Method and System for Personnel Safety Early Warning in Thermal Power Plants Based on Image Recognition

Abstract

The present invention provides a method and system for personnel safety early warning in a thermal power plant based on image recognition. The present invention calculates the safety score of the staff through the personnel scoring module based on the basic information, historical skill training data, number of violations, and number of accidents of the staff. An image danger recognition model is established through the image recognition construction module and trained. The personnel early warning module divides the behavior danger level according to the image label data recognized by the recognition model, the number of violations of the staff, working hours, and safety score. The environment early warning module divides the equipment safety level according to the number of abnormal types of each equipment by obtaining the equipment operation parameters, and divides the environment danger level by obtaining the historical environment data in each area of the power plant. A danger recognition model is established through the early warning prediction module, and dangerous accidents are predicted according to the behavior danger level, equipment safety level, and environment danger level, and early warning is carried out according to the prediction results.

Description

Technical Field

[0001] The present invention relates to the technical field, and specifically to a method and system for personnel safety warning in a thermal power plant based on image recognition. Background Art

[0002] During the operation of a thermal power plant, the safety management of personnel is a challenging task. Traditional safety management methods mainly rely on manual monitoring and written records. This method is not only inefficient but also prone to omissions of safety hazards due to human factors. Therefore, there is an urgent need for a system that can automatically identify risk factors and give real-time warnings to improve the efficiency and effectiveness of safety management.

[0003] In the prior art, the publication number CN109614946B discloses a method and system for personnel safety protection in the coal conveying system of a thermal power plant based on image intelligent recognition technology. The invention analyzes the image to be detected in the working area through a deep convolutional neural network model, and monitors and intelligently evaluates in real time whether a person falls on the coal conveying pulley of the coal conveying device. After detecting the target, it feeds back to the PLC control unit to control the emergency braking of the coal conveying belt, ensuring the personal safety of the staff. However, this invention is only applicable to the personnel safety protection of the coal conveying system in a thermal power plant. In fact, there are also personnel in the boiler system, steam turbine system and generator system in a thermal power plant. Since there are too many types of positions of the personnel who need safety warnings, various factors need to be considered for safety warnings, especially the mutual influence between different factors and dangerous accidents. At the same time, the accuracy and timeliness of warnings are also particularly important.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for personnel safety warning in a thermal power plant based on image recognition to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A method for personnel safety warning in a thermal power plant based on image recognition, the specific steps include:

[0008] Step 1: Distribute unique identification number plates to the power plant staff to be warned, correspond the identification number plates with the identities of the workers one by one, obtain the basic information, historical skill training data, number of violations, and number of accidents of each staff member, and calculate the safety score of the staff member;

[0009] Step 2: Obtain historical images containing types of staff violations, accident types, and equipment anomaly types in a thermal power plant, form an image dataset, box the types of violations, accident types, identification number plates, and equipment anomaly types in the historical images, mark the label data, establish an image hazard recognition model, use the image dataset as the input of the model, and use the label data as the output to train the model;

[0010] Step 3: Obtain real-time images as the input of the trained image hazard recognition model, output the recognized image label data through the recognition model, directly give early warnings for the accident types in the output label data, calculate the behavior hazard coefficient according to the number of staff violations, working hours, and safety scores during daily work, and divide the behavior hazard levels and formulate early warning strategies;

[0011] Step 4: By obtaining equipment operation parameters, calculate the equipment safety valuation according to the quantity of each equipment anomaly type, divide the equipment safety levels, divide the internal area of the power plant, obtain the historical environmental data in each area of the power plant, calculate the regional environmental hazard coefficient according to the environmental data, divide the environmental hazard levels according to the environmental hazard coefficient, and give early warnings;

[0012] Step 5: Obtain the historical accident types in each area, establish a hazard recognition model through Bayes' theorem, form a factor dataset according to the staff identification number plate numbers, behavior hazard levels, equipment safety levels, and environmental hazard levels in each area, predict the types of future hazardous accidents, and give early warnings according to the prediction results.

[0013] Furthermore, the basic information includes the post area, working years, and identification number plate number;

[0014] The post areas include the boiler area, steam turbine area, and power generation area;

[0015] The skill training data includes the skill training duration, number of skill trainings, and assessment scores for each training.

[0016] Furthermore, the specific calculation steps for calculating the safety score of the staff are as follows:

[0017] Calculate the staff's business skill level. The specific calculation formula is:

[0018]

[0019] Where, is the staff's business skill level, is the number of skill trainings, is the th skill training duration, is the The score of the next training assessment, 、 is a positive integer;

[0020] Calculate the safety score of the staff. The specific calculation formula is:

[0021]

[0022] where, is the safety score of the staff, is the influence coefficient of the number of violations on the safety score, is the influence coefficient of the number of accidents on the safety score, is the number of violations, is the number of accidents, .

