Intelligent power distribution room safety monitoring management system

Abstract

The invention discloses an intelligent power distribution room safety monitoring management system, which comprises the following steps: checking states of a sensor, a gateway and a camera, and collecting environment data through the sensor; preprocessing the collected data, identifying a smoke area by using a YOLOv5 model, and starting a linkage safety mechanism according to abnormal environmental parameters; performing personnel detection by using a YOLOv5s model, capturing and analyzing a behavior track through a SimAM mechanism, and realizing multi-target tracking and abnormal behavior detection; grading the abnormal events based on a rule classifier, and triggering equipment linkage control; and storing the data in an NoSQL database, analyzing a trend to predict potential dangers, and optimizing a YOLOv5 model through historical data. The system can monitor the environment and personnel safety conditions in the power distribution room in real time, has efficient response capability and a trend prediction function, and can be widely applied to the field of safety monitoring of intelligent power grids and industrial environments.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to an intelligent power distribution room safety monitoring and management system. Background Art

[0002] As the core link of the power system, the operating environment of the power distribution room directly affects the safety and reliability of power equipment. However, the current monitoring of the power distribution room mainly relies on traditional manual inspections or single parameter monitoring methods, which are inefficient and difficult to detect complex abnormal conditions in time, especially in complex environmental conditions or sudden dangerous situations. It is impossible to respond quickly, which easily leads to major safety hazards.

[0003] At present, some automated monitoring systems have introduced sensor technology to monitor parameters such as temperature, humidity, and smoke in real time, but most systems can only alarm for a single parameter and lack the ability to integrate and analyze multi-dimensional data. In addition, although traditional video surveillance technology can collect images in real time, it has limited performance in detecting and analyzing human behavior and abnormal behavior, and cannot accurately identify complex scenes, such as people not wearing safety equipment or falling to the ground. In addition, the existing system also lags in the classification of abnormal events and response linkage, resulting in the inability to quickly and effectively control dangers.

[0004] The present invention provides an intelligent distribution room safety monitoring and management system, which introduces the deep learning model YOLOv5 and multimodal sensors, and provides a new technical path for the intelligent monitoring of distribution rooms. By integrating multi-source data and combining with behavior analysis models, comprehensive monitoring and abnormal linkage control of the environment and personnel are achieved, thereby improving the safety and reliability of distribution room operation. Summary of the invention

[0005] In view of the above problems, the present invention provides an intelligent distribution room safety monitoring and management system to solve the problems that existing technicians have limited performance in behavior detection and abnormal behavior analysis, cannot accurately identify complex scenarios, and have lags in abnormal event classification and response linkage, resulting in the inability to quickly and effectively control dangers.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: The present invention provides an intelligent power distribution room safety monitoring and management system, which is characterized by comprising the following steps: Step S1, check the status of sensors, gateways, and cameras, and collect environmental data through sensors; Wherein step S1 also includes the following sub-steps: S1-1, device startup, initialization of edge computing devices, and confirmation that sensors, gateways, and cameras are in normal status; S1-2, environmental data acquisition, uses sensors to obtain environmental parameters in real time, including temperature, humidity, and gas concentration.

[0007] Step S2, pre-processing the collected data, using the YOLOv5 model to identify the smoke area, and starting the linkage safety mechanism according to the abnormal environmental parameters; Wherein step S2 also includes the following sub-steps: S2-1, environmental data preprocessing, using a sliding window to calculate the average value, filtering the noise of the collected environmental parameters, and dynamically adjusting the monitoring threshold based on historical data, as shown in formula (1): Formula (1) in, is the smoothed value, is the window size, is the original data; S2-2, use the YOLOv5 model to analyze the smoke diffusion range and determine whether the environmental data exceeds the safety threshold. The backbone network selects CSPDarknet53 to extract features. The feature fusion is based on OANet to enhance the detection accuracy. The loss function is as follows: Formula (2) in, is the basic intersection-and-union ratio, is the Euclidean distance between the center of the prediction box and the target box, b is the center point coordinate of the prediction box, is the center point coordinate of the target box is the minimum diagonal distance of the outer frame, is the balancing factor, is the difference between the length ratio of the prediction box and the target box; S2-3, linkage triggering, when abnormal environmental parameters are detected, the corresponding safety mechanism is activated, and the state triggering logic is as follows: Formula (3) in, It is the specific linkage action that is triggered. is the trigger threshold, Current environment data.

