An intelligent monitoring and early warning system for power distribution cabinets

Abstract

The present invention relates to the technical field of safety monitoring of power equipment, and specifically to an intelligent monitoring and early warning system for a power distribution cabinet, comprising: collecting historical fault data and false alarm data through a data collection and processing module, and identifying key events. The parameter setting and management module obtains a set of abnormal change parameters for each key event, and sets the initial acquisition parameters, priority, acquisition frequency and trigger conditions. The real-time data acquisition module obtains the real-time initial parameters according to the priority and acquisition frequency. If the real-time initial parameters meet its triggering conditions, a set of real-time change parameters are obtained within the impact time; the dust coverage rate calculation module calculates the dust coverage rate based on the real-time change parameters; the risk assessment module calculates the risk assessment value based on the real-time change parameters and the dust coverage rate; the false alarm model training and identification module generates a false alarm prediction label according to the real-time change parameters. Finally, when the risk assessment value exceeds the preset threshold and the false alarm identification result is not a false alarm, the early warning module issues an early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring of electric power equipment, and in particular to an intelligent monitoring and early warning system for a power distribution cabinet. Background Art

[0002] The power distribution cabinet is the terminal equipment in the power system, which is usually connected to multiple power-consuming equipment, protection devices and monitoring systems. It not only undertakes the task of power distribution, but also is responsible for the safety and stability of the entire power system. The power distribution cabinet is widely used in industry, commercial buildings and large-scale infrastructure, and is an important node in the power system.

[0003] With the continuous growth of power demand and the complexity of power systems, the design of distribution cabinets has gradually developed towards intelligence and automation. Distribution cabinets not only have to bear and manage high-power power loads, but also have to cope with the challenges of various external environmental factors, such as humidity, high temperature, dust, etc. If these environmental factors are not monitored and managed, they may lead to aging and poor contact of internal components, and even serious accidents such as electrical fires. Therefore, modern distribution cabinets have gradually integrated environmental monitoring functions and intelligent management systems. These systems can monitor the temperature, humidity, and dust accumulation inside and outside the distribution cabinet in real time, and detect the operating status of internal components, such as the status of circuit breakers and the tightness of cable connections. Once an abnormality is detected, the system will issue an early warning in time to notify maintenance personnel to carry out maintenance and prevent potential failures. This intelligent monitoring system achieves more accurate and efficient power management by improving the ability to perceive the environment and equipment status.

[0004] Existing intelligent monitoring and early warning systems for distribution cabinets usually adopt a comprehensive monitoring strategy, attempting to provide extensive coverage and real-time monitoring of all parameters in the distribution cabinet and its environment. However, this comprehensive monitoring approach has some significant shortcomings in practice. First, comprehensive monitoring lacks specificity and is prone to resource dispersion and waste, because the system needs to process a large amount of data, including many parameters that may be irrelevant or non-critical. Secondly, because the monitoring scope is too wide, the system may not be able to effectively focus on the most critical anomalies, resulting in reduced accuracy of fault warnings. In addition, processing a large amount of irrelevant data will also increase the complexity and operational burden of the system, which may slow down the response to truly important issues and ultimately affect the safety and reliability of the overall power system.

[0005] Therefore, an intelligent monitoring and early warning system for distribution cabinets is proposed. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent monitoring and early warning system for a power distribution cabinet, which includes multiple modules: a data collection and processing module is used to collect historical fault data and false alarm data, and identify key events that cause faults; a parameter setting and management module obtains abnormal change parameters corresponding to key events according to historical fault data, defines initial acquisition parameters, and sets priority, acquisition frequency, trigger conditions and impact duration; a real-time data acquisition module acquires initial parameters in real time according to priority and acquisition frequency, and continuously acquires abnormal change parameters corresponding to key events when the trigger conditions are met to obtain real-time change parameters; a dust coverage rate calculation module calculates dust coverage rate based on real-time change parameters; a risk assessment module calculates risk assessment values ​​in combination with real-time change parameters and dust coverage rate; and a false alarm model training and identification module generates false alarm prediction labels according to real-time change parameters. The early warning module issues an early warning when the risk assessment value exceeds a preset threshold and is predicted to be non-false alarm. By introducing a dust coverage rate prediction model, the system can monitor the dust accumulation on the surface of the equipment in real time, thereby improving the accuracy of equipment fault prediction and avoiding false alarms or equipment failures caused by dust coverage. The calculation method of risk assessment value enables the system to dynamically adjust the weights and parameters in the assessment process, comprehensively consider factors such as equipment operating status, dust coverage and equipment usage time, so as to more accurately conduct risk assessment and early warning. At the same time, the false alarm recognition model is introduced. By independently training the model for different events, false alarms can be effectively identified, false alarms caused by environmental factors or equipment aging can be reduced, and the accuracy and reliability of the system's early warning can be improved.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] An intelligent monitoring and early warning system for a power distribution cabinet, comprising:

[0009] The data collection and processing module is used to collect the historical fault data of the specified power distribution cabinet, including actual fault data and false alarm data; and identify the key events that cause the fault to occur based on the actual fault data;

[0010] A parameter setting and management module, used to obtain a set of abnormal change parameters corresponding to each of the key events according to the actual fault data; define a set of initial acquisition parameters for each of the key events, and set a priority and acquisition frequency for each set of the initial acquisition parameters according to the severity and historical occurrence frequency of the key events; set the triggering conditions and impact duration for each of the key events;

[0011] A real-time data acquisition module is used to acquire multiple groups of the initial acquisition parameters in real time according to the priority and the acquisition frequency to obtain multiple groups of real-time initial parameters; determine whether the real-time initial parameters meet the corresponding trigger conditions, and if so, continuously acquire a group of the abnormal change parameters corresponding to the key event within the corresponding impact time to obtain a group of real-time change parameters;

[0012] A dust coverage rate calculation module, used for obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters;

[0013] A risk assessment module, used for calculating a risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate;

[0014] A false alarm model training and identification module, used to obtain a false alarm prediction label according to the real-time changing parameters;

[0015] The early warning module is used to issue an early warning when the risk assessment value is greater than a preset threshold and the false alarm prediction label is not a false alarm.

[0016] Furthermore, the real-time changing parameters include device surface images acquired by a high-resolution camera at corresponding data collection points; the device surface images are surface images of the device for acquiring the corresponding parameters, and each device surface image contains only one device.

[0017] Furthermore, the data collection point is also equipped with lighting equipment.

[0018] Further, obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters includes:

[0019] A group of device surface images corresponding to a group of the real-time changing parameters are input into a trained dust coverage rate prediction model to obtain a group of corresponding dust coverage rates.

