Automatic intelligent risk grading management and control method and hidden danger checking system

Abstract

The invention provides an automatic intelligent risk grading management and control method and a hidden danger checking system. The method comprises the following steps: firstly, judging whether an equipment operation switch is turned on or not in real time to determine a monitoring scene section; secondly, outputting a first risk value R1 by adopting a kernel ridge regression model in a working period; then, in a time period when no person has a rest, outputting a second risk value R2 by adopting the convolutional neural network model; thirdly, the first risk value R1 or the second risk value R2 is judged with a preset threshold value; if the analysis result is between the minimum threshold value and the maximum threshold value, reducing the operation power of the equipment; if yes, entering the step S5; and finally, the system adopts measures of automatically triggering an alarm system or automatically powering off equipment or starting a fire-fighting device and the like according to the risk level. According to the invention, a kernel ridge regression model and a convolutional neural network model are adopted to calculate a risk value and compare the risk value with a preset threshold range; the accuracy and efficiency of risk early warning are greatly improved, and the user experience is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of video image processing, and in particular to an automated intelligent risk classification management and control method and a hidden danger detection system. Background Art

[0002] With the rapid development of intelligent monitoring technology, risk assessment systems based on video monitoring have been widely used in industrial production, construction, smart home and other fields. Video monitoring technology can timely discover potential risks and hidden dangers by acquiring image data in the monitoring scene and processing the data in combination with intelligent algorithms, thereby reducing the workload of manual inspections and improving the response speed to risks. In recent years, with the development of artificial intelligence algorithms, machine learning and deep learning models such as convolutional neural networks (CNN) and kernel ridge regression (KRR) have gradually been introduced into technical solutions for risk assessment and hidden danger investigation. These intelligent algorithms have shown great advantages in automated data analysis, complex scene processing and abnormal state detection. Traditional monitoring systems rely on regular inspections by personnel or simple preset rules (such as temperature exceeding the limit, abnormal equipment stop, etc.), and it is difficult to detect potential hidden dangers in real time. For example, during the operation of industrial equipment, sensor data such as temperature, humidity, gas concentration and operating speed of the equipment often have nonlinear and complex changes, making it difficult to effectively assess its risks through simple logical rules. However, traditional risk monitoring systems based on video monitoring can only capture explicit faults and cannot conduct in-depth analysis of the operating status of complex equipment. Therefore, relying solely on traditional monitoring systems can easily miss key hidden danger signals, thereby increasing the potential risk of safety hazards.

[0003] On the other hand, with the gradual maturity of deep learning technology, especially the widespread application of convolutional neural networks in computer vision, intelligent monitoring systems based on video images have begun to gradually replace traditional monitoring solutions. Convolutional neural networks have powerful feature extraction capabilities and can automatically identify subtle changes in images, making them suitable for complex scene monitoring and hidden danger identification. For example, in unattended equipment monitoring scenarios, video data obtained through infrared cameras can be used to monitor the operating status of equipment in real time. Convolutional neural network models can identify abnormal conditions in images and output risk assessment values. However, relying solely on image data may still be insufficient to fully evaluate the overall operating status of the equipment.

[0004] In recent years, deep learning artificial intelligence models have been gradually introduced into the field of intelligent monitoring risk warning technology. However, existing risk warnings have limitations. For example, most current intelligent risk warnings are only based on simple neural network or machine learning algorithm control implementations, and fail to fully consider the influencing factors of whether different machines and equipment are running or whether there are people or no one present, resulting in a significant increase in the manpower and material resources invested in monitoring, and cannot be flexibly adjusted according to actual scenarios, thereby improving monitoring accuracy and efficiency. In addition, existing systems usually only rely on current equipment data and cannot dynamically adjust control strategies based on the age of the equipment, resulting in the inability to achieve refined control in actual applications. In addition, the existing monitoring adjustment does not comprehensively consider the impact of equipment status data on the judgment of intelligent algorithms, which greatly reduces the efficiency and accuracy of video monitoring. In view of the limitations of existing risk warnings, a refined and automated real-time risk warning method urgently needs a new solution to improve the efficiency and accuracy of processing, which is conducive to improving user satisfaction. Summary of the invention

[0005] In view of the above problems mentioned in the prior art, the present invention provides an automated intelligent risk classification management method and hidden danger investigation system. The method first determines in real time whether the equipment operation switch is turned on to determine the monitoring scene section; secondly, the kernel ridge regression model is used to output the first risk value during the working period. ; Then, during the rest period when no one is around, the convolutional neural network model is used to output the second risk value ; Again, compare with the threshold to determine the risk level; Finally, the system automatically triggers the alarm system or automatically cuts off the power of the equipment or activates the fire-fighting device according to the risk level. This application adopts the kernel ridge regression model and the convolutional neural network model to calculate the risk value and compare it with the preset threshold range; it greatly improves the accuracy and efficiency of risk warning and greatly increases the user experience.

[0006] This application provides an automated intelligent risk classification management method, including the following steps: S1: Determine in real time whether the equipment operation switch is turned on, and determine the monitoring scene segment according to the equipment operation switch status. The scene segment includes the working period or the unmanned rest period; S2: During the working period, the equipment operation status data X is obtained in real time and input into the trained kernel ridge regression model. The kernel ridge regression model outputs the first risk value ; Equipment operation status data includes temperature, humidity, gas concentration, and equipment operation speed acquired by sensors; S3: During the rest period, the infrared camera is started to obtain infrared image video data in real time and input it into the trained convolutional neural network model. The convolutional neural network model outputs the second risk value. ; S4: The first risk value or the second risk value The preset first minimum threshold , the second minimum threshold To judge, if the first risk value or the second risk value If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to notify them to conduct an inspection. If the first risk value is less than the corresponding threshold, it is judged as no risk and the monitoring is continued in step S1. If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to remind them to conduct an inspection. Greater than the first maximum threshold or the second risk value Greater than the second maximum threshold Then proceed to step S5; S5: The system will automatically trigger the alarm system, the equipment will automatically cut off the power, start the fire-fighting device, and activate the evacuation indicator light and alarm broadcast.

[0007] Preferably, the kernel ridge regression model outputs a first risk value , which is expressed as follows: in, Represents the historical data point, i.e., the equipment operation status data of the past i-th historical data point, is the number of historical data points, Indicates the equipment operation status data at the current moment. is a hyperparameter, Represents the square of the Euclidean distance between the current equipment operation status data and the equipment operation status data at the i-th moment in the past; the weight parameter of the i-th historical data point for: in, is the first adjustment parameter, is the second adjustment parameter.

[0008] Preferably, the trained convolutional neural network model adopts an improved Sigmoid activation function: in, is the output of the convolution operation of the convolution layer, is the first risk value output by the kernel ridge regression model; Adjust the parameters for the first risk value.

[0009] Preferably, the step of reducing the operating power of the equipment and sending SMS reminders to the management personnel at intervals includes adjusting the power according to the magnitude of the risk: in, is the adjusted equipment operating power, is the current operating power of the device, Determined by the equipment operation switch, which is the first risk value or the second risk value ; The first minimum threshold is preset or the second minimum threshold ; The first maximum threshold or the second maximum threshold .

