Power equipment operation inspection method and device based on cloud edge coordination reasoning

Abstract

The invention provides a power equipment operation and maintenance method and device based on cloud edge coordination reasoning, and belongs to the field of intelligent operation and maintenance of power equipment. The method is applied to an edge end and comprises the following steps: acquiring an operation inspection data set of power equipment to be operated and inspected; performing first reasoning on the operation inspection data set to obtain a reasoning result set; determining the confidence of all target reasoning results in the reasoning result set; wherein the target reasoning result is any reasoning result in the reasoning result set; when the confidence coefficient of the target reasoning result is greater than a first preset confidence coefficient and the target reasoning result is greater than a preset threshold value, triggering an alarm response; when the confidence coefficient of the target reasoning result is smaller than or equal to a first preset confidence coefficient, sending the first data to the cloud; wherein the original data corresponding to the target reasoning result is the first data, and the cloud end is used for carrying out second reasoning based on the received first data and carrying out response based on a second reasoning result. According to the invention, data can be processed in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of electric power equipment, and in particular to a method and device for operation and maintenance of electric power equipment based on cloud-edge coordinated reasoning. Background Art

[0002] With the introduction of artificial intelligence technology in power operation and maintenance tasks, the accuracy and efficiency of status monitoring and fault diagnosis of power equipment have been greatly improved.

[0003] However, the traditional power equipment monitoring system mainly relies on the inference mode of centralized processing in the cloud, uploading a large amount of raw data to the cloud for processing and analysis. This mode faces many problems in the actual application of power equipment operation and inspection: First, the amount of power equipment data is huge and the data modality is diverse. Transmitting all data to the cloud will occupy a lot of bandwidth, increase transmission costs, and cause a burden on cloud computing resources. Second, the data collected by edge devices may contain inaccurate or deviated noise information. Directly uploading it to the cloud without preprocessing is likely to cause time delays and data deviations, thereby affecting the accuracy and real-time performance of reasoning.

[0004] In addition, since the power equipment operation and maintenance tasks have high requirements for the real-time and response speed of fault diagnosis, the traditional cloud-based centralized reasoning model is also difficult to meet the needs of emergency fault alarms. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for operation and inspection of electric power equipment based on cloud-edge coordinated reasoning, so as to solve the problem that the current operation and inspection method of electric power equipment cannot meet the real-time requirements of the power grid.

[0006] In a first aspect, an embodiment of the present invention provides a power equipment operation and inspection method based on cloud-edge coordinated reasoning, which is applied to an edge terminal and includes:

[0007] Obtaining an inspection data set of the power equipment to be inspected;

[0008] Perform the first reasoning on the operation and inspection data set to obtain a reasoning result set;

[0009] Determine the confidence of all target reasoning results in the reasoning result set; wherein the target reasoning result is any reasoning result in the reasoning result set;

[0010] When the confidence of the target reasoning result is greater than a first preset confidence and the target reasoning result is greater than a preset threshold, triggering an alarm response;

[0011] When the confidence of the target reasoning result is less than or equal to the first preset confidence, the first data is sent to the cloud; wherein the original data corresponding to the target reasoning result is the first data, and the cloud is used to perform a second reasoning based on the received first data, and respond based on the result of the second reasoning.

[0012] In a possible implementation, a determiner is provided at the edge end;

[0013] Determine the confidence of all target inference results in the inference result set, including:

[0014] The inspection data set and the inference result set are input into the judger to obtain the confidence of all target inference results.

[0015] In one possible implementation, the determiner is an error estimation model;

[0016] Input the inspection data set and the inference result set into the judger to obtain the confidence of all target inference results, including:

[0017] Input the inspection data set and the inference result set into the error estimation model to obtain the estimated values ​​of all target inference results;

[0018] Based on the estimated value of the target reasoning result and the regularized entropy value of the target reasoning result, the confidence of the target reasoning result and the confidence of all target reasoning results are determined.