[0020] Furthermore, the dust coverage prediction model includes:

[0021] An input layer, used for receiving the device surface image;

[0022] A preprocessing layer, used for graying and removing noise from the device surface image;

[0023] Device area segmentation layer, used to identify and segment device surface areas;

[0024] a dust detection layer for identifying dust areas on the surface area of ​​the device;

[0025] A dust coverage calculation layer, used to calculate the proportion of the dust area in the surface area of ​​the device;

[0026] The output layer is used to output the dust coverage.

[0027] Further, obtaining a set of abnormal change parameters corresponding to each key event includes:

[0028] Multidimensional feature parameters are extracted from the actual fault data, including current, voltage, temperature, humidity, vibration and equipment surface image features, and the multidimensional feature parameters are analyzed using Granger causal analysis to determine the causal relationship between the multidimensional feature parameters, identify significant parameter changes caused by the key event, and obtain K significant change parameters. The set of the significant change parameters constitutes a set of abnormal change parameters corresponding to each key event.

[0029] Furthermore, if the real-time initial parameters meet the corresponding trigger conditions, a set of abnormal change parameters corresponding to the key event are continuously collected within the corresponding impact duration, and obtaining a set of real-time change parameters also includes: immediately issuing a potential risk warning to indicate that the key event has occurred.

[0030] Further, calculating the risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate includes:

[0031] Acquire daily maintenance data when the power distribution cabinet is not faulty; extract the normal fluctuation range corresponding to each abnormal change parameter from the daily maintenance data, and establish a baseline reference standard; the baseline reference standard includes the mean, variance and standard deviation of each abnormal change parameter; based on the baseline reference standard, set a threshold for each abnormal change parameter;

[0032] The real-time change parameter that does not exceed the corresponding threshold is used as a safety parameter, and a first risk assessment value of the safety parameter is calculated using the following calculation formula:

[0033]

[0034] in, is the first risk assessment value, is the index of the security parameter, is the total number of the security parameters, is a fixed value, For the The dust coverage weight corresponding to the safety parameter is: For the The dust coverage rate corresponding to the safety parameter; For the The device usage time weight corresponding to the security parameter, For the The device usage time corresponding to each of the security parameters;

[0035] The real-time change parameter exceeding the corresponding threshold is used as a potential risk parameter, and the similarity between each potential risk parameter and the corresponding data in the actual fault data is calculated to obtain Historical risk similarity;

[0036] The second risk assessment value is calculated according to the historical risk similarity, and the calculation formula is:

[0037]

[0038] in, is the second risk assessment value, is the index value of the potential risk parameter, is the total number of potential risk parameters, For the The weight corresponding to each of the potential risk parameters, For the The historical risk similarity corresponding to the potential risk parameters, For the The dust coverage rate corresponding to the potential risk parameters is: For the The equipment usage time weight corresponding to each of the potential risk parameters, For the The device usage time corresponding to each of the potential risk parameters;

[0039] The risk assessment value is calculated according to the first risk assessment value and the second risk assessment value, and the calculation formula is:

[0040]

[0041]

[0042] in, is the risk assessment value of a set of real-time changing parameters, is the weight of the first risk assessment value, is the weight of the second risk assessment value.

[0043] Further, obtaining a false alarm prediction label according to the real-time change parameter includes:

[0044] Training a false alarm recognition model according to the false alarm data includes: identifying false alarm events that cause false alarms according to the false alarm data; dividing the false alarm data into multiple groups of labeled false alarm data according to the false alarm events, each group of labeled false alarm data includes a false alarm event label; dividing the historical fault data into multiple groups of labeled historical fault data according to the key events, each group of labeled historical fault data includes a key event label; integrating multiple groups of labeled historical fault data and multiple groups of labeled historical fault data according to the false alarm event labels and the key event labels to obtain multiple groups of integrated data; dividing each group of the integrated data into a training set and a test set respectively, using the training set to train a random forest model, the random forest model is trained based on feature parameters, the feature parameters including: current, voltage, temperature, humidity, vibration, dust coverage, texture features of the device surface image and color histogram of the device surface image; using the test set to tune the parameters of the random forest model, and after reaching the specified training conditions, the false alarm recognition model is obtained; the output of the false alarm recognition model is a binary classification label, including false alarms and non-false alarms;

[0045] The key event corresponding to the real-time change parameter is identified, and according to the key event, the real-time change parameter is input into the trained corresponding false alarm recognition model.

[0046] Furthermore, when the risk assessment value is greater than the preset threshold and the false alarm prediction label is not a false alarm, issuing an early warning further includes:

[0047] The triggering result of the key event, the corresponding real-time change parameter, the risk assessment value and the response measures taken by the system are stored in the historical fault data for subsequent analysis and system optimization.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention proposes a dust coverage prediction model. First, the preprocessing layer performs grayscale processing and noise removal on the input image, providing high-quality input data for subsequent image analysis. The introduction of the device area segmentation layer ensures that the system can accurately identify and segment the surface area of ​​the device, eliminate background interference, thereby improving the accuracy of dust detection and reducing unnecessary computational burden. The subsequent dust detection layer focuses on the segmented device surface area, further extracts dust-related features, and ensures accurate identification of dust areas. The dust coverage calculation layer integrates the processing results of the previous layers, accurately calculates the dust coverage, and finally outputs stable and reliable results through the output layer. The dust accumulation on the surface of the equipment can be effectively identified, making the intelligent monitoring of the distribution cabinet more reliable.

[0050] 2. The intelligent monitoring and early warning system of the power distribution cabinet has achieved more accurate risk assessment and early warning by introducing multi-dimensional characteristic parameters and multi-level risk assessment mechanisms, combining historical fault data with equipment usage, and comprehensively considering multiple factors such as equipment operating parameters, dust coverage, and equipment usage time. The system not only improves the accuracy of fault prediction, but also effectively reduces false alarms caused by environmental factors or equipment aging. In addition, the weight setting enables the system to flexibly adjust the impact of each parameter in risk assessment, enhancing the accuracy and adaptability of the assessment; by highlighting key parameters and their corresponding equipment, reducing the interference of minor factors, and significantly reducing the probability of false alarms. At the same time, dynamic adjustment of weights ensures that the system adapts to equipment status and environmental changes in a timely manner, maintains efficient and accurate monitoring and early warning capabilities, and ultimately ensures the safe and stable operation of the power distribution cabinet.