[0010] Preferably, the real-time acquisition of infrared image video data further includes, before inputting into the trained convolutional neural network model, using Gaussian filtering to perform denoising on the real-time acquired infrared image video data. The present application also provides an automated intelligent risk classification hidden danger investigation system, comprising: Equipment operation judgment module: judge in real time whether the equipment operation switch is turned on, and determine the monitoring scene section according to the equipment operation switch status. The scene section includes the working period or the rest period when no one is resting; The first calculation module: During the working period, the equipment operation status data X is obtained in real time and input into the trained kernel ridge regression model, which outputs the first risk value ; Equipment operation status data includes temperature, humidity, gas concentration, and equipment operation speed acquired by sensors; Second calculation module: When no one is resting, start the infrared camera to obtain infrared image video data in real time and input it into the trained convolutional neural network model. The convolutional neural network model outputs the second risk value. ; Judgment module: the first risk value or the second risk value The preset first minimum threshold , the second minimum threshold To judge, if the first risk value or the second risk value If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to notify them to conduct an inspection. If the first risk value is less than the corresponding threshold, it is determined to be risk-free, and the monitoring is continued in step S1. If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to remind them to conduct an inspection. Greater than the first maximum threshold or the second risk value Greater than the second maximum threshold Then proceed to step S5; Early warning module: The system will automatically trigger the alarm system, the equipment will automatically cut off power, start the fire-fighting device, and activate the evacuation indicator light and alarm broadcast.

[0011] Preferably, the kernel ridge regression model outputs a first risk value , which is expressed as follows: in, Represents the historical data point, i.e., the equipment operation status data of the past i-th historical data point, is the number of historical data points, Indicates the equipment operation status data at the current moment. is a hyperparameter, Represents the square of the Euclidean distance between the current equipment operation status data and the equipment operation status data at the i-th moment in the past; the weight parameter of the i-th historical data point for: in, is the first adjustment parameter, is the second adjustment parameter.

[0012] Preferably, the trained convolutional neural network model adopts an improved Sigmoid activation function: in, is the output of the convolution operation of the convolution layer, is the first risk value output by the kernel ridge regression model; Preferably, the step of reducing the operating power of the equipment and sending a short message to the management personnel at intervals includes adjusting the power according to the magnitude of the risk: in, is the adjusted equipment operating power, is the current operating power of the device, Determined by the equipment operation switch, which is the first risk value or the second risk value ; The first minimum threshold is preset or the second minimum threshold ; The first maximum threshold or the second maximum threshold .

[0013] Preferably, the real-time acquisition of infrared image video data further includes, before inputting into a trained convolutional neural network model, performing denoising processing on the real-time acquired infrared image video data using Gaussian filtering.

[0014] The present invention provides an automated intelligent risk classification control method and a hidden danger investigation system, which can achieve the following beneficial technical effects: 1. The present invention first determines in real time whether the equipment operation switch is turned on to determine the monitoring scene section; secondly, the kernel ridge regression model is used to output the first risk value during the working period ; Then, during the rest period when no one is around, the convolutional neural network model is used to output the second risk value ; Again, compare with the threshold to determine the risk level; Finally, the system automatically triggers the alarm system or automatically cuts off the power of the equipment or activates the fire-fighting device according to the risk level. This application adopts the kernel ridge regression model and the convolutional neural network model to calculate the risk value and compare it with the preset threshold range; it greatly improves the accuracy and efficiency of risk warning and greatly increases the user experience.

[0015] 2. The convolutional neural network model trained by the present invention adopts an improved Sigmoid activation function, by transforming the first risk value Introduced into the activation function, the system can dynamically perceive the changes in the video image environment and the impact of the equipment operating status, especially the impact of the equipment operating status on the infrared image. The technical solution not only uses sensor data for risk assessment, but also processes the infrared image video data during the unattended period through the convolutional neural network model to output the second risk value; the convolutional neural network can extract and analyze the subtle changes in the video image to achieve real-time monitoring of the external status of the equipment, especially at night or in unattended scenarios, to effectively identify potential hidden dangers in the environment around the equipment (such as abnormal heat sources, fire precursors, illegal intrusion, etc.). The fusion of two data sources (equipment sensor data and video image data) enables the system to perceive the status of the equipment from multiple dimensions, greatly improving the accuracy and comprehensiveness of monitoring.

[0016] 3. This solution processes the multi-dimensional sensor data of the equipment operating status (such as temperature, humidity, gas concentration, equipment operating speed, etc.) through the kernel ridge regression model. The kernel ridge regression model can effectively combine historical data with current sensor data and dynamically output the first risk value. This real-time assessment method can not only capture abnormal equipment status in a timely manner, but also predict possible equipment failures in advance by comparing with historical operating status, thereby reducing the potential risk of long-term operation of the equipment. The combination of the kernel ridge regression model and the convolutional neural network model makes the system more adaptive. Through the continuous accumulation of historical data and the dynamic calculation of risk values, the kernel ridge regression model can automatically adjust the weight parameters to make the risk assessment more in line with the actual operating status of the equipment.

[0017] 4. The present invention dynamically adjusts the operating power of the equipment according to the risk value. When the risk value falls between the preset minimum threshold and maximum threshold, the system will automatically reduce the operating power of the equipment to reduce the load and potential risk of the equipment. The power adjustment is dynamically calculated based on the difference between the risk value and the threshold. The higher the risk value, the greater the amplitude of the equipment power adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a schematic diagram of the steps of an automated intelligent risk classification management and control method of the present invention; Figure 2 It is a schematic diagram of an automated intelligent risk classification hidden danger investigation system of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 creative work are within the scope of protection of the present invention. Example

[0021] In view of the above problems mentioned in the prior art, in order to solve the above technical problems, as shown in the attached Figure 1 As shown: This application provides an automated intelligent risk classification management method, including the steps of: S1: Determine in real time whether the equipment operation switch is turned on, and determine the monitoring scene section according to the equipment operation switch state, and the scene section includes a working period or an unattended rest period; in some embodiments, in an intelligent monitoring system for industrial equipment, the equipment includes multiple sensors for monitoring the real-time operation state of the equipment, such as the power switch state, temperature, humidity, gas concentration, etc. of the equipment. At the same time, the equipment monitoring system is equipped with an infrared camera for monitoring the state of the equipment and its surrounding environment during the unattended period. In this embodiment, the first step of the system is to determine the operation switch state of the equipment in real time. The equipment feeds back the current operation status of the equipment in real time through a switch sensor connected to the main control system. The sensor records the on and off state of the equipment power supply, and the system determines whether the current equipment is in working state by reading the signal of the sensor. If the signal returned by the sensor shows that the switch of the equipment is in the on state, the system determines that the equipment is in working state; if the switch of the equipment is turned off, the system determines that the equipment is in the unattended state.

