GIS (Geographic Information System) fault monitoring method, system, device and equipment based on cloud edge collaboration, and readable storage medium

Abstract

The invention relates to a GIS fault monitoring method, system and device based on cloud edge collaboration, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring a monitoring data sample of GIS equipment, and training a to-be-trained teacher model by using the monitoring data sample to obtain a trained teacher model; based on the teacher model, searching a student model architecture with the optimal performance under the hardware constraint condition of the edge equipment by using a neural architecture search method, and obtaining a basic student model according to the student model architecture; according to the trained teacher model, performing model compression and knowledge distillation processing on the basic student model to obtain a student model corresponding to the edge device; deploying the student model to each edge device; and the edge device is used for obtaining a fault monitoring result of the target GIS device according to the monitoring data of the target GIS device by using the student model. By adopting the method, the fault detection efficiency of the GIS equipment can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power safety, in particular to a GIS fault monitoring method, system and device based on cloud-edge collaboration, computer equipment, computer readable storage medium and computer program product. BACKGROUND

[0002] Gas-Insulated Switchgear (GIS) is a key device in the power system, and the operation reliability of GIS plays a key role in ensuring the safe and stable operation of the power system. Traditional GIS monitoring mainly relies on manual regular inspection and system online monitoring. Among them, the online monitoring system usually uploads all the raw data (such as ultra high frequency (UHF) signals, ultrasonic waves, vibration signals, etc.) of GIS equipment collected by sensors to the cloud or data center through a communication network, and then performs unified analysis and processing in the cloud. However, this method cannot efficiently detect the fault conditions of GIS equipment. SUMMARY

[0003] Therefore, it is necessary to provide a GIS fault monitoring method, system and device based on cloud-edge collaboration, computer equipment, computer readable storage medium and computer program product to solve the above technical problems.

[0004] In a first aspect, the present application provides a GIS fault monitoring method based on cloud-edge collaboration, comprising:

[0005] Obtaining a monitoring data sample of a GIS device, training a teacher model to be trained using the monitoring data sample, and obtaining a trained teacher model;

[0006] Based on the teacher model, a student model architecture with optimal performance under the hardware constraint conditions of the edge device is searched using a neural architecture search method, and a basic student model is obtained according to the student model architecture;

[0007] According to the trained teacher model, the basic student model is subjected to model compression and knowledge distillation processing, and a student model corresponding to the edge device is obtained;

[0008] Deploying the student model to each edge device; the edge device is used to obtain a fault monitoring result of a target GIS device according to the monitoring data of the target GIS device using the student model.

[0009] In one of the embodiments, the monitoring data samples include unlabeled data samples and labeled data samples; the obtaining of the monitoring data samples of the GIS device, the training of the to-be-trained teacher model by using the monitoring data samples, and the obtaining of the trained teacher model include: obtaining a physical law constraint of the GIS device; and training the to-be-trained teacher model by using a semi-supervised learning method according to the physical law constraint and the monitoring data samples, to obtain the trained teacher model.

[0010] In one of the embodiments, the method further includes: receiving model update information from each edge device; the model update information is obtained by updating the student model according to monitoring data of the target GIS device by the edge device; aggregating each model update information by using a federated learning algorithm to obtain an updated student model; and deploying the updated student model to each edge device.

[0011] In a second aspect, the application further provides a GIS fault monitoring method based on cloud-edge collaboration, including:

[0012] Obtaining monitoring data of a target GIS device;

[0013] Obtaining a fault monitoring result of the target GIS device according to the monitoring data by using a student model; the student model is from a cloud server, and the student model is obtained by the cloud server by performing neural architecture search, model compression and knowledge distillation processing under the hardware constraint condition of an edge device according to a trained teacher model; the trained teacher model is trained by the cloud server by using monitoring data samples of a GIS device.

[0014] In one of the embodiments, the obtaining of the fault monitoring result of the target GIS device according to the monitoring data by using the student model includes: performing anomaly detection on the monitoring data; and obtaining the fault monitoring result of the target GIS device according to the anomaly data by using the student model when it is detected that the monitoring data contains anomaly data.

[0015] In one of the embodiments, the obtaining of the fault monitoring result of the target GIS device according to the monitoring data by using the student model includes: obtaining a fault prediction label of the target GIS device and a confidence of the fault prediction label according to the monitoring data by using the student model; obtaining the fault monitoring result of the target GIS device according to the fault prediction label when the confidence is greater than a confidence threshold; and sending the monitoring data to the cloud server when the confidence is not greater than the confidence threshold; and the cloud server is configured to send the monitoring data to a terminal of an operation and maintenance personnel.

[0016] In one of the embodiments, the method further comprises: updating the student model according to the monitoring data of the target GIS device, obtaining model update information of the student model; sending the model update information to the cloud server; receiving an updated student model from the cloud server; the updated student model is obtained by the cloud server by aggregating the model update information from each edge device using a federated learning algorithm.

[0017] In a third aspect, the application further provides a GIS fault monitoring system based on cloud-edge collaboration, comprising: a cloud server and edge devices.

[0018] The cloud server is configured to obtain monitoring data samples of GIS devices, train a teacher model to be trained using the monitoring data samples, and obtain a trained teacher model; based on the teacher model, search for a student model architecture with optimal performance under the hardware constraint condition of the edge device using a neural architecture search method, and obtain a basic student model according to the student model architecture; perform model compression and knowledge distillation processing on the basic student model according to the trained teacher model, and obtain a student model corresponding to the edge device; and deploy the student model to each edge device.

[0019] The edge device is configured to obtain monitoring data of a target GIS device, and obtain a fault monitoring result of the target GIS device according to the monitoring data using the student model.

[0020] In a fourth aspect, the application further provides a GIS fault monitoring device based on cloud-edge collaboration, comprising:

[0021] The model training module is configured to obtain monitoring data samples of GIS devices, train a teacher model to be trained using the monitoring data samples, and obtain a trained teacher model.

