A switch cabinet state monitoring method and device based on multi-sensor space-time calibration and feature fusion

Abstract

The present application relates to the technical field of electrified railway power supply equipment monitoring, in particular to a switch cabinet state monitoring method and device based on multi-sensor space-time calibration and feature fusion, the method comprising: collecting multi-source state data of the switch cabinet through heterogeneous sensors; performing space-time calibration on the data to obtain multi-source synchronous data aligned in space-time; performing feature extraction and multi-scale feature fusion to generate a fusion feature vector; inputting the fusion feature vector into a CNN-LSTM hybrid diagnosis model to output equipment state diagnosis and early warning. The device comprises a sensor module, a data acquisition and processing unit, an encryption communication module, a central diagnosis platform and a power management module. The present application realizes high-precision data alignment through microsecond-level clock synchronization and TDOA space calibration, improves diagnosis accuracy based on attention mechanism-based feature fusion and CNN-LSTM model, and uses national encryption algorithm and key dynamic update to ensure data transmission safety, solving the problems of data asynchronization, unintelligent analysis and unsafe transmission in the prior art.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology for power supply equipment in electrified railways, specifically to a method and device for monitoring the status of switchgear based on multi-sensor spatiotemporal calibration and feature fusion. Background Technology

[0002] With the continuous growth of my country's electrified railway operating mileage and the continuous improvement of train operating speed, the reliability and safety of the power supply system have become the key to ensuring smooth railway transportation. As the core equipment in the railway power supply system, high / low voltage switchgear undertakes important functions such as power distribution, line protection, and equipment control. It operates under complex environmental conditions of high load, strong electromagnetic interference, and large temperature and humidity fluctuations for a long time, which makes it prone to various faults such as partial discharge, contact overheating, insulation aging, and mechanical jamming. In severe cases, it can lead to serious accidents such as equipment burnout and line tripping, affecting the continuity and safety of railway power supply.

[0003] Traditional switchgear maintenance relies mainly on periodic inspections and manual patrols. This model has significant drawbacks: First, it is inefficient, as manual patrols require partial power interruption, demand high levels of expertise from maintenance personnel, have long patrol cycles, and cannot achieve real-time monitoring. Second, it lacks real-time performance, often only able to respond passively after a fault occurs, making early warning difficult. Third, it is prone to missed or false detections, as manual judgment relies on experience and has limited ability to identify hidden faults (such as internal partial discharge or minor contact overheating), making it difficult to meet the needs of modern railway management that emphasizes "prevention first and precise maintenance."

[0004] Currently, domestic railway power supply systems have gradually introduced online monitoring devices, such as single-point temperature monitoring and independent partial discharge detection. However, the following prominent problems generally exist, making it difficult to achieve accurate and comprehensive equipment status monitoring: 1) Monitoring methods are singular, usually targeting only a single physical quantity (such as temperature or partial discharge), lacking the fusion analysis of multi-source heterogeneous information, and unable to comprehensively and objectively assess the overall health status of the equipment; 2) Data analysis relies on human experience, with low levels of intelligence, resulting in high false alarm and false negative rates in fault diagnosis, making it difficult to meet the operation and maintenance needs of large-scale railway power supply equipment; 3) The phenomenon of data silos between systems is serious, with inconsistent data formats and incompatible transmission protocols among different monitoring devices, resulting in weak collaborative monitoring capabilities; 4) Early warning mechanisms mostly use fixed thresholds, which cannot adapt to environmental changes (such as temperature and humidity fluctuations, electromagnetic interference) and feature drift caused by equipment aging, resulting in poor adaptability and easy occurrence of invalid or missed early warnings.

[0005] While there is research abroad on technologies such as multi-sensor fusion, AI diagnostics, and digital twins, the core algorithms and equipment have technical barriers. Furthermore, their adaptability to the complex electromagnetic environment (such as strong electromagnetic interference generated by the traction power supply system) and climatic conditions (such as high temperature, high humidity, and severe cold) of China's railways has not been fully verified, making it difficult to directly apply them to the monitoring scenarios of switchgear in my country's railways.