[0022] According to the equipment operation switch status, the system automatically determines the current monitoring scene segment. The monitoring scene segments are divided into two categories: If the system detects that the equipment switch is in the "on" state, the equipment is operating normally, and the system enters the working period monitoring mode. In this mode, the system collects the equipment's operating data (such as temperature, humidity, gas concentration, operating speed, etc.) in real time through sensors, and uses the kernel ridge regression model to perform real-time risk assessment on the equipment's operating status. If the equipment switch is in the "off" state, the system enters the unmanned rest period monitoring mode. In this mode, the equipment stops running, and the system mainly relies on infrared cameras to obtain infrared video images of the equipment's surrounding environment, and uses convolutional neural networks to identify potential risks in infrared images (such as abnormal temperature, fire hazards, etc.). In the factory workshop, workers turn off the equipment power after get off work every day, and the system uses the equipment switch sensor to determine the equipment's switch status in real time. The working hours are from 8 am to 6 pm every day. During this period, the equipment is turned on, and the system enters the working period monitoring mode to obtain the equipment's operating status data in real time and assess the risks. After get off work (6pm to 8am the next day), the power of the equipment is turned off, the system automatically enters the unmanned rest period monitoring mode, and the infrared camera starts to monitor the environment around the equipment to ensure that potential hidden dangers can still be detected during unmanned periods. This embodiment realizes the automatic conversion between working hours and unmanned rest periods through the real-time connection between the equipment switch sensor and the main control system. The system does not require manual intervention and can autonomously switch different monitoring modes according to the status of the equipment, making the monitoring work more flexible and intelligent.

[0023] S2: During the working period, the equipment operation status data X is obtained in real time and input into the trained kernel ridge regression model. The kernel ridge regression model outputs the first risk value ; Equipment operation status data includes temperature, humidity, gas concentration, and equipment operation speed acquired by sensors; In one embodiment, in the risk monitoring system of industrial equipment, the equipment is equipped with multiple sensors for collecting real-time operation status data of the equipment. These sensors include: temperature sensors detect the operating temperature of the equipment in real time. Humidity sensors monitor the humidity of the environment around the equipment. Gas concentration sensors detect the concentration of harmful gases near the equipment, such as volatile organic compounds (VOC) or carbon dioxide. Equipment operation speed sensors monitor the operation speed of mechanical parts of the equipment (such as motors or fans). Real-time acquisition of equipment operation status data: During the working period of the equipment, the system automatically acquires and summarizes the status data of the equipment through a real-time monitoring module connected to the sensor. These sensors collect temperature, humidity, gas concentration, and equipment operation speed data regularly (such as every second) to form a multi-dimensional data set. The system will process data of different dimensions according to different sensor types to ensure the real-time and accuracy of the collected data.

[0024] For example, during the working hours (8 am to 6 pm), the temperature data collected by the temperature sensor may be 80°C, the humidity sensor monitors the ambient humidity to be 45%, and the gas concentration sensor detects CO 2 The concentration is 300 ppm and the equipment running speed is 1200 RPM (revolutions per minute). These data will be packaged into the current state data X of the equipment and input into the system in the form of a multi-dimensional vector for subsequent processing. After the system obtains the equipment state data X, it will immediately input it into the pre-trained kernel ridge regression model for risk assessment. The kernel ridge regression model is trained based on historical data. It can process multi-dimensional nonlinear data and predict the risk level of the current equipment by comparing the similarity between the current data and the historical operating data. The kernel ridge regression model outputs the first risk value. The kernel ridge regression model processes the current equipment state data X, and compares and analyzes it based on the historical equipment state data, and finally outputs a first risk value. The risk value reflects the safety level of the current equipment state. The higher the risk value, the closer the current operating state of the equipment is to the historical failure or abnormal situation. A low risk value indicates that the current state of the equipment is relatively stable, and the operating state of the equipment is similar to the historical safety state data. A medium risk value indicates that some state parameters of the equipment are abnormal, such as temperature or operating speed deviating from the normal range, but it does not immediately pose a danger. A high risk value indicates that the current state of the equipment is very close to the state where a failure or abnormality occurred in the past, and there is a great safety hazard. The system will further trigger early warning or control measures. When the equipment operates under different conditions, the data fed back by the sensor will change continuously. For example, during a certain working period, the temperature of the equipment gradually rises from the normal 70°C to 90°C, the gas concentration gradually increases, and the operating speed of the equipment is unstable. The kernel ridge regression model analyzes these data in real time, and the output risk value will also increase accordingly. When the risk value exceeds the preset safety threshold, the system can trigger the corresponding early warning mechanism or automatically adjust the equipment operating parameters (such as reducing the equipment operating power). The output first risk value can be used for further decision-making, such as triggering system early warning or automatic equipment adjustment (such as reducing the equipment operating power). If the risk value exceeds the maximum threshold, the system will take further safety measures, such as triggering an emergency shutdown of the equipment or notifying the management personnel to conduct on-site inspections.

[0025] S3: During the rest period, the infrared camera is started to obtain infrared image video data in real time and input it into the trained convolutional neural network model. The convolutional neural network model outputs the second risk value. ; In some embodiments, the equipment can operate normally or automatically during unmanned rest periods; In some embodiments, during unmanned periods (such as at night or during factory closures), in order to ensure the safety of the equipment and its surroundings, the system monitors the equipment through an infrared camera to obtain real-time infrared image video data. Since no one is operating the equipment during this period, the system uses infrared image data to detect potential hidden dangers in the equipment, and performs intelligent analysis through a trained convolutional neural network (CNN) to output a second risk value, thereby achieving safety monitoring in an unmanned state. The infrared camera is started and image data is acquired. When the equipment enters an unmanned rest period (such as 6 pm to 8 am the next day), the system automatically detects that the power of the equipment has been turned off and enters an unmanned state. At this time, the system starts the infrared cameras installed around the equipment through a preset timing trigger or equipment status monitoring.

[0026] The camera can work in low-light environments, capture infrared images of the device and its surroundings, and provide visual data on abnormal device temperature, fire hazards, and temperature rise on the device surface. The image captured by the infrared camera is a real-time continuous video stream. Each frame of the image is a visual expression of temperature data, which can reflect the temperature distribution of the device and its environment. The infrared image data stream continuously captured by the camera forms a set of video frame data, and each frame of the image is used to represent the temperature distribution of the device and the environment. Each frame of the infrared image can be represented as a two-dimensional matrix, and each element in the matrix represents the infrared temperature information of the point. The image size is H×W, where H is the height of the image and W is the width. For example, the size of the infrared image is 256x256 pixels. After being input into the trained convolutional neural network model, the acquired infrared image video data is preprocessed (such as denoising, normalization, etc.) and then input into the pre-trained convolutional neural network (CNN) model in real time for analysis. Convolutional neural networks are specially trained for infrared image data and can identify potential abnormal temperature and heat source changes.

[0027] CNN model structure and processing flow: Input layer: The input data is a single frame infrared image obtained from an infrared camera, with a size of 256×256 pixels. Convolution layer: The first step of CNN is to perform a convolution operation on the image, using multiple convolution kernels to extract features such as edges and textures in the image. Each convolution kernel scans the entire image, performs a weighted sum operation on each local area, and extracts local features. The convolution layer operation is as follows: ; It is the feature map output by the convolutional layer; is the input infrared image, is the convolution kernel, is a bias term; activation function: the feature map output by the convolution layer is nonlinearly mapped through an activation function (such as Sigmoid) so that the model can capture the nonlinear features in the image. Pooling layer: The pooling layer is used to reduce the dimension of the feature map, reduce the amount of calculation, and retain important features. The commonly used maximum pooling operation is: Fully connected layer: After multiple convolution and pooling operations, the feature map is flattened into a one-dimensional vector and input to the fully connected layer for further feature combination; , is the output of the fully connected layer, is the weight matrix of the fully connected layer, is the bias term of the fully connected layer. Output layer: Finally, the convolutional neural network maps the output of the fully connected layer to the interval [0, 1] through the Sigmoid activation function, generating a probability value representing the size of the risk, called the second risk value , ; Larger values ​​indicate higher risk.