[0022] The architecture search module is configured to search for a student model architecture with optimal performance under the hardware constraint condition of the edge device based on the teacher model using a neural architecture search method, and obtain a basic student model according to the student model architecture.

[0023] The model compression module is configured to perform model compression and knowledge distillation processing on the basic student model according to the trained teacher model, and obtain a student model corresponding to the edge device.

[0024] The model distribution module is configured to deploy the student model to each edge device; and the edge device is configured to obtain a fault monitoring result of a target GIS device according to monitoring data of the target GIS device using the student model.

[0025] In a fifth aspect, the application further provides a GIS fault monitoring device based on cloud-edge collaboration, comprising:

[0026] a data acquisition module configured to acquire monitoring data of a target GIS device;

[0027] a result acquisition module configured to obtain a fault monitoring result of the target GIS device according to the monitoring data by using a student model; the student model is obtained from a cloud server, and the student model is obtained by the cloud server from a trained teacher model by performing neural architecture search, model compression and knowledge distillation processing under hardware constraint conditions of an edge device; the trained teacher model is obtained by the cloud server by training using monitoring data samples of GIS devices.

[0028] In a sixth aspect, the application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0029] acquiring monitoring data samples of GIS devices, training a to-be-trained teacher model using the monitoring data samples to obtain a trained teacher model;

[0030] searching for a student model architecture with optimal performance under hardware constraint conditions of an edge device based on the teacher model using a neural architecture search method, and obtaining a basic student model according to the student model architecture;

[0031] performing model compression and knowledge distillation processing on the basic student model according to the trained teacher model to obtain a student model corresponding to the edge device;

[0032] deploying the student model to each edge device; the edge device is configured to obtain a fault monitoring result of a target GIS device according to monitoring data of the target GIS device by using the student model.

[0033] In a seventh aspect, the application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0034] acquiring monitoring data of a target GIS device;

[0035] obtaining a fault monitoring result of the target GIS device according to the monitoring data by using a student model; the student model is obtained from a cloud server, and the student model is obtained by the cloud server from a trained teacher model by performing neural architecture search, model compression and knowledge distillation processing under hardware constraint conditions of an edge device; the trained teacher model is obtained by the cloud server by training using monitoring data samples of GIS devices.

[0036] In an eighth aspect, the present application also provides a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the following steps:

[0037] obtaining monitoring data samples of GIS devices, training a teacher model to be trained by using the monitoring data samples to obtain a trained teacher model;

[0038] based on the teacher model, searching for a student model architecture with optimal performance under the hardware constraint condition of an edge device by using a neural architecture search method, and obtaining a basic student model according to the student model architecture;

[0039] performing model compression and knowledge distillation processing on the basic student model according to the trained teacher model to obtain a student model corresponding to the edge device;

[0040] deploying the student model to each edge device; the edge device is configured to obtain a fault monitoring result of a target GIS device according to monitoring data of the target GIS device by using the student model.

[0041] In a ninth aspect, the present application also provides a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the following steps:

[0042] obtaining monitoring data of a target GIS device;

[0043] obtaining a fault monitoring result of the target GIS device according to the monitoring data by using a student model; the student model is from a cloud server, and the student model is obtained by the cloud server by performing neural architecture search, model compression and knowledge distillation processing on a trained teacher model under the hardware constraint condition of an edge device; the trained teacher model is trained by the cloud server by using monitoring data samples of GIS devices.

[0044] In a tenth aspect, the present application also provides a computer program product comprising a computer program, the computer program being executed by a processor to implement the following steps:

[0045] obtaining monitoring data samples of GIS devices, training a teacher model to be trained by using the monitoring data samples to obtain a trained teacher model;

[0046] based on the teacher model, searching for a student model architecture with optimal performance under the hardware constraint condition of an edge device by using a neural architecture search method, and obtaining a basic student model according to the student model architecture;

[0047] According to the trained teacher model, the base student model is subjected to model compression and knowledge distillation processing to obtain a student model corresponding to the edge device;

[0048] The student model is deployed to each edge device, and the edge device is configured to utilize the student model to obtain a fault monitoring result of a target GIS device according to monitoring data of the target GIS device.

[0049] In a first aspect, the present application provides a GIS fault monitoring method based on cloud-edge collaboration, comprising the following steps:

[0050] Obtaining monitoring data of a target GIS device;

[0051] Utilizing a student model to obtain a fault monitoring result of the target GIS device according to the monitoring data; the student model is from a cloud server, and the student model is obtained by the cloud server through neural architecture search, model compression and knowledge distillation processing under the hardware constraint condition of an edge device, based on a trained teacher model; the trained teacher model is trained by the cloud server using monitoring data samples of a GIS device.

[0052] The above-mentioned GIS fault monitoring method, system, device, computer equipment, computer readable storage medium and computer program product based on cloud-edge collaboration, obtain monitoring data samples of a GIS device and train a teacher model to be trained using the monitoring data samples to obtain a trained teacher model, then search for a student model architecture with optimal performance under the hardware constraint condition of an edge device based on the teacher model using a neural architecture search method and obtain a base student model according to the student model architecture, further perform model compression and knowledge distillation processing on the base student model according to the trained teacher model to obtain a student model corresponding to the edge device, and then deploy the student model to each edge device, wherein the edge device is configured to utilize the student model to obtain a fault monitoring result of a target GIS device according to monitoring data of the target GIS device. This scheme, in view of the high sampling frequency and large data volume of the monitoring data of the GIS device, first trains a teacher model capable of detecting the fault condition of the GIS device with high precision in the cloud, then performs architecture search under the hardware constraint of the edge device using the teacher model as a guide to obtain a base student model, and performs model compression and knowledge distillation processing on the base student model to obtain a student model that is compatible with the hardware condition of the edge device and has high accuracy. Therefore, by deploying the student model to each edge device, each edge device can utilize the student model to efficiently and reliably detect the fault of the corresponding target GIS device, which is conducive to ensuring the safe and stable operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the accompanying drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the accompanying drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0054] Figure 1 The figure is a schematic diagram of the architecture of a GIS fault monitoring system based on cloud-edge collaboration in one embodiment.