[0023] Furthermore, the types of violation behaviors include protection violations, production violations, and maintenance violations;

[0024] The types of accidents include electric shock accidents, mechanical accidents, and scalding accidents;

[0025] The types of equipment abnormalities include equipment displacement, loose connection, pipeline leakage, and line spark;

[0026] The label data includes the types of violation behaviors, the types of accidents, the types of equipment abnormalities, and the identification number plate number.

[0027] Furthermore, the mathematical calculation formula for establishing the image danger recognition model is:

[0028] The danger recognition model is based on the CNN neural network, specifically including a convolutional layer, an activation layer, a pooling layer, and an output layer;

[0029] Convolutional layer; Extract features from the bounding box of the input image and the label data through convolution operations. The calculation formula is:

[0030] ,

[0031] where, is the input image matrix, is the convolutional kernel, is the coordinate of the output matrix, The value of the convolutional kernel in the th row and the th column;

[0032] The calculation formula of the activation layer is:

[0033]

[0034] where, Coordinates of the output matrix output by the convolutional layer;

[0035] The calculation formula of the pooling layer is:

[0036]

[0037] Wherein, is the input matrix, is the maximum pooling output;

[0038] The calculation formula of the output layer is:

[0039]

[0040] Wherein, is the output of the pooling layer, is the result of image recognition, is the weight, is the bias parameter.

[0041] Furthermore, the calculation formula of the behavior risk coefficient is:

[0042]

[0043] Wherein, is the behavior risk coefficient, is the safety score of the staff, is the working hours, is the number of violations during the daily work period;

[0044] The method for dividing the behavior risk level and formulating the warning strategy is:

[0045] When it is determined that the behavior risk level is level 1 and no warning is given;

[0046] When it is determined that the behavior risk level is level 2 and a warning is given;

[0047] When it is determined that the behavior risk level is level 3, a direct warning is given, and the work post is left;

[0048] Wherein, and are the division thresholds of different behavior risk levels respectively, .

[0049] Furthermore, the equipment operation parameters include the operation years, the designed service life, the historical failure times, and the time of failure;

[0050] The calculation formula of the equipment safety evaluation is:

[0051]

[0052] Among them, is the equipment safety valuation, is the designed service life of the equipment, is the operating life of the equipment, is the moment of the i-th failure, is the working duration, is the number of each equipment abnormal type during each working period;

[0053] The specific method for dividing the equipment safety level is as follows:

[0054] When it is judged that the equipment danger level is level 1 and no early warning is given;

[0055] When it is judged that the equipment danger level is level 2 and a warning early warning is given;

[0056] When it is judged that the equipment danger level is level 3, directly give an early warning and the equipment is taken out of the working position;

[0057] Among them, and are the division thresholds of different equipment safety levels respectively, .

[0058] Furthermore, the environmental data includes coal dust particle concentration, temperature, humidity and noise;

[0059] The calculation formula of the environmental danger coefficient is:

[0060]

[0061] Among them, is the environmental danger coefficient, is the coal dust particle concentration, is the temperature, is the humidity, is the noise, are the weights of coal dust particle concentration, temperature, humidity and noise respectively, ;

[0062] The specific method for dividing the environmental danger level is:

[0063] When it is judged that the environmental danger level is level 1 and no early warning is given;

[0064] When it is judged that the environmental danger level is level 2 and a warning early warning is given;

[0065] When it is determined that the environmental danger level is level 3, give a direct warning and the worker shall leave the post;

[0066] Among them, and are the division thresholds for different environmental danger levels respectively, .

[0067] Furthermore, the mathematical calculation formula for the factor data set is:

[0068]

[0069] Among them, is the number of the staff identification number plate, is the behavior danger level, is the equipment safety level, is the environmental danger level, , i is a positive integer;

[0070] The specific steps for establishing the danger identification model through Bayes' theorem are as follows:

[0071] According to the regional location of the dangerous events in the historical data, calculate the probability of each accident type occurring, that is, the prior probability of each accident type:

[0072]

[0073] Among them, is the prior probability of each accident type , N is the number of times the accident type occurred in the historical data, is the indicator function, is the jth accident type in the historical data, , c is the accident type in the historical data, represents an electric shock accident, represents a machinery accident, represents a scald accident, is the smoothing coefficient;

[0074] The formula for the indicator function is:

[0075] ;

[0076] Calculate the likelihood probability of the equipment feature set in each safety state:

[0077]

[0078] Among them, is the probability of observing the factor data set under the given accident type , For a given accident type The probability of observing the i-th factor data where n is the number of factor data and n is a positive integer;

[0079] Calculate the posterior probability of each accident type occurring according to Bayes' law:

[0080]

[0081] where is the accident type The posterior probability of is a normalization constant;

[0082] Establish a decision model for the occurrence of accident types:

[0083]

[0084] where is the finally identified accident type that occurred.