[0051] 3. The present invention trains the integrated data containing different events into multiple independent false alarm identification models, so that each key event can be specifically identified for its specific abnormal change parameters and characteristics. The independently trained model can more accurately capture the specific pattern of each event, avoid feature confusion between different events, and thus improve the accuracy of recognition. Since different events may have similar parameter characteristics but different reasons, separate event training can effectively avoid the risk of the model confusing the characteristics of different events, thereby ensuring that the model is more accurate in identifying false alarms. Independently trained models can better focus on the characteristics and behavior patterns of a certain type of event, improve learning effects, and optimize prediction performance. Separate event training also facilitates independent optimization and tuning of each event, so that the model can better adapt to the characteristics of its respective data sets and improve generalization capabilities. Through accurate false alarm identification, the system can significantly reduce the false alarm rate, improve the accuracy and timeliness of fault warnings, and ensure the safety of distribution cabinet operation and the overall reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A structural diagram of an intelligent monitoring and early warning system for a power distribution cabinet provided by an embodiment of the present invention;

[0053] Figure 2 A structure diagram of a dust coverage prediction model provided by an embodiment of the present invention;

[0054] Figure 3 A flow chart of an intelligent monitoring and early warning method for a power distribution cabinet provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] 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.

[0056] Embodiment 1

[0057] Plant A is a large manufacturing plant with a complex power distribution system, which includes multiple high-voltage distribution cabinets responsible for supplying power to different production lines. Due to the long-term operation of the distribution cabinets and the complex industrial environment, the equipment inside the distribution cabinets is easily affected by the environment, such as temperature rise, humidity fluctuations, and dust accumulation, which may cause electrical failures. In order to ensure the continuous and stable operation of the production line, an intelligent monitoring and early warning system for distribution cabinets is applied to monitor the distribution cabinets in real time and issue early warnings.

[0058] An intelligent monitoring and early warning system for power distribution cabinets, such as Figure 1 As shown, including:

[0059] The data collection and processing module is used to collect the historical fault data of the specified power distribution cabinet, including actual fault data and false alarm data; and identify the key events that cause the fault to occur based on the actual fault data;

[0060] Critical events refer to specific events that cause equipment failure or trigger abnormal conditions during the operation of the distribution cabinet. Through in-depth analysis and evaluation of historical fault data, technicians can identify these critical events, which are the root causes of faults caused by specific parameter changes or environmental factors. The purpose of identifying critical events is to focus on similar situations in future monitoring processes, thereby improving the accuracy and timeliness of fault warnings and ensuring the safe and stable operation of the distribution cabinet. In this embodiment, critical events include voltage surges, equipment overheating, abnormal equipment vibration, excessive humidity, extreme temperature changes in high or low temperature environments, exposure to corrosive gases in the environment, and liquid leakage. These critical events vary depending on the specific parameters monitored. The system identifies these critical events by analyzing historical fault data so that similar events can be warned in future monitoring, and timely measures can be taken to avoid the recurrence of faults.

[0061] The parameter setting and management module is used to obtain a set of abnormal change parameters corresponding to each of the key events based on the actual fault data; in this embodiment, based on the collected actual fault data, the system will identify a set of abnormal change parameters corresponding to each key event (i.e., the specific cause of the equipment failure). These parameters are monitoring data due to abnormalities in specific key events when the failure occurs. For example, if the key event is "a certain device overheats". When the device overheats due to poor heat dissipation or excessively high ambient temperature, the following abnormal changes may occur:

[0062] Temperature parameters: Due to overheating of the device, the temperature sensor on the surface of the device will record abnormally high temperature values.

[0063] Current parameters: Overheating of the device causes the internal resistance to increase, which causes the current value of the device to increase or fluctuate abnormally.

[0064] Voltage parameters: Overheating affects the voltage stability of the equipment, causing abnormal voltage fluctuations or drops.

[0065] Vibration parameters: Overheating also causes thermal expansion and contraction of internal components of the equipment, causing abnormal mechanical vibration of the equipment.

[0066] A set of initial collection parameters is defined for each of the key events, and a priority and collection frequency are set for each set of the initial collection parameters according to the severity and historical occurrence frequency of the key events; a trigger condition and impact duration are set for each of the key events;

[0067] For the initial acquisition parameters, the system needs to predetermine which specific monitoring parameters should be collected and analyzed in order to determine whether the key event has occurred. For example, the initial acquisition parameters for the key event of equipment overheating include: the temperature, current, voltage of the equipment surface, and the speed of the equipment's cooling fan, etc. The initial acquisition parameters for the key event of voltage swell include: input voltage, output voltage, and current load conditions, etc.

[0068] A real-time data acquisition module is used to acquire multiple groups of the initial acquisition parameters in real time according to the priority and the acquisition frequency to obtain multiple groups of real-time initial parameters; determine whether the real-time initial parameters meet the corresponding trigger conditions, and if so, continuously acquire a group of the abnormal change parameters corresponding to the key event within the corresponding impact time to obtain a group of real-time change parameters;

[0069] After a critical event is identified and triggered, the system will start real-time monitoring based on the preset list of abnormal change parameters. During this period, the system will continuously collect data through the corresponding sensors for each abnormal change parameter within the specified impact duration. The acquired data are the real-time change parameters. Since the real-time change parameters are directly derived from the specified abnormal change parameters, the two have a one-to-one correspondence in terms of collection objects, collection methods, and data types. Ultimately, through this real-time monitoring process, the real-time change parameters can accurately reflect the changes in equipment status after a critical event occurs, ensuring that the system can make risk assessment and early warning decisions in a timely manner.

[0070] As shown in Table 1, Table 1 is the various indicators obtained from the actual fault data, among which, key events refer to specific events that may cause equipment failures or abnormal conditions; initial acquisition parameters refer to parameters that the system needs to collect in real time during the monitoring process to determine whether key events have occurred; the impact duration is the time period for continuous monitoring of abnormal change parameters, which is used to ensure that the impact of key events is fully captured. In this table, "trigger conditions (part)" refers to a part of the main conditions in the initial acquisition parameters used to determine whether a key event has occurred. Because the trigger conditions of some key events are more complex, multiple conditions may be involved in actual applications to determine the occurrence of an event. Therefore, only some exemplary conditions are listed in the table, and do not cover all possible trigger conditions.

[0071] Table 1. Example table of key event monitoring and response parameters

[0072]

[0073] A dust coverage rate calculation module, used for obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters;

[0074] A risk assessment module, used for calculating a risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate;

[0075] A false alarm model training and identification module, used to obtain a false alarm prediction label according to the real-time changing parameters;

[0076] The early warning module is used to issue an early warning when the risk assessment value is greater than a preset threshold and the false alarm prediction label is not a false alarm.

[0077] Furthermore, the real-time changing parameters include device surface images acquired by a high-resolution camera at corresponding data collection points; the device surface images are surface images of the device for acquiring the corresponding parameters, and each device surface image contains only one device.