[0028] The convolutional neural network model generates a second risk value in real time based on the input infrared image, which is used to assess the risk level of the device during the unattended period. The risk value reflects whether there are potential safety hazards in the device or environment (such as abnormal temperature rise, uneven heating of the device, external fire signs, etc.). Low risk value: indicates that the temperature and environment around the device are within the normal range and no abnormality is detected. Medium risk value: indicates that the temperature of the device is slightly abnormal, or there is a local temperature rise in the environment, which requires further monitoring. High risk value: indicates that there are obvious abnormalities in the device or environment, such as a sharp rise in temperature, the proximity of an external heat source, etc., and emergency measures need to be taken immediately. In some embodiments, the device stops running at night, and the system starts an infrared camera to monitor the external state of the device. The infrared camera captures the temperature distribution image of the device and inputs it into the convolutional neural network model for risk assessment. If CNN detects that the temperature of a certain part of the device is abnormally high, there may be a fire hazard, and the system will output a high risk value, trigger an alarm and notify relevant personnel. If the temperature distribution in the image is normal, CNN will output a low risk value, indicating that the device and environment are in a safe state.

[0029] S4: The first risk value or the second risk value The preset first minimum threshold , the second minimum threshold To judge, if the first risk value or the second risk value If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to notify them to conduct an inspection. If the first risk value is less than the corresponding threshold, it is judged as no risk and the monitoring is continued in step S1. If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to remind them to conduct an inspection. Greater than the first maximum threshold or the second risk value Greater than the second maximum threshold Then proceed to step S5; in the industrial equipment monitoring system, the status of the equipment is monitored in real time by the kernel ridge regression model and the convolutional neural network model, respectively outputting the first risk value (based on sensor data) and the second risk value (based on infrared camera images). The system automatically performs corresponding control and early warning measures based on the comparison of these two risk values ​​with the preset risk thresholds to ensure the safe operation of the equipment under different risk conditions. In the intelligent monitoring system of industrial equipment, the status of the equipment is monitored in real time through sensor data and infrared camera images, and the system uses the kernel ridge regression model and the convolutional neural network model to output the first risk value R1 and the second risk value R2 respectively. The system performs a series of automated risk processing and emergency measures based on the size of the two risk values.

[0030] Processing of risk-free status: When the device is operating normally, the system continuously monitors the sensor data and infrared camera images of the device. The system compares the first risk value R1 output by the kernel ridge regression model and the second risk value R2 output by the convolutional neural network model with the preset first minimum threshold and second minimum threshold. If both risk values ​​are lower than the preset minimum threshold, the system determines that the current device status is safe and no further measures are required. For example: R1=0.3 (less than the first minimum threshold 0.5), R2=0.2 (less than the second minimum threshold 0.4). In this case, the system determines that the device is operating normally, continues to maintain real-time monitoring, and returns to step S1. The system will not perform any adjustment or warning operations.

[0031] For the processing of medium risk status, if the analysis results of the equipment's operating data or infrared image data show that the first risk value R1 or the second risk value R2 falls between the preset minimum threshold and the maximum threshold, the system will determine that the equipment is in a medium risk state. At this time, the system will automatically take the following measures: The system will automatically reduce the equipment's operating power through the intelligent control module according to the current risk value. This can reduce the load of the equipment in a potential risk state and prevent the equipment from failing due to continuous operation in a high-risk state. While reducing the equipment power, the system will send SMS notifications to managers to remind them that there is a certain risk in the current equipment and that manual inspection is recommended. The system will adjust the time interval of SMS reminders according to the size of the risk value. The closer the risk value is to the maximum threshold, the shorter the reminder interval is to ensure that managers can take inspection measures in time. For example: R1=0.7 falls between the first minimum threshold of 0.5 and the first maximum threshold of 0.9. The system reduces the equipment's operating power to 70% and sends SMS reminders to managers every 30 minutes until the risk situation is improved or resolved.

[0032] If the system detects that the first risk value R1 exceeds the first maximum threshold, or the second risk value R2 exceeds the second maximum threshold, the system determines that the equipment is in a high-risk state and immediately triggers emergency response measures to ensure the safety of equipment and personnel. At this time, the system will take the following measures: The system will immediately cut off the power supply of the equipment and stop the operation of the equipment to prevent the equipment from continuing to operate in a high-risk state and prevent further damage or accidents. If the system detects that there may be a fire hazard (for example, the infrared image shows that the temperature of the equipment is abnormally high), the system will automatically activate the fire protection system, such as starting the sprinkler or triggering the fire extinguisher. The system will issue a sound warning through the alarm device on site and activate the evacuation indicator light to remind the staff on site to quickly evacuate the high-risk area to ensure the safety of personnel. For example: R2 = 1.1, exceeding the second maximum threshold of 1.0, the system determines that the equipment is at risk of serious failure, immediately cuts off the power supply of the equipment and starts the fire sprinkler system, and at the same time issues an emergency alarm through the broadcasting system to notify the on-site personnel to evacuate.

[0033] In one embodiment, during the operation of the equipment, the sensor monitors that the temperature of the equipment is gradually rising. At the same time, the infrared camera captures the thermal image of the surface of the equipment, showing that the temperature distribution is abnormal. The system outputs the second risk value R2 of 0.85 through model analysis. At this time, the system determines that the risk value is between the preset minimum threshold and maximum threshold. To prevent accidents, the system automatically reduces the operating power of the equipment to 60%, and sends text message reminders to the on-duty manager every 20 minutes, suggesting equipment inspection. If the risk value further rises to more than 1.0, the system will trigger power outages and fire prevention measures.

[0034] S5: The system will automatically trigger the alarm system, the equipment will automatically power off, start the fire fighting device, and activate the evacuation indicator light and alarm broadcast. After the equipment has been running for a long time, the sensor data and infrared image data show abnormalities. The system detects that the equipment has serious safety hazards through the kernel ridge regression model and the convolutional neural network model. At this time, the system triggers a series of emergency response measures based on the risk assessment results to ensure the safety of the equipment and surrounding personnel. When the system detects that the first risk value R1 or the second risk value R2 of the equipment exceeds the preset maximum risk threshold, the system immediately triggers the alarm system to notify nearby staff of an emergency. The alarm system includes the following parts: Sound alarm: The system triggers the alarm in the workshop or equipment area to issue a harsh alarm to remind people working nearby to immediately pay attention to the abnormal situation of the equipment and prepare for emergency evacuation. Alarm information display: On the display screen in the control room or workshop, the system will pop up a red emergency alarm message to remind managers that the equipment has entered a high-risk state and require them to take immediate action. For example, when the system detects that R2=1.1, which exceeds the maximum threshold of 1.0, the system immediately activates the sound and light alarm system, sounds an alarm, and displays "Equipment overheating, fire hazard, please evacuate immediately" on the electronic screen in the workshop. In order to prevent the equipment from continuing to operate in a high-risk state and causing further deterioration of the accident, the system will automatically cut off the power supply of the equipment after the alarm is triggered. The power-off operation ensures that the equipment does not continue to operate, preventing further temperature rise, mechanical wear or other dangerous situations. The system is connected to the control interface of the equipment. When a high risk is detected, the system automatically sends a power-off signal to the control unit of the equipment to cut off the main power supply of the equipment. The power-off operation is usually linked with the emergency shutdown program of the equipment to ensure that the internal components of the equipment are not damaged when the power is off. When the system triggers a high-risk alarm, the power supply system of the equipment receives a power-off signal, the equipment immediately stops all operations, and the main power is cut off. At this time, the equipment is in a completely shut down state to prevent further risks.