[0085] The present invention also provides an image recognition-based personnel safety early warning system for thermal power plants. The image recognition-based personnel safety early warning system for thermal power plants is used to execute the above-mentioned image recognition-based personnel safety early warning method for thermal power plants, including:

[0086] A personnel scoring module, which is used to distribute unique identification number plates to the power plant staff to be warned, correspond the identification number plates with the worker identities one by one, obtain the basic information, historical skill training data, number of violations, and number of accidents of each staff member, and calculate the safety score of the staff;

[0087] An image recognition construction module, which is used to obtain historical images in the thermal power plant including staff violation behavior types, accident types, and equipment abnormality types, form an image data set, and frame the violation behavior types, accident types, identification number plates, and equipment abnormality types in the historical images, mark the label data, establish an image danger recognition model, and use the image data set as the input of the model and the label data as the output to train the model;

[0088] A personnel early warning module, which is used to obtain real-time images as the input of the trained image danger recognition model, output the recognized image label data through the recognition model, directly warn the accident types in the output label data, calculate the behavior danger coefficient according to the number of violations, working hours, and safety score of the staff during the daily work period, and divide the behavior danger level and formulate an early warning strategy;

[0089] An environmental warning module, which is used to calculate the equipment safety valuation according to the number of abnormal types of each device by obtaining the equipment operation parameters, divide the equipment safety level, divide the internal area of the power plant, obtain the historical environmental data in each area of the power plant, calculate the regional environmental hazard coefficient according to the environmental data, divide the environmental hazard level, and issue a warning;

[0090] A warning prediction module, which is used to obtain the historical accident types in each area, establish a hazard identification model through Bayes' theorem, form a factor data set according to the staff identification number plate numbers, behavior hazard levels, equipment safety levels, and environmental hazard levels in each area, predict the types of future hazardous accidents, and issue a warning according to the prediction results.

[0091] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention calculates the safety score of the staff through the personnel scoring module based on the basic information of the staff, historical skill training data, number of violations, and number of accidents, establishes an image hazard identification model through the image recognition construction module and trains the recognition model. The personnel warning module divides the behavior hazard level according to the image label data identified by the recognition model, the number of violations of the staff, working hours, and safety score. The environmental warning module divides the equipment safety level according to the number of abnormal types of each device by obtaining the equipment operation parameters, divides the environmental hazard level by obtaining the historical environmental data in each area of the power plant, establishes a hazard identification model through the warning prediction module, predicts the hazardous accidents according to the behavior hazard level, equipment safety level, and environmental hazard level, and issues a warning according to the prediction results;

[0092] The present invention quantitatively scores and risk-assesses the violation behaviors of the staff, and combines the equipment operation parameters and environmental historical data to comprehensively analyze possible potential safety hazards. Through the multi-dimensional analysis and the hazard identification model created by the warning prediction module, potential accidents can be predicted more accurately, and a warning can be issued before the hazard occurs. This method not only greatly improves the accuracy and timeliness of the warning, but also effectively reduces the potential safety hazards caused by human factors and management oversights;

[0093] The present invention establishes a hazard identification model through Bayes' theorem, which can consider the mutual influence between different factors and hazardous accidents. Through this information fusion, the actual risks in the power plant can be evaluated more accurately. Brief Description of the Drawings

[0094] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0095] Figure 2 It is a schematic diagram of the overall system structure of the present invention. Detailed implementation manners

[0096] To make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with specific embodiments.

[0097] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those with ordinary skills in the field to which the present invention pertains. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left" and "right" are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0098] Embodiment:

[0099] Please refer to Figure 1 , the present invention provides a technical solution:

[0100] A method for personnel safety warning in a thermal power plant based on image recognition, the specific steps include:

[0101] Step 1: Distribute unique identification number plates to the power plant workers to be warned, correspond the identification number plates with the worker identities one by one, obtain the basic information, historical skill training data, violation times, and accident times of each worker, and calculate the safety scores of the workers.

[0102] Each worker is distributed a unique identification number plate to ensure that each person can be accurately identified. This enables the system to clearly associate data and events with specific workers during monitoring and analysis. The identification number plate is bound to the worker's identity information, basic information and historical records. This means that any monitored behavior data can be immediately associated with the comprehensive history of the person for personalized analysis and warning.

[0103] In this embodiment, the basic information includes the post area, working years, and identification number plate number;

[0104] The post area includes the boiler area, the steam turbine area, and the power generation area;

[0105] The skill training data includes the skill training duration, the number of skill trainings, and the assessment scores for each training.

[0106] The safety issues faced by staff in different areas vary. For example, areas with high temperature, high pressure, and mechanical operations are generally considered to be at higher risk. Another example is that when staff in the boiler area move to the steam turbine area, due to different equipment, the staff in the boiler area are unable to operate the equipment in the steam turbine area proficiently, which is dangerous for both personnel and equipment.

[0107] A longer working tenure of staff usually reflects rich experience, which helps to more effectively identify and avoid potential safety risks. Therefore, they have a higher ability to avoid danger. A longer training duration generally enables better mastery of safety skills and improves the ability to identify and avoid danger. Multiple training opportunities contribute to consolidating safety awareness and skill operations. A higher training assessment score indicates good mastery of safety knowledge and implementation ability, which will positively affect the scoring.