[0078] Each real-time changing parameter corresponds to a specific parameter acquisition device. For example, humidity corresponds to a humidity sensor, and the humidity parameter can be obtained through the humidity sensor. The high-resolution camera in the system takes an image of the surface of the humidity sensor, which only contains the humidity sensor, and inputs it into the system as a real-time changing parameter.

[0079] Acquiring the surface image of the parameter acquisition device can not only provide numerical data on the operating status of the device, but also capture the appearance of the device, such as dust accumulation and other external factors that may affect the normal operation of the device. This appearance information may be ignored when relying solely on parameter data. By combining the device surface image with the parameter data, the system can more comprehensively evaluate the health of the equipment and improve the accuracy of fault diagnosis. In addition, the device surface image can help identify and eliminate false alarms caused by environmental factors (such as dust blocking the sensor), thereby further enhancing the reliability and accuracy of the system. The device surface image corresponds one-to-one with the device of the corresponding parameter, ensuring the accurate match of the image data and the parameter data, and avoiding data confusion caused by the inclusion of multiple devices in the image. This precise mapping relationship helps to more effectively identify and diagnose equipment failures, improve the monitoring accuracy of the system and the accuracy of early warning.

[0080] Furthermore, the data collection point is also equipped with lighting equipment.

[0081] Data collection points equipped with lighting equipment can ensure that the image of the equipment surface can be clearly captured under various lighting conditions, especially in poorly lit or dark environments, which is crucial for accurately monitoring the appearance of the equipment. The introduction of lighting equipment can reduce image blur or information loss caused by insufficient light, ensure that the details of the equipment surface are clearly visible, and thus improve the system's ability to detect dust accumulation.

[0082] Further, obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters includes:

[0083] A group of device surface images corresponding to a group of the real-time changing parameters are input into a trained dust coverage rate prediction model to obtain a group of corresponding dust coverage rates.

[0084] By inputting a set of equipment surface images corresponding to real-time changing parameters into a trained dust coverage prediction model, the dust coverage of the equipment surface can be calculated quickly and accurately. Using this model, dust accumulation can be monitored in real time without interfering with the normal operation of the equipment, laying a data foundation for the subsequent calculation of risk assessment values.

[0085] Furthermore, if Figure 2 As shown, the dust coverage prediction model includes:

[0086] An input layer, used for receiving the device surface image;

[0087] A preprocessing layer, used for graying and removing noise from the device surface image;

[0088] Device area segmentation layer, used to identify and segment device surface areas;

[0089] a dust detection layer for identifying dust areas on the surface area of ​​the device;

[0090] A dust coverage calculation layer, used to calculate the proportion of the dust area in the surface area of ​​the device;

[0091] The output layer is used to output the dust coverage.

[0092] As a feasible method in this embodiment, the actual fault data and the false alarm data both contain their corresponding device surface images when collected; firstly, the dust coverage rate of each device surface image in the actual fault data and the false alarm data is annotated to obtain an annotated device surface data set; the annotation process is specifically as follows: Step 1: Annotate the device surface area to determine the image part that the model needs to focus on. This step can help the model ignore irrelevant background areas (such as the ground and walls). Step 2: In the device surface area, mark the specific coverage area of ​​dust.

[0093] The preprocessing layer includes a 5x5 Gaussian convolution layer, which is used to smooth the image and remove noise in the image.

[0094] The device area segmentation layer is used to identify and segment the device surface area and exclude the background (such as the ground, walls, etc.).

[0095] Includes: Convolution block 1: includes 3 3x3 convolution layers, each with 32 filters, ReLU activation function, followed by 1 2x2 maximum pooling layer (MaxPooling); Convolution block 2: includes 3 3x3 convolution layers, each with 64 filters, ReLU activation function, followed by 1 2x2 maximum pooling layer; Output layer: uses 1 1x1 convolution layer for channel compression, and uses Sigmoid activation function to output the binary mask of the device surface area.

[0096] A dust detection layer is used to identify dust areas on the surface area of ​​the device.

[0097] Includes: Convolutional block 3: includes 3 3x3 convolutional layers, each with 64 filters, ReLU activation function, followed by 1 2x2 maximum pooling layer for extracting dust-related features; Convolutional block 4: includes 3 3x3 convolutional layers, each with 128 filters, ReLU activation function, for capturing more complex dust patterns; Output layer: uses 1 1x1 convolutional layer for channel compression, and uses Sigmoid activation function to output the binary mask of the dust area.

[0098] The dust coverage calculation layer is used to calculate the proportion of the dust area in the surface area of ​​the device, including: an area calculation module: calculating the area of ​​the dust area by counting the number of activated pixels in the dust area mask.

[0099] Coverage calculation module: Divide the dust area by the equipment surface area to obtain the dust coverage.

[0100] The output layer is used to output the dust coverage rate, including: a fully connected layer and an activation function. The fully connected layer is used to convert the dust coverage rate value into the final output. Multiple fully connected layers are selected for regression output. The activation function is used to output the dust coverage rate.

[0101] By analyzing the surface image of the equipment, the dust coverage prediction model can accurately identify and calculate the dust coverage on the equipment surface, thereby providing important information about the cleanliness of the equipment surface. By introducing steps such as preprocessing, region segmentation, and dust detection, the model can effectively eliminate noise and interference in the image to ensure the accuracy of dust detection. The calculated dust coverage can help the system detect potential problems caused by dust accumulation earlier and avoid equipment failures or false alarms caused by dust coverage. By providing a quantitative dust coverage indicator, the system can better evaluate the surface condition of the equipment and support more accurate preventive maintenance and fault warning decisions, thereby significantly improving the reliability and warning capabilities of the system.

[0102] Further, obtaining a set of abnormal change parameters corresponding to each key event includes:

[0103] Extract multi-dimensional characteristic parameters from the actual fault data, including current, voltage, temperature, humidity, vibration and equipment surface image features, and use Granger causal analysis to analyze the multi-dimensional characteristic parameters, determine the causal relationship between the multi-dimensional characteristic parameters, identify the significant parameter changes caused by the key events, and obtain K significant change parameters. In this embodiment, K 1. The set of the significant change parameters constitutes a set of the abnormal change parameters corresponding to each key event.

[0104] By introducing Granger causal analysis to analyze multi-dimensional characteristic parameters, the system can accurately identify the causal relationship between each characteristic parameter, and then accurately determine the significant parameter changes caused by key events. This method not only improves the accuracy of abnormal change parameter identification, but also effectively filters irrelevant or noisy data to ensure that the extracted abnormal change parameters are highly correlated with key events. Ultimately, this causal relationship-based parameter extraction method helps to more accurately locate potential risk factors, improve the reliability and accuracy of fault warnings, and thus enhance the safety of distribution cabinets and the stability of the system.

