CNN-LSTM deep learning architecture-based switch cabinet insulation state monitoring method and system

Abstract

The invention discloses a CNN-LSTM deep learning architecture-based switch cabinet insulation state monitoring method and system, and the method gives full play to the respective advantages of a CNN convolutional neural network and an LSTM long and short term memory network in data processing, and overcomes the limitation of a single model. By learning historical data and analyzing latest data in real time, the system can give out an insulation degradation early warning prompt in advance, the reliability and timeliness of fault early warning are improved, and switch cabinet fault power failure accidents caused by insulation degradation are remarkably reduced. According to the method, the concentration of various collected gases is processed to generate a unified defect index, and the index can reflect the damage degree of the insulation state of the switch cabinet, so that the data processing flow is simplified, the workload is remarkably reduced, the prediction efficiency is improved, and the practicability and operability of the system are enhanced.

Description

Technical field

[0001] The present invention belongs to the technical field of switchgear insulation state monitoring, and particularly relates to a switchgear insulation state monitoring method and system based on a CNN-LSTM deep learning architecture. Background technology

[0002] As a core device for the power grid to directly face users, metal-enclosed switchgear is widely used in substations to control circuits and play a protective role. Its safe operation directly affects the power supply reliability of substations and is related to the power supply indicators of the power grid. Once a switchgear fails, it will cause large-scale power outages, resulting in major losses to production and life.

[0003] There are many factors that can cause switchgear failures. A large number of failure cases show that insulation defects caused during the manufacturing and assembly of metal devices in the switchgear or generated during operation can cause partial discharges inside the switchgear; long-term exposure of the switchgear to too high or too low temperature environments can lead to deterioration of insulating materials, significantly reducing the insulation performance; deposition of contaminants on the insulation surface will affect its power frequency withstand voltage characteristics. Especially in a humid environment, the electrolyte decomposed from the contaminant layer forms a conductive liquid film on the insulation surface, further reducing the insulation strength of the equipment; during long-term operation, the switchgear will experience insulation deterioration due to partial discharges, resulting in a reduction in electrical insulation strength. Therefore, various conditions need to be considered comprehensively to effectively avoid the occurrence of switchgear failures.

[0004] Currently, existing insulation deterioration monitoring methods such as partial discharge detection and environmental monitoring can reflect the insulation state, but most of them can only provide monitoring results individually, rely on manual judgment, lack multi-parameter fusion calculation and intelligent analysis, and affect the reliability of diagnosis. In contrast, monitoring methods based on neural networks have significant advantages. They can fuse various monitoring data, accurately judge the degree of insulation deterioration by learning historical data patterns, and achieve accurate prediction of faults. The intelligent algorithm not only simplifies the operation process but also improves the accuracy and stability of monitoring through self-optimization, providing a more economical and efficient switchgear insulation state monitoring solution.

[0005] Compared with traditional monitoring methods, the deep learning monitoring method based on neural network has significant advantages. Traditional monitoring mainly relies on preset thresholds or simple linear models, making it difficult to accurately capture the complex changes of insulating deterioration gases in switchgear. Deep learning, by constructing a neural network, can handle the non-linear correlations of multi-dimensional data, automatically extract features and continuously optimize the model. This not only significantly improves the judgment accuracy of the degree of insulating deterioration, but also can predict the degree of insulating deterioration in real time to achieve early warning. In addition, the neural network has an adaptive ability to cope with the uncertainties in environmental changes, ensuring safer maintenance and repair work. Therefore, it is necessary to propose a switchgear insulation state monitoring method and system based on the CNN-LSTM deep learning architecture to solve the above problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a switchgear insulation state monitoring method and system based on the CNN-LSTM deep learning architecture. By comprehensively processing the collected multiple gas concentrations, a unified defect index is generated, which can reflect the damage degree of the switchgear insulation state. In subsequent insulation state prediction, only key parameters such as the defect index, temperature, and humidity need to be input to complete the prediction task. This method not only simplifies the data processing flow, significantly reduces the workload, improves the prediction efficiency, but also enhances the practicability and operability of the system.