[0019] In a possible implementation, an inference model is provided at the edge, and the inference model is a compressed model;

[0020] Perform the first inference on the operation and inspection data set to obtain the inference result set, including:

[0021] Preprocessing the operation and inspection data set to obtain a processed operation and inspection data set;

[0022] The processed operation and inspection data set is input into the inference model to obtain the inference result set.

[0023] In a possible implementation, the operation inspection data set is multi-source data;

[0024] Input the processed operation and inspection data set into the inference model to obtain the inference result set, including:

[0025] Performing a first aggregation process on the multi-source data in the operation and inspection data set to obtain operation and inspection aggregated data;

[0026] Input the operation and inspection aggregated data into the inference model to obtain the inference result set;

[0027] Among them, when the first data is multi-source data, the cloud is used to perform a second aggregation process on the first data to obtain a high-dimensional feature data set; the cloud is also used to perform a second inference on the high-dimensional feature data set.

[0028] In a possible implementation, the first aggregation process includes maximum pooling, averaging or serial aggregation, and the second aggregation process is serial aggregation.

[0029] In one possible implementation, the preprocessing includes at least one of noise elimination, outlier processing, integrity checking, or sensitive information filtering.

[0030] In a possible implementation, a first reasoning model is provided at the edge, and a second reasoning model is provided at the cloud;

[0031] When the edge load is less than a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model, and the first reasoning model and the second reasoning model are two complete and independent neural network models;

[0032] When the edge load is greater than or equal to a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model. The first reasoning model is the first network layer of the reasoning model, and the second reasoning model is the second network layer of the reasoning model. The first network layer and the second network layer constitute the reasoning model.

[0033] In a possible implementation, the operation and inspection data set includes at least one of operation data, inspection records, maintenance status, temperature, voltage, current, or inspection pictures of the equipment.

[0034] In a second aspect, an embodiment of the present invention provides an electric power equipment operation and inspection device based on cloud-edge coordinated reasoning, which is applied to an edge terminal and includes:

[0035] An acquisition module is used to acquire an operation and inspection data set of the power equipment to be inspected;

[0036] The reasoning module is used to perform a first reasoning on the operation and inspection data set to obtain a reasoning result set;

[0037] A determination module, used to determine the confidence of all target reasoning results in the reasoning result set; wherein the target reasoning result is any reasoning result in the reasoning result set;

[0038] An alarm module is used to trigger an alarm response when the confidence of the target reasoning result is greater than a first preset confidence and the target reasoning result is greater than a preset threshold;

[0039] A sending module is used to send the first data to the cloud when the confidence of the target reasoning result is less than or equal to the first preset confidence; wherein the original data corresponding to the target reasoning result is the first data, and the cloud is used to perform a second reasoning based on the received first data and respond based on the result of the second reasoning.

[0040] The embodiment of the present invention provides an electric power equipment operation and inspection method based on cloud-edge coordinated reasoning. In order to improve the speed of operation and inspection response, after obtaining the operation and inspection data set of the electric power equipment to be inspected, the edge end performs a first reasoning on the operation and inspection data set to obtain a reasoning result set. Then, the confidence of all target reasoning results in the reasoning result set is determined. When the confidence of the target reasoning result is greater than the first preset confidence, and the target reasoning result is greater than the preset threshold, the edge end can directly trigger an alarm response without waiting for cloud processing, thereby improving the response speed. In addition, when the confidence of the target reasoning result is less than or equal to the first preset confidence, in order to improve the accuracy of the analysis, the first data needs to be sent to the cloud, and the cloud performs a second reasoning based on the received first data, and responds based on the result of the second reasoning. Thus, the edge end can analyze the data together with the cloud end, and the data with high confidence can be directly processed at the edge end without uploading to the cloud end to wait for cloud processing, thereby improving the response speed. The edge end only needs to upload the data with low confidence to the cloud end for processing, saving bandwidth, improving the processing speed, and realizing real-time and efficient processing of the operation and inspection data. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a flow chart for implementing the power equipment operation and inspection method based on cloud-edge coordinated reasoning provided by an embodiment of the present invention;

