Sf6 sensor long-term operation reliability evaluation method and system based on pinn

Abstract

The application discloses a PINN-based SF6 sensor long-term operation reliability evaluation method and system, belongs to the technical field of sensor reliability evaluation, collects SF6 sensor operation data and environmental stress data of the installation position of the SF6 sensor, respectively generates time series and completes preprocessing, obtains actual values of sensor performance degradation, determines real distribution parameters of a monotonic irreversible random process to which long-term operation is subjected through statistical estimation, constructs a double parallel path PINN model, inputs preprocessed operation and environmental data on one path to output performance degradation estimation values, inputs environmental stress data on the other path to output prediction distribution parameters of the random process, completes model training by using a joint loss function, calculates performance degradation and corresponding prediction distribution parameters by using the trained PINN model, combines preset failure thresholds and probability distribution, and obtains long-term reliability and residual service life of the SF6 sensor, thereby meeting the evaluation requirement of long-term operation reliability.

Description

Technical Field

[0001] This invention belongs to the field of sensor reliability evaluation technology, specifically relating to a method and system for evaluating the long-term operational reliability of SF6 sensors based on PINN. Background Technology

[0002] GIS equipment uses SF6 gas as an arc-extinguishing and insulating medium. Real-time monitoring and alarm functions for this medium's status are achieved through SF6 gas state sensors. This is a key technology for ensuring reliable GIS operation and supporting intelligent operation and maintenance in digital power grid construction. The complex electromagnetic fields within substations, changes in ambient temperature and humidity, and strong vibrations generated by GIS switching can cause multiple environmental stresses, including internal temperature rises, structural stress deformation, or chemical reactions in components, accelerating the degradation of measurement performance and even leading to failure. Simultaneously, the current lack of comprehensive evaluation theories and standards for power sensors, coupled with insufficient long-term operational testing and verification, results in frequent sensor failures, false alarms, and missed alarms under actual operating conditions. These problems not only prevent timely warnings when SF6 pressure drops to alarm values, causing GIS to lose its switching function and creating significant accident hazards, but also severely impair the early defect identification and fault evolution prediction capabilities of GIS. Therefore, an effective assessment of the long-term operational reliability of SF6 sensors is urgently needed.

[0003] Currently, there is no unified standard for sensor reliability evaluation, and related research is scarce. Furthermore, traditional reliability research theories rely on a large amount of lifetime and failure data, but in the actual long-term operation of sensors, it is difficult to collect sufficient data to meet the evaluation requirements in a short period of time, making conventional evaluation methods difficult to apply.

[0004] Traditional sensor reliability evaluation methods are mainly divided into two categories: one is the accelerated life test method, which accelerates sensor aging by applying stresses higher than normal operating conditions, such as high temperature, high humidity, and mechanical vibration, and calculates the life under normal conditions by combining the acceleration factor. After the test, the failure data is analyzed using Weibull distribution and normal distribution to calculate reliability indicators such as failure rate and mean time between failures; the other is the degradation trajectory method, which constructs a degradation model and predicts the failure threshold by monitoring the long-term changes in performance parameters such as sensor output signal drift and sensitivity reduction.

[0005] Among the existing related patent technologies, Chinese patent CN111814403B, published on October 23, 2020, adopts the deep learning SASTGCN model, utilizing its adaptive and nonlinear feature extraction capabilities to adapt to the characteristics of high sensor data complexity and significant spatiotemporal characteristics, thereby improving the evaluation accuracy; Chinese patent application CN120180978A, published on June 20, 2025, combines physical equations and physical information neural networks (PINN), achieving accurate damage prediction through dual-drive loss functions and adaptive training; Chinese patent CN115962797B, published on April 14, 2023, applies temperature stress to weak modules of sensors to conduct accelerated degradation experiments, constructs a reliability model based on the degradation trajectory model and incorporates electric field stress to improve prediction reliability and confidence; and Chinese patent application CN117191105A, published on December 8, 2023, integrates weighted coefficients of historical sensor observation data, static and dynamic reliability, to achieve reliability and safety level assessment of converter valve sensors.