[0006] Therefore, there is a need for a switchgear status monitoring solution that can achieve accurate synchronization of multi-source data, deep feature fusion, high-security data transmission and intelligent diagnostic capabilities, and is adapted to the complex operating environment of my country's railways. This solution would address the problems of asynchronous monitoring data, unintelligent analysis, insecure transmission, and poor adaptability in existing technologies, and provide technical support for the safe and stable operation of railway power supply systems. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a switchgear status monitoring method and device based on multi-sensor spatiotemporal calibration and feature fusion. The aim is to solve the technical problems in existing technologies, such as monitoring data from different sources, asynchronous time and space, unintelligent analysis, insecure transmission, and poor adaptability. This enables accurate perception, intelligent diagnosis, and proactive early warning of switchgear status, thereby improving the intelligence level and reliability of railway power supply equipment operation and maintenance.

[0008] A switchgear status monitoring method based on multi-sensor spatiotemporal calibration and feature fusion includes the following steps: Step S1: Collect multi-source status data of the switchgear in real time during operation by deploying multiple heterogeneous sensors in key parts of the switchgear; the heterogeneous sensors include at least a self-powered wireless temperature sensor, an ultra-high frequency partial discharge sensor, an ultrasonic partial discharge sensor, a temperature and humidity sensor, and an image acquisition device. Step S2: Perform spatiotemporal calibration on the collected multi-source state data to obtain spatiotemporally aligned multi-source synchronized data; the spatiotemporal calibration includes timestamp synchronization calibration based on a unified clock source, and spatial position calibration based on sensor installation location coordinates and signal propagation speed; Step S3: Perform feature extraction and multi-scale feature fusion on the spatiotemporally aligned multi-source synchronization data to generate a fusion feature vector that can comprehensively characterize the health status of the equipment; Step S4: Input the fused feature vector into the pre-trained deep learning-based state diagnosis model to perform real-time evaluation and fault prediction of the switchgear's operating status, and output diagnostic results and early warning information.

[0009] Further, in step S2, the timestamp synchronization calibration specifically involves: periodically broadcasting synchronization signals to all sensors through a locally deployed high-precision time synchronization module; each sensor calibrates its local clock according to the signal and adds a unified time tag to the collected data; the high-precision time synchronization module supports BeiDou / GPS dual-mode time synchronization with a time synchronization accuracy of ≤1μs and a synchronization signal broadcast period of 100ms; for sensor data with inconsistent sampling frequencies, linear interpolation is used to interpolate to a unified sampling frequency.

[0010] Further, in step S2, the spatial position calibration specifically involves: pre-measuring and recording the three-dimensional spatial coordinates of each sensor; when the UHF partial discharge sensor and the ultrasonic partial discharge sensor simultaneously detect a suspected partial discharge signal, utilizing the difference in propagation speed between the UHF signal and the ultrasonic signal, and combining the coordinates of the two sensors, calculating the specific spatial coordinates of the partial discharge source using the Time Difference of Arrival (TDOA) positioning algorithm; and based on the coordinates of the partial discharge source, correlating and calibrating the temperature data and image data within the same spatial area to eliminate signal misalignment.

[0011] Furthermore, step S3 specifically includes: Step S31: Perform feature extraction on different types of sensor data; extract time-domain features such as mean, maximum, minimum, peak-to-peak value, rate of change, and fluctuation variance from temperature data; extract statistical features such as maximum discharge amplitude, average discharge amplitude, discharge frequency, and discharge phase distribution, as well as texture features of the PRPD spectrum from partial discharge data; extract visual features such as meter readings and insulator appearance from image data using a lightweight convolutional neural network; extract mean, fluctuation range, and trend features from temperature and humidity data. Step S32: An attention-based feature fusion network is used to adaptively weight and fuse feature vectors from different sources. The network first maps the feature vectors from each source to a unified dimension through a fully connected layer, then calculates the attention weights of each source feature through a sigmoid activation function, and finally sums the weighted feature vectors element by element to generate the fused feature vector.

[0012] Furthermore, the deep learning-based state diagnosis model in step S4 is a CNN-LSTM hybrid network, which includes convolutional layers, LSTM layers, and fully connected layers. The convolutional layers are used to extract local correlation features from the fused feature vector, the LSTM layers are used to mine the dynamic change patterns of data over time, and the fully connected layers are used to output the device health score, fault type, fault probability, and remaining life prediction value. The fault diagnosis accuracy of the model is ≥96%, the false alarm rate is ≤4%, and the remaining life prediction error is ≤10%.