[0007] To achieve the above technical effects, the technical solutions adopted by the present invention are as follows: A switchgear insulation state monitoring method based on the CNN-LSTM deep learning architecture includes the following steps: S1, model development and training: According to historical data, establish a training data set; Perform data preprocessing; Establish a CNN-LSTM insulation deterioration degree monitoring model; Use the training data set to train the CNN-LSTM insulation deterioration degree monitoring model to obtain a monitoring model for monitoring the insulation deterioration degree of air switchgear; S2, model deployment and application: Real-time monitor relevant data, and use the monitoring model to monitor the data and output the monitoring results.

[0008] Preferably, in step S1, the data set includes: Gas concentration, temperature, and humidity data inside the air switchgear during normal operation, as well as gas concentration, temperature, and humidity data inside the air switchgear during partial discharge at different insulation defect levels; the gas types include CO, NO2, and NO gases.

[0009] Preferably, in step S1, the data preprocessing includes: Performing data cleaning and data normalization on the gas concentration, temperature, and humidity data; among them, the gas concentrations of CO, NO2, and NO are comprehensively processed to generate a unified defect index to reflect the degree of damage to the insulation state of the switchgear.

[0010] Preferably, the data cleaning includes: Removing outliers, checking whether there are outliers in the data, and correcting or deleting them according to the actual situation; Missing value processing, using the mean filling method to process the missing values in the data. Preferably, Min-Max is used for data normalization: ; Among them, is the original data value, is the minimum data value, is the maximum data value, is the normalized value, with a range of [0, 1].

[0011] Furthermore, the gas types include carbon monoxide (CO), nitrogen dioxide (NO2), and nitric oxide (NO). Since insulating deterioration gases such as CO, NO2, and NO will be released when the air switchgear suffers insulation damage, the monitored target gases can be determined as CO, NO2, and NO to characterize the degree of insulation deterioration of the air switchgear.

[0012] Preferably, when calculating the defect index, first determine the weight distribution of each gas through data analysis, and then calculate the defect index based on the normalized gas concentration and weights: ; Among them, is the defect index, , and are the weight coefficients, , and are the normalized gas concentration values.

[0013] Preferably, in step S1, the CNN-LSTM insulation deterioration degree monitoring model includes an input layer, a CNN-LSTM layer, and an output layer; the input layer includes 3 input nodes, namely the defect index, temperature, and humidity; the CNN-LSTM layer is a convolutional neural network-long short-term memory network layer for feature extraction; the output layer includes N output nodes, and N is equal to the number of insulation defect levels.

[0014] Preferably, the insulation defect levels include five levels: normal, slightly deteriorated, moderately deteriorated, severely deteriorated, and critical.

[0015] Preferably, the data of the data set is collected by gas sensors, temperature sensors, and humidity sensors, and the gas sensors transmit data to the monitoring model through antennas; the gas sensors, temperature sensors, and humidity sensors are installed in the air switchgear to be detected, and the gas concentration, temperature, and humidity inside the switchgear are measured by the sensors.

[0016] Preferably, a switchgear insulation status monitoring system based on the CNN-LSTM deep learning architecture is used to execute the switchgear insulation status monitoring method based on the CNN-LSTM deep learning architecture. The system includes: The switchgear is an air switchgear; The power supply module provides stable power support for the normal operation of the device; The host module specifically includes: a data processing module, an insulation status monitoring module, and an insulation defect recording module; The data acquisition module is used to obtain the environmental data and gas composition information inside the air switchgear in real time, including the concentration changes of CO, NO2, and NO, and at the same time collect the temperature and humidity environmental parameters related to the insulation status; The communication module is responsible for transmitting the collected data to the host module or a remote server and supports multiple wireless communication methods; The insulation defect recording module records the insulation defects found during the monitoring process, including the time when the defect occurred, the gas concentration change situation, and the defect level information.