[0055] Figure 2 The figure is a schematic diagram of the architecture of a GIS fault monitoring system based on cloud-edge collaboration in another embodiment.

[0056] Figure 3 The figure is a schematic diagram of the process of GIS fault monitoring by an edge device in one embodiment.

[0057] Figure 4 The figure is a schematic diagram of the process of a GIS fault monitoring method based on cloud-edge collaboration in one embodiment.

[0058] Figure 5 The figure is a schematic diagram of the process of a GIS fault monitoring method based on cloud-edge collaboration in another embodiment.

[0059] Figure 6 The figure is a structural block diagram of a GIS fault monitoring device based on cloud-edge collaboration in one embodiment.

[0060] Figure 7 The figure is a structural block diagram of a GIS fault monitoring device based on cloud-edge collaboration in another embodiment.

[0061] Figure 8 The figure is an internal structure diagram of a computer device in one embodiment.

[0062] Figure 9 The figure is an internal structure diagram of a computer device in another embodiment. DETAILED DESCRIPTION

[0063] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0064] It should be noted that the terms "first", "second", etc. used in the present application can be used to describe various objects, but these objects are not limited by these terms. These terms are only used to distinguish the first object from the second object. The terms "include" and "have" and any variations thereof used in the present application are intended to cover non-exclusive inclusion. The term "a plurality of" used in the present application means two or more. The term "and / or" used in the present application means one of the options or any combination of the options.

[0065] In an exemplary embodiment, as shown in Figure 1 a cloud edge collaborative based GIS fault monitoring system is provided, which comprises a cloud server and an edge device. The cloud server is in communication connection with the edge device. The number of edge devices can be multiple, and each edge device can be used for fault monitoring of one or more target GIS devices. The cloud server can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The edge device can be a standalone physical server or a terminal device.

[0066] The cloud server is configured to obtain monitoring data samples of the GIS devices, train a teacher model to be trained using the monitoring data samples, and obtain a trained teacher model.

[0067] Specifically, the cloud server can obtain monitoring data samples of the GIS devices. The monitoring data samples can include multiple sets of monitoring data collected from multiple GIS devices in a historical period. For example, a set of monitoring data of a GIS device can include data collected by a sensor deployed on the GIS device, such as vibration, ultra-high frequency signal, temperature, humidity, etc. The monitoring data samples further include label information corresponding to at least part of the monitoring data. The label information corresponding to a set of monitoring data can be used to indicate whether the GIS device corresponding to the set of monitoring data has failed.

[0068] The cloud server can train the teacher model to be trained using the monitoring data samples to obtain a trained teacher model. The teacher model to be trained can be a pre-constructed neural network model, such as a convolutional neural network, a Transformer, etc.

[0069] Exemplarily, when the monitoring data samples are labeled data samples, a supervised learning method can be used to train the teacher model to be trained, so that the model can learn the relationship between the monitoring data of the GIS device and whether the device fails. Exemplarily, when the monitoring data samples include unlabeled data samples and labeled data samples, a semi-supervised learning method can be used to train the teacher model to be trained, so that the model first learns feature representation according to the unlabeled data samples, and then performs supervised fine-tuning on the model according to the labeled data samples, to train a teacher model with high precision and generalization ability. Exemplarily, when training the teacher model, the model can also be trained in combination with the physical law constraint of the GIS device.

[0070] The cloud server is further configured to search, based on the teacher model, a student model architecture with optimal performance under hardware constraint conditions of the edge device by using a neural architecture search method, and obtain a basic student model according to the student model architecture.

[0071] The cloud server can obtain the hardware constraint conditions of the edge device, and search the architecture of the student model to be deployed on the edge device by using a hardware-aware neural architecture search (NAS) method under the guidance of the teacher model. The hardware constraint conditions of the edge device can be obtained according to the hardware parameters of the edge device, for example, can include delay requirements, memory limitations, and power consumption budgets.

[0072] Exemplarily, when performing neural architecture search, the overall architecture of the student model can be determined under the guidance of the teacher model, and then the operation and parameters of each layer of the model can be determined by using the neural architecture search method under the hardware constraint conditions. Exemplarily, the hardware constraint conditions can be used to optimize the search space of the NAS, for example, the number of layers of the student model or the optional operations of each layer can be limited according to the hardware constraint conditions; the hardware constraint conditions can also be used to construct the loss function of the NAS, for example, a loss function including loss terms corresponding to model delay, energy consumption, and memory occupancy can be constructed, so that architecture search can be performed under hardware constraint conditions. Exemplarily, the student model architecture with optimal performance can be the model architecture with the highest accuracy under the condition of meeting the hardware constraint conditions of the edge device. According to the student model architecture, a basic student model can be obtained.

[0073] The cloud server is further configured to perform model compression and knowledge distillation processing on the basic student model according to the trained teacher model, to obtain a student model corresponding to the edge device; and deploy the student model to each edge device.

[0074] The cloud server can input the basic student model into an automated pipeline, which can integrate model compression and knowledge distillation techniques. The model compression and knowledge distillation techniques can be used to compress and distill the knowledge of the trained teacher model to the basic student model to obtain a student model that can be deployed to the edge device.

[0075] The model compression process can include removing redundant convolution kernels and neuron channels in the basic student model by a structured pruning method, quantization-aware training of the basic student model, and the like. The knowledge distillation process can be based on a teacher-student architecture and use soft labels to transfer the knowledge of the trained teacher model to the basic student model. The knowledge distillation process can be performed before or after the model compression process to further improve the accuracy of the student model. The automated pipeline can also use a Bayesian optimization algorithm to automatically search for an optimal compression parameter combination and use the parameter combination to compress the basic student model.

[0076] After obtaining the student model, the cloud server can send the model file to each edge device, so that each edge device can deploy the student model.

[0077] The edge device is configured to obtain monitoring data of a target GIS device.