[0013] Furthermore, after step S1 and before step S2, the method further includes a step of encrypting and transmitting the collected multi-source state data: encrypting the multi-source state data using a lightweight encryption protocol based on Chinese cryptographic algorithms, with the encryption algorithm being the SM4 block cipher algorithm; the encrypted data is then transmitted to a local edge computing node or a cloud data center via a wireless transmission module; the wireless transmission module supports one or more communication methods among LoRa, ZigBee, and 5G, and has adaptive anti-interference capabilities and a dynamic key update strategy.

[0014] A switchgear status monitoring device based on multi-sensor spatiotemporal calibration and feature fusion, used to implement the method described in any of the above-mentioned embodiments, comprising: Sensor module: includes multiple heterogeneous sensors deployed in key parts of the switch cabinet for collecting multi-source status data; all heterogeneous sensors are waterproof, dustproof, and resistant to electromagnetic interference, with a protection level ≥ IP65; Data acquisition and processing unit: connected to the sensor module, used for spatiotemporal calibration and feature fusion processing of the acquired multi-source state data; the data acquisition and processing unit adopts a microcontroller with a main frequency of ≥480MHz and built-in storage space of ≥16GB; Encrypted communication module: connected to the data acquisition and processing unit, used to encrypt the processed data or raw data based on the national cryptographic algorithm and transmit it through a wireless network; the encrypted communication module has a built-in national cryptographic algorithm engine and supports SM2 / SM3 / SM4 national cryptographic algorithms; Central diagnostic platform: Deployed with a deep learning-based state diagnostic model, used to receive data transmitted by the encrypted communication module and perform state diagnosis and early warning; the central diagnostic platform includes a data receiving module, a data decryption module, a model diagnostic module, an early warning push module, and a data storage module; Power management module: used to provide self-powered or auxiliary power to the sensor module and data acquisition and processing unit; the power management module includes a current transformer power supply unit and a backup lithium battery, and has overvoltage, overcurrent and undervoltage protection functions.

[0015] Furthermore, the data acquisition and processing unit includes: The clock synchronization module adopts a Beidou / GPS dual-mode timing module to provide a unified clock source and broadcast synchronization signals to the sensor module; The data preprocessing module is used to filter, denoise, and perform spatiotemporal registration on the received multi-source data; the filtering uses the Kalman filter algorithm, and the denoising uses the wavelet denoising algorithm. The feature fusion module is used to perform feature extraction from multi-source data and feature fusion based on attention mechanisms.

[0016] Furthermore, the encrypted communication module has a built-in ASR3603 chip, which supports LoRa, ZigBee and 5G dual-mode communication, and has signal strength detection function and adaptive communication switching capability; the key dynamic update strategy includes periodic update - update cycle of 24 hours and triggered update - update immediately when abnormal data transmission is detected.

[0017] Furthermore, the sensor module can also be equipped with vibration sensors and insulation resistance sensors to expand the monitoring dimensions; the backup lithium battery has a capacity of ≥5000mAh and a battery life of ≥72 hours to ensure uninterrupted operation of the device.

[0018] The technical effects and advantages of this invention are as follows: 1. High-precision data fusion: By introducing a spatiotemporal calibration mechanism, the problem of asynchronous data from multiple sensors in time and space is solved. The time synchronization accuracy reaches the microsecond level, and the spatial positioning error is ≤5cm. This lays a solid foundation for subsequent accurate feature fusion and diagnostic analysis, and avoids misjudgment and missed judgment caused by data misalignment.

[0019] 2. High diagnostic accuracy: The model adopts a feature fusion based on attention mechanism and a CNN-LSTM hybrid diagnostic model, which can effectively integrate complementary information from multi-source heterogeneous data, mine the deep features and evolution patterns of equipment status, achieve a fault diagnosis accuracy of ≥96%, a false alarm rate of ≤4%, and a remaining life prediction error of ≤10%, significantly improving the accuracy of fault diagnosis and the precision of remaining life prediction, realizing the transformation from "passive response" to "proactive prevention", and reducing the occurrence rate of railway power supply failures.