[0105] Furthermore, if the real-time initial parameters meet the corresponding trigger conditions, a set of abnormal change parameters corresponding to the key event are continuously collected within the corresponding impact duration, and obtaining a set of real-time change parameters also includes: immediately issuing a potential risk warning to indicate that the key event has occurred.

[0106] By immediately issuing a potential risk warning when the real-time initial parameters meet the trigger conditions, the present invention can quickly respond to critical events at an early stage of occurrence and prompt critical events that have occurred. In this way, preventive measures can be taken in time to reduce potential damage before abnormal change parameters further evolve into serious faults. This mechanism ensures early identification and handling of abnormal situations, improves the response speed and safety of the system, thereby effectively reducing the risk of accidents and ensuring the safe operation of the distribution cabinet.

[0107] Further, calculating the risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate includes:

[0108] Acquire daily maintenance data when the power distribution cabinet is not faulty; extract the normal fluctuation range corresponding to each abnormal change parameter from the daily maintenance data, and establish a baseline reference standard; the baseline reference standard includes the mean, variance and standard deviation of each abnormal change parameter; based on the baseline reference standard, set a threshold for each abnormal change parameter;

[0109] The real-time change parameter that does not exceed the corresponding threshold is used as a safety parameter, and a first risk assessment value of the safety parameter is calculated using the following calculation formula:

[0110]

[0111] in, is the first risk assessment value, is the index of the security parameter, is the total number of the security parameters, is a fixed value, For the The dust coverage weight corresponding to the safety parameter is: For the The dust coverage rate corresponding to the safety parameter; For the The device usage time weight corresponding to the security parameter, For the The device usage time corresponding to each of the security parameters;

[0112] The real-time change parameter exceeding the corresponding threshold is used as a potential risk parameter, and the similarity between each potential risk parameter and the corresponding data in the actual fault data is calculated to obtain Historical risk similarity;

[0113] The second risk assessment value is calculated according to the historical risk similarity, and the calculation formula is:

[0114]

[0115] in, is the second risk assessment value, is the index value of the potential risk parameter, is the total number of potential risk parameters, For the The weight corresponding to each of the potential risk parameters, For the The historical risk similarity corresponding to the potential risk parameters, For the The dust coverage rate corresponding to the potential risk parameters is: For the The equipment usage time weight corresponding to each of the potential risk parameters, For the The device usage time corresponding to each of the potential risk parameters;

[0116] The risk assessment value is calculated according to the first risk assessment value and the second risk assessment value, and the calculation formula is:

[0117]

[0118]

[0119] in, is the risk assessment value of a set of real-time changing parameters, is the weight of the first risk assessment value, is the weight of the second risk assessment value.

[0120] The present invention realizes accurate evaluation of the operating status of the power distribution cabinet by introducing a multi-level risk assessment mechanism and combining the weighted calculation of safety parameters and potential risk parameters. By comparing the real-time changing parameters with historical data, the system can effectively identify potential risk factors. In addition, key influencing factors such as the dust coverage rate and usage time of the equipment are also taken into consideration during the evaluation process, further improving the accuracy of risk assessment. The dust coverage rate reflects the impact of the surface condition and environment of the equipment on the operation of the equipment. By monitoring the dust coverage rate, the system can identify potential problems caused by dust accumulation in advance, thereby preventing the occurrence of failures. The use time of the equipment is another important factor. As the use time increases, the equipment components used to obtain parameters may age and wear, resulting in an increased risk of failure. Therefore, incorporating the use time into the evaluation helps to more accurately judge the health status of the equipment and predict the possibility of failure. By comprehensively considering multiple key parameters in the operation of the equipment and assigning different weights, the system can flexibly adjust the contribution of each parameter in the overall risk assessment, making the identification of potential failures more accurate, effectively improving the early warning capability of the system, and ensuring the safe operation of the equipment.

[0121] Further, obtaining a false alarm prediction label according to the real-time change parameter includes:

[0122] Training a false alarm identification model according to the false alarm data includes:

[0123] Identify the false alarm events that cause the false alarm according to the false alarm data; divide the false alarm data into multiple groups of labeled false alarm data according to the false alarm events, each group of labeled false alarm data includes a false alarm event label; divide the historical fault data into multiple groups of labeled historical fault data according to the key events, each group of labeled historical fault data includes a key event label; integrate multiple groups of labeled historical fault data and multiple groups of labeled historical fault data according to the false alarm event labels and the key event labels to obtain multiple groups of integrated data; divide each group of the integrated data into a training set and a test set respectively, use the training set to train a random forest model, the random forest model is trained based on feature parameters, and the feature parameters include: current, voltage, temperature, humidity, vibration, dust coverage, texture features of the device surface image and color histogram of the device surface image; use the test set to tune the parameters of the random forest model, and obtain the false alarm recognition model after reaching the specified training conditions; the output of the false alarm recognition model is a binary classification label, including false alarms and non-false alarms;

[0124] In this embodiment, the training conditions are specified as follows: the model error is reduced to less than 0.01 within 100 iterations, and the error change does not exceed 0.001 in 10 consecutive iterations.

[0125] The key event corresponding to the real-time change parameter is identified, and according to the key event, the real-time change parameter is input into the trained corresponding false alarm recognition model.

[0126] In this embodiment, the false alarm model training and identification module includes a false alarm model training unit and a false alarm identification unit;

[0127] The false alarm model training unit is used to train the false alarm recognition model according to the false alarm data;

[0128] The false alarm identification unit is used to identify the key event corresponding to the real-time change parameter, and according to the key event, input the real-time change parameter into the trained corresponding false alarm identification model.

[0129] The present invention trains the integrated data containing different events into multiple independent false alarm identification models, so that each key event can be specifically identified for its specific abnormal change parameters and characteristics. This independent training method helps to more accurately capture the specific pattern of each event, reduce feature confusion between events, and improve recognition accuracy. Event-based training can prevent the model from being confused when processing events with similar features but different causes, making the model more accurate in false alarm identification. Through this method, the model focuses more on the characteristics of specific events, optimizes the learning effect and prediction performance, thereby effectively reducing the system's false alarm rate and improving the accuracy of fault warnings.

[0130] Furthermore, when the risk assessment value is greater than the preset threshold and the false alarm prediction label is not a false alarm, issuing an early warning further includes:

[0131] The triggering result of the key event, the corresponding real-time change parameter, the risk assessment value and the response measures taken by the system are stored in the historical fault data for subsequent analysis and system optimization.

[0132] This data storage and recording mechanism provides rich data support for subsequent fault analysis and system optimization. By analyzing these records, the system can continuously optimize the warning threshold, false alarm identification model and response measures, thereby improving the intelligence and responsiveness of the entire system, further reducing the occurrence of future faults, improving the accuracy and timeliness of warnings, and enhancing the safety and reliability of distribution cabinet operation.