[0008] Step S3, using the YOLOv5s model to detect people, capturing and analyzing behavior trajectories through the SimAM mechanism, and realizing multi-target tracking and abnormal behavior detection; Wherein step S3 also includes the following sub-steps: S3-1, safety clothing detection, uses the improved YOLOv5s model to detect whether personnel wear safety helmets and work clothes. For those who do not wear or wear them, their behavior is recorded and a voice warning is triggered. The specific method is as follows: Input video frames of size 640*640*3, mosaic stitching, and multi-lighting HSV adjustment for data enhancement; S3-2, behavior monitoring, uses the SimAM attention mechanism to capture whether a person is close to a high-voltage area or performing a dangerous operation, dynamically analyzes the action trajectory, and determines whether it complies with the regulations, as shown in formulas (4)-(6): Formula (4) Formula (5) Formula (6) in, Pixel The energy is the total number of pixels in the feature map, is the value of the current pixel, is the average value of the feature map; is the target state, is the next target state, is the state transfer matrix, is the input matrix, Input control quantity, System noise; is the observed value, is the state-to-observation transfer matrix, Measurement noise; S3-3, Multi-target Tracking, identifies and tracks multiple targets in complex scenes, using optimized feature fusion techniques for identification.

[0009] Step S4, classifying abnormal events based on the rule classifier and triggering equipment linkage control; Wherein step S4 also includes the following sub-steps: S4-1, abnormal analysis, integrates environmental monitoring and personnel behavior data, and determines the abnormal event level through a rule-based classifier, as shown in formula (7): Formula (7) in, is the level of abnormal events, is the threshold for level classification, is the current data value; According to the severity and impact of the event, abnormal events can be divided into the following three categories: low-level abnormalities, medium-level abnormalities, and high-level abnormalities. The rule classifier integrates various monitoring data and classifies the levels in the following way: Temperature and humidity abnormalities: temperature > 30℃ and ≤ 35℃ is considered a low-level abnormality; temperature > 35℃ and ≤ 45℃ is considered a medium-level abnormality; temperature > 45℃ is considered a high-level abnormality; Smoke diffusion range, smoke area <10 , slow growth rate is considered a low-level anomaly; smoke area ≥ 10 , fast growth rate is considered as intermediate abnormality; smoke area ≥30 Considered as a high-level anomaly; Abnormal behavior of personnel, not wearing a helmet, not wearing protective clothing is considered an intermediate abnormality; entering a high-voltage area or falling to the ground is considered a high-level abnormality; In addition, different response mechanisms are triggered according to the abnormality level: low-level abnormalities alert staff through the platform or on-site voice; medium-level abnormalities trigger relevant equipment and send alarms to the monitoring center to prompt the on-duty personnel to handle; high-level abnormalities activate emergency plans, notify all relevant personnel through sound and light alarms, and send real-time alarms to the monitoring center and emergency management personnel; S4-2, linkage mechanism, realizes the linkage between equipment and system, adjusts the air conditioning operation mode, quickly switches the external power supply through the fast interface box, and uses the state machine algorithm to control the air conditioning, power supply, and ventilation system. Specifically, as shown in formula (8): , event) Formula (8) in, is the next state of the system, is the state transition function, event is the event currently triggered; Trigger conditions: when abnormal data exceeds the set threshold or abnormal behavior is detected, the linkage response process is initiated; Air conditioning operation mode adjustment: adjust the air conditioning operation parameters according to the temperature and humidity sensor data. When the temperature is higher than the set range, turn on the cooling mode and adjust the wind speed to medium; when the humidity is higher than the set range, turn on the dehumidification mode; Fast power switching: when a main power failure is detected, it quickly switches to the backup power supply. Using a fast interface box, the main and backup power supply switching can be completed within 0.1 seconds. When the main power supply voltage is lower than the safety threshold or is completely interrupted, the interface box is triggered to switch to the backup power supply, and the time and trigger reason of the power switching are recorded at the same time. Ventilation system control: when the SF6 concentration or smoke concentration is detected to be excessive, the ventilation equipment is turned on to dilute the gas. When the SF6 concentration is greater than the safety threshold and the gas concentration continues to rise, the exhaust fans in the corresponding area are turned on. When the smoke concentration detects a diffusion trend, the exhaust equipment near the smoke area is turned on first. The ventilation equipment operation mode can dynamically adjust the wind speed according to the gas concentration. S4-3, alarm and notification, triggers different levels of alarms according to the event level. The alarm information includes video streams and environmental data snapshots, which are sent to the monitoring platform in real time; Alarm classification: Low-level abnormality, only prompting local alarm; medium-level abnormality: triggering sound and light alarm and notifying the on-duty personnel; high-level abnormality: initiating a comprehensive alarm and sending emergency notifications to all relevant parties.