[0020] 3. High security: It adopts a lightweight encryption communication mechanism based on the national cryptographic algorithms - SM2 / SM3 / SM4-, and combined with a dynamic key update strategy, which effectively ensures the confidentiality, integrity and availability of monitoring data during transmission in the complex electromagnetic environment of the railway, prevents data from being stolen or tampered with, and meets the security and confidentiality requirements of the railway power supply system.

[0021] 4. Strong adaptability: The spatial calibration algorithm corrects the error caused by the signal propagation path, and the attention mechanism dynamically adjusts the feature weights, making the system highly robust to sensor position changes, environmental noise interference, temperature and humidity fluctuations, etc. It is suitable for the complex electromagnetic environment and climate conditions of my country's railways and can be widely used in different types of railway switchgear monitoring scenarios.

[0022] 5. Convenient and efficient operation and maintenance: Enables 24 / 7 unmanned and intelligent monitoring of switchgear, eliminating the need for frequent manual inspections, significantly reducing the workload of maintenance personnel and maintenance costs. At the same time, through an active early warning mechanism, it can detect equipment anomalies in advance, shorten fault repair time, and improve the operation and maintenance efficiency and reliability of railway power supply systems. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a switchgear status monitoring method based on multi-sensor spatiotemporal calibration and feature fusion provided in an embodiment of this application. Figure 2 This is a structural block diagram of a switchgear status monitoring device based on multi-sensor spatiotemporal calibration and feature fusion provided in an embodiment of this application; In the picture: 10. Sensor Module; 101. Self-Powered Wireless Temperature Sensor; 102. UHF Partial Discharge Sensor; 103. Ultrasonic Partial Discharge Sensor; 104. Temperature and Humidity Sensor; 105. Image Acquisition Unit; 20. Data Acquisition and Processing Unit; 201. Clock Synchronization Module; 202. Data Preprocessing Module; 203. Feature Fusion Module; 30. Encrypted Communication Module; 40. Central Diagnostic Platform; 401. Data Receiving Module; 402. Data Decryption Module; 403. Model Diagnostic Module; 404. Early Warning Push Module; 405. Data Storage Module; 50. Power Management Module. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and design various embodiments with various modifications suitable for a particular purpose.

[0025] Example 1

[0026] Please see Figures 1-2 This embodiment provides a switchgear status monitoring method and device based on multi-sensor spatiotemporal calibration and feature fusion, which is applied to the status monitoring of a 27.5kV switchgear in a railway traction substation. This switchgear is mainly used for power distribution and line protection in the railway traction power supply system. It operates under high load and strong electromagnetic interference environment for a long time and is prone to faults such as busbar overheating and partial discharge. The solution of this invention can achieve accurate monitoring and active early warning.

[0027] A method for monitoring the status of switchgear based on multi-sensor spatiotemporal calibration and feature fusion; like Figure 1 As shown, the method includes the following steps: Step S1, Multi-source data acquisition Sensor modules 10 are deployed in key areas of the switchgear, such as the cable compartment, circuit breaker compartment, and busbar compartment. The specific deployment method is as follows: Self-powered wireless temperature sensor 101: A total of 6 sensors are deployed, which are respectively attached to 3 busbar connections, 2 circuit breaker contacts, and 1 cable terminal. The sampling frequency is set to 5Hz, the temperature measurement range is -40℃ to 150℃, the measurement accuracy is ±0.5℃, and they are magnetically installed without external power supply. They transmit data wirelessly.

[0028] Ultra-high frequency partial discharge sensor 102: One unit is deployed and fixed to the opening in the wall of the circuit breaker compartment via a flange. The detection frequency band is 300MHz~3GHz, and the sensitivity is ≤5pC. It is used to detect ultra-high frequency signals generated by partial discharge in the cabinet.

[0029] Ultrasonic partial discharge sensor 103: One unit is deployed and attached to the surface of the circuit breaker compartment wall with a coupling agent, corresponding to the position of the ultra-high frequency partial discharge sensor 102. The detection frequency band is 20kHz~200kHz, which is used to assist in the detection of partial discharge signals and eliminate electromagnetic interference.

[0030] Temperature and humidity sensor 104: One unit is deployed and fixed in the middle of the switch cabinet with a clip. The sampling frequency is 1Hz, the temperature measurement accuracy is ±0.3℃, and the humidity measurement accuracy is ±5%RH. It is used to collect temperature and humidity data of the cabinet environment.