[0043] Figure 2 is a schematic diagram of the interaction between the edge terminal and the cloud provided by an embodiment of the present invention;

[0044] Figure 3 is a schematic diagram of edge-side and cloud-side execution time provided by an embodiment of the present invention;

[0045] Figure 4 It is a structural diagram of an electric power equipment operation and maintenance device based on cloud-edge coordinated reasoning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0047] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0048] As introduced in the background technology, with the development of edge computing, edge devices have certain data processing capabilities and can pre-process data locally and perform preliminary reasoning. By deploying small models at the edge for real-time pre-processing and reasoning, the cloud load can be reduced, low-quality data transmission can be reduced, and efficient use of data can be achieved.

[0049] However, the inventors found that due to the limited computing power and storage resources of edge devices, the accuracy and real-time performance of reasoning cannot be guaranteed. It is still necessary to coordinate and distribute reasoning tasks between the edge and the cloud through cloud-edge collaboration to ensure the accuracy and response efficiency of reasoning.

[0050] Figure 1 The implementation flow chart of the power equipment operation and inspection method based on cloud-edge coordinated reasoning provided in an embodiment of the present invention is applied at the edge end, and is described in detail as follows:

[0051] S110, obtaining an operation and inspection data set of the power equipment to be operated and inspected.

[0052] The inspection data set may include at least one or more of the operation data, inspection records, maintenance status, temperature, voltage, current or inspection pictures of the equipment to be inspected.

[0053] Temperature, voltage and current can directly reflect the operating status of the equipment to be inspected, and other electrical parameters can also be selected according to actual application conditions.

[0054] S120: Perform a first inference on the operation and inspection data set to obtain an inference result set.

[0055] Since the modality of power equipment operation and inspection data is complex and the data volume is large, if all data is directly uploaded to the cloud for processing, it will not only increase bandwidth consumption, but also may cause deviations in data accuracy. Therefore, in the present invention, the operation and inspection data set can be preprocessed at the edge first to achieve real-time data processing.

[0056] In some embodiments, in order to ensure the accuracy of the acquired data, the data needs to be preprocessed. First, the operation inspection data set can be preprocessed to obtain a processed operation inspection data set. Then, the processed operation inspection data set is input into the reasoning model to obtain a reasoning result set.

[0057] In this embodiment, preprocessing may include at least one of noise elimination, outlier processing, integrity check, or sensitive information filtering. The data is subjected to noise elimination and outlier processing to ensure the initial accuracy of the data. The integrity check is used to check whether the data contains missing or abnormal fields to ensure that the data uploaded to the edge and the cloud is complete and reliable. Sensitive information filtering is used to filter sensitive information in the data to ensure data privacy and security. By filtering sensitive information in the data, it can be ensured that sensitive information is not leaked when the data is inferred or uploaded to the cloud. For example, the geographic location of the device or other sensitive information that is not convenient to disclose. For sensitive information, information encryption, replacement or masking of sensitive parts, and deletion of unnecessary sensitive information can be used for processing.

[0058] By preprocessing data at the edge, high-quality data can be acquired in real time, while reducing the burden of data cleaning and preprocessing on the cloud and avoiding uploading duplicate or low-quality data to the cloud.

[0059] In some embodiments, since the operation and inspection data includes many types, when the operation and inspection data set is multi-source data, the multi-source data in the operation and inspection data set can first be aggregated to obtain operation and inspection aggregated data. Then, the operation and inspection aggregated data is input into the reasoning model to obtain a reasoning result set.

[0060] In this embodiment, the first aggregation process includes maximum pooling process, averaging process or serial aggregation.

[0061] When the edge receives a large amount of multi-source data in a short period of time, reasoning one by one may be inefficient. Therefore, the data can be aggregated by performing maximum pooling, average pooling, or series aggregation to reduce the data size and complete the detection task in a timely manner.

[0062] In some embodiments, in order to further improve the real-time response capability of power equipment operation and maintenance, a small artificial intelligence inference model with model compression is deployed at the edge to complete preliminary reasoning locally.