[0078] Each edge device in the system can be configured to monitor the fault of one or more target GIS devices. The monitoring data of the target GIS device can include, but is not limited to, real-time collected vibration, ultra-high frequency signal, temperature, humidity, and the like. These data can be collected by sensors deployed on the target GIS device and sent to the corresponding edge device, so that the edge device can obtain the monitoring data of the target GIS device.

[0079] The edge device is further configured to use the student model to obtain a fault monitoring result of the target GIS device based on the monitoring data.

[0080] The edge device can input the monitoring data of the target GIS device into the student model deployed locally to the edge device, so that the student model can infer whether the target GIS device has a fault based on the monitoring data, and then obtain the fault monitoring result of the target GIS device. The fault monitoring result can be used to indicate whether the target GIS device has a fault.

[0081] The GIS fault monitoring system based on cloud-edge collaboration can obtain a student model that is compatible with the hardware conditions of the edge device and has high accuracy by training a teacher model that can detect the fault condition of the GIS device with high precision on the cloud, guiding architecture search of the teacher model to obtain a basic student model under the hardware constraints of the edge device, and performing model compression and knowledge distillation processing on the basic student model. By deploying the student model to each edge device, each edge device can use the student model to efficiently and reliably detect the fault of the corresponding target GIS device, which is conducive to ensuring the safe and stable operation of the power system.

[0082] In an exemplary embodiment, the edge device can also be configured to update the student model according to the monitoring data of the target GIS device, obtain model update information of the student model, and send the model update information to the cloud server. The cloud server can also be configured to receive the model update information from each edge device, aggregate the model update information using a federated learning algorithm to obtain an updated student model, and deploy the updated student model to each edge device. The edge device can also be configured to receive the updated student model from the cloud server.

[0083] Specifically, in this embodiment, the collaborative evolution of the model can be achieved through a federated learning algorithm.

[0084] The edge device can update the student model locally using the collected monitoring data of the target GIS device and obtain model update information corresponding to the student model. For example, the edge device can fine-tune the student model using the monitoring data of the target GIS device through self-supervised learning. The model update information can include adjusted model parameters or changes in model parameters, etc. The edge device can encrypt the model update information and upload the encrypted model update information to the cloud server.

[0085] The cloud server can receive the model update information from the edge devices, aggregate the model update information using a federated learning algorithm to obtain a global model. For example, assuming that the model update information of the edge devices includes adjusted model parameters of the student model deployed by the edge devices, the cloud server can aggregate the adjusted model parameters using a federated averaging algorithm (FedAvg), and obtain an updated student model according to the aggregation result. Subsequently, the cloud server can send the data of the updated student model to the edge devices. Alternatively, the cloud server can also compress the updated student model through an automated pipeline before sending the updated student model to the edge devices, and then send the compressed model to the edge devices.

[0086] The edge devices in the system can receive the data of the updated student model sent by the cloud server, and locally deploy the updated student model according to the received data. The data sent by the cloud server can be model parameters of the updated student model, or parameter update values of the updated student model relative to the student model deployed by the edge devices. The cloud server can encrypt the data and send the encrypted data to the edge devices.

[0087] For example, the above updating process can be periodically performed according to a preset federated learning update period.

[0088] In this embodiment, the student model is updated locally by the edge devices and the model update information is uploaded to the cloud server, and the cloud server aggregates the model update information uploaded by multiple edge devices through federated learning, which can integrate the learned knowledge and features of multiple edge devices to obtain an updated student model with higher performance. Therefore, by deploying the updated student model to each edge device, each edge device can use the new student model to monitor the fault of the target GIS device, which is beneficial to improve the fault monitoring accuracy and reliability of the GIS device.

[0089] In an exemplary embodiment, the monitoring data samples include unlabeled data samples and labeled data samples; the cloud server can also be configured to obtain physical law constraints of the GIS device; and the trained teacher model is obtained by training the to-be-trained teacher model using a semi-supervised learning method according to the physical law constraints and the monitoring data samples.

[0090] The monitoring data samples used to train the teacher model can include unlabeled data samples and labeled data samples. The unlabeled data samples can include monitoring data corresponding to the GIS device, and the labeled data samples can include monitoring data corresponding to the GIS device and label information labeled by an expert. The number of the unlabeled data samples can be greater than the number of the labeled data samples. In this embodiment, the teacher model to be trained can be trained based on the monitoring data samples containing a large number of unlabeled data samples and a small number of labeled data samples, and the physical law constraint of the GIS device to obtain the trained teacher model.

[0091] In the semi-supervised learning, the model can be first trained by using the contrast learning method to learn feature representation, and then the model can be supervised trained based on the physical law constraint of the GIS device and the labeled data samples to obtain the trained teacher model. The physical law constraint of the GIS device can reflect the physical relationship between the monitoring data of the GIS device and the device fault, for example, can include mechanical vibration law, UHF signal law and the like in the normal operation state and the fault state of the GIS device. When the model is supervised trained, the physical law constraint of the GIS device can be constructed as a regularization term added to the loss function, so that the inference of the model on the fault condition of the GIS device conforms to the physical law.

[0092] In this embodiment, for the case that the monitoring data samples only contain a small amount of labeled data, the teacher model to be trained can be trained by using the semi-supervised learning method, so that a large amount of unlabeled monitoring data samples can be fully utilized, the model can still obtain rich feature representation under limited supervision information, and the generalization ability and accuracy of the model can be improved. At the same time, by adding the physical law constraint of the GIS device to the loss function, the model can learn the physical relationship between the monitoring data of the GIS device and the device fault, and the generalization ability of the model can be further improved.

[0093] In an exemplary embodiment, the edge device can also be used for anomaly detection of the monitoring data. When it is detected that the monitoring data contains abnormal data, the student model is used to obtain the fault monitoring result of the target GIS device according to the abnormal data.