[0035] In the event that there is a fire hazard in the equipment, the system will automatically activate the fire-fighting device to prevent the fire from spreading further. The fire-fighting device may include an automatic sprinkler system, a fire extinguisher, etc. If the system detects that the temperature inside or around the equipment is abnormally high and exceeds the safe range (for example, the infrared camera detects that the local temperature around the equipment is much higher than the normal value), the system will automatically activate the fire sprinkler system to cool the equipment by spraying water or fire-extinguishing foam to prevent the occurrence or spread of fire. In some high-risk scenarios, automatic fire-fighting devices (such as carbon dioxide fire extinguishers, dry powder fire extinguishers, etc.) are installed around the equipment. When a fire risk is detected, the system will automatically trigger the fire-fighting device to release fire extinguishing agents near the equipment. When the system detects that the infrared image shows that the surface temperature of the equipment reaches 300°C or above, the system determines that there is a fire risk and automatically starts the fire sprinkler system in the workshop to spray water around the equipment to quickly reduce the temperature.

[0036] In the case of high-risk equipment, in order to ensure that on-site workers can quickly evacuate the dangerous area, the system will activate the evacuation indicator light and issue an emergency evacuation order through the alarm broadcast system. In the area where the equipment is located, the system will light up the green emergency evacuation indicator light indicating the evacuation route, instructing workers to evacuate towards the safe area. These indicators will flash automatically to attract the attention of workers and ensure that they can quickly find the evacuation path. The system will issue a voice alarm through the broadcast system to notify all on-site workers to immediately evacuate the area around the equipment. The broadcast content will remind workers of the type of danger currently existing and instruct them to evacuate along the designated evacuation route. When the risk value of the equipment exceeds the threshold, the system will not only trigger an alarm, but also issue a notice through voice broadcast "Emergency evacuation, all workers are requested to leave the equipment area immediately", and the green evacuation indicator light will begin to flash in the workshop, indicating the nearest safe exit.

[0037] When the factory's automated production line is unattended at night, the system detects that the equipment is operating abnormally. The infrared image shows that the temperature of the equipment has risen sharply. The system uses the convolutional neural network model to determine that the second risk value R2 has exceeded the preset maximum threshold. At this time, the system immediately triggers a high-risk alarm and initiates the following emergency measures: the sound and light alarm system is activated, a high-pitched alarm sounds, and the control screen displays "Equipment abnormality, deal with it immediately". The equipment automatically powers off, the main power is cut off, and the equipment stops running. The fire sprinkler system starts and sprays fire extinguishing agent around the equipment. The evacuation indicator light flashes, and the nearby staff is notified to evacuate immediately through the broadcasting system.

[0038] In some embodiments, the kernel ridge regression model outputs a first risk value , which is expressed as follows: in, Represents the historical data point, i.e., the equipment operation status data of the past i-th historical data point, is the number of historical data points, Indicates the equipment operation status data at the current moment. is a hyperparameter, Represents the square of the Euclidean distance between the current equipment operation status data and the equipment operation status data at the i-th moment in the past; the weight parameter of the i-th historical data point for: in, is the first adjustment parameter, is the second adjustment parameter.

[0039] In some embodiments, the trained convolutional neural network model uses an improved Sigmoid activation function: in, is the output of the convolution operation of the convolution layer, is the first risk value output by the kernel ridge regression model; Adjust the parameters for the first risk value.

[0040] In some embodiments, the step of reducing the operating power of the device and sending a short message to the administrator at intervals includes adjusting the power according to the magnitude of the risk: in, is the adjusted equipment operating power, is the current operating power of the device, Determined by the equipment operation switch, which is the first risk value or the second risk value ; The first minimum threshold is preset or the second minimum threshold ; The first maximum threshold or the second maximum threshold .

[0041] In some embodiments, the real-time acquisition of infrared image video data further includes, before inputting into a trained convolutional neural network model, using Gaussian filtering to perform denoising processing on the real-time acquired infrared image video data.

[0042] This application also provides an automated intelligent risk classification hidden danger investigation system, such as Figure 2The following are included: Sensor module: The temperature sensor is used to monitor the real-time temperature of the equipment to help detect overheating or potential fire hazards. The humidity sensor is used to monitor the humidity of the environment around the equipment to prevent equipment failure due to excessive humidity. The gas concentration sensor is used to detect the concentration of harmful gases (such as volatile organic compounds or carbon monoxide, etc.) in the environment around the equipment to determine whether there is a harmful gas leak in the equipment or environment. The equipment operation speed sensor is used to monitor the operating speed of the mechanical parts of the equipment in real time to determine whether the equipment is in normal operation. The infrared camera module is used to monitor the equipment and the surrounding environment during unattended periods, obtain infrared image video data, monitor the temperature distribution of the external environment of the equipment in real time, and identify potential high temperature or fire hazards.

[0043] Control and Processing Unit (CPU): The main control system is responsible for managing and processing data from sensors and infrared cameras and performing equipment risk assessment. The main control system integrates a kernel ridge regression model and a convolutional neural network model to analyze sensor data and infrared image data and output a first risk value R1 and a second risk value R2. The network communication module is used to connect sensors, cameras and control units to achieve real-time data transmission. It supports wired network or wireless network (Wi-Fi) communication to ensure that sensors and cameras can transmit the collected data to the main control system in a timely manner. The sound and light alarm is used to issue sound and light alarms when the system detects a high risk, reminding on-site personnel to pay attention and take emergency measures. The SMS alarm module notifies managers of the risk status of the equipment through SMS or email, especially when the risk is medium or high, the system will send intermittent SMS reminders according to the risk level. Power control unit: Controls the power management of the equipment, including the automatic power-off function. When the system detects that the equipment is in a high-risk state, the power control unit will automatically cut off the power supply of the equipment to prevent further damage to the equipment or cause a safety accident. Fire protection system: When the system detects a fire risk, it automatically starts the sprinkler system for fire extinguishing. Automatic fire extinguisher: If a carbon dioxide or dry powder fire extinguisher is installed around the equipment, the system can automatically trigger the fire extinguishing device to extinguish a local fire. Evacuation indicator light and broadcasting system: The evacuation indicator light is installed around the equipment. When the system detects a high risk, the indicator light will automatically light up and indicate the evacuation direction. The broadcasting system notifies on-site personnel to evacuate the dangerous area immediately through a voice alarm system.