[0063] In this embodiment, the inference model is implemented through compression technologies such as pruning and quantization to adapt to the limited computing resources at the edge.

[0064] In this embodiment, the first reasoning refers to the reasoning model deployed at the edge end performing reasoning on the preprocessed data. The reasoning here can be the identification of the inspection image or the classification of the device status. The reasoning result can be the event type and the corresponding probability.

[0065] Take the inspection picture as an example to explain the first reasoning process:

[0066] First, the inspection image may be subjected to pre-processing such as denoising, resolution checking, and image resizing to 640*640 to obtain a processed image.

[0067] Next, the processed image is input into the inference model. After feature extraction, feature fusion and other processing, the position of the detection frame and the probability distribution of the detection frame will be output.

[0068] S130: Determine the confidence of all target reasoning results in the reasoning result set.

[0069] The target reasoning result is any reasoning result in the reasoning result set.

[0070] In some embodiments, a determiner may be provided at the edge, which is a machine learning model, and is a supervised learning model for learning the relationship between input data and inference results.

[0071] In this embodiment, the inspection data set and the reasoning result set may first be input into the determiner to obtain the confidence of all target reasoning results.

[0072] In order to evaluate the prediction probability of the inference after the inference model is inferred, the determiner can be set as an error evaluation model. The error estimation model uses the XGBoost algorithm to learn based on the input and prediction errors in the training data to estimate the prediction deviation of the edge inference model. The determiner can identify the uncertainty of the inference model under specific input conditions, so as to better grasp the reliability of the inference model.

[0073] Ultimately, the confidence level is determined by the predicted probability obtained by the inference model and the estimated value of the error assessment model.

[0074] In addition, in order to quantify the uncertainty of the probability distribution predicted by the inference model, the inference results obtained by inference can be processed by regularizing the entropy value. Specifically:

[0075] First, the inspection data set and the inference result set may be input into the error estimation model to obtain estimated values ​​of all target inference results.

[0076] Then, based on the estimated value of the target reasoning result and the regularized entropy value of the target reasoning result, the confidence of the target reasoning result and the confidence of all target reasoning results are determined.

[0077] In addition, the confidence level may also be determined based on the estimated value of the target inference result, the regularized entropy value of the target inference result, and the predicted probability of the target result.

[0078] S140: When the confidence of the target reasoning result is greater than a first preset confidence and the target reasoning result is greater than a preset threshold, trigger an alarm response.

[0079] When the confidence of the target reasoning result is greater than the first preset confidence, it means that the data is suitable for edge processing and the prediction result of the first reasoning result is reliable. When the target reasoning result is greater than the preset threshold, the edge can directly trigger an alarm response and send a notification to the maintenance personnel. As a result, the edge can respond quickly to high-confidence data in an emergency without waiting for cloud processing, greatly improving the system's fault response speed.

[0080] S150: When the confidence level of the target reasoning result is less than or equal to a first preset confidence level, the first data is sent to the cloud.

[0081] The first data obtains a target reasoning result after a first reasoning, and the cloud is used to perform a second reasoning based on the received first data and respond based on the result of the second reasoning.

[0082] like Figure 2 As shown in the figure, in the power equipment operation and inspection tasks, due to the limited computing power of the edge devices, it is difficult to support the full amount of reasoning of complex models. The reasoning model deployed internally is small and cannot handle complex processing. In order to improve the real-time and accuracy of the overall reasoning, a fine reasoning model can also be set up in the cloud to collaborate with the edge, so that data can be processed in real time.

[0083] The fine inference model set in the cloud is used to perform fine inference on the received first data. Compared with the inference model set in the edge, the fine inference model in the cloud has higher inference and computing capabilities, larger model parameters and higher complexity, and is used to perform inference processing on data with low confidence.

[0084] When the first data received by the cloud is multi-source data, due to the cloud's stronger processing capabilities and more sufficient computing power, it can be inferred according to the input order of the data in the first data, realize serial inference, and perform aggregate analysis on the results to complete the detection task. Aggregate analysis is to take the value with the highest probability as the final result of the inference result.