[0017] The beneficial effects of the present invention are as follows: 1. By collecting environmental data such as gas composition, temperature, and humidity inside the switchgear, the present invention constructs and trains a CNN-LSTM deep learning model, thereby forming a monitoring model capable of identifying the insulation deterioration degree of the switchgear. For the switchgear to be detected, the system first collects its internal environmental data, analyzes it using the trained model, and finally outputs the monitoring result and generates a warning message. Since the monitoring data comes from inside the switchgear, the influence of external gas interference is avoided, effectively improving the accuracy of insulation deterioration detection.

[0018] 2. The present invention adopts an insulation status monitoring method based on the CNN-LSTM deep learning architecture, combining the advantages of CNN in feature extraction and the excellent performance of LSTM in time series modeling. This method overcomes the limitations of a single model in processing complex environmental data, significantly improving the accuracy of insulation deterioration detection. In addition, while optimizing the model training efficiency, this hybrid architecture enhances the system's learning ability for complex data features and is suitable for the multi-scenario requirements of switchgear insulation status monitoring.

[0019] 3. The present invention also proposes a method for preprocessing relevant data to enhance the practicability of the patent and better meet the technical requirements in this field. This method comprehensively processes the collected concentrations of various gases to generate a unified defect index. This index can reflect the degree of damage to the insulation state of the switchgear cabinet. This method not only simplifies the data processing flow, but also significantly reduces the workload, improves the prediction efficiency, and enhances the practicability and operability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a schematic diagram of the architecture of the convolutional neural network CNN of the present invention; Figure 2 is a schematic diagram of the internal structure of the long short-term memory network LSTM mentioned in the present invention; Figure 3 is a schematic diagram of the deep learning framework based on CNN-LSTM proposed by the present invention; Figure 4 is a flowchart of the switchgear cabinet insulation state monitoring method based on the CNN-LSTM deep learning architecture provided in the example of the present invention; Figure 5 is a block diagram of the switchgear cabinet insulation state monitoring system based on the CNN-LSTM deep learning architecture in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Embodiment 1: The switchgear cabinet insulation state monitoring method based on the CNN-LSTM deep learning architecture includes the following steps: S1. Model development and training: According to historical data, establish a training data set; Perform data preprocessing; Establish a CNN-LSTM insulation deterioration degree monitoring model; Use the training data set to train the CNN-LSTM insulation deterioration degree monitoring model to obtain a monitoring model for monitoring the insulation deterioration degree of the air switchgear cabinet; S2. Model deployment and application: Real-time monitor relevant data, and use the monitoring model to monitor the data and output the monitoring results.

[0022] Preferably, in step S1, the data set includes: The gas concentration, temperature, and humidity data inside the air switchgear cabinet during normal operation, as well as the gas concentration, temperature, and humidity data inside the air switchgear cabinet during partial discharge under different insulation defect levels; the gas types include CO, NO2, and NO gases.

[0023] Preferably, in step S1, the data preprocessing includes: Performing data cleaning and data normalization on the gas concentration, temperature, and humidity data; among them, comprehensively processing the CO, NO2, and NO gas concentrations to generate a unified defect index to reflect the degree of damage to the insulation state of the switchgear.

[0024] Preferably, the data cleaning includes: Removing outliers, checking whether there are outliers in the data, and correcting or deleting them according to the actual situation; Missing value processing, using the mean filling method to process the missing values in the data. Preferably, use Min - Max for data normalization: ; Among them, is the original data value, is the minimum data value, is the maximum data value, is the value after normalization, with a range of [0, 1].

[0025] Furthermore, the types of gases include carbon monoxide (CO), nitrogen dioxide (NO2), and nitric oxide (NO). Since insulating deterioration gases such as CO, NO2, and NO will be released when the air switchgear suffers insulation damage, the monitoring target gases can be determined as CO, NO2, and NO to characterize the degree of insulating deterioration of the air switchgear.