[0044] In some embodiments, the device operation judgment module determines in real time whether the device operation switch is turned on, and determines the monitoring scene section according to the device operation switch state, and the scene section includes a working period or an unmanned rest period; the connection between the sensor module and the control unit: each sensor (temperature, humidity, gas concentration, operating speed) is connected to the main control system by wired or wireless means, and transmits the operation data of the device to the main control system in real time. The sensor module uses the RS485 communication interface or the Zigbee wireless protocol to ensure that multi-sensor data can be transmitted synchronously. The connection between the infrared camera module and the control unit: the infrared camera is connected to the main control system through a network interface (such as Ethernet or Wi-Fi) to transmit infrared image video data. The image data of the camera is sent to the main control system through a dedicated video processing interface (such as RTSP or HTTP protocol). The connection between the control unit and the early warning system: when the control unit detects that there is a risk in the equipment, it will activate the sound and light alarm through the built-in interface (such as GPIO interface or RS485 interface), and send a reminder to the management personnel through the SMS alarm module. The connection between the control unit and the power control unit: the main control system directly controls the power-off operation of the equipment through the power control interface (such as the relay interface). When a high risk is detected, the main control system sends a signal to the power control unit to cut off the power supply to the equipment. Connection between the control unit and the fire protection system: The fire protection system is connected to the main control system through a digital signal interface. The main control system triggers the start signal of the fire protection system based on infrared image data or sensor data. Connection between the evacuation indicator light and the broadcasting system: The evacuation indicator light and the broadcasting system are connected to the main control system through a digital output interface. When a high risk is detected, the main control system will trigger the indicator light to light up and the broadcasting system to broadcast the evacuation command.

[0045] The first calculation module: During the working period, the equipment operation status data X is obtained in real time and input into the trained kernel ridge regression model, which outputs the first risk value ; Equipment operation status data includes temperature, humidity, gas concentration, and equipment operation speed acquired by sensors; Second calculation module: When no one is resting, start the infrared camera to obtain infrared image video data in real time and input it into the trained convolutional neural network model. The convolutional neural network model outputs the second risk value. ; Judgment module: the first risk value or the second risk value The preset first minimum threshold , the second minimum threshold To judge, if the first risk value or the second risk value If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to notify them to conduct an inspection. If the first risk value is less than the corresponding threshold, it is judged as no risk and the monitoring is continued in step S1. If the analysis result is between the minimum threshold and the maximum threshold, the equipment operating power is reduced and a text message is sent to the management personnel to remind them to conduct an inspection. Greater than the first maximum threshold or the second risk value Greater than the second maximum threshold Then proceed to step S5; Early warning module: The system will automatically trigger the alarm system, the equipment will automatically cut off power, start the fire-fighting device, and activate the evacuation indicator light and alarm broadcast.

[0046] In some embodiments, the kernel ridge regression model outputs a first risk value , which is expressed as follows: in, Represents the historical data point, i.e., the equipment operation status data of the past i-th historical data point, is the number of historical data points, Indicates the equipment operation status data at the current moment. is a hyperparameter, Represents the square of the Euclidean distance between the current equipment operation status data and the equipment operation status data at the i-th moment in the past; the weight parameter of the i-th historical data point for: in, is the first adjustment parameter, is the second adjustment parameter.

[0047] In some embodiments, kernel ridge regression is an extension of ridge regression, which combines kernel methods with ridge regression techniques to handle nonlinear regression problems. The kernel ridge regression model can map data from low-dimensional space to high-dimensional space by introducing kernel functions, and realize linear regression in high-dimensional space, thereby solving complex nonlinear problems. The structure and working principle of the kernel ridge regression model are described in detail below. The input layer is the starting point of the kernel ridge regression model, which receives input data, usually represented as a set of feature vectors. The input data can be multidimensional, representing various sensor data of the device (such as temperature, humidity, running speed, etc.). The input layer accepts data in the form of vectors, and each input data point contains several features. For example, for device monitoring scenarios, the input may include multidimensional data such as temperature, humidity, and device speed. Each input vector represents a data point of a device state. The core part of the kernel ridge regression model is the kernel function, which maps the input data from low-dimensional space to high-dimensional space. The role of the kernel function is to make linearly inseparable problems become linearly separable in high-dimensional space. In kernel ridge regression, the kernel function converts the linear regression problem into a nonlinear regression problem by calculating the similarity between input data points.

[0048] Common kernel functions are as follows: Linear kernel functions are used for linear regression problems and are suitable for situations where features are linearly related to target variables. Polynomial kernel functions perform polynomial expansion on the features of input data to capture nonlinear relationships. Gaussian kernel function (RBF, radial basis function): One of the most commonly used kernel functions, it can map data to an infinite-dimensional space and is suitable for scenarios with strong nonlinear relationships. It measures the distance between input data points. The closer the distance, the higher the similarity. Sigmoid kernel function is similar to the activation function in neural networks and is used to simulate nonlinear relationships between data. Ridge regression The ridge regression part of the kernel ridge regression model is used to solve the problem of multicollinearity and prevent the model from overfitting. Ridge regression constrains the parameters of the model by adding regularization terms to the loss function, which reduces the complexity of the model and improves the generalization ability. Regularization term In the kernel ridge regression model, the regularization term is a penalty term. Its function is to limit the size of the regression coefficient and prevent the model from overfitting the training data. The regularization term is automatically adjusted according to the distribution of the training data to ensure that the model can maintain a stable regression effect in high-dimensional space.

[0049] Kernel Matrix In kernel ridge regression, all input data points are computed similarly using kernel functions to form a kernel matrix (also called Gram matrix). Each element of the kernel matrix represents the similarity between two data points in high-dimensional space. Construction of the kernel matrix The kernel matrix compares all input data pairs to obtain the similarity between each pair of data points. For example, if there are N data points, the size of the kernel matrix is ​​N×N. This matrix is ​​used in subsequent regression calculations and determines the contribution of each input data point to the target variable. Weight Calculation In kernel ridge regression, the prediction results of the model are calculated using a set of weight parameters. The weight parameters reflect the influence of each data point on the final prediction. By combining the kernel matrix and the target value, kernel ridge regression calculates the weight of each training data point. The role of weights The weight of each data point determines the contribution of the data point to the prediction result. The larger the weight, the greater the influence of the data point on the final prediction. The calculation of weights combines the similarity of the kernel function and the regularization term in ridge regression to ensure that the model can output stably in high-dimensional space. Prediction output In the last step of kernel ridge regression, the model uses the calculated weights and kernel matrix to predict new input data points. The similarity between the new data point and the training data point is calculated by the kernel function, and then the weights are combined to generate the prediction result. For each new input data point, kernel ridge regression first calculates the similarity between the data point and the training data point through the kernel function. Then, the model obtains the prediction result based on the weighted average of the similarity and weight. Since kernel ridge regression can capture nonlinear relationships, it performs well in dealing with complex regression problems. The regularization control term in kernel ridge regression is used to adjust the complexity of the model, thereby improving the generalization ability of the model. The size of the regularization control term determines how well the model fits the data. If the regularization control term is too large, the model may underfit; if the regularization control term is too small, the model may overfit. Therefore, regularization control is a key step in ridge regression. Through the control of the regularization term, kernel ridge regression can achieve a good balance between training data and test data, ensuring that the model can also perform well on new data. This ability makes the kernel ridge regression model suitable for actual industrial equipment monitoring, complex system state prediction and other scenarios.

[0050] The training process of the kernel ridge regression model involves determining the regularization parameter, kernel function type, and kernel parameters. These parameters need to be tuned according to the specific data characteristics and tasks to obtain the best regression effect. The training process of the kernel ridge regression model is as follows: Select the kernel function First, select a suitable kernel function, such as the Gaussian kernel function is suitable for data with strong nonlinear relationships, and the polynomial kernel function is suitable for relationships with polynomial distributions. Determine the regularization parameter The optimal value of the regularization parameter is determined by methods such as cross-validation to prevent the model from overfitting or underfitting. The training model uses the training data to calculate the kernel matrix and weight parameters, and completes the model training based on the kernel function and regularization term. Model tuning adjusts the kernel parameters (such as the width parameter of the Gaussian kernel) and the regularization parameters so that the model can achieve the best prediction performance in practical applications.