[0006] However, existing technologies still have significant limitations: First, most existing technologies are designed for accelerated life testing and accelerated degradation testing scenarios, and do not focus on the long-term actual operational reliability assessment of sensors; second, they are highly dependent on accelerated test data and cannot adapt to the actual situation where data is difficult to collect during long-term operation of sensors, thus failing to meet the evaluation requirements for their long-term operational reliability. Summary of the Invention

[0007] The purpose of this invention is to overcome the problem that existing sensor reliability assessment technologies do not focus on the long-term actual operation reliability assessment of sensors and are highly dependent on accelerated test data. This invention proposes a method and system for evaluating the long-term operation reliability of SF6 sensors based on PINN.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for evaluating the long-term operational reliability of an SF6 sensor based on PINN, comprising the following steps: The operation data of the SF6 sensor and the environmental stress data of the installation location of the SF6 sensor were collected and time series were generated respectively. The time series were preprocessed to obtain the preprocessed operation data and the preprocessed environmental stress data. The actual value of the performance degradation of the SF6 sensor is obtained. Based on the actual value of the performance degradation, the true distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation are obtained through statistical estimation. A physical neural network (PINN) model with dual parallel processing paths is constructed. The first path receives preprocessed operating data and preprocessed environmental stress data, and outputs an estimate of the performance degradation of the SF6 sensor during long-term operation. The second path receives preprocessed environmental stress data and outputs the predicted distribution parameters of the monotone irreversible random process that the SF6 sensor follows during long-term operation. The PINN model is trained using a joint loss function that integrates data loss and physical loss. The data loss is calculated based on the estimated and actual performance degradation values, while the physical loss is calculated based on the true and predicted distribution parameters. Using the trained PINN model, the performance degradation of the SF6 sensor and the predicted distribution parameters of the corresponding monotonically irreversible random process are calculated. Based on the preset performance degradation failure threshold and the probability distribution of the monotonically irreversible random process determined by the predicted distribution parameters, the long-term reliability and remaining service life of the SF6 sensor are calculated.

[0009] Furthermore, a monotonically irreversible random process is an inverse Gaussian process; The statistical estimate is the maximum likelihood estimate.

[0010] Furthermore, the steps to obtain the actual value of the performance degradation of the SF6 sensor are as follows: perform regular offline calibration of the SF6 sensor to obtain the actual value of the performance degradation at different operating time points.

[0011] Furthermore, environmental stress data includes ambient temperature data, humidity data, electromagnetic field strength data, and vibration data.

[0012] Furthermore, preprocessing includes temporal alignment, missing value handling, outlier handling, and normalization.

[0013] Furthermore, the first path adopts a gated cyclic unit network structure, while the second path adopts a fully connected network structure.

[0014] Furthermore, the weight coefficients of data loss and physical loss in the joint loss function are determined using a progressive learning method. In the initial training phase, the weight coefficient of physical loss is set to 0, and the model is trained to convergence using only data loss. Subsequently, the weight coefficient of physical loss is gradually increased until the physical loss is lower than a preset threshold.

[0015] Secondly, this invention provides a long-term operational reliability evaluation system for SF6 sensors based on PINN, comprising: The data preprocessing module is used to collect the operating data of the SF6 sensor and the environmental stress data of the SF6 sensor installation location, and generate time series respectively; the time series are preprocessed to obtain the preprocessed operating data and the preprocessed environmental stress data. The distribution parameter baseline value acquisition module is used to obtain the actual value of the performance degradation of the SF6 sensor. Based on the actual value of the performance degradation, the true distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation are obtained through statistical estimation. The distributed parameter output module is used to construct a physical neural network (PINN) model with dual parallel processing paths. The first path receives preprocessed operating data and preprocessed environmental stress data, and outputs an estimate of the performance degradation of the SF6 sensor during long-term operation. The second path receives preprocessed environmental stress data and outputs the predicted distribution parameters of the monotone irreversible random process that the SF6 sensor follows during long-term operation. The PINN model optimization module is used to train the PINN model using a joint loss function that integrates data loss and physical loss, resulting in a trained PINN model. The data loss is calculated based on the estimated and actual values ​​of performance degradation, while the physical loss is calculated based on the true and predicted distribution parameters. The reliability calculation module is used to calculate the performance degradation of the SF6 sensor and the predicted distribution parameters of the corresponding monotonically irreversible random process using the trained PINN model. Based on the preset performance degradation failure threshold and the probability distribution of the monotonically irreversible random process determined by the predicted distribution parameters, the long-term reliability and remaining service life of the SF6 sensor are calculated.