[0026] Preferably, when calculating the defect index, first determine the weight distribution of each gas through data analysis, and then calculate the defect index based on the normalized gas concentration and weights: ; Among them, is the defect index, , and are the weight coefficients, , and are the normalized gas concentration values.

[0027] Furthermore, when establishing the CNN - LSTM neural network model, CNN is responsible for extracting local spatial features in the time - series data, while LSTM is good at capturing the long - term dependence relationships of the time - series.

[0028] Such as Figure 1As shown, the network architecture of the one-dimensional CNN includes a Convolutional Layer (CL), a Pooling Layer (PL), and a Fully Connected Layer (FC); among them, the main functions of CL and PL are to extract key local features from the time series and reduce the dimension to reduce data redundancy and improve computational efficiency; through this feature extraction method, the CNN can effectively capture the local patterns of the time series data, provide more representative input features for the subsequent LSTM module, and further model the dynamic characteristics and global dependencies of the time series to achieve better sequence analysis performance; Figure 2 It is a schematic diagram of the internal structure of the long short-term memory network LSTM mentioned in the present invention.

[0029] Figure 3 It is a CNN-LSTM deep learning framework. The one-dimensional CNN performs excellently in time series analysis because its convolutional kernel slides along a fixed direction and can automatically extract the hidden feature patterns on the time axis. Subsequently, the LSTM structure takes the features extracted from the CNN as input. Through the gating mechanism of the LSTM (such as the input gate, forget gate, and output gate), combined with continuously optimized training data, the model can effectively learn the complex relationship between the input and output sequences, thereby establishing an accurate mapping.

[0030] Preferably, in step S1, the CNN-LSTM insulation deterioration degree monitoring model includes an input layer, a CNN-LSTM layer, and an output layer; the input layer includes 3 input nodes, namely the defect index, temperature, and humidity; the CNN-LSTM layer is a convolutional neural network-long short-term memory network layer for feature extraction; the output layer includes N output nodes, and N is equal to the number of insulation defect levels.

[0031] Preferably, the insulation defect levels include five levels: normal, slightly deteriorated, moderately deteriorated, severely deteriorated, and critical.

[0032] Furthermore, normal means that the insulation state of the switchgear is good, and indicators such as insulation resistance, capacitance, and leakage current are within the normal range, and no maintenance is required; slightly deteriorated means that the insulation state is slightly deteriorated, and some parameters exceed the normal range but do not reach the alarm threshold, and regular tracking and monitoring are required; moderately deteriorated means that the insulation material is significantly deteriorated, and multiple insulation indicators exceed the normal range, and it is recommended to plan maintenance or replace relevant components; severely deteriorated means that the insulation state is severely deteriorated, there is a greater risk of failure, and operation must be stopped immediately and emergency maintenance must be carried out; critical means that the insulation condition has deteriorated extremely, all main insulation parameters seriously exceed the standard, the equipment is in a high-risk state, and operation must be stopped immediately and comprehensive maintenance must be carried out to prevent equipment failures or accidents.

[0033] Preferably, the data of the data set is collected by a gas sensor, a temperature sensor, and a humidity sensor, and the data is transmitted between the gas sensor and the monitoring model through an antenna. The gas sensor, the temperature sensor, and the humidity sensor are installed inside the air switchgear to be detected, and the gas concentration, temperature, and humidity inside the switchgear are measured by the sensors.

[0034] Preferably, a switchgear insulation state monitoring system based on the CNN-LSTM deep learning architecture is used to execute the switchgear insulation state monitoring method based on the CNN-LSTM deep learning architecture. The system includes: The switchgear is an air switchgear; The power supply module provides stable power support for the normal operation of the device; an efficient power management technology is adopted to ensure that the device can continue to work under different environmental conditions. The power supply module has overload protection, short-circuit protection, and power storage functions to enhance the safety and stability of the device.