[0133] In this embodiment, the real-time change parameter is collected based on the abnormal change parameter and is a real-time parameter corresponding to each abnormal change parameter one by one.

[0134] In this embodiment, there are 580 distribution cabinets in Factory A. In order to verify the effect of the false alarm identification model in the intelligent monitoring and early warning system of the distribution cabinets of the present invention, these distribution cabinets are randomly divided into two groups, each containing 290 distribution cabinets. The distribution cabinets in Group 1 use this system for intelligent monitoring and early warning; the other group of distribution cabinets (Group 2) uses a system that does not contain a false alarm identification model (other designs are the same as this system). The operating conditions of the two groups of distribution cabinets are monitored and recorded over a period of time, and the actual number of faults, number of false alarms, and fault identification rate of each group of distribution cabinets in key events are counted, and the average data of the two groups of distribution cabinets are calculated.

[0135] Table 2. Comparative experimental results of fault identification and early warning of power distribution cabinet

[0136]

[0137] By comparing the data of Group 1 and Group 2 in Table 2, it can be seen that the system including the false alarm recognition model significantly reduces the number of false alarms and improves the accuracy of fault recognition. The number of successfully identified faults and the average fault recognition rate are better than those of the system without the false alarm recognition model. This shows that the introduction of the false alarm recognition model can effectively improve the performance of the intelligent monitoring and early warning system of the distribution cabinet, making it more reliable and accurate.

[0138] The intelligent monitoring and early warning system of the power distribution cabinet can accurately identify and monitor abnormal change parameters related to key events through the event-driven design concept, greatly improving the accuracy and timeliness of fault warning. Based on historical fault data and false alarm data, the system customizes initial acquisition parameters and real-time monitoring plans for each key event, ensuring efficient use of resources and accurate monitoring of key parameters. In addition, by introducing a false alarm identification model, the system can effectively filter false alarm information, reduce unnecessary alarms, and improve the reliability of early warnings. The linkage of the real-time data acquisition module, the dust coverage calculation module, and the risk assessment module enables the system to have dynamic adjustment capabilities under different working conditions, realize flexible evaluation of different parameters and corresponding equipment, and further ensure the safe operation of the power distribution cabinet and the overall stability of the system.

[0139] Embodiment 2

[0140] In Plant B, a large commercial complex with multiple high-rise buildings, the power distribution cabinet is responsible for supplying power to lighting, air conditioning, elevators and office equipment on different floors. Due to the wide distribution of power distribution cabinets and the difficulty of maintenance, coupled with the frequent start and stop of the air conditioning system, the current and temperature fluctuate greatly, which can easily cause electrical failures. In order to ensure the safe operation of the commercial complex, Plant B applied a distribution cabinet intelligent monitoring and early warning method and deployed a distribution cabinet intelligent monitoring and early warning system to monitor the operating status of each distribution cabinet in real time, especially for abnormal detection of key events during the operation of the air conditioning system, and timely warn of potential risks to avoid safety hazards and economic losses caused by electrical failures.

[0141] An intelligent monitoring and early warning system for a power distribution cabinet, comprising:

[0142] The data collection and processing module is used to collect the historical fault data of the specified power distribution cabinet, including actual fault data and false alarm data; and identify the key events that cause the fault to occur based on the actual fault data;

[0143] A parameter setting and management module, used to obtain a set of abnormal change parameters corresponding to each of the key events according to the actual fault data; define a set of initial acquisition parameters for each of the key events, and set a priority and acquisition frequency for each set of the initial acquisition parameters according to the severity and historical occurrence frequency of the key events; set the triggering conditions and impact duration for each of the key events;

[0144] A real-time data acquisition module is used to acquire multiple groups of the initial acquisition parameters in real time according to the priority and the acquisition frequency to obtain multiple groups of real-time initial parameters; determine whether the real-time initial parameters meet the corresponding trigger conditions, and if so, continuously acquire a group of the abnormal change parameters corresponding to the key event within the corresponding impact time to obtain a group of real-time change parameters;

[0145] A dust coverage rate calculation module, used for obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters;

[0146] A risk assessment module, used for calculating a risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate;

[0147] A false alarm model training and identification module, used to obtain a false alarm prediction label according to the real-time changing parameters;

[0148] The early warning module is used to issue an early warning when the risk assessment value is greater than a preset threshold and the false alarm prediction label is not a false alarm.

[0149] Furthermore, the real-time changing parameters include device surface images acquired by a high-resolution camera at corresponding data collection points; the device surface images are surface images of the device for acquiring the corresponding parameters, and each device surface image contains only one device.

[0150] Furthermore, the data collection point is also equipped with lighting equipment.

[0151] Further, obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters includes:

[0152] A group of device surface images corresponding to a group of the real-time changing parameters are input into a trained dust coverage rate prediction model to obtain a group of corresponding dust coverage rates.

[0153] Furthermore, the dust coverage prediction model includes:

[0154] An input layer, used for receiving the device surface image;

[0155] A preprocessing layer, used for graying and removing noise from the device surface image;

[0156] Device area segmentation layer, used to identify and segment device surface areas;

[0157] a dust detection layer for identifying dust areas on the surface area of ​​the device;

[0158] A dust coverage calculation layer, used to calculate the proportion of the dust area in the surface area of ​​the device;

[0159] The output layer is used to output the dust coverage.

[0160] Further, obtaining a set of abnormal change parameters corresponding to each key event includes:

[0161] Multidimensional feature parameters are extracted from the actual fault data, including current, voltage, temperature, humidity, vibration and equipment surface image features, and the multidimensional feature parameters are analyzed using Granger causal analysis to determine the causal relationship between the multidimensional feature parameters, identify significant parameter changes caused by the key event, and obtain K significant change parameters. The set of the significant change parameters constitutes a set of abnormal change parameters corresponding to each key event.

[0162] Furthermore, if the real-time initial parameters meet the corresponding trigger conditions, a set of abnormal change parameters corresponding to the key event are continuously collected within the corresponding impact duration, and obtaining a set of real-time change parameters also includes: immediately issuing a potential risk warning to indicate that the key event has occurred.