[0016] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a long-term reliability evaluation method for an SF6 sensor based on PINN.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for evaluating the long-term operational reliability of an SF6 sensor based on PINN.

[0018] Compared with the prior art, the present invention has the following beneficial technical effects: This invention proposes a long-term reliability evaluation method for SF6 sensors based on PINN (Physical Neural Network). This method utilizes measurement data transmitted by the sensor and the environmental stress conditions to evaluate the sensor's reliability at the measurement point. The proposed method does not rely on manual observation; the resulting model can run autonomously, facilitating power plant personnel's understanding of sensor performance degradation and enabling them to rationally plan sensor maintenance and upgrades, thus providing strong support for the construction of smart grid sensor networks. This invention employs a physical neural network to achieve sensor reliability evaluation. The network consists of a time-series prediction path and a distribution fitting path. These two paths are merged during loss function calculation to ensure that the evaluation results fall within the region consistent with the sensor degradation mechanism.

[0019] Furthermore, the inverse Gaussian process itself is strictly monotonically increasing, consistent with the degradation characteristics of sensor performance. The two parameters in the inverse Gaussian distribution correspond to two typical failure mechanisms: cumulative damage and random impact. Sensor degradation often exhibits a right skew (early stability, then a sharp decline), and the inverse Gaussian distribution naturally matches this "slow at first, then rapid" pattern, making it particularly suitable for describing sensor degradation. The model is highly interpretable. Moreover, the inverse Gaussian process is additive, facilitating the calculation of the first-arrival time of sensor failure and reliability, thus simplifying the sensor reliability evaluation process.

[0020] Furthermore, the GRU network structure used in the time prediction path of physical neural networks can be replaced by other time prediction networks, such as LSTM and Informer. The time prediction path adopts the GRU structure, which is simpler, easier to train, and has a faster initial convergence speed than the LSTM structure. Therefore, it is suitable for use as a time prediction network in physical neural networks.

[0021] Furthermore, this invention is driven by mechanism data fusion, with a small sample size and high modeling accuracy. This method integrates the physical interpretability of the inverse Gaussian process with the data fitting ability of the neural network, incorporates the deviation of the parameters of the inverse Gaussian process into the loss function, avoids the neural network output from violating physical laws, achieves high-precision modeling in small sample scenarios, and is suitable for sparse failure data scenarios of sensors.

[0022] Furthermore, this approach simultaneously considers multiple factors that superimpose stresses and affect sensor reliability, such as temperature, humidity, electromagnetic interference, and mechanical vibration. By leveraging the data fitting capabilities of physical neural networks, it addresses the extremely complex and difficult-to-describe failure mechanisms under multi-stress coupling, enabling a more comprehensive assessment of the sensor's long-term reliability in complex operating environments.

[0023] Furthermore, the method of this invention allows for rapid transfer learning, possesses strong model interpretability, and its physically constrained neural network structure enables rapid adaptation through fine-tuning across similar operating conditions, reducing reliance on target scenario data. By using transfer learning or fixing the weights of other layers, the fine-tuning method for learning the weights of the last layer can be extended to the long-term operational reliability evaluation of other types of sensors. Attached Figure Description

[0024] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely schematic to aid in understanding the invention and do not specifically limit the shapes and proportions of the components. In the drawings: Figure 1 This is a flowchart of the long-term operational reliability evaluation method for SF6 sensors based on PINN according to the present invention.

[0025] Figure 2 This is a simplified structural diagram of the PINN-based SF6 sensor long-term operational reliability evaluation system of the present invention.

[0026] Figure 3 This is an electronic device diagram of the long-term operational reliability evaluation method for SF6 sensors based on PINN according to the present invention.

[0027] Figure 4 This is a flowchart of the long-term operational reliability evaluation method for SF6 sensors based on PINN in an embodiment of the present invention.

[0028] Figure 5 This refers to the physical neural network in this embodiment of the invention.