[0051] The workflow of the kernel ridge regression model is as follows: the device sensor or other data source provides input data, and each data point consists of multiple features, representing the state of the device or the environmental condition. The input data is mapped to a high-dimensional space, and the similarity between each data point is calculated through the kernel function. The kernel matrix of all data points is calculated to capture the nonlinear relationship of the data. Regularization control terms are used to prevent overfitting, and the weight parameters are calculated by combining the target value and the kernel matrix. The weight parameters and the kernel matrix are used to predict new data to generate regression outputs to predict the device state or environmental risk.

[0052] In some embodiments, the trained convolutional neural network model uses an improved Sigmoid activation function: in, is the output of the convolution operation of the convolution layer, is the first risk value output by the kernel ridge regression model; Adjust the parameters for the first risk value.

[0053] In some embodiments, the structure of a convolutional neural network (CNN) consists of multiple different types of layers, each of which has a specific function and works together to achieve feature extraction and classification of images. The following is a detailed description of the structure of a typical convolutional neural network: The input layer is the first layer of the convolutional neural network, which is used to receive the original image data. The image can be a grayscale image (single channel) or a color image (multiple channels, such as RGB three channels). The data received by the input layer is usually a three-dimensional tensor, and the dimension of the tensor is height × width × number of channels (for example, 256×256×3 represents a 256×256 pixel color image, and each pixel has three channels corresponding to red, green, and blue). The convolution layer is the core part of the CNN and is responsible for extracting local features from the input image. By using multiple convolution kernels (also called filters), the convolution layer can capture basic features such as edges and textures in the image. The convolution kernel is a small-sized matrix (such as 3×3 or 5×5), which slides on the input image (also called a convolution operation), performs weighted summation on each local area, and generates a feature map. Activation function Usually after the convolution operation, the convolution layer will apply a nonlinear activation function (such as ReLU) to introduce nonlinear characteristics to the network to help the model process complex features. The pooling layer is used to reduce the dimension of the feature map output by the convolution layer, reduce the amount of calculation and retain important features. Pooling can reduce the size of the feature map while avoiding overfitting. Max pooling: This is a commonly used pooling method that selects the maximum value in a window area as the feature of the area. For example, the maximum value of the area is selected in a 2×2 pooling window to reduce the size of the feature map. Average pooling: Another pooling method that reduces the dimension by calculating the average of all values ​​in the pooling window. Compared with maximum pooling, average pooling is less sensitive to subtle changes. The fully connected layer is a layer used for feature combination in the middle and late stages of CNN. It flattens the high-dimensional feature map obtained after the previous convolution and pooling operations into a one-dimensional vector and performs a linear transformation with the weight matrix. The fully connected layer is usually used to generate classification results or make predictions. The fully connected layer is similar to the traditional artificial neural network layer, where each neuron is connected to all neurons in the previous layer. Through these connections, the network can make a comprehensive judgment based on the previously extracted features and output the prediction results. The output layer is the last layer of CNN, which is used to generate the final prediction results of the model. In classification tasks, the number of neurons in the output layer is usually equal to the number of categories to be classified. CNN uses activation functions to map the output of the fully connected layer to the final result. Softmax activation function: commonly used in multi-classification tasks, the Softmax function converts the output of the fully connected layer into a probability distribution and predicts the maximum probability of the category corresponding to the result. Sigmoid activation function: used for binary classification tasks or generating probability values, it limits the output value to between 0 and 1, indicating the probability of a certain category. The activation function is a nonlinear transformation function applied after the convolutional layer and the fully connected layer.Activation functions introduce nonlinearity to the network, enabling the model to learn complex feature patterns. ReLU truncates negative values ​​in the input value to 0 and retains positive values. ReLU is the most commonly used activation function in CNN, which can effectively solve the gradient vanishing problem and speed up convergence. Sigmoid: Maps the input value to a range between 0 and 1, often used in the output layer or binary classification tasks. Softmax: Used for multi-classification tasks, it converts the output value into the probability of multiple categories. Dropout layer is one of the techniques to prevent overfitting. During the training process, Dropout randomly "discards" a part of the neurons so that the network does not rely on certain specific neurons during training, thereby enhancing the generalization ability of the model. Dropout is usually applied after the fully connected layer. The batch normalization layer is used to normalize the input data in each batch to ensure that the input distribution of each neuron remains stable. This can speed up the training process and prevent the problem of gradient vanishing or exploding. The normalization layer is applied when the image is input, which is used to normalize the input data (such as subtracting the mean and dividing by the standard deviation) to standardize the pixel values ​​of the input image to a uniform range, thereby speeding up training. Convolutional neural network workflow: Input layer: receives an image with a shape of height × width × channels. Convolution layer: extracts local features of the image through multiple convolution kernels and generates multiple feature maps. Activation function (such as ReLU): performs nonlinear mapping on the feature map output by the convolution layer. Pooling layer: performs dimensionality reduction operations on the feature map to reduce the amount of calculation and retain important features. Repeat "convolution layer + activation function + pooling layer" many times: In CNN, this combination is usually stacked repeatedly many times to gradually extract higher-level features. Fully connected layer: flattens the previously extracted features into a one-dimensional vector and combines the features through the fully connected layer. Output layer: generates classification results or prediction values ​​based on the final fully connected layer output through the activation function.

[0054] In some embodiments, the step of reducing the operating power of the device and sending a short message to the administrator at intervals includes adjusting the power according to the magnitude of the risk: in, is the adjusted equipment operating power, is the current operating power of the device, Determined by the equipment operation switch, which is the first risk value or the second risk value ; The first minimum threshold is preset or the second minimum threshold ; The first maximum threshold or the second maximum threshold .

[0055] In some embodiments, the real-time acquisition of infrared image video data further includes, before inputting into a trained convolutional neural network model, using Gaussian filtering to perform denoising processing on the real-time acquired infrared image video data.

[0056] The present invention provides an automated intelligent risk classification control method and a hidden danger investigation system, which can achieve the following beneficial technical effects: 1. The present invention first determines in real time whether the equipment operation switch is turned on to determine the monitoring scene section; secondly, the kernel ridge regression model is used to output the first risk value during the working period ; Then, during the rest period when no one is around, the convolutional neural network model is used to output the second risk value ; Again, compare with the threshold to determine the risk level; Finally, the system automatically triggers the alarm system or automatically cuts off the power of the equipment or activates the fire-fighting device according to the risk level. This application adopts the kernel ridge regression model and the convolutional neural network model to calculate the risk value and compare it with the preset threshold range; it greatly improves the accuracy and efficiency of risk warning and greatly increases the user experience.

[0057] 2. The convolutional neural network model trained by the present invention adopts an improved Sigmoid activation function, by transforming the first risk value Introduced into the activation function, the system can dynamically perceive the changes in the video image environment and the impact of the equipment operating status, especially the impact of the equipment operating status on the infrared image. The technical solution not only uses sensor data for risk assessment, but also processes the infrared image video data during the unattended period through the convolutional neural network model to output the second risk value; the convolutional neural network can extract and analyze the subtle changes in the video image to achieve real-time monitoring of the external status of the equipment, especially at night or in unattended scenarios, to effectively identify potential hidden dangers in the environment around the equipment (such as abnormal heat sources, fire precursors, illegal intrusion, etc.). The fusion of two data sources (equipment sensor data and video image data) enables the system to perceive the status of the equipment from multiple dimensions, greatly improving the accuracy and comprehensiveness of monitoring.