[0031] Image acquisition device 105: Deploy one unit, fixed to the top of the switch cabinet by a bracket, aligned with the instrument panel and mechanical indicators, with a resolution of 1080P and a frame rate of 3fps, for acquiring meter readings such as voltage and current readings and images of the appearance and condition of insulators.

[0032] All sensors collect corresponding status data in real time, and the collected data is transmitted wirelessly to the data acquisition and processing unit 20.

[0033] Step S2, Data Encryption Transmission and Spatiotemporal Calibration Encrypted transmission: After receiving the raw data collected by the sensor, the data acquisition and processing unit 20 uses the SM4 block cipher algorithm with a key length of 128 bits and encryption mode CBC to encrypt the data. The encrypted data is then sent to the local edge computing node via the 5G communication mode of the encrypted communication module 30 with a communication rate of 100Mbps, and is simultaneously backed up to the cloud data center. The key of the encrypted communication module 30 is automatically updated every 24 hours. If abnormal data transmission is detected, such as data loss or tampering, the key is immediately updated to ensure data transmission security.

[0034] Timestamp synchronization calibration: The clock synchronization module 201 of the data acquisition and processing unit 20 adopts a Beidou / GPS dual-mode timing module with a timing accuracy of ≤1μs. It broadcasts a synchronization signal to all sensors every 100ms. After receiving the synchronization signal, each sensor calibrates its local clock to ensure that the time stamps of all data are consistent. For low sampling frequency data of temperature and humidity sensor (sampling frequency 1Hz) and image acquisition device (frame rate 3fps), linear interpolation is used to interpolate to 5Hz, which is consistent with the sampling frequency of temperature sensor, to achieve time synchronization.

[0035] Spatial position calibration: The three-dimensional spatial coordinates of each sensor are measured and recorded in advance using a laser rangefinder. With the bottom left corner of the switchgear as the origin, the X-axis runs along the length of the switchgear, the Y-axis along the height, and the Z-axis along the depth. The specific coordinates are as follows: Temperature sensor at busbar connection: X=50cm, Y=120cm, Z=30cm; Circuit breaker contact temperature sensor: X=80cm, Y=100cm, Z=30cm; Ultra-high frequency partial discharge sensor X=60cm, Y=80cm, Z=25cm; Ultrasonic partial discharge sensor X=60cm, Y=80cm, Z=26cm; Temperature and humidity sensor: X=70cm, Y=100cm, Z=25cm; Image acquisition unit: X=70cm, Y=150cm, Z=25cm.

[0036] When both the UHF partial discharge sensor 102 and the ultrasonic partial discharge sensor 103 simultaneously detect a suspected partial discharge signal, the time difference Δt (approximately 1000µs~1500µs) between the signals received by the two sensors is recorded, combined with the UHF signal propagation speed v1 = 3 × 10⁻⁶. 8 Given that the ultrasonic signal propagation speed is v2 = 340 m / s, the coordinates (X0, Y0, Z0) of the partial discharge source are calculated using the TDOA localization algorithm. The calculation formula is as follows:

[0037] Where (X1, Y1, Z1) are the coordinates of the UHF partial discharge sensor, and (X2, Y2, Z2) are the coordinates of the ultrasonic partial discharge sensor. After calculating the coordinates of the partial discharge source, spatial calibration is performed by correlating temperature sensor data and image data near those coordinates to eliminate interference data from uncorrelated areas.

[0038] Step S3, Feature Extraction and Multi-Scale Feature Fusion Feature extraction: The data preprocessing module 202 filters and denoises the spatiotemporally calibrated multi-source data, using a Kalman filter coefficient of 0.2, wavelet denoising with a db4 basis, and 3-level decomposition, and then extracts features from each type of data. (1) Temperature data characteristics: Six time-domain characteristics were extracted, namely mean - such as the mean temperature at the busbar connection point of 35℃, maximum value -42℃, minimum value -32℃, peak value -10℃, rate of change -0.5℃ / min, and fluctuation variance -0.8.

[0039] (2) Partial discharge data characteristics: 10 features were extracted, including 6 statistical features: maximum discharge amplitude 500pC, average discharge amplitude 120pC, discharge frequency 15 times / min, discharge phase distribution concentrated in 0°-90° and 180°-270°, discharge amplitude standard deviation 80pC, and 4 PRPD spectrum texture features: contrast 0.6, entropy 1.2, energy 0.8, and correlation 0.7.