[0029] Figure 6 This is a simplified structural diagram of the PINN-based SF6 sensor long-term operational reliability evaluation system in this embodiment of the invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] Example 1 Based on the failure mechanism and performance degradation law of the sensor, measurement data and environmental stress data of the sensor are collected. A physical neural network is used to evaluate the performance degradation of the sensor, further revealing changes in sensor reliability. In the physical neural network, an inverse Gaussian process is used to describe the sensor degradation process, comprehensively considering both the prediction loss of the data and the deviation loss of the sensor degradation mechanism. The entire method is as follows: S1. Sensor data and stress acquisition on the sensor S2. Data Preprocessing S3. Construct a sensor reliability evaluation network S4. Complete the sensor reliability evaluation. Explanation of relevant terms: SF6 gas sensor: A specialized sensor used to monitor and manage the state of sulfur hexafluoride (SF6) gas in high-voltage electrical equipment. SF6, as a highly efficient insulating and arc-quenching medium, is widely used in high-voltage electrical equipment such as circuit breakers, disconnect switches, and transformers. The SF6 gas state sensor assesses its state by detecting the gas's physical and chemical properties. Common detection parameters include: Pressure: Pressure sensors monitor changes in SF6 gas pressure to determine if a leak exists.

[0032] Slight moisture: SF6 gas may contain a certain amount of moisture. Excessive moisture content can lead to a decrease in the insulation and arc-quenching capabilities of SF6.

[0033] Temperature: The temperature sensor monitors the gas temperature to prevent overheating that could lead to gas decomposition.

[0034] Decomposition products: Infrared sensors or spectrometers are used to detect the decomposition products of SF6 gas, such as SOF2 and SO2, to determine whether the gas has undergone chemical changes.

[0035] Physical neural networks are a new type of neural network architecture that combines physical laws with deep learning. They aim to solve problems in scientific computing, engineering modeling, and prediction of complex systems by embedding known physical knowledge (such as differential equations, conservation laws, etc.) into the training process of neural networks.

[0036] Inverse Gaussian process: This is a stochastic process model based on the inverse Gaussian distribution, widely used in reliability analysis, lifetime prediction, performance degradation modeling, and other fields. Its core idea is to utilize the properties of the inverse Gaussian distribution to describe the monotonic degradation process, and then perform statistical inference and prediction based on this.

[0037] Reliability: Reliability refers to the ability of a product to perform its intended function within a specified time and under specified conditions.

[0038] See Figure 1The long-term operational reliability evaluation method for SF6 sensors based on PINN includes the following steps: Operational data and environmental stress data of the SF6 sensor's installation location were collected and generated as time series. Each time series was preprocessed to obtain preprocessed operational data and preprocessed environmental stress data. The actual performance degradation of the SF6 sensor was obtained, and based on this actual value, the true distribution parameters of the monotonically irreversible random process followed by the SF6 sensor during long-term operation were statistically estimated. A physical neural network (PINN) model with dual parallel processing paths was constructed. The first path received the preprocessed operational data and preprocessed environmental stress data, and output the estimated performance degradation of the SF6 sensor during long-term operation. The second path received the preprocessed environmental stress data... Stress data is used to output the predicted distribution parameters of the monotone irreversible random process that the SF6 sensor follows during long-term operation. The PINN model is trained using a joint loss function that integrates data loss and physical loss. The data loss is calculated based on the estimated and actual values ​​of performance degradation, while the physical loss is calculated based on the true and predicted distribution parameters. Using the trained PINN model, the performance degradation of the SF6 sensor and the corresponding predicted distribution parameters of the monotone irreversible random process are calculated. Based on the preset performance degradation failure threshold and the probability distribution of the monotone irreversible random process determined by the predicted distribution parameters, the long-term reliability and remaining service life of the SF6 sensor are calculated.

[0039] See Figure 4 The flowchart of the long-term operational reliability evaluation method for SF6 sensors based on PINN is shown, and it includes the following steps: S1: Sensor data and stress acquisition on the sensor Data collected by sensors and stress measurements at sensor installation locations were performed. The sensors being evaluated primarily included SF6 gas state sensors. These sensors were installed at corresponding locations on the GIS (Gas Insulated Switchgear), and the collected SF6 gas state data within the GIS was transmitted back via a data acquisition device. The collected stresses mainly included ambient temperature, humidity, electromagnetic field, and vibration at the sensor installation location. Each collected data point constituted a time series for subsequent data processing.