[0085] In some embodiments, in order to distribute reasoning tasks between the edge and the cloud and reduce bandwidth consumption caused by data transmission, the reasoning model can be segmented.

[0086] The segmentation methods include horizontal segmentation and vertical segmentation. Horizontal segmentation: segmentation is performed at the neural network level. According to factors such as resource consumption and data transmission volume of each layer of the network, appropriate segmentation points are selected to deploy some network layers on the edge and the remaining layers on the cloud to balance inference efficiency and resource utilization. Vertical segmentation: vertically divide the layers of the neural network into independently executable task partitions, and reuse existing partitions by sending partitions to the edge one by one, thereby reducing network bandwidth usage.

[0087] During the reasoning process, a heuristic algorithm can be used to predict the execution time of each layer of network partition in the cloud and edge, and select the optimal execution location based on the reasoning task and network latency. Figure 3 As shown in the figure, if the cloud execution time is short, the data runs directly on the cloud; if the edge reuse or loading execution time is short, the task is completed on the edge. This adaptive model partitioning strategy improves the overall reasoning performance and data processing efficiency of the system.

[0088] In this embodiment, the edge is provided with a first reasoning model, and the cloud is provided with a second reasoning model. When the edge load is less than a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model, and the first reasoning model and the second reasoning model are two complete and independent neural network models.

[0089] When the edge load is greater than or equal to a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model. The first reasoning model is the first network layer of the reasoning model, and the second reasoning model is the second network layer of the reasoning model. The first network layer and the second network layer constitute the reasoning model.

[0090] In this embodiment, the edge can execute the first few layers of the reasoning model (horizontal segmentation) or be responsible for a specific task branch (vertical segmentation). Then, the processed intermediate features (such as convolutional feature maps or reduced-dimensional data) are compressed and uploaded to the cloud. The edge reduces the computational burden and response time through segmentation. After receiving the intermediate features from the edge, the cloud restores the subsequent network structure and completes the remaining deep reasoning. The cloud is able to run more complex networks and provide refined reasoning results.

[0091] For vertical segmentation, for example, a power inspection robot needs to detect whether there are various problems on the surface of the equipment, such as cracks, rust, missing components, etc. A small inference model can be deployed on the edge to quickly locate the bounding box of the detection device in the image and achieve fast positioning. The cloud then receives the content in the bounding box and performs further inference on the area.

[0092] Specifically, according to the load status and network status of the edge, a heuristic choice can be made to perform complete reasoning, vertical split reasoning, or horizontal split reasoning.

[0093] The edge load status includes the current CPU or GPU utilization and the remaining memory on the edge. The network status is the current bandwidth and network latency. When the edge load is small, the complete inference model is run directly on the edge to complete the inference. When the edge load is moderate, the edge runs the segmented partial model to complete the area detection and uploads the results to the cloud to complete the inference. When the edge load is high, only shallow feature extraction is run, and the feature map is transmitted to the cloud, which completes the inference.

[0094] The power equipment operation and inspection method based on cloud-edge coordinated reasoning provided by the present invention, in order to improve the speed of operation and inspection response, after the edge end obtains the operation and inspection data set of the power equipment to be inspected, the first reasoning is performed on the operation and inspection data set to obtain the reasoning result set. Then, the confidence of all target reasoning results in the reasoning result set is determined. When the confidence of the target reasoning result is greater than the first preset confidence, and the target reasoning result is greater than the preset threshold, the edge end can directly trigger the alarm response without waiting for cloud processing, thereby improving the response speed. In addition, when the confidence of the target reasoning result is less than or equal to the first preset confidence, in order to improve the accuracy of the analysis, the first data needs to be sent to the cloud, and the cloud performs a second reasoning based on the received first data, and responds based on the result of the second reasoning. Thus, the edge end can analyze the data together with the cloud end, and the data with high confidence can be directly processed at the edge end without uploading to the cloud end to wait for cloud processing, thereby improving the response speed. The edge end only needs to upload the data with low confidence to the cloud end for processing, saving bandwidth, improving the processing speed, and realizing real-time and efficient processing of the operation and inspection data.

