constant, approximately 2.718;

[0212] The output range is 0-1 (when x→+∞, the output is →1; when x→-∞, the output is →0; when x=0, the output is =0.5).

[0213] Based on the above, this environmental analysis unit introduces individual violation risk from the behavior identification unit to link risks in the personnel dimension and the environmental dimension, avoiding isolated assessment of environmental risks. The core function of the environmental analysis unit is to construct a coupled risk assessment model of personnel violations, environmental factors and equipment status, solving the shortcomings of traditional environmental risk assessment that considers environmental parameters in isolation and ignores the interaction between people and the environment.

[0214] By incorporating individual violation risks into environmental risk assessment through the E1×IV item, the interactive effect of personnel violations exacerbating environmental risk consequences is quantified (e.g., humidity increases the threat level to personnel when not wearing insulated clothing), thus avoiding severing the dynamic relationship between personnel and the environment.

[0215] By introducing sensitivity coefficients for different work process stages and correction coefficients for equipment insulation aging, environmental risk assessments can be dynamically adjusted according to work stages (such as the live operation stage being more sensitive to humidity) and equipment conditions (such as the reduced temperature tolerance of aging equipment), thus aligning with the stage-specific nature of power work scenarios and the life cycle characteristics of equipment.

[0216] By using the Sigmoid function, the environmental risk of linear combinations is compressed to the 0-1 range, simulating the risk threshold effect (such as the risk increasing exponentially after environmental parameters exceed the critical value). This is more in line with the physical law that small deviations in the power system may lead to major accidents. Traditional environmental risk assessment assumes that environmental factors have the same impact on all equipment and all operation stages, ignoring the dynamic differences in equipment status and operation stages.

[0217] The equipment insulation aging correction coefficient integrates data such as equipment operating years and insulation resistance test values ​​to quantify the amplification effect of insulation aging on environmental risks (e.g., the humidity risk of aging equipment is more than 50% higher than that of new equipment), solves the blind spot in the assessment of the risk difference between new and old equipment for the same environmental parameter, and makes the environmental risk assessment more in line with the actual tolerance capacity of the equipment.

[0218] The sensitivity coefficient of the operation process stage dynamically adjusts the weight of environmental factors according to the operation stage (such as live operation and power outage maintenance), so that the sensitivity of high-risk stages (such as live operation) to environmental parameters (such as humidity and temperature) is increased, and the sensitivity of low-risk stages (such as closing and evacuation) is reduced. This avoids a one-size-fits-all environmental risk assessment and is more in line with the characteristics of the phased safety requirements of power operation.

[0219] Please see Figure 1 , Figure 2 , Figure 3 and Figure 4 The processing flow of the safety early warning unit is as follows:

[0220] I. By incorporating the individual violation risk index and the environmental coupling risk coefficient into this safety early warning unit, the risk quantification values ​​of personnel violations and environmental coupling are provided respectively, forming the core data source for comprehensive assessment;

[0221] 2. By analyzing the synergistic effect between individual violation risks and environmental risks, the risk synergistic amplification coefficient can be calculated;

[0222] Third, by considering the speed at which the current risk spreads to the surrounding area over time, the physical diffusion effect of the risk is reflected. Based on the duration of the risk, the half-life of the risk diffusion, the width of the safety isolation zone, and the actual distance between the current risk point and the nearest surrounding equipment, the spatiotemporal diffusion coefficient of the risk is comprehensively analyzed, and finally, a real-time safety early warning index is output.

[0223] More specifically, the calculation process of the safety early warning unit is as follows:

[0224] ;

[0225] in:

[0226] RS stands for Real-time Safety Early Warning Index, representing a comprehensive safety early warning quantitative index for power operation sites. It integrates multi-dimensional risks such as personnel violations, environmental coupling, and risk diffusion, and outputs graded early warning results (safety / early warning / emergency) in the range of 0-2. It directly guides on-site safety response decisions (such as activating audible and visual alarms and emergency shutdowns), realizing a closed loop from local risk detection to system risk early warning. It solves the limitations of traditional monitoring that emphasizes detection over assessment and provides a practical early warning basis for power operation safety.

[0227] R1 refers to the risk synergy amplification coefficient, which is used to quantify the synergistic effect between individual violation risk and environmental risk (when both are high, the overall risk exhibits non-linear amplification), and its value ranges from 0.2 to 0.4.