[0040] (3) Image data features: The preprocessed image is extracted using the MobileNetV3 network, and a 128-dimensional visual feature vector is output, which includes meter readings and insulator appearance features.

[0041] (4) Temperature and humidity data characteristics: Three features were extracted, namely, the average temperature of 32℃, the average humidity of 60%RH, and the temperature and humidity fluctuation range (temperature ±2℃, humidity ±5%RH).

[0042] Feature fusion: The source feature vectors are mapped to 128 dimensions through a fully connected layer and input into a feature fusion network based on an attention mechanism. The attention weights of each source feature are calculated: temperature feature weight 0.3, partial discharge feature weight 0.4, image feature weight 0.2, and temperature and humidity feature weight 0.1. The weighted feature vectors are summed element by element to obtain the 128-dimensional fused feature vector F_fused.

[0043] Step S4, Intelligent Diagnosis and Proactive Early Warning The fused feature vector F_fused is fed into the CNN-LSTM hybrid diagnostic model deployed on the edge computing node in real time. The model has been trained on a training dataset containing 10,000 sets of data in different states, with a validation accuracy of 97.2% and a false positive rate of 3.5%.

[0044] The model extracts local correlation features from the fused features through convolutional layers, captures the state evolution patterns over time using LSTM layers, and finally outputs diagnostic results. Health score: 88 points (normal range); Fault type: Normal; Failure probability: 98% for normal operation, 1% for busbar overheating, 1% for partial discharge, and 0% for mechanical jamming; Life expectancy prediction: 180 days.

[0045] When poor contact of the simulated busbar leads to abnormal temperature, such as a temperature rise to 55℃ and a partial discharge signal amplitude rise to 800pC, the model outputs: health score 58 points - moderate abnormality, fault type is busbar overheating, fault probability 85%, remaining lifespan 30 days, and a moderate warning is triggered. The warning information and maintenance suggestions are pushed to the maintenance personnel through the APP - regularly check the busbar connection bolts and tighten loose parts.

[0046] A switchgear status device based on multi-sensor spatiotemporal calibration and feature fusion like Figure 2 As shown, the device for implementing the above method includes a sensor module 10, a data acquisition and processing unit 20, an encrypted communication module 30, a central diagnostic platform 40, and a power management module 50. The working process of each module is as follows: Sensor module 10: Real-time acquisition of multi-source status data such as temperature, partial discharge, temperature and humidity, and images of the switchgear, and wireless transmission to data acquisition and processing unit 20; Among them, the self-powered wireless temperature sensor 101 uses a current transformer for power, eliminating the need for an external power supply and ensuring long-term stable operation.

[0047] Data acquisition and processing unit 20: adopts STM32H743 microcontroller with a main frequency of 480MHz and built-in 16GB storage space; clock synchronization module 201 provides a unified clock source to realize sensor clock synchronization; data preprocessing module 202 performs filtering, noise reduction and spatiotemporal calibration on the data; feature fusion module 203 performs feature extraction and fusion, and outputs fused feature vector.

[0048] Encrypted communication module 30: It has a built-in ASR3603 chip and SM2 / SM3 / SM4 national cryptographic algorithm engine to encrypt the fused feature vector and transmit it to the central diagnostic platform 40 through 5G / LoRa dual-mode communication; it has a dynamic key update function to ensure data transmission security; when the 5G signal strength is lower than -80dBm, it automatically switches to LoRa communication mode - communication distance 1km to avoid data loss.

[0049] Central diagnostic platform 40: Data receiving module 401 receives encrypted data, data decryption module 402 decrypts data using SM4 algorithm, model diagnostic module 403 calls CNN-LSTM hybrid diagnostic model to perform status diagnosis, early warning push module 404 pushes early warning information based on diagnostic results, and data storage module 405 stores monitoring data and diagnostic results for 1 year, supporting maintenance personnel to query and analyze historical data.

[0050] Power management module 50: includes a current transformer power supply unit with a power supply range of 10-100A and a backup lithium battery with a capacity of 5000mAh, providing stable power supply for sensor module 10 and data acquisition and processing unit 20; when the bus current is lower than 10A, it automatically switches to backup lithium battery power supply, with a battery life of ≥72 hours, ensuring uninterrupted operation of the device.