[0108] Multiple violation records may reflect weak safety awareness or inadequate implementation of safety regulations by staff in the power plant, increasing the probability of dangerous accidents. Multiple accident records may indicate that the worker has a higher risk propensity and is more likely to have dangerous accidents.

[0109] In this embodiment, the specific calculation steps for calculating the safety score of the staff are as follows:

[0110] Calculate the business skill level of the staff. The specific calculation formula is:

[0111]

[0112] Wherein, is the business skill level of the staff, is the number of skill training sessions, is the th skill training duration, is the th training assessment score, , are positive integers;

[0113] Calculate the safety score of the staff. The specific calculation formula is:

[0114]

[0115] Wherein, is the safety score of the staff, is the influence coefficient of the number of violations on the safety score, is the influence coefficient of the number of accidents on the safety score, is the number of violations, is the number of accidents, .

[0116] The level of safety score can reflect the safety awareness of the staff, and it is also the ability to respond to and avoid dangers. A high score may indicate that the worker has received sufficient training, has few records of violations and accidents, has a high safety awareness, and has a high ability to respond to and avoid dangers. On the contrary, a low score indicates weak safety awareness and an inability to effectively handle dangerous accidents. The safety score is not just a static value, but a dynamic indicator that changes over time and events. This helps to record and reflect the changing trends of the staff's safety awareness and the ability to respond to dangers during a specific period.

[0117] Step 2: Obtain historical images in the thermal power plant that contain types of staff violation behaviors, accident types, and equipment anomaly types, form an image dataset, and perform bounding box selection on the types of violation behaviors, accident types, identification number plates, and equipment anomaly types in the historical images to mark the label data. Establish an image danger recognition model, use the image dataset as the input of the model, and the label data as the output to train the model.

[0118] In this embodiment, the types of violation behaviors include protection violations, production violations, and maintenance violations;

[0119] The accident types include electric shock accidents, mechanical accidents, and scalding accidents;

[0120] The equipment anomaly types include equipment displacement, loose connection, pipeline leakage, and line spark;

[0121] The label data includes types of violation behaviors, accident types, equipment anomaly types, and identification number plate numbers.

[0122] Protection violations refer to the situation where the staff fails to correctly wear or use necessary protective equipment, or fails to comply with the rules of the protection area, such as not wearing a safety helmet, goggles, or protective gloves, and not wearing the specified protective clothing in areas with high temperature, high pressure, or electrical hazards. Production violations refer to the situation where the staff violates the procedures during actual production operations, which may directly threaten the safety of equipment and personnel, such as operating key equipment such as high-pressure boilers and turbines not according to the procedures. Maintenance violations refer to the situation where the staff fails to comply with safety procedures during equipment maintenance and repair, such as not setting warning signs during the repair process and not preventing others from accidentally touching.

[0123] Obtain historical images in the thermal power plant that contain types of staff violation behaviors, accident types, and equipment anomaly types, form an image dataset, that is, in one historical image, it contains one or more of these features of types of staff violation behaviors, accident types, and equipment anomaly types. Use the historical images as the input of the image danger recognition model, and the label data as the output to train the model, so as to identify these features in the images through real-time images.

[0124] The image hazard recognition model based on the CNN neural network is established because the CNN neural network has powerful feature extraction capabilities, efficient multi-label data classification capabilities, and robustness to adapt to complex environments when processing image data. The environment of a thermal power plant is complex, with possible dynamic light changes, background interference (such as equipment operation, smoke), and various types of dangerous behaviors. The CNN neural network can capture spatial local features through the convolutional layer, reduce background interference through the pooling layer, and gradually abstract key features to adapt to these complex environments.

[0125] For the application scenario of personnel safety warning in a thermal power plant, the CNN neural network can efficiently identify violation behaviors, accident types, and equipment abnormalities, helping to achieve automated and real-time safety monitoring, and providing an important guarantee for accident prevention and production safety.

[0126] In this embodiment, the mathematical calculation formula for establishing the image hazard recognition model is as follows:

[0127] The hazard recognition model is based on the CNN neural network and specifically includes a convolutional layer, an activation layer, a pooling layer, and an output layer;

[0128] Convolutional layer; extract features from the bounding box and label data of the input image through convolution operations, and the calculation formula is:

[0129]

[0130] Among them, is the input image matrix, is the convolution kernel, is the coordinate of the output matrix, The value of the convolution kernel in the th row and the th column;

[0131] The calculation formula of the activation layer is:

[0132]

[0133] Among them, is the coordinate of the output matrix output by the convolutional layer;

[0134] The calculation formula of the pooling layer is:

[0135]

[0136] Among them, is the input matrix, is the maximum pooling output;

[0137] The calculation formula of the output layer is:

[0138]

[0139] Among them, is the output of the pooling layer, is the result of image recognition, is the weight, is the bias parameter.