[0040] S2: Data Preprocessing Since the sampling rates of various sensors and stress measurements may be inconsistent, it is necessary to use methods such as downsampling or interpolation to ensure that the lengths of each time series are consistent, thus achieving time series alignment. The collected data may contain missing values ​​or outliers, which also need to be processed through methods such as missing value imputation and outlier handling before being used as input to the physical neural network.

[0041] S3: Constructing a sensor reliability evaluation network This section proposes a sensor reliability evaluation network based on a physical neural network. The network's input consists of a set of sensor-acquired data sequences X and a set of stress sequences applied to the sensor S, and its output is the sensor's performance degradation. d The stresses experienced by the sensor typically include temperature T, humidity H, electromagnetic field U, vibration V, etc., i.e.: S=[T,H,U,V] Sensor reliability is assessed through sensor performance failure. The failure mechanism of sensors is complex, subject to various stresses, and degradation often exhibits an irreversible, monotonic trend. The degradation process of sensors is best described by an inverse Gaussian process, i.e., the amount of sensor degradation... d ( t It follows an inverse Gaussian process:

[0042] Its probability density function is:

[0043] in These are undetermined constants and need to be obtained through parameter estimation.

[0044] The commonly used method is the maximum likelihood estimation method, which involves constructing a likelihood function:

[0045] To each Taking the partial derivative and maximizing it, we get:

[0046]

[0047] See Figure 5 The physical neural network is divided into two paths: a time-series prediction path that calculates the performance degradation amount based on the sensor's acquired data sequence, and a distribution fitting path that fits the sensor degradation path using sensor data and stress data.

[0048] The main structure of the time series prediction path is a special type of recurrent neural network—the Gated Recurrent Unit (GRU), which is a simplification of the classic Long Short-Term Memory (LSTM) network, resulting in higher computational efficiency. This part can also employ other network structures for time series prediction. The input is the set of sensor-acquired data sequences X and the stress set S, and the output is the sensor's performance degradation. d .

[0049] The loss function for this path is the data loss part of the entire physical neural network:

[0050] in This is an estimate of the sensor degradation output by this path. This represents the actual degradation value of the sensor.

[0051] The distribution fitting path employs a Fully Connected Network (FCN) to fit the sensor's degradation distribution, specifically an inverse Gaussian distribution. Once the parameters in the distribution are determined, the sensor's degradation path is also determined. This path ensures that the sensor's degradation prediction has a mechanistic basis, avoiding the problem of the entire evaluation method becoming completely black-boxed and the output results being uninterpretable. The FCN's input is the sensor stress set S, and its output is two undetermined parameters of the inverse Gaussian distribution. μ and λ .

[0052] The loss function for this path is the physical loss component of the entire physical neural network:

[0053] in , The following is obtained through maximum likelihood estimation:

[0054]

[0055] The loss function of the entire physical neural network is a fusion of data loss and physical loss.

[0056] in The constant value needs to be determined through incremental learning. Specifically, at the start of training, let... This means training the network to convergence using only data fitting. Then, the step size is gradually increased according to a certain increment. The value is adjusted until the physical loss falls below a certain threshold. The advantage of this approach is that it avoids excessively strong initial physical constraints that could slow down the network's convergence speed in fitting the data, thus achieving a balance between data fitting and physical constraints in the later stages.

[0057] The results in S3 provide an estimate of the sensor's degradation and the parameter estimates of the inverse Gaussian distribution that the sensor follows.

[0058] S4: Complete sensor reliability evaluation There is a failure threshold for the performance degradation of sensors. d t When the degradation of a sensor reaches the failure threshold, the sensor fails. Therefore, the sensor's lifespan is limited.T The failure threshold can be reached for the first time by the amount of degradation. Time calculation:

[0059] Based on the properties of the inverse Gaussian distribution, the performance degradation path of the sensor follows a strictly increasing function. Therefore, the reliability of the sensor is:

[0060] This concludes the sensor reliability evaluation process, as we now know the first arrival time of failure and the sensor's reliability.