[0051] Performance verification A six-month field test was conducted on a 27.5kV switchgear in a railway traction substation, collecting 12,000 sets of valid data. The model diagnostic accuracy was 97.8%, the false alarm rate was 3.2%, and the average error in remaining lifetime prediction was 8.7%, representing a significant improvement over traditional methods. The encryption module operated stably within a temperature range of -40℃ to 85℃ and under strong electromagnetic interference, with a 100% success rate in dynamic key updates. Spatial positioning calibration ensured that the partial discharge source positioning error was ≤5cm and the time synchronization error was ≤1μs, effectively solving the problem of multi-source data asynchrony.

[0052] Through the above solution, this embodiment can perform 24 / 7 unmanned and intelligent monitoring of switchgear, effectively solving the problems of asynchronous data, unintelligent analysis, and insecure transmission in existing monitoring methods. The fault diagnosis accuracy rate is ≥96%, and the false alarm rate is ≤4%, which can significantly improve the operation and maintenance efficiency and reliability of switchgear in railway traction substations and avoid the occurrence of serious power supply failures.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0054] Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art and related fields based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention, unless otherwise specified or limited, shall be implemented according to conventional means in the art.

Claims

1. A method for monitoring the status of switchgear based on multi-sensor spatiotemporal calibration and feature fusion, characterized in that, Includes the following steps: Step S1: Collect multi-source status data of the switch cabinet in real time by deploying multiple heterogeneous sensors in key parts of the switch cabinet; the heterogeneous sensors include at least a self-powered wireless temperature sensor (101), an ultra-high frequency partial discharge sensor (102), an ultrasonic partial discharge sensor (103), a temperature and humidity sensor (104), and an image acquisition device (105). Step S2: Perform spatiotemporal calibration on the collected multi-source state data to obtain spatiotemporally aligned multi-source synchronized data; the spatiotemporal calibration includes timestamp synchronization calibration based on a unified clock source, and spatial position calibration based on sensor installation location coordinates and signal propagation speed; Step S3: Perform feature extraction and multi-scale feature fusion on the spatiotemporally aligned multi-source synchronization data to generate a fusion feature vector that can comprehensively characterize the health status of the equipment; Step S4: Input the fused feature vector into the pre-trained deep learning-based state diagnosis model to perform real-time evaluation and fault prediction of the switchgear's operating status, and output diagnostic results and early warning information.

2. The switchgear status monitoring method based on multi-sensor spatiotemporal calibration and feature fusion according to claim 1, characterized in that, In step S2, the timestamp synchronization calibration specifically includes: By using a high-precision time synchronization module deployed locally, synchronization signals are periodically broadcast to all sensors. Each sensor calibrates its local clock according to the signal and adds a unified time tag to the collected data. The high-precision timing module supports BeiDou / GPS dual-mode timing, with a timing accuracy of ≤1μs and a synchronization signaling broadcast period of 100ms. For sensor data with inconsistent sampling frequencies, linear interpolation is used to interpolate to a unified sampling frequency.

3. The switchgear status monitoring method and device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 1, characterized in that, In step S2, the spatial position calibration specifically involves: The three-dimensional spatial coordinates of each sensor are measured and recorded in advance. When the ultra-high frequency partial discharge sensor (102) and the ultrasonic partial discharge sensor (103) detect a suspected partial discharge signal at the same time, the specific spatial coordinates of the partial discharge source are calculated by using the difference in propagation speed between the ultra-high frequency signal and the ultrasonic signal, combined with the coordinates of the two sensors, and through the time difference of arrival positioning algorithm. Based on the coordinates of the partial discharge source, the temperature data and image data in the same spatial area are correlated and calibrated to eliminate signal misalignment.

4. The switchgear status monitoring method and device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 1, characterized in that, Step S3 specifically includes: Step S31: Extract features from different types of sensor data respectively; Extract time-domain features of temperature data, including mean, maximum, minimum, peak-to-peak value, rate of change, and variance of fluctuation. Statistical features of the maximum discharge amplitude, average discharge amplitude, discharge frequency, discharge phase distribution, and texture features of the PRPD spectrum were extracted from the partial discharge data. Visual features of meter readings and insulator appearance are extracted from image data using a lightweight convolutional neural network; mean, fluctuation range, and trend features are extracted from temperature and humidity data. Step S32: Use an attention-based feature fusion network to adaptively weight and fuse feature vectors from different sources; The network first maps the source feature vectors to a unified dimension through a fully connected layer, then calculates the attention weights of each source feature through a sigmoid activation function, and finally sums the weighted feature vectors element by element to generate the fused feature vector.