[0140] Step 3: Obtain a real-time image as the input of the trained image danger recognition model. Output the recognized image label data through the recognition model, directly give early warnings for the accident types in the output label data, calculate the behavioral danger coefficient based on the number of violations, working hours, and safety scores of the staff during daily work, and divide the behavioral danger levels and formulate early warning strategies.

[0141] The frequency of violations is an important indicator to measure the safety awareness and actual operation standardization of the staff. Those with violations often have higher potential safety hazards in operations. By counting the number of violations, it can intuitively reflect the safety operation performance of the staff within a specific period. A high number of violations can trigger a higher behavioral danger level and requires key attention to avoid further expanding the hidden dangers.

[0142] Working for a long time will lead to the accumulation of fatigue and increase the risk of personnel negligence and misoperation. Introducing the working hours as a calculation factor can dynamically reflect the impact of fatigue on safety risks.

[0143] By comprehensively considering the number of violations, working hours, and safety scores, calculating the behavioral danger coefficient is more comprehensive and scientific, avoiding the one-sidedness of single-factor evaluation. This method can dynamically reflect the safety risks of employees in different working states and improve the accuracy of risk assessment.

[0144] In this embodiment, the calculation formula of the behavioral danger coefficient is:

[0145]

[0146] Among them, is the behavioral danger coefficient, is the safety score of the staff, is the working hours, is the number of violations during daily work;

[0147] The method for dividing the behavioral danger levels and formulating early warning strategies is:

[0148] When it is determined that the behavioral danger level is level 1 and no early warning is given;

[0149] When it is determined that the behavioral danger level is level 2 and a warning early warning is given;

[0150] When it is the case, the judgment behavior danger level is level 3, directly give an early warning, and the worker leaves the post;

[0151] Among them, and are respectively the division thresholds of different behavior danger levels, .

[0152] Step 4: By obtaining the device operation parameters, calculate the device safety valuation according to the number of each device abnormal type, divide the device safety level, divide the internal area of the power plant, obtain the historical environmental data in each area of the power plant, calculate the area environmental danger coefficient according to the environmental data, divide the environmental danger level according to the environmental danger coefficient, and give an early warning.

[0153] In this embodiment, the device operation parameters include the operation years, the designed service life, the number of historical faults, and the time when the fault occurs;

[0154] The calculation formula of the device safety valuation is:

[0155]

[0156] Among them, is the device safety valuation, is the designed service life of the device, is the operation years of the device, is the time when the i-th fault occurs, is the working duration, is the number of each device abnormal type during each working period.

[0157] The device safety valuation reflects the current safe operation state of the device. The higher the device safety valuation, the safer the operation state; the lower it is, the more an early warning is needed. As the operation years of the device increase, the physical wear and aging degree of the device intensifies, the performance of key components (such as pipelines, bearings, circuits) may decline, the probability of failure increases, and the failures of aging devices may cause fires, explosions or electrical hazards, directly threatening the lives of the staff. The comparison between the designed service life and the actual operation years can predict the future risks of the device. Devices with frequent failures may have potential systematic defects or fatigue damage. If not repaired in time, more serious accidents will occur. Frequently failing devices may be unstable, increasing the uncertainty of the operation environment and posing a threat to the personnel working near the devices.

[0158] The specific method for dividing the device safety level is:

[0159] When it is the case, judge that the device danger level is level 1 and no early warning is given;

[0160] When occurs, the equipment danger level is determined to be level 2, and a warning is issued.

[0161] When occurs, the equipment danger level is determined to be level 3, and a direct warning is issued to leave the working position.

[0162] Among them, and are the classification thresholds for different equipment safety levels respectively. .

[0163] In this embodiment, the environmental data includes coal dust particle concentration, temperature, humidity, and noise.

[0164] The calculation formula for the environmental danger coefficient is:

[0165]

[0166] Among them, is the environmental danger coefficient, is the coal dust particle concentration, is the temperature, is the humidity, is the noise, are the weights of the coal dust particle concentration, temperature, humidity, and noise respectively. .

[0167] Coal dust is a common risk factor in thermal power plants. Long-term exposure to a high-concentration coal dust environment may lead to serious health problems such as respiratory diseases (such as pneumoconiosis) and lung cancer. In addition, the tiny particles of coal dust can penetrate deep into the lungs, increasing the risk of chronic respiratory diseases. When coal dust is in a high-concentration environment and encounters a fire source (such as an electrical equipment failure, heat source, etc.), an explosion may occur. A coal dust explosion can cause equipment damage and serious casualties.

[0168] The temperature inside a thermal power plant is usually high, especially in areas such as boiler rooms and unit equipment. A high-temperature environment is likely to cause heat stress reactions such as heatstroke and heat exhaustion, especially when personnel are exposed for a long time. High temperature may also lead to safety accidents such as equipment overheating and fires.