[0035] The host module specifically includes: a data processing module, an insulation state monitoring module, and an insulation defect recording module; the data processing module preprocesses the collected multi-dimensional data, including clearing abnormal data, normalizing, and calculating the defect index, to provide high-quality input data for the subsequent deep learning model; the insulation state monitoring module analyzes the processed data based on the CNN-LSTM deep learning model to evaluate the insulation state of the air switchgear and identify potential insulation defects in real time; the strategy matching module performs corresponding strategy matching according to the output result of the insulation state monitoring module to adjust the state.

[0036] The data acquisition module is used to obtain the environmental data and gas composition information inside the air switchgear in real time, including the concentration changes of CO, NO2, and NO, and at the same time collect the temperature and humidity environmental parameters related to the insulation state; this module adopts high-sensitivity sensors and anti-interference design to ensure the accuracy and reliability of data acquisition.

[0037] The communication module is responsible for transmitting the collected data to the host module or a remote server, supporting multiple wireless communication methods; such as Wi-Fi, Bluetooth, 4G / 5G networks, etc., combined with an antenna to achieve efficient signal transmission and reception. The antenna is an important part of the communication module, used to enhance the signal stability and coverage; it adopts a high-gain, low-noise design to ensure that the device can still maintain a high-quality communication connection in a complex electromagnetic environment, especially suitable for environments with high interference in the power system. In addition, this module supports the two-way communication function, can receive instructions issued by the server, and perform operations such as system update and parameter adjustment to improve the intelligence level of the device.

[0038] The insulation defect recording module records the insulation defects found during the monitoring process, including the time when the defects occur, the gas concentration changes, and the defect level information, providing data support for subsequent maintenance and fault diagnosis; among them, the insulation defect recording device is an SD memory card, and the detection model is on the host computer. Just insert the SD memory card into the host computer.

[0039] Embodiment 2: As Figure 4 shown, the embodiment of the present invention provides a switchgear insulation status monitoring method based on the CNN-LSTM deep learning architecture. The switchgear is an air switchgear, and the method includes the following steps: S1: Collect samples, store and classify the collected samples, so as to establish a training data set. The training data set includes the gas types and contents, temperature, and humidity data in the switchgear during normal operation, as well as the gas types and contents, temperature, and humidity data in the switchgear during partial discharge under different insulation defect degrees. The different insulation defect levels of high-voltage switchgear include five levels: normal, slightly deteriorated, moderately deteriorated, severely deteriorated, and critical; the main gases monitored in the switchgear are CO, NO2, and NO gases. Perform data preprocessing, and perform data cleaning and data normalization on the gas concentration, temperature, and humidity data. Among them, the CO, NO2, and NO gas concentrations are comprehensively processed to generate a unified defect index, which can reflect the damage degree of the switchgear insulation status.

[0040] S2: Construct a CNN-LSTM neural network model; the neural network model includes an input layer, a CNN-LSTM layer, and an output layer. The input layer includes 3 input nodes, namely the defect index, temperature, and humidity; the CNN-LSTM layer is a convolutional neural network-long short-term memory network layer for feature extraction; the output layer includes N output nodes, and N is equal to the number of insulation defect degree levels. In this embodiment, the insulation defect levels are divided into five levels: normal, slightly deteriorated, moderately deteriorated, severely deteriorated, and critical, so N = 5.

[0041] S3: Input the training data set into the CNN-LSTM neural network model, train the CNN-LSTM neural network model, and obtain a monitoring model for monitoring the insulation status of the switchgear.

[0042] S4: Real-time monitor the gas components and contents, temperature, and humidity data in the switchgear to be monitored; In this embodiment, gas sensors, temperature sensors, and humidity sensors are also provided. These sensors are connected to the monitoring model; the sensors are installed in the switchgear to be monitored, and the gas components and contents, temperature, and humidity data in the switchgear are measured by the sensors. The sensors are powered by a power supply module.