[0094] The edge device can perform abnormality detection on the monitoring data before inferring according to the monitoring data of the target GIS device by using the student model, and filter out obviously normal data segments. For example, the edge device can calculate the signal effective value, peak value, waveform factor and other data characteristics of the target GIS device according to the monitoring data of the target GIS device in real time, and then match the calculated characteristics with the pre-stored abnormality judgment rule. When the data characteristics of a segment of monitoring data match the abnormality judgment rule, it can be determined that the segment of monitoring data is abnormal data, otherwise it can be determined that the segment of monitoring data is normal data.

[0095] The edge device can input the abnormal data segment to the student model and infer according to the abnormal data by using the student model to obtain the fault monitoring result of the target GIS device when it is detected that the monitoring data contains abnormal data.

[0096] In this embodiment, by performing abnormality detection on the monitoring data before inferring by using the student model, filtering out normal data and inputting only abnormal data to the student model for inference, the data processing amount of the model can be reduced, the computational burden and energy consumption of the edge device can be reduced, and the fault detection efficiency of the GIS device as a whole can be improved.

[0097] In an exemplary embodiment, the edge device can also be configured to use the student model to obtain a fault prediction label and a confidence of the fault prediction label of the target GIS device according to the monitoring data; obtain a fault monitoring result of the target GIS device according to the fault prediction label when the confidence is greater than a confidence threshold; and send the monitoring data to a cloud server when the confidence is not greater than the confidence threshold.

[0098] The student model deployed by the edge device can infer whether the target GIS device has failed according to the input monitoring data and output a corresponding fault prediction label and a confidence, wherein the fault prediction label can be used to indicate whether the target GIS device has failed.

[0099] The edge device can compare the confidence with a preset confidence threshold after obtaining the fault prediction label and the confidence output by the model. When the confidence is greater than the confidence threshold, the fault prediction label can be used as the fault monitoring result of the target GIS device, and when the confidence is not greater than the confidence threshold, the segment of monitoring data can be uploaded to the cloud server for expert intervention.

[0100] When the cloud server receives the monitoring data from the edge device, it can send the data to the terminal of the operation and maintenance personnel, and the operation and maintenance personnel can determine whether the target GIS device has failed according to the segment of monitoring data.

[0101] In the embodiment, after the student model outputs the failure prediction label and the confidence of the target GIS device, the confidence is evaluated, and when the confidence exceeds the confidence threshold, the label is adopted as the failure monitoring result, and when the confidence does not exceed the confidence threshold, expert intervention is requested, which can avoid that the failure prediction label with low reliability is directly adopted, and is beneficial to improve the reliability and robustness of failure monitoring.

[0102] In one exemplary embodiment, as shown in Figure 2 a cloud edge collaborative GIS failure monitoring system is provided.

[0103] Specifically, as shown in Figure 2 the GIS failure monitoring system in the embodiment can include a cloud collaborative evolution platform (hereinafter referred to as a "cloud server") and an edge intelligent monitoring terminal (hereinafter referred to as an "edge device").

[0104] The cloud server can include a semi-supervised training module, an automated NAS module, a one-stop compression pipeline module, and a federated learning aggregation module.

[0105] The semi-supervised training module can be used to train a teacher model to be trained by using the monitoring data samples of the GIS device in combination with physical law constraints to obtain a trained teacher model. Exemplarily, the semi-supervised training module can include: a data collection unit for collecting massive unlabeled data samples and expert-annotated labeled data samples of the GIS device; a feature learning unit for processing the unlabeled data samples using a contrastive learning algorithm to extract feature representations; a physical constraint unit for adding the physical laws of the discharge, mechanical vibration and other monitoring data of the GIS device as regularization terms to the loss function; a model training unit for supervised training of the model in combination with the labeled data samples and the loss function containing the physical law constraints, and outputting the trained teacher model.

[0106] The automated NAS module can be used to automatically search for an optimal student model architecture under the hardware constraint condition according to the hardware constraint of the edge device, and obtain a basic student model according to the student model architecture. Exemplarily, the automated NAS module can include: a hardware constraint definition unit for setting hardware parameters of the edge device, including delay requirements, memory limitations and power consumption budgets; an architecture search unit for automatically searching for an optimal student model architecture using a one-time NAS method based on a performance predictor; and an architecture evaluation unit for performance evaluation and verification of the searched architecture.

[0107] A one-stop compression pipeline module can be used to automatically complete the pruning, quantization and distillation of the base student model to generate a lightweight student model. Exemplarily, the one-stop compression pipeline module can include a pruning processing unit configured to remove redundant convolution kernels and neuron channels using a structured pruning method; a quantization training unit configured to implement quantization-aware training which can support multiple precision conversions such as FP16, INT8, etc.; a knowledge distillation unit configured to use a teacher-student framework to transfer the knowledge of the teacher model to the student model through soft labels; and a hyperparameter optimization unit configured to use a Bayesian optimization algorithm to automatically search for the optimal compression parameter combination.

[0108] A federated learning aggregation module can be used to securely aggregate the model update quantities uploaded by each terminal to generate a new generation of global model. Exemplarily, the federated learning aggregation module can include a secure communication unit configured to establish an encrypted communication channel with the edge device and receive the model update information uploaded by the edge device through the channel; a model aggregation unit configured to aggregate the model update information from multiple edge devices using a FedAvg algorithm to obtain an updated student model; a version management unit configured to manage model versions and distribution records; and a quality monitoring unit configured to monitor model update quality and performance indicators.

[0109] The edge device can include a signal acquisition and preprocessing module, a hierarchical diagnosis engine, an explainability processing module, a local learning and communication module, and a security protection module.

[0110] The signal acquisition and preprocessing module can be used to acquire monitoring data of the target GIS device and perform preliminary filtering and standardization on the data to obtain processed monitoring data. Exemplarily, the signal acquisition and preprocessing module can include a multi-sensor interface unit configured to access vibration, UHF, temperature, humidity and other sensors and acquire data collected by these sensors; a signal conditioning unit configured to provide signal amplification, filtering and isolation functions; an ADC conversion unit configured to implement high-precision conversion of analog signals to digital signals; and a cache management unit configured to manage data buffering and storage.