[0010] In step S5, the data is stored in a NoSQL database, trends are analyzed to predict potential dangers, and the YOLOv5 model is optimized through historical data.

[0011] Wherein, in step S5, the following sub-steps are also included: S5-1, event record, stores all abnormal events and operation records in the NoSQL database, and manages abnormal events by time and type, and stores them in JSON data format; S5-2, based on historical data feedback, use new abnormal data to fine-tune the YOLOv5 model and optimize the target detection model. Specific steps: Data preparation, extracting data on abnormal events from NoSQL databases; Freeze the backbone network. The backbone network of YOLOv5 contains a large number of pre-trained parameters, which are frozen to retain the existing feature extraction capabilities. Only the classification head is trained for specific tasks. Retrain the classification head and use the newly added dataset to train the output layer of the model; Loss function adjustment: adjust the weight of CIoU loss function; S5-3, regularly generate safety operation reports, including statistics on safety attire rate, distribution and handling of abnormal events; S5-4, analyze data trends, predict potential risks and adjust monitoring strategies.

[0012] Compared with the prior art, the present invention has the following beneficial effects: The present invention integrates multiple sensors, including SF6 gas sensors, acoustic cameras, temperature and humidity sensors, and combines them with the YOLOv5 model to achieve multi-dimensional monitoring and analysis of environmental parameters. It can accurately identify complex scenes and effectively improve the accuracy and response speed of abnormal event detection.

[0013] The present invention utilizes the YOLOv5s model and SimAM attention mechanism, combined with a multi-target tracking algorithm, to achieve real-time detection and behavior monitoring of personnel in the distribution room. It can identify abnormal behaviors such as not wearing a safety helmet and falling to the ground, significantly improving the intelligence level of behavior analysis and reducing the probability of safety accidents.

[0014] The present invention is based on a rule classifier, performs graded processing on abnormal events, triggers corresponding safety mechanisms, and can dynamically respond according to the danger level, thereby reducing the risk of accidents.

[0015] The present invention uses a NoSQL database to store historical data, and combines time series analysis and machine learning methods to predict potential dangers. At the same time, the YOLOv5 model is optimized through feedback data, so that the system has the ability of continuous learning and improvement, and can better adapt to complex and changing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It is understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but is only for the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0019] Please refer to Figure 1 The present invention provides a flow chart of a smart power distribution room safety monitoring and management system, which includes the following steps: Step S1, check the status of sensors, gateways, and cameras, and collect environmental data through sensors; Wherein step S1 also includes the following sub-steps: S1-1, device startup, initialization of edge computing devices, and confirmation that sensors, gateways, and cameras are in normal status; S1-2, environmental data acquisition, uses sensors to obtain environmental parameters in real time, including temperature, humidity, and gas concentration.

[0020] It should be noted that the sensors include SF6 gas sensors, acoustic cameras, temperature sensors, and humidity sensors; SF6 gas sensor status confirmation is achieved by starting the self-test function of the SF6 gas sensor to confirm that its power supply, communication interface, and data return status are normal, and initial SF6 concentration data is collected to verify whether it is within a reasonable range; acoustic camera status confirmation is achieved by checking the device status and collecting initial sound field data to verify whether it is working normally; temperature sensor and humidity sensor status confirmation is achieved by collecting initial environmental parameters and comparing them with historical data to confirm whether the data is reasonable.