[0169] An environment with too high humidity (especially in a humid climate or a high-humidity environment near equipment) may cause electrical equipment to short-circuit and corrode, thereby triggering fires or electric shock accidents. In addition, too high humidity will also exacerbate the discomfort of personnel, increasing health problems such as heatstroke and eczema.

[0170] The noise inside a thermal power plant usually comes from equipment such as steam turbines, pumps, and fans. Prolonged exposure to a high-noise environment can easily lead to hearing damage (such as noise-induced deafness), fatigue, and stress, and may even cause work mistakes by staff, increasing the probability of accidents. Excessive noise also means there are potential safety hazards in the equipment.

[0171] By real-time monitoring of risk factors such as coal dust concentration and temperature, measures can be taken before an explosion, fire, or overheating to prevent accidents. Temperature and humidity monitoring helps identify potential overload or short-circuit risks in equipment, and timely maintenance measures can be taken to reduce the equipment failure rate and the probability of danger.

[0172] The specific method for dividing the environmental risk level is as follows:

[0173] When , it is determined that the environmental risk level is level 1, and no warning is given.

[0174] When , it is determined that the environmental risk level is level 2, and a warning is given.

[0175] When , it is determined that the environmental risk level is level 3, and a direct warning is given to evacuate from the work post.

[0176] Among them, , are the division thresholds for different environmental risk levels respectively, .

[0177] Step 5: Obtain the historical accident types in each area, establish a risk identification model through Bayes' theorem, form a factor data set based on the staff identification number plate numbers, behavioral risk levels, equipment safety levels, and environmental risk levels in each area, predict the types of future dangerous accidents, and give warnings according to the prediction results.

[0178] In this embodiment, the mathematical calculation formula for the factor data set is:

[0179]

[0180] Among them, is the staff identification number plate number, is the behavioral risk level, is the equipment safety level, is the environmental risk level, , and i is a positive integer.

[0181] Factors such as staff identification number plate numbers, behavioral risk levels, equipment safety levels, and environmental risk levels do not exist independently but are interrelated and influential. For example, poor environmental conditions can exacerbate the risk of equipment failure, which in turn requires more manual intervention by workers, increasing the likelihood of accidents caused by human error. At the same time, the intrusion of people outside the post area into the post area will increase the danger of accidents. The accumulation of multiple factors may lead to a higher overall risk level. For example, although an employee has a high score for safe behavior, if he works near a constantly malfunctioning device and the surrounding environment is also poor, the overall risk is still high.

[0182] By quantitatively evaluating these factors, high-risk areas and operations can be identified in advance, specific safety countermeasures can be formulated, and early warnings can be given to personnel to reduce the occurrence of accidents.

[0183] The Bayesian model is based on probability theory and can effectively handle uncertainty and incomplete data, which is the core ability for prediction in complex environments. The prediction of accident occurrence is based on known probabilities rather than deterministic events, making the prediction more flexible and adaptable.

[0184] Bayes' theorem can well combine historical accident data as prior knowledge and update it based on new observed data. This dynamic update ability helps to maintain the accuracy and relevance of the prediction when the environment changes.

[0185] In a thermal power plant, the occurrence of accidents is often affected by various factors, including personnel behavior, equipment status, environmental conditions, etc. The Bayesian model can integrate these different types of variables and provide a comprehensive risk assessment. By comprehensively considering various influencing factors and their interrelationships, the Bayesian model can reduce the false alarm rate and missed alarm rate and improve the reliability of safety warnings. It effectively improves the accuracy of prediction, reduces the likelihood of accidents, and ensures the safety of personnel and equipment.

[0186] The specific steps to establish a hazard identification model through Bayes' theorem are as follows:

[0187] According to the regional location of hazard events in historical data, calculate the probability of occurrence of each accident type, that is, the prior probability of each accident type:

[0188]

[0189] Among them, is the prior probability of each accident type N is the number of times the accident type occurs in historical data, is the indicator function, is the jth accident type in historical data, , is the accident type in historical data, represents an electric shock accident, represents a mechanical accident, represents a scald accident, is the smoothing coefficient;

[0190] The formula of the said indicator function is:

[0191] ;

[0192] Calculate the likelihood probability of the device feature set in each safety state:

[0193]

[0194] where, is the probability of observing the factor data set x under the given accident type , is the given accident type under which the i-th factor data is observed is the number of factor data, is a positive integer;

[0195] Calculate the posterior probability of the occurrence of each accident type according to Bayes' law:

[0196]

[0197] where, is the posterior probability of the accident type , is a normalization constant;

[0198] Establish a decision model for the occurrence of accident types:

[0199]

[0200] where, is the finally identified accident type that occurred.