[0111] The layered diagnosis engine can be internally provided with a rule filtering layer and an AI diagnosis layer, which can be used for rapid screening and fine diagnosis of the processed monitoring data. Illustratively, the rule filtering layer can include: a feature extraction subunit for calculating effective values, peak values, waveform factors and other features of the monitoring data in real time; a rule library subunit for storing expert experience and diagnosis rules; a rapid decision subunit for performing millisecond-level anomaly detection and filtering on the monitoring data according to the features of the monitoring data obtained by the feature extraction subunit and the expert experience and diagnosis rules stored in the rule library subunit. Illustratively, the AI diagnosis layer can include: a model loading subunit for loading and initializing a student model from a cloud server; a dynamic reasoning subunit for dynamically selecting a calculation path according to the input complexity by using a multi-branch structure; a confidence evaluation subunit for calculating a confidence score of the output result.

[0112] The explainability processing module can be used to generate a key feature heat map during diagnosis. Illustratively, the explainability processing module can include: a feature visualization unit for integrating a lightweight explainability algorithm (such as the gradient-weighted class activation mapping method) to generate a diagnosis heat map identifying key features; a result interpretation unit for providing visual interpretation and feature importance analysis of the diagnosis result; a data compression unit for efficiently compressing the visualization result.

[0113] The local learning and communication module can be used to fine-tune the student model locally and communicate with the cloud server in an encrypted manner, upload model update information or receive an updated student model. Illustratively, the local learning and communication module can include: a local training unit for fine-tuning the student model according to new data; a secure encryption unit for implementing encryption processing of model update information; a communication protocol unit for communicating with the cloud server, which can support multiple communication methods such as 4G / 5G, Ethernet, etc.; a power consumption management unit for optimizing communication power consumption and supporting low-power mode.

[0114] The security protection module can be used to ensure the security, reliability and stability of the edge device throughout its entire life cycle from startup to operation. Illustratively, the security protection module can include: a secure startup unit for ensuring the security of the edge device startup process; an integrity verification unit for verifying the integrity and authenticity of the model file of the student model; an anomaly detection unit for monitoring the running state of the edge device and detecting abnormal behavior; a fault tolerance processing unit for providing fault recovery and degraded operation capability.

[0115] Illustratively, in the present embodiment, the edge device can perform fault monitoring on the target GIS device through layered diagnosis and dynamic reasoning, and the process can be as shown in Figure 3 .

[0116] Specifically, the edge device can collect the monitoring data of the target GIS device in real time through the signal collection and preprocessing module, and perform abnormality detection on the raw data by using a lightweight signal processing algorithm and a rule engine through a rule filtering layer of the layered diagnosis engine. The lightweight signal processing algorithm can include calculating the effective value, peak value or waveform factor of the signal.

[0117] Subsequently, the edge device can determine a processing path according to the abnormality detection result output by the rule filtering layer through an AI diagnosis layer in the layered diagnosis engine. When the abnormality detection result indicates that the monitoring data is normal data, the normal data can be filtered out. When the abnormality detection result indicates that the monitoring data is abnormal data, a student model can be used to infer according to the abnormal data to obtain a fault prediction label of the target GIS device and a confidence of the label. When the student model is used for inference, an explainability module can be used to generate a diagnosis heat map identifying key features by using an explainability algorithm. The diagnosis heat map can be used to visually represent the features that the student model focuses on when making a fault prediction.

[0118] The AI diagnosis layer of the edge device can also determine whether the confidence corresponding to the fault prediction label of the target GIS device is greater than a confidence threshold. When the confidence is greater than the threshold, the fault monitoring result of the target GIS device can be obtained according to the fault prediction label. When the confidence is not greater than the threshold, the edge device can trigger uploading of the abnormal data segment to the cloud server to request expert intervention.

[0119] The edge device can also collect environmental data synchronously, and use the environmental data as auxiliary input features of the student model or for post-calibration of the fault monitoring result of the target GIS device.

[0120] In this embodiment, the model is trained on the cloud server, and then deployed to the edge device for inference. This can make full use of the computing resources of the cloud server to fully train the model, and can also enable the edge device to complete the fault monitoring of the target GIS device locally without uploading the monitoring data of the device to the cloud, thereby effectively avoiding the problems of large communication bandwidth pressure, large storage pressure, poor real-time performance, waste of computing resources, and data security risks in the traditional centralized processing mode, and avoiding the problems of limited edge computing power and storage resources in the full edge training and inference mode. In the model training process of the cloud server, the architecture of the student model is searched through the hardware-aware neural architecture search method under the guidance of the teacher model, and the basic student model constructed based on the architecture is subjected to model compression and knowledge distillation processing, so that the student model is adapted to the hardware conditions of the edge device and has better performance. At the same time, the student model is evolved in a collaborative manner through the use of federated learning, which can integrate the learned knowledge and features of multiple edge devices to obtain an updated student model with higher performance. Moreover, in this embodiment, the hierarchical diagnosis and dynamic inference of the monitoring data of the target GIS device by the edge device can reduce the redundant processing of normal data by the edge device and improve the reliability of the fault monitoring results. Therefore, the system provided in this embodiment not only realizes the lightweight deployment of the artificial intelligence model, but also builds a "cloud collaborative evolution-edge intelligent perception" closed-loop system to fundamentally solve the core problems of limited computing power and storage resources, data scarcity, model generalization, update and maintenance, result credibility, and system robustness in GIS fault monitoring.

[0121] Based on the same inventive concept, the embodiments of the present application also provide a GIS fault monitoring method based on cloud-edge collaboration. The implementation scheme for solving the problem provided by the method is similar to the implementation scheme described in the above system embodiments, so the specific limitations in one or more GIS fault monitoring methods based on cloud-edge collaboration provided below can refer to the limitations of the GIS fault monitoring system based on cloud-edge collaboration described above, which will not be repeated here.