[0043] S5: The monitoring model receives the measured gas composition and content, temperature and humidity data, and identifies the measured gas composition and content, thereby outputting the monitoring results.

[0044] S6: Provide an insulation defect recording device to record the insulation defects found during the monitoring process, including the time when the defect occurred, the change in gas concentration, the defect level and other information; regularly check the records of the insulation defect recording device, and maintenance personnel can refer to the records to repair the switch cabinet.

[0045] In step S1, the number of samples collected in the normal state and each insulation defect state should be sufficient to ensure the accuracy of the monitoring model.

[0046] Furthermore, 10-35kV switch cabinets usually use air filling as insulating gas to improve insulation performance. When partial discharge occurs in the air switch cabinet, the insulation material will be damaged and accompanied by obvious temperature rise, followed by the slow generation of CO, CO2, NO, NO2 and other gases. Therefore, CO, NO and NO2 can be used as characteristic components for fault judgment of damaged solid insulating materials in the switch cabinet (because CO2 is a relatively large gas in the air, it cannot be used as a characteristic component for detection of damaged solid insulating materials), and temperature is used as a characteristic parameter for judging the stage of partial discharge fault in the switch cabinet. The present invention collects the gas in the air switch cabinet to avoid interference from external gases, thereby improving the accuracy of the monitoring results.

[0047] In this embodiment, judgment is made through the trained monitoring model to reduce the number of false alarms. Maintenance personnel can grasp the insulation defect information of the switch cabinet and solve the existing safety hazards through the recorded judgment results, reduce the number of 10-35kV switch cabinet trips caused by partial discharge, ensure that the switch cabinet is not harmed by partial discharge caused by insulation defects, and ensure the safety and stability of the switch cabinet, the reliability of power supply and the safe operation of the power grid.

[0048] See also Figure 5 As shown, this embodiment also provides a switch cabinet insulation status monitoring system based on CNN-LSTM neural network, the switch cabinet is a ring network cabinet, which includes a power module, a host module, a data acquisition module, and a communication module, wherein: Install the power module and host module of the insulation monitoring equipment. According to the survey configuration status, determine that the 220V operating power supply of the insulation monitoring equipment is drawn from the instrument box of the voltage transformer cabinet (PT cabinet), and fix the power strip on the top of the PT cabinet, and also fix the host module of the insulation monitoring equipment on the top of the PT cabinet.

[0049] Install the sensor module of the insulation monitoring device. According to the actual status of the ring main unit, layout the host module and sensor module of the insulation monitoring device. The sensor module is installed in the instrument box of the ring main unit and is placed near the left secondary wire threading hole.

[0050] Install the antenna of the insulation monitoring device. Install the antenna near the ventilation grille hole of the ring main unit cabinet.

[0051] The sensor module of the insulation monitoring device monitors the gas concentration, temperature and humidity information inside the ring main unit cabinet and transmits it to the data receiving module of the host module.

[0052] The data receiving module receives the gas concentration, temperature and humidity data inside the switch cabinet transmitted from the monitoring module, processes the data and inputs it into the monitoring module.

[0053] The monitoring model receives the data, outputs the monitoring results and transmits them to the data recording module.

[0054] The data recording module records the insulation defects found during the monitoring process, including information such as the time when the defect occurred, the change in gas concentration, and the defect level.

Claims

1. A method for monitoring the insulation state of switchgear based on the CNN-LSTM deep learning architecture, characterized in that, It includes the following steps: S1, Model Development and Training: Based on historical data, establish a training dataset; Perform data preprocessing; Establish a CNN-LSTM insulation degradation degree monitoring model; Use the training dataset to train the CNN-LSTM insulation degradation degree monitoring model to obtain a monitoring model for monitoring the insulation degradation degree of air switchgear; S2, Model Deployment and Application: Monitor relevant data in real time, and use the monitoring model to monitor the data and output the monitoring results.