[0228] This is used to quantify the nonlinear synergistic effect between personnel violations and environmental risks (the risk is amplified exponentially when both are high).

[0229] RSA refers to the risk spatiotemporal diffusion coefficient, used to quantify the rate at which a current risk (such as equipment overheating and personnel violations) spreads to the surrounding area over time, reflecting the physical diffusion effect of the risk (such as overheated equipment causing cascading failures in adjacent equipment). Its value ranges from 0 to 0.3 (with an upper limit of 0.3 to avoid excessive amplification of the comprehensive index by a single risk). The equipment location coordinates and the width D of the safety buffer zone can be obtained from the topology map of the work area. 安全 The duration T of the risk point can be obtained from the AI ​​target tracking results. 当前 The risk diffusion half-life (RSR) can be obtained from the equipment type parameter table, such as transformer RSR = 60 seconds and cable RSR = 40 seconds.

[0230] ;

[0231] Among them, T 当前 Indicates the duration of the risk (e.g., the equipment has been overheating for 30 seconds).

[0232] RSR refers to the risk diffusion half-life (determined by the type of equipment, such as cable overheating RSR = 40 seconds).

[0233] Among them, D 安全 Refers to the width of the safety isolation zone (e.g., the safety distance requirement between equipment rooms is 2 meters);

[0234] DSD refers to the actual distance between the current risk point and the nearest surrounding device (which can be obtained through AI image ranging).

[0235] For example, the duration T of the risk point (cable overheating) 当前 =40 seconds, RSR=40 seconds, D 安全 =2 meters, DSD=1.5 meters, then:

[0236] ;

[0237] The risk spatiotemporal diffusion coefficient (RSA) is used to reflect the physical diffusion effect of risk in time and space (such as the speed at which overheated equipment spreads to the surrounding area), enabling the prediction from local risk to regional risk.

[0238] Based on the above, the Individual Violation Risk Index (IV) and the Environmental Coupling Risk Coefficient (ED) serve as basic risk inputs, providing quantitative values ​​of the risk of personnel violations coupled with the environment, respectively. These constitute the core data source for comprehensive assessment. The core function of the safety early warning unit is to output system-level comprehensive safety early warning results, thereby achieving a closed loop from local risk assessment to overall safety decision-making.

[0239] This safety early warning unit achieves deep coupling between individual violation risk and environmental risk through a product term (IV×ED), and further amplifies the risk through a synergistic amplification term. When combined with the risk spatiotemporal diffusion coefficient, a comprehensive risk index covering personnel, environment, equipment, and time and space is formed, avoiding the one-sidedness of single-dimensional risk assessment;

[0240] By outputting numerical values ​​in the 0-2 range, corresponding to multi-level response standards such as safety, early warning, and emergency, it provides clear decision-making basis for on-site safety management personnel, solving the problems of ambiguous risk levels and unclear response measures in traditional early warning systems.

[0241] By introducing a risk spatiotemporal diffusion coefficient, the potential diffusion trend of risks in physical space and time (such as the risk of equipment overheating being transmitted to surrounding equipment) is quantified, enabling early warning to extend from current risks to potential risks and improving the system's ability to predict accident chains.

[0242] Traditional comprehensive risk assessments often simply superimpose individual risks with environmental risks, neglecting the diffusion effects of risks in physical space and time (such as local overheating causing cascading failures of adjacent equipment). The creative optimization of the risk spatiotemporal diffusion coefficient is reflected in:

[0243] By integrating physical parameters such as risk duration, safety isolation zone width, and equipment spacing, the speed at which risk spreads to the surrounding area over time is quantified (e.g., after a cable overheats for 30 seconds, the failure risk of adjacent equipment within 1.5 meters increases significantly). This addresses the limitations of traditional assessments that isolate local risks and ignore physical diffusion effects, making comprehensive risk assessment more consistent with the physical characteristics of spatial coupling and chain reactions in power systems.

[0244] By calculating the risk diffusion half-life and equipment spacing, the possibility of a risk evolving from a local hazard into a system failure can be predicted in advance, enabling early warning to shift from passive response to proactive prevention and giving on-site personnel time to intervene.