[0021] Gateway status confirmation: check whether the gateway is successfully connected to edge computing devices and sensors, verify whether the data upload channel is unobstructed, and receive test data from the gateway to ensure that the data collection and return delays are within the allowable range.

[0022] Confirm the camera status, check the physical connection of the camera, ensure that the device is powered properly, capture an initial frame of the image and perform quality checks, including: image clarity, image brightness, and use a simple algorithm to confirm that the camera is functioning properly.

[0023] Step S2, pre-processing the collected data, using the YOLOv5 model to identify the smoke area, and starting the linkage safety mechanism according to the abnormal environmental parameters; Wherein step S2 also includes the following sub-steps: S2-1, environmental data preprocessing, using a sliding window to calculate the average value, filtering the noise of the collected environmental parameters, and dynamically adjusting the monitoring threshold based on historical data, as shown in formula (1): Formula (1) in, is the smoothed value, is the window size, is the original data; S2-2, use the YOLOv5 model to analyze the smoke diffusion range and determine whether the environmental data exceeds the safety threshold. The backbone network selects CSPDarknet53 to extract features. The feature fusion is based on OANet to enhance the detection accuracy. The loss function is as follows: Formula (2) in, is the basic intersection-and-union ratio, is the Euclidean distance between the center of the prediction box and the target box, b is the center point coordinate of the prediction box, is the center point coordinate of the target box is the minimum diagonal distance of the outer frame, is the balancing factor, is the difference between the length ratio of the prediction box and the target box; S2-3, linkage triggering, when abnormal environmental parameters are detected, the corresponding safety mechanism is activated, and the state triggering logic is as follows: Formula (3) in, It is the specific linkage action that is triggered. is the trigger threshold, Current environment data.

[0024] It should be noted that YOLO is an end-to-end target detection algorithm that can complete target detection and positioning in a single forward pass. YOLOv5 is an optimized version in the YOLO series. YOLOv5 can output high-quality bounding boxes. YOLOv5 has multi-target detection capabilities and comes with a data enhancement mechanism.

[0025] The process of using the YOLOv5 model to detect the range of smoke diffusion is as follows: Input the real-time video stream captured by the environmental camera; image resolution preprocessing to ensure that the input size meets the requirements of YOLOv5; feature extraction, extract the visual features of smoke, such as transparency and blurred boundaries, through CSPDarknet53; use SPP to process multi-scale features to ensure that the model adapts to different smoke sizes; detection output, output the bounding box and classification confidence of each smoke area, and use the CIoU loss function to optimize the overlap between the predicted box and the real box; diffusion analysis, combine the detected bounding box with time series analysis to determine the diffusion trend and speed of the smoke.

[0026] Abnormal conditions include: abnormal SF6 gas concentration, abnormal smoke diffusion range, abnormal temperature and humidity, and abnormal acoustic camera detection; When it is detected that the SF6 concentration exceeds the safety threshold or the concentration trend rises rapidly, the ventilation equipment in the distribution room is automatically started to accelerate the dilution of the gas and reduce the concentration. The sound and light alarm equipment is activated to remind the on-site staff to evacuate immediately or take measures. Real-time alarm information is sent to the monitoring center, including the current SF concentration, concentration change trend and leakage location. If the concentration reaches a dangerous level, the power supply is automatically cut off to prevent safety hazards.

[0027] When the YOLOv5 model detects a smoke area and the diffusion range exceeds the safety threshold or the smoke concentration increases rapidly, the early warning module of the fire protection system is automatically triggered to confirm whether further response is needed, activate sound and light alarms in the smoke area to alert nearby staff, call the camera to monitor the smoke diffusion area in real time, send the video stream to the monitoring center, start local ventilation equipment, try to reduce the smoke concentration, and send alarm information in real time to indicate the possible risk of combustion or gas leakage.

[0028] When it is detected that the temperature exceeds the set range or the humidity exceeds the standard, the air conditioning operation mode is dynamically adjusted according to the abnormal temperature and humidity, and the on-site staff are reminded to check the equipment that may be overheated or damp, record the abnormal temperature and humidity parameters, and generate an alarm report. If the humidity continues to exceed the standard, the system will send a notification to the monitoring center for further processing.