[0122] In one exemplary embodiment, as shown in Figure 4 , a GIS fault monitoring method based on cloud-edge collaboration is provided. Taking the cloud server in Figure 1 as an example, the method includes the following steps:

[0123] Step S401, obtaining monitoring data samples of a GIS device, training a teacher model to be trained using the monitoring data samples to obtain a trained teacher model.

[0124] Step S402, based on the teacher model, search for a student model architecture with optimal performance under the hardware constraint condition of the edge device by using a neural architecture search method, and obtain a basic student model according to the student model architecture.

[0125] Step S403, according to the trained teacher model, perform model compression and knowledge distillation processing on the basic student model to obtain a student model corresponding to the edge device.

[0126] Step S404, deploy the student model to each edge device; the edge device is used to obtain a fault monitoring result of the target GIS device according to the monitoring data of the target GIS device by using the student model.

[0127] In an exemplary embodiment, the monitoring data samples include unlabeled data samples and labeled data samples; obtaining monitoring data samples of the GIS device, and training the to-be-trained teacher model by using the monitoring data samples to obtain the trained teacher model, comprising: obtaining the physical law constraint of the GIS device; and training the to-be-trained teacher model by using a semi-supervised learning method according to the physical law constraint and the monitoring data samples to obtain the trained teacher model.

[0128] In an exemplary embodiment, the method can further comprise: receiving model update information from each edge device; the model update information is obtained by updating the student model according to the monitoring data of the target GIS device by the edge device; aggregating each model update information by using a federated learning algorithm to obtain an updated student model; and deploying the updated student model to each edge device.

[0129] In an exemplary embodiment, as shown in Figure 5 , a GIS fault monitoring method based on cloud-edge collaboration is provided, which is applied to edge devices in Figure 1 as an example for illustration, comprising the following steps:

[0130] Step S501, obtaining monitoring data of a target GIS device.

[0131] Step S502, obtaining a fault monitoring result of the target GIS device according to the monitoring data by using a student model; the student model comes from a cloud server, and the student model is obtained by the cloud server by performing neural architecture search, model compression and knowledge distillation processing on the trained teacher model under the hardware constraint condition of the edge device; the trained teacher model is trained by the cloud server by using monitoring data samples of the GIS device.

[0132] In an example embodiment, the student model is used to obtain the fault monitoring result of the target GIS device according to the monitoring data, including: performing anomaly detection on the monitoring data; when it is detected that the monitoring data contains abnormal data, using the student model to obtain the fault monitoring result of the target GIS device according to the abnormal data.

[0133] In an example embodiment, the student model is used to obtain the fault monitoring result of the target GIS device according to the monitoring data, including: using the student model to obtain a fault prediction label of the target GIS device and a confidence of the fault prediction label according to the monitoring data; in a case where the confidence is greater than a confidence threshold, obtaining the fault monitoring result of the target GIS device according to the fault prediction label; in a case where the confidence is not greater than the confidence threshold, sending the monitoring data to a cloud server; and the cloud server is configured to send the monitoring data to a terminal of an operation and maintenance personnel.

[0134] In an example embodiment, the method further includes: updating the student model according to the monitoring data of the target GIS device to obtain model update information of the student model; sending the model update information to the cloud server; receiving an updated student model from the cloud server; and the updated student model is obtained by the cloud server by aggregating the model update information from each edge device using a federated learning algorithm.

[0135] It should be understood that, although each step in the flowchart involved in each of the above embodiments is shown in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each of the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or stages. It can be understood that the steps in different embodiments can be freely combined as needed, and various non-contradictory schemes formed by the combination are within the scope of protection of the present application.

[0136] Based on the same inventive concept, the embodiments of the present application also provide a cloud-edge collaborative GIS fault monitoring device for implementing the above-mentioned cloud-edge collaborative GIS fault monitoring method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more cloud-edge collaborative GIS fault monitoring device embodiments provided below can refer to the limitations of the cloud-edge collaborative GIS fault monitoring method described above, which will not be repeated here.

[0137] In one example embodiment, as shown in Figure 6 A cloud-edge collaborative GIS fault monitoring device is provided, comprising:

[0138] A model training module 601 is configured to obtain monitoring data samples of GIS equipment, train a teacher model to be trained by using the monitoring data samples, and obtain a trained teacher model.

[0139] An architecture search module 602 is configured to search for a student model architecture with optimal performance under hardware constraint conditions of an edge device by using a neural architecture search method based on the teacher model, and obtain a basic student model according to the student model architecture.

[0140] A model compression module 603 is configured to perform model compression and knowledge distillation processing on the basic student model according to the trained teacher model, and obtain a student model corresponding to the edge device.

[0141] A model distribution module 604 is configured to deploy the student model to each edge device. The edge device is configured to obtain a fault monitoring result of a target GIS device by using the student model according to monitoring data of the target GIS device.

[0142] In one example embodiment, the monitoring data samples include unlabeled data samples and labeled data samples. The model training module 601 is configured to obtain a physical law constraint of the GIS equipment, and train the teacher model to be trained by using a semi-supervised learning method according to the physical law constraint and the monitoring data samples, and obtain the trained teacher model.

[0143] In one example embodiment, the device further comprises an information receiving module configured to receive model update information from each edge device, the model update information being obtained by updating the student model by the edge device according to monitoring data of the target GIS equipment, an information aggregation module configured to aggregate each model update information by using a federated learning algorithm to obtain an updated student model, and an update issuing module configured to deploy the updated student model to each edge device.

[0144] In one example embodiment, as shown in Figure 7 A cloud-edge collaborative GIS fault monitoring device is provided, comprising:

[0145] A data acquisition module 701 is configured to obtain monitoring data of a target GIS device.

[0146] The result obtaining module 702 is configured to obtain a fault monitoring result of the target GIS device according to the monitoring data by using a student model, wherein the student model is obtained from a cloud server, and the student model is obtained by the cloud server through neural architecture search, model compression and knowledge distillation processing under hardware constraints of the edge device based on a trained teacher model, and the trained teacher model is obtained by the cloud server based on monitoring data samples of GIS devices.