[0058] 3. This solution processes the multi-dimensional sensor data of the equipment operating status (such as temperature, humidity, gas concentration, equipment operating speed, etc.) through the kernel ridge regression model. The kernel ridge regression model can effectively combine historical data with current sensor data and dynamically output the first risk value. This real-time assessment method can not only capture abnormal equipment status in a timely manner, but also predict possible equipment failures in advance by comparing with historical operating status, thereby reducing the potential risk of long-term operation of the equipment. The combination of the kernel ridge regression model and the convolutional neural network model makes the system more adaptive. Through the continuous accumulation of historical data and the dynamic calculation of risk values, the kernel ridge regression model can automatically adjust the weight parameters to make the risk assessment more in line with the actual operating status of the equipment.

[0059] 4. The present invention dynamically adjusts the operating power of the equipment according to the risk value. When the risk value falls between the preset minimum threshold and maximum threshold, the system will automatically reduce the operating power of the equipment to reduce the load and potential risk of the equipment. The power adjustment is dynamically calculated based on the difference between the risk value and the threshold. The higher the risk value, the greater the amplitude of the equipment power adjustment.

[0060] The above is a detailed introduction to an automated intelligent risk grading control method and a hidden danger inspection system. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention; at the same time, for general technical personnel in this field, according to the ideas and methods of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. An automated intelligent risk classification management and control method, characterized in that: Includes steps: S1: Determine in real time whether the equipment operation switch is turned on, and determine the monitoring scene segment according to the equipment operation switch status. The scene segment includes the working period or the unmanned rest period; S2: During the working period, the equipment operation status data X is obtained in real time and input into the trained kernel ridge regression model. The kernel ridge regression model outputs the first risk value ; Equipment operation status data includes temperature, humidity, gas concentration, and equipment operation speed acquired by sensors; S3: During the rest period, the infrared camera is started to obtain infrared image video data in real time and input it into the trained convolutional neural network model. The convolutional neural network model outputs the second risk value. ; S4: The first risk value or the second risk value The preset first minimum threshold , the second minimum threshold To judge, if the first risk value or the second risk value If the values ​​are less than the corresponding thresholds, it is determined to be risk-free, and the process returns to step S1 to continue monitoring; If the analysis result is between the minimum threshold and the maximum threshold, the device operating power is reduced and SMS reminders are sent to the management personnel at intervals to notify them to conduct inspections. Greater than the first maximum threshold or the second risk value Greater than the second maximum threshold Then proceed to step S5; S5: The system will automatically trigger the alarm system, the equipment will automatically cut off the power, start the fire-fighting device, and activate the evacuation indicator light and alarm broadcast.

2. The automated intelligent risk classification management and control method according to claim 1, characterized in that: The kernel ridge regression model outputs a first risk value , which is expressed as follows: in, Represents the historical data point, i.e., the equipment operation status data of the past i-th historical data point, is the number of historical data points, Indicates the equipment operation status data at the current moment. is a hyperparameter, Represents the square of the Euclidean distance between the current equipment operation status data and the equipment operation status data at the i-th moment in the past; the weight parameter of the i-th historical data point for: in, is the first adjustment parameter, is the second adjustment parameter.

3. The automated intelligent risk classification management and control method according to claim 2, characterized in that: The trained convolutional neural network model adopts an improved Sigmoid activation function: in, is the output of the convolution operation of the convolution layer, is the first risk value output by the kernel ridge regression model; Adjust the parameters for the first risk value.

4. The automated intelligent risk classification management and control method according to claim 1, characterized in that: The above mentioned method of reducing the operating power of the equipment and sending SMS reminders to the management personnel at intervals includes adjusting the power according to the magnitude of the risk: in, is the adjusted equipment operating power, is the current operating power of the device, Determined by the equipment operation switch, which is the first risk value or the second risk value ; The first minimum threshold is preset or the second minimum threshold ; The first maximum threshold or the second maximum threshold .

5. The automated intelligent risk classification management and control method according to claim 1, characterized in that: The real-time acquisition of infrared image video data also includes, before inputting into a trained convolutional neural network model, using Gaussian filtering to perform denoising processing on the real-time acquired infrared image video data.

6. An automated intelligent risk classification hidden danger investigation system, characterized in that: include: Equipment operation judgment module: judge in real time whether the equipment operation switch is turned on, and determine the monitoring scene section according to the equipment operation switch status. The scene section includes the working period or the rest period when no one is resting; The first calculation module: During the working period, the equipment operation status data X is obtained in real time and input into the trained kernel ridge regression model, which outputs the first risk value ; Equipment operation status data includes temperature, humidity, gas concentration, and equipment operation speed acquired by sensors; Second calculation module: When no one is resting, start the infrared camera to obtain infrared image video data in real time and input it into the trained convolutional neural network model. The convolutional neural network model outputs the second risk value. ; Judgment module: the first risk value or the second risk value The preset first minimum threshold , the second minimum threshold To judge, if the first risk value or the second risk value If the values ​​are less than the corresponding thresholds, it is determined to be risk-free, and the process returns to step S1 to continue monitoring; If the analysis result is between the minimum threshold and the maximum threshold, the device operating power is reduced and SMS reminders are sent to the management personnel at intervals to notify them to conduct inspections. Greater than the first maximum threshold or the second risk value Greater than the second maximum threshold Then proceed to step S5; Early warning module: The system will automatically trigger the alarm system, the equipment will automatically cut off power, start the fire-fighting device, and activate the evacuation indicator light and alarm broadcast.

7. The automated intelligent risk classification hidden danger investigation system according to claim 6, characterized in that: The kernel ridge regression model outputs a first risk value , which is expressed as follows: in, Represents the historical data point, i.e., the equipment operation status data of the past i-th historical data point, is the number of historical data points, Indicates the equipment operation status data at the current moment. is a hyperparameter, Represents the square of the Euclidean distance between the current equipment operation status data and the equipment operation status data at the i-th moment in the past; the weight parameter of the i-th historical data point for: in, is the first adjustment parameter, is the second adjustment parameter.

8. The automated intelligent risk classification hidden danger investigation system according to claim 7, characterized in that: The trained convolutional neural network model adopts an improved Sigmoid activation function: in, is the output of the convolution operation of the convolution layer, is the first risk value output by the kernel ridge regression model; Adjust the parameters for the first risk value.

9. The automated intelligent risk classification hidden danger investigation system according to claim 6, characterized in that: The above mentioned method of reducing the operating power of the equipment and sending SMS reminders to the management personnel at intervals includes adjusting the power according to the magnitude of the risk: in, is the adjusted equipment operating power, is the current operating power of the device, Determined by the equipment operation switch, which is the first risk value or the second risk value ; The first minimum threshold is preset or the second minimum threshold ; The first maximum threshold or the second maximum threshold .

10. The automated intelligent risk classification hidden danger investigation system according to claim 6, characterized in that: The real-time acquisition of infrared image video data also includes, before inputting into a trained convolutional neural network model, using Gaussian filtering to perform denoising processing on the real-time acquired infrared image video data.

Images