[0245] It is worth noting that this embodiment presents an iterative approach, using the real-time safety warning index RS to optimize and iterate the amplification factor E1 of individual violations on environmental risk in the environmental analysis unit, thereby optimizing the environmental coupling risk coefficient ED calculated by the environmental analysis unit. The specific iterative process is as follows:

[0246] ;

[0247] in:

[0248] E1 k+1 This refers to the amplification factor of individual violations on environmental risk after the (k+1)th iteration;

[0249] E1 k This refers to the amplification factor of individual violations on environmental risk after the k-th iteration;

[0250] α refers to the adjustment step size coefficient, which is set to 0.2 in this embodiment to control the magnitude of a single adjustment and avoid drastic fluctuations in the amplification factor E1 of individual violations on environmental risks;

[0251] ESE refers to the quantified value of the real-time safety warning index RS, which is an indicator of the degree of deviation between the real-time safety warning index RS calculated by the system and the actual risk state RR on site. It is used to determine whether it is necessary to adjust the amplification factor E1 of individual violations on environmental risks. The larger the deviation, the worse the adaptability of the current amplification factor E1 of individual violations on environmental risks to the actual scenario. It is necessary to iteratively optimize the amplification factor E1 of individual violations on environmental risks to reduce false alarms or missed alarms.

[0252] Sign (ESE) refers to the sign function (which is negative for Type I deviation and positive for Type II deviation, controlling the direction of increase or decrease of E1).

[0253] It needs to be said that:

[0254] When the actual risk status RR=1, it indicates that a risk event requiring intervention has occurred on site (such as electric shock to personnel, insulation breakdown of equipment, short circuit caused by misoperation, etc., triggered directly by manual inspection records or sensors, such as the operation of leakage current protection device or the detection of equipment over-temperature fault by infrared thermometer).

[0255] When the actual risk status RR=0 on site, it means that no risk event has occurred on site (the equipment is operating normally, the personnel are operating in compliance with regulations, and the operation is confirmed by the person in charge of the operation or automatically determined by the system, such as no abnormal sensor alarm for 10 consecutive minutes).

[0256] Based on the matching relationship between the real-time safety warning index (RS) warning level and the actual risk status (RR), the deviations are divided into two categories.

[0257] Type I bias (indicating false alarm: the system overestimates the risk);

[0258] Triggering scenario: The system calculates RS ≥ 0.5 (reaching the warning or emergency level), but the actual risk status RR = 0 (no risk event occurs);

[0259] The nature of the deviation: The amplification factor E1 of individual violations on environmental risk may currently be too high, leading to an over-calculation of the amplification effect of individual violation risk on environmental risk, resulting in an artificially high real-time safety warning index RS. Therefore, the calculation formula is: ESE I类 =0.5-RS;

[0260] Type II bias (indicating underreporting: the system underestimates the risk);

[0261] Triggering scenario: The real-time security warning index RS calculated by the system is less than 0.5 (in a safe level), but the actual risk status RR is 1, indicating that a risk event has occurred;

[0262] The nature of the deviation: The amplification factor E1 of individual violations on environmental risk may currently be too low, leading to an underestimation of the amplification effect of individual violation risk on environmental risk. This causes the real-time safety warning index RS to fail to reach the warning threshold. The calculation formula is: ESE II类 =0.5 + RS;

[0263] ESQ refers to the deviation threshold, which is set to 0.3 in this embodiment. When |ESE|>ESQ, it is calculated according to ESQ to avoid excessive adjustment of the amplification factor E1 for environmental risks caused by extreme deviations.

[0264] This refers to the upper limit of a single adjustment, ensuring that the amplification factor E1 of individual violations on environmental risks does not increase or decrease by more than 10% each time, thus guaranteeing system stability.

[0265] It is worth noting that an iteration termination condition also needs to be set. This embodiment uses two termination conditions to achieve iteration convergence. The iteration terminates when either of the following two conditions is met:

[0266] Condition 1: The number of iterations reaches the upper limit. In this embodiment, the maximum number of iterations is set to 50 to avoid the iteration from getting stuck in an infinite loop and to ensure the real-time performance of the system.

[0267] Condition 2: In five consecutive iterations, the change in the amplification factor E1 of individual violations on environmental risk is ≤0.01.