[0147] In an exemplary embodiment, the result obtaining module 702 is configured to perform anomaly detection on the monitoring data, and obtain a fault monitoring result of the target GIS device according to abnormal data in the monitoring data by using the student model when the abnormal data is detected.

[0148] In an exemplary embodiment, the result obtaining module 702 is configured to obtain a fault prediction label of the target GIS device and a confidence of the fault prediction label according to the monitoring data by using the student model, obtain a fault monitoring result of the target GIS device according to the fault prediction label when the confidence is greater than a confidence threshold, and send the monitoring data to the cloud server when the confidence is not greater than the confidence threshold, wherein the cloud server is configured to send the monitoring data to a terminal of an operation and maintenance personnel.

[0149] In an exemplary embodiment, the apparatus further includes a model updating module configured to update the student model according to monitoring data of the target GIS device to obtain model updating information of the student model, an information sending module configured to send the model updating information to the cloud server, and an information receiving module configured to receive an updated student model from the cloud server, wherein the updated student model is obtained by the cloud server by aggregating the model updating information from each edge device by using a federated learning algorithm.

[0150] The above-mentioned various modules of the GIS fault monitoring apparatus based on cloud-edge collaboration can be realized by software, hardware and combinations thereof in whole or in part. The above-mentioned various modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations corresponding to the above-mentioned various modules.

[0151] In an exemplary embodiment, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in Figure 8As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control ability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store monitoring data samples, hardware constraint conditions of edge devices and other data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a GIS fault monitoring method based on cloud edge collaboration.

[0152] In an exemplary embodiment, a computer device, which can be a terminal, is provided, and an internal structure diagram thereof can be as shown in the figure. Figure 9 As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control ability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be realized through WIFI, mobile cellular network, near field communication (Near Field Communication, NFC) or other technologies. The computer program is executed by the processor to implement a GIS fault monitoring method based on cloud edge collaboration. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0153] Those skilled in the art can understand that, Figure 8 and Figure 9The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0154] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0155] In one embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implements the steps in the above method embodiments when executed by a processor.

[0156] In one embodiment, a computer program product is provided, including a computer program, and the computer program implements the steps in the above method embodiments when executed by a processor.

[0157] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0158] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0159] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0160] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A GIS fault monitoring method based on cloud-edge collaboration, characterized in that, The method includes: Acquire monitoring data samples from GIS equipment, and use the monitoring data samples to train the teacher model to be trained, thereby obtaining a trained teacher model; Based on the teacher model, the neural architecture search method is used to search for the student model architecture with the best performance under the hardware constraints of the edge device, and the basic student model is obtained according to the student model architecture. Based on the trained teacher model, the basic student model is subjected to model compression and knowledge distillation to obtain the student model corresponding to the edge device; The student model is deployed to each of the edge devices; the edge devices are used to obtain the fault monitoring results of the target GIS device based on the monitoring data of the target GIS device using the student model.

2. The method according to claim 1, characterized in that, The monitoring data samples include unlabeled data samples and labeled data samples; the acquisition of monitoring data samples from GIS equipment, and the use of these monitoring data samples to train the teacher model to obtain the trained teacher model, includes: Obtain the physical constraints of the GIS equipment; Based on the physical constraints and the monitoring data samples, a semi-supervised learning method is used to train the teacher model to be trained, resulting in a trained teacher model.

3. The method according to claim 1, characterized in that, The method further includes: Receive model update information from each edge device; the model update information is obtained by the edge device updating the student model based on the monitoring data of the target GIS device; The updated student model is obtained by aggregating the update information of each model using a federated learning algorithm. The updated student model is deployed to each of the edge devices.

4. A GIS fault monitoring method based on cloud-edge collaboration, characterized in that, The method includes: Acquire monitoring data from the target GIS equipment; Using a student model, the fault monitoring results of the target GIS device are obtained based on the monitoring data; the student model comes from a cloud server, and is obtained by the cloud server through neural architecture search, model compression, and knowledge distillation under the hardware constraints of the edge device, based on a trained teacher model; the trained teacher model is obtained by the cloud server using monitoring data samples from the GIS device.

5. The method according to claim 4, characterized in that, The method of using a student model to obtain fault monitoring results for the target GIS equipment based on the monitoring data includes: Anomaly detection is performed on the monitoring data; When abnormal data is detected in the monitoring data, the student model is used to obtain the fault monitoring results of the target GIS equipment based on the abnormal data.

6. The method according to claim 4, characterized in that, The method of using a student model to obtain fault monitoring results for the target GIS equipment based on the monitoring data includes: Using a student model, the fault prediction label and the confidence level of the fault prediction label of the target GIS equipment are obtained based on the monitoring data. If the confidence level is greater than the confidence threshold, the fault monitoring result of the target GIS equipment is obtained based on the fault prediction label; If the confidence level is not greater than the confidence level threshold, the monitoring data is sent to the cloud server; the cloud server is used to send the monitoring data to the terminal of the operation and maintenance personnel.

7. The method according to any one of claims 4 to 6, characterized in that, The method further includes: The student model is updated based on the monitoring data of the target GIS device to obtain the model update information of the student model; The model update information is sent to the cloud server; The updated student model is received from the cloud server; the updated student model is obtained by the cloud server aggregating the model update information from each edge device using a federated learning algorithm.

8. A GIS fault monitoring system based on cloud-edge collaboration, characterized in that, The system includes: a cloud server and edge devices; The cloud server is used to acquire monitoring data samples from GIS devices, and to train the teacher model to be trained using the monitoring data samples to obtain a trained teacher model. Based on the teacher model, a neural architecture search method is used to search for the student model architecture with the best performance under the hardware constraints of the edge devices, and a basic student model is obtained according to the student model architecture. Based on the trained teacher model, the basic student model is subjected to model compression and knowledge distillation to obtain the student model corresponding to the edge device. The student model is then deployed to each of the edge devices. The edge device is used to acquire monitoring data of the target GIS device; and using the student model, to obtain the fault monitoring results of the target GIS device based on the monitoring data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images