[0029] When the acoustic camera detects an abnormal sound source, such as high-frequency arc discharge or short-circuit noise, it will be called upon to locate the abnormal sound source, mark the specific location and notify the monitoring personnel. It will also call upon the camera to conduct visual monitoring of the abnormal sound source area and remind maintenance personnel to check related equipment.

[0030] Step S3, using the YOLOv5s model to detect people, capturing and analyzing behavior trajectories through the SimAM mechanism, and realizing multi-target tracking and abnormal behavior detection; Wherein step S3 also includes the following sub-steps: S3-1, safety clothing detection, uses the improved YOLOv5s model to detect whether personnel wear safety helmets and work clothes. For those who do not wear or wear them, their behavior is recorded and a voice warning is triggered. The specific method is as follows: Input video frames of size 640*640*3, mosaic stitching, and multi-lighting HSV adjustment for data enhancement; S3-2, behavior monitoring, uses the SimAM attention mechanism to capture whether a person is close to a high-voltage area or performing a dangerous operation, dynamically analyzes the action trajectory, and determines whether it complies with the regulations, as shown in formulas (4)-(6): Formula (4) Formula (5) Formula (6) in, Pixel The energy is the total number of pixels in the feature map, is the value of the current pixel, is the average value of the feature map; is the target state, is the next target state, is the state transfer matrix, is the input matrix, Input control quantity, System noise; is the observed value, is the state-to-observation transfer matrix, Measurement noise; S3-3, Multi-target Tracking, identifies and tracks multiple targets in complex scenes, using optimized feature fusion techniques for identification.

[0031] It should be noted that multi-target tracking is the real-time identification and tracking of multiple personnel in the distribution room, ensuring that the location and behavior of each person can be accurately monitored, and the dynamic information of the personnel can be identified and recorded, including entry into the high-voltage area, stay time, and trajectory.

[0032] By using the improved YOLOv5s model, human targets in each video frame are detected, target boxes and category information are generated, each target is assigned a unique ID, tracking consistency is maintained through inter-frame matching, deep features are used to avoid tracking loss due to short-term occlusion, and the prediction box is dynamically adjusted according to the speed and direction of the target to reduce missed detections. If a target is not detected for a long time, the system will record it as "target lost" and send a notification to the administrator.

[0033] Behavior classification: normal behavior and abnormal behavior. Normal behavior includes standing, moving, and working; abnormal behavior includes falling down, entering high-voltage areas, and staying in unauthorized areas for a long time.

[0034] Based on the skeleton key point detection model, the posture of the person is analyzed to determine whether the conditions for falling down are met, such as sudden deviation of the center of gravity and horizontal state; if falling down behavior is detected: an alarm is triggered immediately, and the acoustic camera is called to capture the on-site audio to further confirm the abnormal situation.

[0035] Through feature positioning technology, it analyzes whether there are people entering the dangerous area. When people cross the cordon or stay for a long time, a warning is triggered and the monitoring center is notified.