[0268] Based on the above, in traditional safety monitoring systems, the coupling coefficients such as the amplification factor E1 of individual violations on environmental risks rely on static settings (e.g., E1=0.6 for not wearing insulated clothing), which cannot adapt to the differences in risk characteristics of different work scenarios (e.g., high humidity environment in summer or dry environment in winter) and different skill levels of personnel (new or old employees). By iterating the amplification factor E1 of individual violations on environmental risks through real-time safety early warning index RS feedback, the coupling assessment of personnel violations and environmental risks can be dynamically aligned with the actual situation on site.

[0269] When the real-time safety warning index RS frequently reports false alarms (Type I deviation), reduce E1 to reduce the excessive amplification of environmental risks and avoid warning fatigue;

[0270] When the real-time security warning index RS misses a report (Type II bias), E1 is increased to enhance sensitivity to potential coupling risks and avoid underestimation of risks.

[0271] The amplification factor E1 of individual violations on environmental risks serves as the core link between individual violations and environmental risks. Its dynamic optimization upgrades the environmental analysis unit from a static coupling model to a dynamic learning model.

[0272] After a single iteration, the environmental coupling risk coefficient ED is more accurate in assessing the risk coupling of the current scenario (for example, when a group of experienced employees are working together, the decrease in E1 weakens the amplification effect of the individual violation risk index I on the environmental coupling risk coefficient ED, which is consistent with the actual situation that "the consequences of violations by experienced employees are relatively controllable").

[0273] The system accumulates the optimal value of the amplification factor E1 of individual violations on environmental risks in different scenarios through multiple rounds of iteration. For example, the amplification factor E1 of individual violations on environmental risks in scenarios such as live-line work in 220kV substations and maintenance of 10kV distribution rooms is adapted differently to form a scenario-based risk assessment knowledge base and improve the generalization ability of the overall system.