[0095] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0096] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0097] Figure 4The structural diagram of the power equipment operation and inspection device based on cloud-edge coordinated reasoning provided by an embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0098] like Figure 4 As shown, the power equipment operation and inspection device 400 based on cloud-edge coordinated reasoning is applied at the edge, including

[0099] An acquisition module 410 is used to acquire an operation and inspection data set of the power equipment to be inspected;

[0100] The reasoning module 420 is used to perform a first reasoning on the operation and inspection data set to obtain a reasoning result set;

[0101] A determination module 430 is used to determine the confidence of all target reasoning results in the reasoning result set; wherein the target reasoning result is any reasoning result in the reasoning result set;

[0102] An alarm module 440 is used to trigger an alarm response when the confidence of the target reasoning result is greater than a first preset confidence and the target reasoning result is greater than a preset threshold;

[0103] The sending module 450 is used to send the first data to the cloud when the confidence of the target reasoning result is less than or equal to the first preset confidence; wherein the first data obtains the target reasoning result after the first reasoning, and the cloud is used to perform a second reasoning based on the received first data and respond based on the result of the second reasoning.

[0104] In a possible implementation, a determiner is provided at the edge end;

[0105] The determination module 430 is used to input the operation and inspection data set and the reasoning result set into the determiner to obtain the confidence of all target reasoning results.

[0106] In one possible implementation, the determiner is an error estimation model;

[0107] A determination module 430 is used to input the operation and inspection data set and the reasoning result set into the error estimation model to obtain the estimated values ​​of all target reasoning results;

[0108] Based on the estimated value of the target reasoning result and the regularized entropy value of the target reasoning result, the confidence of the target reasoning result and the confidence of all target reasoning results are determined.

[0109] In a possible implementation, an inference model is provided at the edge, and the inference model is a compressed model;

[0110] The reasoning module 420 is used to pre-process the operation and inspection data set to obtain a processed operation and inspection data set;

[0111] The processed operation and inspection data set is input into the inference model to obtain the inference result set.

[0112] In a possible implementation, the operation inspection data set is multi-source data;

[0113] The reasoning module 420 is used to perform a first aggregation process on the multi-source data in the operation and inspection data set to obtain operation and inspection aggregated data;

[0114] Input the operation and inspection aggregated data into the inference model to obtain the inference result set;

[0115] Among them, when the first data is multi-source data, the cloud is used to perform a second aggregation process on the first data to obtain a high-dimensional feature data set; the cloud is also used to perform a second inference on the high-dimensional feature data set.

[0116] In a possible implementation, the first aggregation process includes maximum pooling, averaging or serial aggregation, and the second aggregation process is serial aggregation.

[0117] In one possible implementation, the preprocessing includes at least one of noise elimination, outlier processing, integrity checking, or sensitive information filtering.

[0118] In a possible implementation, a first reasoning model is provided at the edge, and a second reasoning model is provided at the cloud;

[0119] When the edge load is less than a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model, and the first reasoning model and the second reasoning model are two complete and independent neural network models;

[0120] When the edge load is greater than or equal to a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model. The first reasoning model is the first network layer of the reasoning model, and the second reasoning model is the second network layer of the reasoning model. The first network layer and the second network layer constitute the reasoning model.

[0121] In a possible implementation, the operation and inspection data set includes at least one of operation data, inspection records, maintenance status, temperature, voltage, current, or inspection pictures of the equipment.

[0122] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0123] Those of ordinary skill in the art will appreciate that the templates, units, and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0124] If the module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned embodiments of the power equipment operation and inspection method based on cloud-edge coordinated reasoning. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium.