[0036] Step S4, classifying abnormal events based on the rule classifier and triggering equipment linkage control; Wherein step S4 also includes the following sub-steps: S4-1, abnormal analysis, integrates environmental monitoring and personnel behavior data, and determines the abnormal event level through a rule-based classifier, as shown in formula (7): Formula (7) in, is the level of abnormal events, is the threshold for level classification, is the current data value; According to the severity and impact of the event, abnormal events can be divided into the following three categories: low-level abnormalities, medium-level abnormalities, and high-level abnormalities. The rule classifier integrates various monitoring data and classifies the levels in the following way: Temperature and humidity abnormalities: temperature > 30℃ and ≤ 35℃ is considered a low-level abnormality; temperature > 35℃ and ≤ 45℃ is considered a medium-level abnormality; temperature > 45℃ is considered a high-level abnormality; Smoke diffusion range, smoke area <10 , slow growth rate is considered a low-level anomaly; smoke area ≥ 10 , fast growth rate is considered as intermediate abnormality; smoke area ≥30 Considered as a high-level anomaly; Abnormal behavior of personnel, not wearing a helmet, not wearing protective clothing is considered an intermediate abnormality; entering a high-voltage area or falling to the ground is considered a high-level abnormality; In addition, different response mechanisms are triggered according to the abnormality level: low-level abnormalities alert staff through the platform or on-site voice; medium-level abnormalities trigger relevant equipment and send alarms to the monitoring center to prompt the on-duty personnel to handle; high-level abnormalities activate emergency plans, notify all relevant personnel through sound and light alarms, and send real-time alarms to the monitoring center and emergency management personnel; S4-2, linkage mechanism, realizes the linkage between equipment and system, adjusts the air conditioning operation mode, quickly switches the external power supply through the fast interface box, and uses the state machine algorithm to control the air conditioning, power supply, and ventilation system. Specifically, as shown in formula (8): , event) Formula (8) in, is the next state of the system, is the state transition function, event is the event currently triggered; Trigger conditions: when abnormal data exceeds the set threshold or abnormal behavior is detected, the linkage response process is initiated; Air conditioning operation mode adjustment: adjust the air conditioning operation parameters according to the temperature and humidity sensor data. When the temperature is higher than the set range, turn on the cooling mode and adjust the wind speed to medium; when the humidity is higher than the set range, turn on the dehumidification mode; Fast power switching: when a main power failure is detected, it quickly switches to the backup power supply. Using a fast interface box, the main and backup power supply switching can be completed within 0.1 seconds. When the main power supply voltage is lower than the safety threshold or is completely interrupted, the interface box is triggered to switch to the backup power supply, and the time and trigger reason of the power switching are recorded at the same time. Ventilation system control: when the SF6 concentration or smoke concentration is detected to be excessive, the ventilation equipment is turned on to dilute the gas. When the SF6 concentration is greater than the safety threshold and the gas concentration continues to rise, the exhaust fans in the corresponding area are turned on. When the smoke concentration detects a diffusion trend, the exhaust equipment near the smoke area is turned on first. The ventilation equipment operation mode can dynamically adjust the wind speed according to the gas concentration. S4-3, alarm and notification, triggers different levels of alarms according to the event level. The alarm information includes video streams and environmental data snapshots, which are sent to the monitoring platform in real time; Alarm classification: Low-level abnormality, only prompting local alarm; medium-level abnormality: triggering sound and light alarm and notifying the on-duty personnel; high-level abnormality: initiating a comprehensive alarm and sending emergency notifications to all relevant parties.

[0037] It should be noted that low-level anomalies include slight excess of temperature and humidity or SF6 concentration approaching the threshold. The system records the data and reminds relevant personnel to check. Intermediate anomalies include significant excess of environmental parameters or detection of non-standard operation by personnel. The system activates linkage equipment (such as ventilation or area alarms) and sends warning information to the monitoring platform. Advanced anomalies include SF6 concentration exceeding the standard, smoke concentration increasing or equipment short circuit. The system immediately cuts off the power supply, starts the ventilation system, and notifies maintenance personnel for emergency processing.

[0038] In step S5, the data is stored in a NoSQL database, trends are analyzed to predict potential dangers, and the YOLOv5 model is optimized through historical data.

[0039] Wherein, in step S5, the following sub-steps are also included: S5-1, event record, stores all abnormal events and operation records in the NoSQL database, and manages abnormal events by time and type, and stores them in JSON data format; S5-2, based on historical data feedback, use new abnormal data to fine-tune the YOLOv5 model and optimize the target detection model. Specific steps: Data preparation, extracting data on abnormal events from NoSQL databases; Freeze the backbone network. The backbone network of YOLOv5 contains a large number of pre-trained parameters, which are frozen to retain the existing feature extraction capabilities. Only the classification head is trained for specific tasks. Retrain the classification head and use the newly added dataset to train the output layer of the model; Loss function adjustment: adjust the weight of the CIoU loss function.

[0040] S5-3, regularly generate safety operation reports, including statistics on safety attire rate, distribution and handling of abnormal events; S5-4, analyze data trends, predict potential risks and adjust monitoring strategies.

[0041] It should be noted that NoSQL databases are mainly used to store unstructured or semi-structured data. They are suitable for processing large-scale, dynamically changing data, and usually store data in JSON format. The stored data are classified into: sensor data, event records, and model optimization data.