[0274] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI image recognition-based power operation site safety monitoring system, characterized in that, The method comprises the following steps: a data acquisition component is used to acquire power operation individual behavior data, power environment coupling correlation data and safety warning related data, and input the acquired data into a data processing component; the data processing component is used to clean the input power operation individual behavior data, power environment coupling correlation data and safety warning related data, and input the cleaned data into a data processing component; the data processing component comprises the following steps: based on the multiple power operation violation behavior data in the power operation individual behavior data, and combined with the risk weight of the i-th type of violation behavior, the detection confidence of the i-th type of violation behavior, the qualification task matching coefficient, the duration of the i-th type of violation behavior and the maximum duration threshold of single type violation, an individual violation risk index is output to accurately locate the high-risk violation behavior from the personnel dimension; based on the risk weight of the j-th type of environmental factor, the deviation degree of the j-th type of environmental factor, the operation process stage sensitive coefficient and the equipment insulation aging correction coefficient in the power environment coupling correlation data, the influence of the environment on the safety of the operation site is analyzed, and combined with the individual violation risk index, an environment coupling risk coefficient is output to reflect the dynamic risk level under the synergistic action of environmental over-standard and personnel violation; based on the risk synergy amplification coefficient and the risk space-time diffusion coefficient in the safety warning related data, the real-time safety warning situation of the power site is analyzed, and combined with the individual violation risk index and the environment coupling risk coefficient, a real-time safety warning index is output as the comprehensive safety warning quantitative index of the power operation site; the warning execution component is used to input the individual violation risk index, the environment coupling risk coefficient and the real-time safety warning index output by the data calculation component, and trigger a hierarchical warning based on the input data, and execute targeted safety management measures. 2.The AI image recognition-based electric power work site safety monitoring system according to claim 1, characterized by: The data calculation component comprises a behavior recognition unit, an environment analysis unit and a safety warning unit. 3.The AI image recognition-based electric power work site safety monitoring system according to claim 2, characterized by: The processing flow of the behavior recognition unit is as follows: S1, the probability that a target belongs to a violation behavior is judged by AI image recognition to reflect the reliability of the detection result, to analyze the detection confidence of the i-th type of violation behavior, and combined with the risk weight of the i-th type of violation behavior, the risk of multiple individuals is analyzed to output the risk analysis result; S2, the matching degree of the qualification of the operation personnel and the current task is quantified to output the corresponding evaluation qualification task matching coefficient; S3, the accumulated time from the first detection of the violation behavior by AI to the current time is analyzed to reflect the timeliness risk of the violation behavior, and the longest time allowed for a certain type of violation behavior in the safety regulations is analyzed to calculate the duration of the i-th type of violation behavior and the maximum duration threshold of single type violation, and the ratio, minimum value function and summation function are used to finally output the individual violation risk index. 4.The AI image recognition-based electric power work site safety monitoring system of claim 3, wherein: The processing flow of the environment analysis unit is as follows: S1, the deviation relationship between the measured values of multiple types of environmental factors and the safety values is analyzed to calculate the deviation degree of the j-th type of environmental factor, and combined with the risk weight of the j-th type of environmental factor, the risk of the environmental factors in the power operation site is analyzed; S2, obtaining the operation procedure stage sensitivity coefficient by analyzing the sensitivity of the current power operation procedure stage to the jth environmental factor, and combining the equipment operation life, and the insulation condition of the equipment to correct the amplification effect of the environmental factor on the equipment failure risk through the insulation aging degree of the equipment to obtain the equipment insulation aging correction coefficient; S3, the amplification coefficient of the individual violation to the environmental risk is combined with the individual violation risk index, and is output in the form of a summation function and an activation function to output the environmental coupling risk coefficient. 5.The AI image recognition-based electric power work site safety monitoring system according to claim 4, characterized by: The processing flow of the safety warning unit is as follows: S1, the individual violation risk index and the environmental coupling risk coefficient are introduced into the safety warning unit to provide risk quantitative values of personnel violation and environmental coupling, which constitute the core data source of comprehensive evaluation; S2, the synergistic amplification coefficient of the individual violation risk and the environmental risk is calculated by analyzing the synergistic effect of the individual violation risk and the environmental risk; S3, the speed of the current risk spreading to the surrounding area over time is considered to reflect the physical diffusion effect of the risk, and the risk space-time diffusion coefficient is analyzed based on the risk duration, the risk diffusion half-life, the safety isolation belt width, and the actual distance between the current risk point and the nearest device in the surrounding area, so as to finally output the real-time safety warning index. 6.The AI image recognition-based electric power work site safety monitoring system according to claim 1, characterized by: The trigger hierarchical warning in the warning execution component is specifically: When the real-time safety warning index is in [0, 0.3), it represents the safety level, at this time there is no obvious violation, and the environmental risk is controllable; When the real-time safety warning index is in [0.3, 0.5), it represents the attention level, at this time there is slight violation or environmental deviation, and monitoring should be strengthened; When the real-time safety warning index is in [0.5, 0.7), it represents the warning level, at this time there is moderate violation and high environmental risk, and intervention should be carried out; When the real-time safety warning index is in [0.7, 1.0], it represents the emergency level, at this time there is serious violation and extremely high environmental risk, and immediate shutdown is required. 7.The AI image recognition-based electric power work site safety monitoring system according to claim 6, characterized by: The safety management measures for the safety level and the attention level are: When in the safety level, the system continuously monitors, does not trigger the alarm, and generates a safety state report every hour, which is pushed to the management personnel terminal; When in the attention level; Acoustic and optical prompt, on-site voice broadcast risk situation, and operation area display screen displays specific violation points; Auxiliary correction, AI locates the violation personnel through the camera, and sends a vibration reminder to the smart safety wear equipment worn by the personnel; Record and archive, the system automatically screenshots the violation behavior, and stores the personnel ID into the safety log. 8.The AI image recognition-based electric power work site safety monitoring system according to claim 6, characterized by: The safety management measures for the warning level and the emergency level are: When in the warning level; Strengthen the alarm, the on-site loudspeaker cyclically broadcasts the warning, and sends the alarm information to the mobile phone of the on-site person in charge at the same time; Local intervention, if the operation violation involves the device, the system sends an instruction to the intelligent lock of the device to temporarily lock the key operation button; Video linkage, the management personnel call the real-time video of the violation point through the background to remotely guide the correction; When in the emergency level; Emergency shutdown, the system automatically cuts off the power supply of non-key devices in the operation area, triggers the sound and light alarm of the whole field, and flashes the red warning light; Personnel evacuation, voice instruction operation personnel to evacuate urgently, and send evacuation route navigation to all operation personnel smart safety wear equipment; Emergency response, automatically dial the telephone number of the power emergency command center, push the data of the on-site video and personnel position to start the emergency repair process.

Images