[0125] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A power equipment operation and inspection method based on cloud-edge coordinated reasoning, applied to the edge, characterized in that: include: Obtaining an inspection data set of the power equipment to be inspected; Performing a first reasoning on the operation and inspection data set to obtain a reasoning result set; Determine the confidence of all target reasoning results in the reasoning result set; wherein the target reasoning result is any one reasoning result in the reasoning result set; When the confidence of the target reasoning result is greater than a first preset confidence and the target reasoning result is greater than a preset threshold, triggering an alarm response; When the confidence of the target reasoning result is less than or equal to the first preset confidence, the first data is sent to the cloud; the original data corresponding to the target reasoning result is the first data, and the cloud is used to perform a second reasoning based on the received first data and respond based on the result of the second reasoning.

2. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 1 is characterized in that: The edge end is provided with a determiner; Determining the confidence of all target reasoning results in the reasoning result set includes: The inspection data set and the reasoning result set are input into the determiner to obtain the confidence of all the target reasoning results.

3. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 2 is characterized in that: The determiner is an error estimation model; The step of inputting the inspection data set and the reasoning result set into the determiner to obtain the confidence of all the target reasoning results includes: Inputting the operation and inspection data set and the inference result set into the error estimation model to obtain estimated values ​​of all the target inference results; Based on the estimated value of the target reasoning result and the regularized entropy value of the target reasoning result, the confidence of the target reasoning result and the confidences of all the target reasoning results are determined.

4. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 1 is characterized in that: The edge end is provided with an inference model, and the inference model is a compressed model; The first reasoning is performed on the operation inspection data set to obtain a reasoning result set, including: Preprocessing the operation and inspection data set to obtain a processed operation and inspection data set; The processed operation and inspection data set is input into the reasoning model to obtain the reasoning result set.

5. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 4 is characterized in that: The operation and inspection data set is multi-source data; The step of inputting the processed operation and inspection data set into the reasoning model to obtain the reasoning result set includes: Performing a first aggregation process on the multi-source data in the operation and inspection data set to obtain operation and inspection aggregated data; Inputting the operation and inspection aggregated data into the reasoning model to obtain the reasoning result set; Among them, when the first data is multi-source data, the cloud is used to perform a second aggregation process on the first data to obtain a high-dimensional feature data set; the cloud is also used to perform a second inference on the high-dimensional feature data set.

6. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 5 is characterized in that: The first aggregation processing includes maximum pooling processing, averaging processing or serial aggregation, and the second aggregation processing is serial aggregation.

7. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 4 is characterized in that: The preprocessing includes at least one of noise elimination, outlier processing, integrity checking or sensitive information filtering.

8. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to claim 1 is characterized in that: The edge end is provided with a first reasoning model, and the cloud end is provided with a second reasoning model; When the edge load is less than a first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model, and the first reasoning model and the second reasoning model are two complete and independent neural network models; When the edge load is greater than or equal to the first preset value, the edge performs reasoning based on the first reasoning model, and the cloud performs reasoning based on the second reasoning model, and the first reasoning model is the first network layer of the reasoning model, and the second reasoning model is the second network layer of the reasoning model, and the first network layer and the second network layer constitute the reasoning model.

9. The power equipment operation and inspection method based on cloud-edge coordinated reasoning according to any one of claims 1 to 8, characterized in that: The operation and inspection data set includes at least one of the operation data, inspection records, maintenance status, temperature, voltage, current or inspection pictures of the equipment.

10. A power equipment operation and inspection device based on cloud-edge coordinated reasoning, applied to the edge, characterized in that: include: An acquisition module is used to acquire an operation and inspection data set of the power equipment to be inspected; An inference module, configured to perform a first inference on the operation and inspection data set to obtain an inference result set; A determination module, used to determine the confidence of all target reasoning results in the reasoning result set; wherein the target reasoning result is any one reasoning result in the reasoning result set; An alarm module, configured to trigger an alarm response when the confidence of the target reasoning result is greater than a first preset confidence and the target reasoning result is greater than a preset threshold; A sending module, used to send first data to the cloud when the confidence of the target reasoning result is less than or equal to the first preset confidence; wherein the original data corresponding to the target reasoning result is the first data, and the cloud is used to perform a second reasoning based on the received first data, and respond based on the result of the second reasoning.

Images