[0061] Taking an SF6 sensor as an example, the sensor is installed on a GIS (Gas Insulation System), and the sensor body is directly exposed to the environment. Therefore, the sensor is affected by ambient temperature and humidity. Furthermore, GIS is typically deployed within substations, where electromagnetic fields are strong, resulting in high electromagnetic stress on the sensor. When the GIS switches off, it generates particularly strong vibrations, which remain strong when transmitted to the sensor's installation location. This strong vibration may cause measurement errors in the SF6 sensor.

[0062] The measurement error of the SF6 sensor is taken as the degradation factor. The measurement error gradually increases with the cumulative usage time of the sensor. When the measurement error of the sensor is > ±10%, the sensor fails. Using a temperature sensor to collect ambient temperature, a humidity sensor to collect ambient humidity, an electromagnetic sensor to collect electromagnetic field intensity, and a vibration sensor to collect vibration, the reliability and remaining lifespan of the SF6 sensor can be obtained according to the method described above.

[0063] See Figure 6 The long-term reliability evaluation system for SF6 sensors based on PINN, as proposed in this embodiment, mainly includes a data acquisition module, a data processing module, a degradation evaluation module, and a reliability evaluation module. The specific functions of each module are as follows: Data acquisition module: Collects characteristic data of the GIS under test through sensor measurement data acquisition device, and measures the stress at the sensor installation location through environmental stress measurement (sensor, measuring device, etc.), including environmental temperature, humidity, electromagnetic field, vibration, etc. The collected sensor data and stress data are transmitted to the data processing module.

[0064] Data processing module: Completes data preprocessing, including alignment of time series data, handling of missing values, outliers, and normalization.

[0065] Degradation evaluation module: Equipped with a sensor performance degradation evaluation model, it calculates the amount of sensor performance degradation based on the processed sensor measurement data and environmental stress data, and fits the trajectory of the inverse Gaussian process that the sensor degradation conforms to.

[0066] Reliability evaluation module: Based on the predicted performance degradation of the sensor output from the degradation evaluation module and the fitted inverse Gaussian process, calculate the first arrival time of sensor failure and the reliability of the sensor.

[0067] Example 2 See Figure 2 A long-term reliability evaluation system for SF6 sensors based on PINN includes: The data preprocessing module is used to collect the operating data of the SF6 sensor and the environmental stress data of the SF6 sensor installation location, and generate time series respectively; the time series are preprocessed to obtain the preprocessed operating data and the preprocessed environmental stress data. The distribution parameter baseline value acquisition module is used to obtain the actual value of the performance degradation of the SF6 sensor. Based on the actual value of the performance degradation, the true distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation are obtained through statistical estimation. The distributed parameter output module is used to construct a physical neural network (PINN) model with dual parallel processing paths. The first path receives preprocessed operating data and preprocessed environmental stress data, and outputs an estimate of the performance degradation of the SF6 sensor during long-term operation. The second path receives preprocessed environmental stress data and outputs the predicted distribution parameters of the monotone irreversible random process that the SF6 sensor follows during long-term operation. The PINN model optimization module is used to train the PINN model using a joint loss function that integrates data loss and physical loss, resulting in a trained PINN model. The data loss is calculated based on the estimated and actual values ​​of performance degradation, while the physical loss is calculated based on the true and predicted distribution parameters. The reliability calculation module is used to calculate the performance degradation of the SF6 sensor and the predicted distribution parameters of the corresponding monotonically irreversible random process using the trained PINN model. Based on the preset performance degradation failure threshold and the probability distribution of the monotonically irreversible random process determined by the predicted distribution parameters, the long-term reliability and remaining service life of the SF6 sensor are calculated.

[0068] Example 3 See Figure 3 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for evaluating the long-term operational reliability of an SF6 sensor based on PINN.

[0069] Example 4 A computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for evaluating the long-term operational reliability of an SF6 sensor based on PINN.