[0201] The present invention also provides a personnel safety early warning system for a thermal power plant based on image recognition. The personnel safety early warning system for a thermal power plant based on image recognition is used to execute the above-mentioned personnel safety early warning method for a thermal power plant based on image recognition, including:

[0202] Personnel scoring module, which is used to distribute unique identification number plates to the staff of the power plant to be warned, correspond the identification number plates with the identities of the workers one by one, obtain the basic information, historical skill training data, number of violations, and number of accidents of each staff member, and calculate the safety score of the staff;

[0203] Image recognition construction module, which is used to obtain historical images containing the types of staff violations, accident types, and equipment abnormality types in the thermal power plant, form an image data set, box the types of violations, accident types, identification number plates, and equipment abnormality types in the historical images, mark the label data, establish an image hazard recognition model, and use the image data set as the input of the model and the label data as the output to train the model;

[0204] Personnel warning module, which is used to obtain real-time images as the input of the trained image hazard recognition model, output the recognized image label data through the recognition model, directly warn of the accident types in the output label data, calculate the behavior hazard coefficient according to the number of violations, working hours, and safety score of the staff during each working day, and divide the behavior hazard level and formulate a warning strategy;

[0205] Environmental warning module, which is used to obtain equipment operation parameters, calculate the equipment safety valuation according to the number of each equipment abnormality type, divide the equipment safety level, divide the internal area of the power plant, obtain the historical environmental data in each area of the power plant, calculate the regional environmental hazard coefficient according to the environmental data, divide the environmental hazard level according to the environmental hazard coefficient, and issue a warning;

[0206] Warning prediction module, which is used to obtain the historical accident types in each area, establish a hazard recognition model through Bayes' theorem, form a factor data set according to the staff identification number plate numbers, behavior hazard levels, equipment safety levels, and environmental hazard levels in each area, predict the types of future hazardous accidents, and issue a warning according to the prediction results.

[0207] The above formulas are all calculated by taking the numerical values after dimensionless. The formula is a formula obtained by collecting a large amount of data for software simulation to get the closest real situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0208] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0209] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0210] As described above, the above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered within the protection scope of the present application.

Claims

1. A personnel safety early warning method for a thermal power plant based on image recognition, characterized in that: The specific steps include: Step 1: Distribute unique identification number plates to power plant workers to be warned, match the identification number plates with the workers' identities one by one, obtain each worker's basic information and historical skill training data, number of violations, number of accidents, and calculate the worker's safety score; Step 2: Obtain historical images of the thermal power plant containing types of staff violations, accident types, and equipment abnormalities to form an image dataset, select the types of violations, accident types, identification number plates, and equipment abnormalities in the historical images, mark the label data, and establish an image hazard recognition model. Use the image dataset as the input of the model and the label data as the output to train the model. Step 3: Obtain real-time images as input to the trained image hazard recognition model, output the recognized image label data through the recognition model, directly warn of the accident type in the output label data, calculate the behavior risk coefficient according to the number of violations, working hours and safety scores of the staff during the daily working period, and classify the behavior risk level and formulate a warning strategy; Step 4: By obtaining the equipment operating parameters, the equipment safety valuation is calculated according to the number of abnormal types of each equipment, and the equipment safety level is divided. The power plant is divided into regions, and the historical environmental data in each region of the power plant is obtained. The regional environmental risk coefficient is calculated based on the environmental data, and the environmental risk level is divided according to the environmental risk coefficient, and an early warning is issued; Step 5: Obtain the historical accident types in each area, establish a hazard identification model through Bayesian theorem, and form a factor data set based on the staff identification plate number, behavioral hazard level, equipment safety level, and environmental hazard level in each area to predict the type of future dangerous accidents and issue warnings based on the prediction results.

2. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The basic information includes job location, years of service, and identification plate number; The post areas include boiler area, steam turbine area, and power generation area; The skill training data includes the skill training duration, the number of skill training times, and the assessment score of each training.

3. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 2 is characterized in that: The specific calculation steps for calculating the safety score of the staff member are: Calculate the staff's business skill level. The specific calculation formula is: ， in, For the staff's business skills level, The number of skill training sessions, For the Length of skill training, For the Training assessment scores, , is a positive integer; Calculate the safety score of the staff. The specific calculation formula is: ， in, Score the safety of the staff, is the influence coefficient of the number of violations on the safety score, is the influence coefficient of the number of accidents on the safety score, is the number of violations, is the number of accidents, .

4. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The types of violations include protection violations, production violations, and maintenance violations; The types of accidents include electric shock accidents, mechanical accidents, and scalding accidents; The equipment abnormality types include equipment displacement, loose connection, pipeline leakage, and line sparks; The tag data includes violation type, incident type, equipment anomaly type, and identification plate number.

5. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The mathematical calculation formula for establishing the image hazard recognition model is: The hazard identification model is based on a CNN neural network, which specifically includes a convolutional layer, an activation layer, a pooling layer, and an output layer; Convolutional layer: extracts features from the bounding box and label data of the input image through convolution operation. The calculation formula is: ， in, is the input image matrix, is the convolution kernel, are the coordinates of the output matrix, The convolution kernel is Row and The value of the column; The activation layer is calculated as: ， in, The coordinates of the output matrix of the convolutional layer output; The calculation formula of the pooling layer is: ， in, is the input matrix, is the maximum pooling output; The calculation formula of the output layer is: ， in, is the output of the pooling layer, is the result of image recognition, is the weight, is the bias parameter.

6. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The calculation formula of the behavior risk coefficient is: ， in, is the behavioral risk factor, Score the safety of the staff, For working hours, The number of violations during each working day; The method for classifying behavior risk levels and formulating early warning strategies is as follows: when When the behavior is judged to be at level 1, no warning is issued; when When the behavior is judged to be at level 2, a warning is issued; when When the behavior is judged to be at level 3, a warning will be issued immediately and the person will be removed from work. in, , are the thresholds for different behavior risk levels. .

7. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The equipment operation parameters include operating life, design service life, number of historical failures, and time of failure; The calculation formula for the equipment safety valuation is: ， in, To provide equipment safety valuation, Design service life of the equipment. is the equipment operating life, is the time when the i-th failure occurs, For working hours, The number of each equipment abnormality type during each working period; The specific method for dividing the equipment security level is: when When the equipment is judged to be at level 1, no warning is issued; when When the equipment is judged to be at level 2, a warning is issued; when When the equipment is judged to be at level 3, an immediate warning is issued and the worker is asked to leave the work station; in, , They are the thresholds for different equipment security levels. .

8. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The environmental data include coal dust particle concentration, temperature, humidity and noise; The calculation formula of the environmental risk factor is: ， in, is the environmental risk factor, is the concentration of coal dust particles, is the temperature, For humidity, For noise, are the weights of coal dust particle concentration, temperature, humidity and noise respectively. ; The specific method for classifying environmental hazard levels is as follows: when When the environmental danger level is judged to be level 1, no warning is issued; when When the environmental danger level is judged to be level 2, a warning is issued; when When the environmental danger level is judged to be level 3, an immediate warning is issued and the worker is asked to leave the work station; in, , They are the thresholds for different environmental hazard levels. .

9. The method for early warning of personnel safety in a thermal power plant based on image recognition according to claim 1, characterized in that: The mathematical calculation formula of the factor data set is: ， in, Identify the number plate number for the staff, The behavior risk level is is the device safety level, is the environmental hazard level, , i is a positive integer; The specific steps of establishing the hazard identification model by Bayesian theorem are: According to the regional location of dangerous events in historical data, the probability of each accident type is calculated, that is, the prior probability of each accident type: ， in, For each accident type The prior probability of is the number of times the accident type occurred in historical data, is the indicative function, is the jth accident type in the historical data, , c is the accident type in historical data, Indicates an electric shock accident. Represents a mechanical accident. Represents a scalding accident. is the smoothing coefficient; The formula of the indicative function is: ； Calculate the likelihood probability of the device feature set in each safety state: ； in, For a given accident type The probability of observing the factor dataset x under For a given accident type The observed data of the i-th factor The probability of , n is the number of factor data, n is a positive integer; Calculate the posterior probability of each accident type according to Bayes' theorem: ， in, For accident type The posterior probability of is a normalizing constant; Establish a decision model for accident type occurrence: ， in, The type of accident that was finally identified.

10. A thermal power plant personnel safety warning system based on image recognition, characterized in that: The thermal power plant personnel safety warning system based on image recognition is used to execute the thermal power plant personnel safety warning method based on image recognition according to any one of claims 1 to 9, comprising: A personnel scoring module is used to distribute unique identification number plates to the power plant staff to be warned, match the identification number plates with the workers' identities one by one, obtain the basic information and historical skill training data, number of violations, number of accidents of each staff member, and calculate the safety score of the staff member; An image recognition building module, which is used to obtain historical images containing types of violations by staff, types of accidents, and types of equipment abnormalities in a thermal power plant, form an image data set, select the types of violations, types of accidents, identification number plates, and types of equipment abnormalities in the historical images, mark label data, establish an image hazard recognition model, and use the image data set as the input of the model and the label data as the output to train the model; A personnel warning module, which is used to obtain real-time images as input to a trained image hazard recognition model, output the recognized image label data through the recognition model, directly warn of the accident type in the output label data, calculate the behavior risk coefficient according to the number of violations, working hours and safety scores of the staff during the daily working period, and classify the behavior risk level and formulate a warning strategy; An environmental early warning module is used to obtain equipment operating parameters, calculate equipment safety estimates according to the number of abnormal types of each equipment, and classify equipment safety levels, divide the power plant into regions, obtain historical environmental data in each region of the power plant, calculate regional environmental risk factors according to the environmental data, classify environmental risk levels according to the environmental risk factors, and issue early warnings; The early warning prediction module is used to obtain the historical accident types in each area, establish a hazard identification model through Bayesian theorem, form a factor data set based on the staff identification plate number, behavior hazard level, equipment safety level, and environmental hazard level in each area, predict the type of future dangerous accidents, and issue an early warning based on the prediction results.

Images