[0163] Further, calculating the risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate includes:

[0164] Acquire daily maintenance data when the power distribution cabinet is not faulty; extract the normal fluctuation range corresponding to each abnormal change parameter from the daily maintenance data, and establish a baseline reference standard; the baseline reference standard includes the mean, variance and standard deviation of each abnormal change parameter; based on the baseline reference standard, set a threshold for each abnormal change parameter;

[0165] The real-time change parameter that does not exceed the corresponding threshold is used as a safety parameter, and a first risk assessment value of the safety parameter is calculated using the following calculation formula:

[0166]

[0167] in, is the first risk assessment value, is the index of the security parameter, is the total number of the security parameters, is a fixed value, For the The dust coverage weight corresponding to the safety parameter is: For the The dust coverage rate corresponding to the safety parameter; For the The device usage time weight corresponding to the security parameter, For the The device usage time corresponding to each of the security parameters;

[0168] The real-time change parameter exceeding the corresponding threshold is used as a potential risk parameter, and the similarity between each potential risk parameter and the corresponding data in the actual fault data is calculated to obtain Historical risk similarity;

[0169] The second risk assessment value is calculated according to the historical risk similarity, and the calculation formula is:

[0170]

[0171] in, is the second risk assessment value, is the index value of the potential risk parameter, is the total number of potential risk parameters, For the The weight corresponding to each of the potential risk parameters, For the The historical risk similarity corresponding to the potential risk parameters, For the The dust coverage rate corresponding to the potential risk parameters is: For the The equipment usage time weight corresponding to each of the potential risk parameters, For the The device usage time corresponding to each of the potential risk parameters;

[0172] The risk assessment value is calculated according to the first risk assessment value and the second risk assessment value, and the calculation formula is:

[0173]

[0174]

[0175] in, is the risk assessment value of a set of real-time changing parameters, is the weight of the first risk assessment value, is the weight of the second risk assessment value.

[0176] Further, obtaining a false alarm prediction label according to the real-time change parameter includes:

[0177] Training a false alarm identification model according to the false alarm data includes:

[0178] Identify the false alarm events that cause the false alarm according to the false alarm data; divide the false alarm data into multiple groups of labeled false alarm data according to the false alarm events, each group of labeled false alarm data includes a false alarm event label; divide the historical fault data into multiple groups of labeled historical fault data according to the key events, each group of labeled historical fault data includes a key event label; integrate multiple groups of labeled historical fault data and multiple groups of labeled historical fault data according to the false alarm event labels and the key event labels to obtain multiple groups of integrated data; divide each group of the integrated data into a training set and a test set respectively, use the training set to train a random forest model, the random forest model is trained based on feature parameters, and the feature parameters include: current, voltage, temperature, humidity, vibration, dust coverage, texture features of the device surface image and color histogram of the device surface image; use the test set to tune the parameters of the random forest model, and obtain the false alarm recognition model after reaching the specified training conditions; the output of the false alarm recognition model is a binary classification label, including false alarms and non-false alarms;

[0179] The key event corresponding to the real-time change parameter is identified, and according to the key event, the real-time change parameter is input into the trained corresponding false alarm recognition model.

[0180] Furthermore, when the risk assessment value is greater than the preset threshold and the false alarm prediction label is not a false alarm, issuing an early warning further includes:

[0181] The triggering result of the key event, the corresponding real-time change parameter, the risk assessment value and the response measures taken by the system are stored in the historical fault data for subsequent analysis and system optimization.

[0182] In this embodiment, a method for intelligent monitoring and early warning of power distribution cabinets applied in Plant B is as follows: Figure 3 shown.

[0183] The intelligent monitoring and early warning system for power distribution cabinets adopts an event-driven approach. First, the system can identify key events based on actual fault data and set initial acquisition parameters and trigger conditions for these events. Through the dynamic acquisition and analysis of real-time data, the system can quickly capture abnormal change parameters when key events occur and calculate risk assessment values ​​based on these parameters. The introduction of the false alarm identification model further reduces the possibility of false alarms and ensures the accuracy of the early warning. In addition, the system also combines historical fault data with real-time change parameters for storage, which is convenient for subsequent analysis and optimization, and improves the system's self-learning and optimization capabilities.

[0184] In order to verify the superiority of the intelligent monitoring and early warning system for distribution cabinets in the event-driven mode of the present invention, a comparative experiment was carried out to compare the performance of the system with that of the traditional comprehensive early warning system in multiple dimensions.

[0185] In the experiment, several distribution cabinets were randomly selected, and the event-driven early warning system and the comprehensive early warning system were applied respectively. The early warning trigger frequency, false alarm rate, response speed, system resource utilization, and fault detection accuracy of the two systems were monitored and recorded. The experiment lasted for three months to ensure the adequacy and representativeness of the data.

[0186] Table 3. Comparative experimental table of event-driven early warning and comprehensive early warning

[0187]

[0188] As shown in Table 3, the experimental results show that the event-driven early warning system is significantly better than the comprehensive early warning system in multiple key performance indicators. The event-driven system has lower warning trigger frequency and false alarm rate, and more efficient system resource utilization and data collection. At the same time, the average response time of the event-driven system is only 1 second, which is much faster than the 5 seconds of the comprehensive early warning system, and the accuracy of abnormal warning is as high as 95%. In addition, the event-driven system has lower average maintenance costs, lower fault missed detection rate, and lighter system load. These results show that the event-driven early warning system significantly reduces system maintenance costs and resource consumption while improving monitoring accuracy and efficiency.

[0189] Unlike previous comprehensive early warning systems, this intelligent monitoring and early warning system for power distribution cabinets uses event-driven methods to manage early warnings, rather than continuous full coverage monitoring of all parameters. Traditional comprehensive early warning systems tend to monitor all parameters at the same time, regardless of their importance, which not only leads to excessive consumption of system resources, but also increases the probability of false alarms. This system achieves more accurate monitoring by focusing on key events and setting initial collection parameters and trigger conditions based on actual fault data.

[0190] The event-driven early warning method allows the system to respond quickly when a critical event occurs, only capturing abnormal change parameters related to the event and calculating the corresponding risk assessment value. This event-driven parameter monitoring method avoids unnecessary monitoring and is more targeted. Compared with the large number of false alarms and redundant data that may appear in traditional comprehensive early warnings, the system introduces a false alarm identification model, which effectively reduces the possibility of false alarms and further improves the accuracy of early warnings. In addition, the system combines historical fault data with real-time change parameters for storage, making subsequent analysis and optimization more convenient, gradually improving the system's self-learning and optimization capabilities, thereby further enhancing the system's intelligence and reliability.

[0191] This method avoids the waste of resources and false alarms that exist in comprehensive early warning systems, making early warnings more efficient and accurate. It is particularly suitable for intelligent monitoring scenarios of distribution boxes that require high reliability and real-time response.