[0042] The method for predicting potential dangers uses the ARIMA model to predict the future trend of time series data. The SF6 concentration data of the past 7 days is input to predict the concentration change trend in the next 24 hours. Based on the predicted value and the warning threshold, it is judged whether there is a potential danger. If the predicted value continues to approach or exceed the warning threshold, it is judged as a potential danger. The system triggers the alarm mechanism or linkage equipment in advance to reduce the probability of potential danger.

[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention has various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent power distribution room safety monitoring and management system, characterized in that: The following steps are involved: Step S1, check the status of sensors, gateways, and cameras, and collect environmental data through sensors; Step S2, pre-processing the collected data, using the YOLOv5 model to identify the smoke area, and starting the linkage safety mechanism according to the abnormal environmental parameters; Step S3, using the YOLOv5s model to detect people, capturing and analyzing behavior trajectories through the SimAM mechanism, and realizing multi-target tracking and abnormal behavior detection; Step S4, classifying abnormal events based on the rule classifier and triggering equipment linkage control; In step S5, the data is stored in a NoSQL database, trends are analyzed to predict potential dangers, and the YOLOv5 model is optimized through historical data.

2. According to claim 1, the intelligent power distribution room safety monitoring and management system is characterized by: Wherein step S1 also includes the following sub-steps: S1-1, device startup, initialization of edge computing devices, and confirmation that sensors, gateways, and cameras are in normal status; S1-2, environmental data acquisition, uses sensors to obtain environmental parameters in real time, including temperature, humidity, and gas concentration.

3. According to claim 1, the intelligent power distribution room safety monitoring and management system is characterized by: Wherein step S2 also includes the following sub-steps: S2-1, environmental data preprocessing, using a sliding window to calculate the average value, filtering the noise of the collected environmental parameters, and dynamically adjusting the monitoring threshold based on historical data, as shown in formula (1): Formula (1) in, is the smoothed value, is the window size, is the original data; S2-2, use the YOLOv5 model to analyze the smoke diffusion range and determine whether the environmental data exceeds the safety threshold. The backbone network selects CSPDarknet53 to extract features. The feature fusion is based on OANet to enhance the detection accuracy. The loss function is as follows: Formula (2) in, is the basic intersection-over-union ratio, is the Euclidean distance between the center of the prediction box and the target box, b is the center point coordinate of the prediction box, is the center point coordinate of the target box is the minimum diagonal distance of the bounding box, is the balancing factor, is the difference between the length ratio of the prediction box and the target box; S2-3, linkage triggering, when abnormal environmental parameters are detected, the corresponding safety mechanism is activated, and the state triggering logic is as follows: Formula (3) in, It is the specific linkage action that is triggered. is the trigger threshold, Current environment data.

4. The intelligent power distribution room safety monitoring and management system according to claim 1 is characterized by: Wherein step S3 also includes the following sub-steps: S3-1, safety clothing detection, uses the improved YOLOv5s model to detect whether personnel wear safety helmets and work clothes. For those who do not wear or wear them, their behavior is recorded and a voice warning is triggered. The specific method is as follows: Input video frames of size 640*640*3, mosaic stitching, and multi-lighting HSV adjustment for data enhancement; S3-2, behavior monitoring, uses the SimAM attention mechanism to capture whether a person is close to a high-voltage area or performing a dangerous operation, dynamically analyzes the action trajectory, and determines whether it complies with the regulations, as shown in formulas (4)-(6): Formula (4) Formula (5) Formula (6) in, Pixel The energy is the total number of pixels in the feature map, is the value of the current pixel, is the average value of the feature map; is the target state, is the next target state, is the state transfer matrix, is the input matrix, Input control quantity, System noise; is the observed value, is the state-to-observation transfer matrix, Measurement noise; S3-3, Multi-target Tracking, identifies and tracks multiple targets in complex scenes, using optimized feature fusion techniques for identification.