[0070] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, read-only optical discs, optical storage, etc.) containing computer-usable program code.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for evaluating the long-term operational reliability of SF6 sensors based on PINN, characterized in that, Includes the following steps: The operational data of the SF6 sensor and the environmental stress data of the installation location of the SF6 sensor are collected and time series are generated respectively. Each of the time series is preprocessed to obtain preprocessed operational data and preprocessed environmental stress data; Obtain the actual value of the performance degradation of the SF6 sensor, and based on the actual value of the performance degradation, obtain the true distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation through statistical estimation; A physical neural network PINN model with dual parallel processing paths is constructed. The first path receives the preprocessed operating data and the preprocessed environmental stress data, and outputs the estimated value of the performance degradation of the SF6 sensor during long-term operation. The second path receives the preprocessed environmental stress data and outputs the predicted distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation. The PINN model is trained using a joint loss function that integrates data loss and physical loss to obtain a trained PINN model. The data loss is calculated based on the estimated value of the performance degradation and the actual value of the performance degradation, and the physical loss is calculated based on the true distribution parameters and the predicted distribution parameters. Using the trained PINN model, the performance degradation of the SF6 sensor and the corresponding predicted distribution parameters of the monotonic irreversible random process are calculated. Based on the preset performance degradation failure threshold and the probability distribution of the monotonic irreversible random process determined by the predicted distribution parameters, the long-term reliability and remaining service life of the SF6 sensor are calculated.

2. The method for evaluating the long-term operational reliability of an SF6 sensor based on PINN according to claim 1, characterized in that, The monotonically irreversible random process is an inverse Gaussian process; The statistical estimate is a maximum likelihood estimate.

3. The method for evaluating the long-term operational reliability of an SF6 sensor based on PINN according to claim 1, characterized in that, The step of obtaining the actual value of the performance degradation of the SF6 sensor specifically involves: performing periodic offline calibration on the SF6 sensor to obtain the actual value of the performance degradation at different operating time points.

4. The method for evaluating the long-term operational reliability of an SF6 sensor based on PINN according to claim 1, characterized in that, The environmental stress data includes ambient temperature data, humidity data, electromagnetic field strength data, and vibration data.

5. The method for evaluating the long-term operational reliability of an SF6 sensor based on PINN according to claim 1, characterized in that, The preprocessing includes time alignment, missing value handling, outlier handling, and normalization.

6. The method for evaluating the long-term operational reliability of an SF6 sensor based on PINN according to claim 1, characterized in that, The first path adopts a gated cyclic cell network structure, and the second path adopts a fully connected network structure.

7. The method for evaluating the long-term operational reliability of an SF6 sensor based on PINN according to claim 1, characterized in that, The weight coefficients of data loss and physical loss in the joint loss function are determined using a progressive learning method. In the initial training phase, the weight coefficient of physical loss is set to 0, and the model is trained to convergence using only data loss. Subsequently, the weight coefficient of physical loss is gradually increased until the physical loss is lower than a preset threshold.

8. A long-term reliability evaluation system for SF6 sensors based on PINN, characterized in that, include: The data preprocessing module is used to collect the operating data of the SF6 sensor and the environmental stress data of the installation location of the SF6 sensor, and generate time series data respectively. Each of the time series is preprocessed to obtain preprocessed operational data and preprocessed environmental stress data; The distribution parameter benchmark value acquisition module is used to acquire the actual value of the performance degradation of the SF6 sensor, and based on the actual value of the performance degradation, to obtain the true distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation through statistical estimation. The distributed parameter output module is used to construct a physical neural network PINN model with dual parallel processing paths. The first path receives the preprocessed running data and the preprocessed environmental stress data, and outputs the estimated value of the performance degradation of the SF6 sensor during long-term operation. The second path receives the preprocessed environmental stress data and outputs the predicted distribution parameters of the monotonous irreversible random process that the SF6 sensor follows during long-term operation. The PINN model optimization module is used to train the PINN model using a joint loss function that integrates data loss and physical loss to obtain a trained PINN model. The data loss is calculated based on the estimated value of the performance degradation and the actual value of the performance degradation, and the physical loss is calculated based on the true distribution parameters and the predicted distribution parameters. The reliability calculation module is used to calculate the performance degradation of the SF6 sensor and the corresponding predicted distribution parameters of the monotone irreversible random process using the trained PINN model. Based on the preset performance degradation failure threshold and the probability distribution of the monotone irreversible random process determined by the predicted distribution parameters, the module calculates the long-term reliability and remaining service life of the SF6 sensor.

9. An electronic device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the long-term operational reliability evaluation method for the PINN-based SF6 sensor as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the long-term operational reliability evaluation method for SF6 sensors based on PINN as described in any one of claims 1-7.

Images