[0192] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring and early warning system for a power distribution cabinet, characterized in that: include: Data collection and processing module, used to collect historical fault data of the specified distribution cabinet, including actual fault data and false alarm data; identifying a key event causing the failure according to the actual failure data; A parameter setting and management module, used for obtaining a set of abnormal change parameters corresponding to each of the key events according to the actual fault data; A set of initial collection parameters is defined for each of the key events, and a priority and collection frequency are set for each set of the initial collection parameters according to the severity and historical occurrence frequency of the key events; a trigger condition and impact duration are set for each of the key events; A real-time data acquisition module, used for acquiring multiple groups of the initial acquisition parameters in real time according to the priority and the acquisition frequency, to obtain multiple groups of real-time initial parameters; Determine whether the real-time initial parameter satisfies the corresponding trigger condition, and if so, continuously collect a set of the abnormal change parameters corresponding to the key event within the corresponding impact duration to obtain a set of real-time change parameters; A dust coverage rate calculation module, used for obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters; A risk assessment module, used for calculating a risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate; A false alarm model training and identification module, used to obtain a false alarm prediction label according to the real-time changing parameters; An early warning module, used for issuing an early warning when the risk assessment value is greater than a preset threshold and the false alarm prediction label is not a false alarm; Further, calculating the risk assessment value according to a set of the real-time changing parameters and the corresponding dust coverage rate includes: Acquire daily maintenance data when the power distribution cabinet is not faulty; extract the normal fluctuation range corresponding to each abnormal change parameter from the daily maintenance data, and establish a baseline reference standard; the baseline reference standard includes the mean, variance and standard deviation of each abnormal change parameter; based on the baseline reference standard, set a threshold for each abnormal change parameter; The real-time change parameter that does not exceed the corresponding threshold is used as a safety parameter, and a first risk assessment value of the safety parameter is calculated using the following calculation formula: Among them, R normal is the first risk assessment value, i is the index of the security parameter, N is the total number of the security parameters, p normal is a fixed value, is the dust coverage weight corresponding to the i-th safety parameter, C i is the dust coverage rate corresponding to the i-th safety parameter; is the device usage time weight corresponding to the i-th security parameter, T i is the device usage time corresponding to the i-th security parameter; Taking the real-time change parameter exceeding the corresponding threshold as a potential risk parameter, calculating the similarity between each potential risk parameter and the corresponding data in the actual fault data, and obtaining M historical risk similarities; The second risk assessment value is calculated according to the historical risk similarity, and the calculation formula is: Among them, R risk is the second risk assessment value, j is the index value of the potential risk parameter, M is the total number of the potential risk parameters, ω j is the weight corresponding to the jth potential risk parameter, p j is the historical risk similarity corresponding to the jth potential risk parameter, C j is the dust coverage rate corresponding to the jth potential risk parameter, is the equipment usage time weight corresponding to the jth potential risk parameter, T j is the equipment usage time corresponding to the jth potential risk parameter; The risk assessment value is calculated according to the first risk assessment value and the second risk assessment value, and the calculation formula is: R=α·R normal +β·R risk ; α+β＝1； Wherein, R is the risk assessment value of a set of real-time changing parameters, α is the weight of the first risk assessment value, and β is the weight of the second risk assessment value; Further, obtaining a false alarm prediction label according to the real-time change parameter includes: Training a false alarm recognition model according to the false alarm data includes: identifying false alarm events that cause false alarms according to the false alarm data; dividing the false alarm data into multiple groups of labeled false alarm data according to the false alarm events, each group of labeled false alarm data includes a false alarm event label; dividing the historical fault data into multiple groups of labeled historical fault data according to the key events, each group of labeled historical fault data includes a key event label; integrating multiple groups of labeled historical fault data and multiple groups of labeled historical fault data according to the false alarm event labels and the key event labels to obtain multiple groups of integrated data; dividing each group of the integrated data into a training set and a test set respectively, using the training set to train a random forest model, the random forest model is trained based on feature parameters, the feature parameters including: current, voltage, temperature, humidity, vibration, dust coverage, texture features of the device surface image and color histogram of the device surface image; using the test set to tune the parameters of the random forest model, and after reaching the specified training conditions, the false alarm recognition model is obtained; the output of the false alarm recognition model is a binary classification label, including false alarms and non-false alarms; The key event corresponding to the real-time change parameter is identified, and according to the key event, the real-time change parameter is input into the trained corresponding false alarm recognition model.

2. According to claim 1, the intelligent monitoring and early warning system for power distribution cabinets is characterized in that: The real-time changing parameters include device surface images acquired by a high-resolution camera at corresponding data acquisition points; the device surface images are surface images of the device for acquiring the corresponding parameters, and each device surface image contains only one device.

3. The intelligent monitoring and early warning system for power distribution cabinet according to claim 2 is characterized in that: The data collection point is also equipped with lighting equipment.

4. The intelligent monitoring and early warning system for power distribution cabinet according to claim 1 is characterized in that: Obtaining a set of corresponding dust coverage rates according to a set of real-time changing parameters includes: A group of device surface images corresponding to a group of the real-time changing parameters are input into a trained dust coverage rate prediction model to obtain a group of corresponding dust coverage rates.

5. The intelligent monitoring and early warning system for power distribution cabinet according to claim 4 is characterized in that: The dust coverage prediction model includes: An input layer, used for receiving the device surface image; A preprocessing layer, used for graying and removing noise from the device surface image; Device area segmentation layer, used to identify and segment device surface areas; a dust detection layer for identifying dust areas on the surface area of ​​the device; A dust coverage calculation layer, used to calculate the proportion of the dust area in the surface area of ​​the device; The output layer is used to output the dust coverage.

6. The intelligent monitoring and early warning system for power distribution cabinet according to claim 1 is characterized in that: Acquiring a set of abnormal change parameters corresponding to each key event includes: Multidimensional feature parameters are extracted from the actual fault data, including current, voltage, temperature, humidity, vibration and equipment surface image features, and the multidimensional feature parameters are analyzed using Granger causal analysis to determine the causal relationship between the multidimensional feature parameters, identify significant parameter changes caused by the key event, and obtain K significant change parameters. The set of the significant change parameters constitutes a set of abnormal change parameters corresponding to each key event.

7. The intelligent monitoring and early warning system for power distribution cabinet according to claim 1 is characterized in that: If the real-time initial parameters meet the corresponding trigger conditions, a set of abnormal change parameters corresponding to the key event are continuously collected within the corresponding impact duration, and obtaining a set of real-time change parameters also includes: immediately issuing a potential risk warning to indicate that the key event has occurred.

8. The intelligent monitoring and early warning system for power distribution cabinet according to claim 1 is characterized in that: When the risk assessment value is greater than the preset threshold and the false alarm prediction label is not a false alarm, issuing an early warning further includes: The triggering result of the key event, the corresponding real-time change parameter, the risk assessment value and the response measures taken by the system are stored in the historical fault data for subsequent analysis and system optimization.

Images