5. The intelligent power distribution room safety monitoring and management system according to claim 1 is characterized by: Wherein step S4 also includes the following sub-steps: S4-1, abnormal analysis, integrates environmental monitoring and personnel behavior data, and determines the abnormal event level through a rule-based classifier, as shown in formula (7): Formula (7) in, is the level of abnormal events, is the threshold for level classification, is the current data value; According to the severity and impact of the event, abnormal events can be divided into the following three categories: low-level abnormalities, medium-level abnormalities, and high-level abnormalities. The rule classifier integrates various monitoring data and classifies the levels in the following way: Temperature and humidity abnormalities: temperature > 30℃ and ≤ 35℃ is considered a low-level abnormality; temperature > 35℃ and ≤ 45℃ is considered a medium-level abnormality; temperature > 45℃ is considered a high-level abnormality; Smoke diffusion range, smoke area <10 , slow growth rate is considered a low-level anomaly; smoke area ≥ 10 , fast growth rate is considered as intermediate abnormality; smoke area ≥30 Considered as a high-level anomaly; Abnormal behavior of personnel, not wearing a helmet, not wearing protective clothing is considered an intermediate abnormality; entering a high-voltage area or falling to the ground is considered a high-level abnormality; In addition, different response mechanisms are triggered according to the abnormality level: low-level abnormalities alert staff through the platform or on-site voice; medium-level abnormalities trigger relevant equipment and send alarms to the monitoring center to prompt the on-duty personnel to handle; high-level abnormalities activate emergency plans, notify all relevant personnel through sound and light alarms, and send real-time alarms to the monitoring center and emergency management personnel; S4-2, linkage mechanism, realizes the linkage between equipment and system, adjusts the air conditioning operation mode, quickly switches the external power supply through the fast interface box, and uses the state machine algorithm to control the air conditioning, power supply, and ventilation system. Specifically, as shown in formula (8): , event) Formula (8) in, is the next state of the system, is the state transition function, event is the currently triggered event; Trigger conditions: when abnormal data exceeds the set threshold or abnormal behavior is detected, the linkage response process is initiated; Air conditioning operation mode adjustment: adjust the air conditioning operation parameters according to the temperature and humidity sensor data. When the temperature is higher than the set range, turn on the cooling mode and adjust the wind speed to medium; when the humidity is higher than the set range, turn on the dehumidification mode; Fast power switching: when a main power failure is detected, it quickly switches to the backup power supply. Using a fast interface box, the main and backup power supply switching can be completed within 0.1 seconds. When the main power supply voltage is lower than the safety threshold or is completely interrupted, the interface box is triggered to switch to the backup power supply, and the time and trigger reason of the power switching are recorded at the same time. Ventilation system control: when the SF6 concentration or smoke concentration is detected to be excessive, the ventilation equipment is turned on to dilute the gas. When the SF6 concentration is greater than the safety threshold and the gas concentration continues to rise, the exhaust fans in the corresponding area are turned on. When the smoke concentration detects a diffusion trend, the exhaust equipment near the smoke area is turned on first. The ventilation equipment operation mode can dynamically adjust the wind speed according to the gas concentration. S4-3, alarm and notification, triggers different levels of alarms according to the event level. The alarm information includes video streams and environmental data snapshots, which are sent to the monitoring platform in real time; Alarm classification: Low-level abnormality, only prompting local alarm; medium-level abnormality: triggering sound and light alarm and notifying the on-duty personnel; high-level abnormality: initiating a comprehensive alarm and sending emergency notifications to all relevant parties.

6. The intelligent power distribution room safety monitoring and management system according to claim 1 is characterized by: Wherein, in step S5, the following sub-steps are also included: S5-1, event record, stores all abnormal events and operation records in the NoSQL database, and manages abnormal events by time and type, and stores them in JSON data format; S5-2, based on historical data feedback, use new abnormal data to fine-tune the YOLOv5 model and optimize the target detection model. Specific steps: Data preparation, extracting data on abnormal events from NoSQL databases; Freeze the backbone network. The backbone network of YOLOv5 contains a large number of pre-trained parameters, which are frozen to retain the existing feature extraction capabilities. Only the classification head is trained for specific tasks. Retrain the classification head and use the newly added dataset to train the output layer of the model; Loss function adjustment: adjust the weight of CIoU loss function; S5-3, regularly generate safety operation reports, including statistics on safety attire rate, distribution and handling of abnormal events; S5-4, analyze data trends, predict potential risks and adjust monitoring strategies.

Images