5. The switchgear status monitoring method and device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 1, characterized in that, The state diagnosis model based on deep learning in step S4 is a CNN-LSTM hybrid network, which includes convolutional layers, LSTM layers and fully connected layers. Convolutional layers are used to extract local correlation features from the fused feature vectors, LSTM layers are used to mine the dynamic change patterns of data over time, and fully connected layers are used to output device health scores, fault types, fault probabilities, and remaining life predictions. The model has a fault diagnosis accuracy of ≥96%, a false alarm rate of ≤4%, and a remaining life prediction error of ≤10%.

6. The switchgear status monitoring method and device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 1, characterized in that, After step S1 and before step S2, the method further includes encrypting and transmitting the collected multi-source state data. A lightweight encryption protocol based on Chinese national cryptographic algorithms is used to encrypt multi-source state data. The encryption algorithm is the SM4 block cipher algorithm. The encrypted data is transmitted wirelessly to the local edge computing node or cloud data center. The wireless transmission module supports one or more communication methods, including LoRa, ZigBee, and 5G, and has adaptive anti-interference function and dynamic key update strategy.

7. A switchgear status monitoring device based on multi-sensor spatiotemporal calibration and feature fusion, characterized in that, The method for implementing any one of claims 1 to 6 includes: Sensor module (10): includes multiple heterogeneous sensors deployed in key parts of the switch cabinet for collecting multi-source status data; all heterogeneous sensors are waterproof, dustproof, and anti-electromagnetic interference capable, with a protection level ≥ IP65; Data acquisition and processing unit (20): connected to the sensor module (10), used to perform spatiotemporal calibration and feature fusion processing on the acquired multi-source state data; the data acquisition and processing unit (20) adopts a microcontroller with a main frequency of ≥480MHz and has a built-in storage space of ≥16GB; Encrypted communication module (30): connected to the data acquisition and processing unit (20), used to encrypt the processed data or the original data based on the national cryptographic algorithm, and transmit it through a wireless network; the encrypted communication module (30) has a built-in national cryptographic algorithm engine and supports SM2 / SM3 / SM4 national cryptographic algorithms; Central diagnostic platform (40): Deployed with a state diagnostic model based on deep learning, used to receive data transmitted by the encrypted communication module (30) and perform state diagnosis and early warning; the central diagnostic platform (40) includes a data receiving module (401), a data decryption module (402), a model diagnostic module (403), an early warning push module (404), and a data storage module (405). Power management module (50): used to provide self-powered or auxiliary power to the sensor module (10) and the data acquisition and processing unit (20); the power management module (50) includes a current transformer power supply unit and a backup lithium battery, and has overvoltage, overcurrent and undervoltage protection functions.

8. The switchgear status monitoring device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 7, characterized in that, The data acquisition and processing unit (20) includes: The clock synchronization module (201) adopts a Beidou / GPS dual-mode timing module to provide a unified clock source and broadcast synchronization signaling to the sensor module (10); The data preprocessing module (202) is used to filter, denoise and perform spatiotemporal registration on the received multi-source data; the filtering adopts the Kalman filter algorithm and the denoising adopts the wavelet denoising algorithm. The feature fusion module (203) is used to perform feature extraction from multi-source data and feature fusion based on attention mechanism.

9. A switchgear status monitoring device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 7, characterized in that, The encrypted communication module (30) has a built-in ASR3603 chip, supports LoRa, ZigBee and 5G dual-mode communication, and has signal strength detection function and adaptive communication switching capability; the key dynamic update strategy includes periodic update and triggered update.

10. A switchgear status monitoring device based on multi-sensor spatiotemporal calibration and feature fusion according to claim 7, characterized in that, The sensor module (10) can also be equipped with vibration sensors and insulation resistance sensors to expand the monitoring dimensions; The backup lithium battery has a capacity of ≥5000mAh and a battery life of ≥72 hours, ensuring uninterrupted operation of the device.

Images