2. The method for monitoring the insulation state of a switchgear cabinet based on a CNN-LSTM deep learning architecture according to claim 1, wherein In step S1, the dataset includes: Gas concentration, temperature, and humidity data inside the air switchgear during normal operation, and gas concentration, temperature, and humidity data inside the air switchgear during partial discharge at different insulation defect levels; the gas types include CO, NO2, and NO gases.

3. The method for monitoring the insulation state of a switchgear based on a CNN-LSTM deep learning architecture according to claim 1, characterized in that, In step S1, the data preprocessing includes: Perform data cleaning and data normalization on the gas concentration, temperature, and humidity data; among them, perform comprehensive processing on the CO, NO2, and NO gas concentrations to generate a unified defect index to reflect the damage degree of the switchgear insulation state.

4. The method for monitoring the insulation state of a switchgear cabinet based on a CNN-LSTM deep learning architecture according to claim 3, characterized in that, Data cleaning includes: Remove outliers, check whether there are outliers in the data, and correct or delete them according to the actual situation; Missing value processing, use the mean filling method to process the missing values in the data.

5. The method for monitoring the insulation state of a switchgear based on a CNN-LSTM deep learning architecture according to claim 3, characterized in that, Use Min-Max for data normalization: ； Among them, is the original data value, is the minimum data value, is the maximum data value, is the value after normalization, and the range is [0, 1].

6. The method for monitoring the insulation state of a switchgear based on a CNN-LSTM deep learning architecture according to claim 3, characterized in that When calculating the defect index, first determine the weight distribution of each gas through data analysis, and then calculate the defect index based on the normalized gas concentration and weights: ； Among them, is the defect index, , and are the weight coefficients, , and are the normalized gas concentration values.

7. The switchgear insulation state monitoring method based on the CNN-LSTM deep learning architecture according to claim 1, characterized in that In step S1, the CNN-LSTM insulation degradation degree monitoring model includes an input layer, a CNN-LSTM layer, and an output layer; the input layer includes 3 input nodes, namely the defect index, temperature, and humidity; the CNN-LSTM layer is a convolutional neural network-long short-term memory network layer for feature extraction; the output layer includes N output nodes, and N is equal to the number of insulation defect levels.

8. The insulation state monitoring method of the switchgear based on the CNN-LSTM deep learning architecture according to claim 7, characterized in that, The insulation defect levels include five levels: normal, slightly deteriorated, moderately deteriorated, severely deteriorated, and critical.

9. The method for monitoring the insulation state of a switchgear based on a CNN-LSTM deep learning architecture according to claim 1, characterized in that The data in the dataset is collected through gas sensors, temperature sensors, and humidity sensors, and the gas sensors transmit data to the monitoring model through an antenna; the gas sensors, temperature sensors, and humidity sensors are installed inside the air switchgear to be detected, and the gas concentration, temperature, and humidity inside the switchgear are measured through the sensors.

10. A switchgear insulation condition monitoring system based on a CNN-LSTM deep learning architecture, characterized in that, A system for implementing the switchgear insulation state monitoring method based on the CNN-LSTM deep learning architecture described in claims 1 to 9, the system includes: The switchgear is an air switchgear; The power supply module provides stable power support for the normal operation of the device; The host module specifically includes: a data processing module, an insulation state monitoring module, and an insulation defect recording module; The data acquisition module is used to obtain the environmental data and gas component information inside the air switchgear in real time, including the concentration changes of CO, NO2, and NO, and at the same time collect the temperature and humidity environmental parameters related to the insulation state; The communication module is responsible for transmitting the collected data to the host module or a remote server, supporting multiple wireless communication methods; The insulation defect recording module records the insulation defects found during the monitoring process, including the time when the defect occurred, the change in gas concentration, and the defect level information.

Images