A method for monitoring power equipment faults based on edge computing

Abstract

The present invention discloses a method for monitoring power equipment faults based on edge computing, including edge computing, deep learning classification, transfer learning, data fusion and time series information prediction modeling; edge computing is used to process and analyze data on edge nodes close to power equipment to improve the real-time performance and efficiency of the system; deep learning classification is used to extract feature information from data and process noise and outliers in complex scenarios; transfer learning is used to reduce the amount of data and training time required for target tasks or fields to improve the generalization ability and accuracy of the model. The present invention utilizes edge computing nodes deployed near power equipment to collect and process data from multiple data sources such as sensors and cameras, and uses transfer learning technology to achieve rapid adaptation and accurate identification of power equipment of different types and sizes, and realize intelligent monitoring and fault warning of power equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment fault monitoring, and specifically relates to a power equipment fault monitoring method based on edge computing. Background Art

[0002] Power equipment is an important infrastructure for national economic and social development, and its safe and stable operation is related to the national economy and people's livelihood. However, due to the complexity, diversity and distribution of power equipment, traditional monitoring methods are difficult to meet the requirements of real-time, accuracy and universality. Therefore, it is necessary to develop an intelligent monitoring method that uses artificial intelligence technology to achieve real-time monitoring and fault diagnosis of power equipment. Summary of the invention

[0003] In order to address the shortcomings of the prior art, the present invention aims to provide a method for monitoring power equipment faults based on edge computing. The method utilizes edge computing nodes deployed near the power equipment to collect and process data from various data sources such as sensors and cameras. The method uses transfer learning to quickly adapt to and accurately identify power equipment of different types and sizes, thereby realizing intelligent monitoring and fault warning of power equipment.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A method for monitoring power equipment faults based on edge computing is implemented based on the following steps:

[0006] 1) Edge computing: Data processing and analysis are performed on edge nodes close to power equipment;

[0007] 2) Deep learning classification: extracting feature information from data and handling noise and outliers in complex scenarios;

[0008] 3) Transfer learning: Migrate the existing model to the new power equipment, adjust the model through the labeled data on the new power equipment, and realize the adaptation and recognition of power equipment of different types and sizes;

[0009] 4) Data fusion: Fusion of multi-source data including sensors, cameras, radar and sonar to obtain more information about power equipment;

[0010] 5) Time series information prediction modeling: Add timestamp information to each data point, use recursive neural network time series model for training, capture the time series dependency of data, continuously perform training and loss analysis, build a power equipment prediction model; and use the prediction modeling model to identify the sequence of data;

[0011] 6) Strengthen adaptability and self-learning capabilities: Design an adaptive mechanism to monitor the operating status and fault conditions of power equipment in real time; obtain real-time performance indicators based on the monitoring results, and summarize and analyze the real-time monitoring information and evaluation, conduct a comprehensive assessment of the power system performance, and adaptively adjust the system operating status.

[0012] Furthermore, step 1) edge computing specifically includes:

[0013] 1-1) Data collection and data preparation: Install sensors, cameras and other related equipment near the power equipment, and write corresponding data collection programs; collect a series of data containing different power equipment scenarios, and annotate the operating status and fault type of the power equipment for each data as training sets and test sets;

[0014] 1-2) Edge computing data processing: Data processing and analysis are performed at edge nodes close to power equipment; methods including image processing and signal processing are used to extract characteristic parameters from the data, and data cleaning and normalization operations are performed.

[0015] Furthermore, the deep learning in step 2) specifically includes:

[0016] 2-1) Establish a classification model for the operating status and fault types of power equipment; specifically including:

[0017] Define the power plant power equipment operation state variables to ensure that the power plant power equipment operation state is linearly related to the analysis, including the voltage amplitude value, current value and voltage value of the power plant power equipment operation;

[0018] The power plant power equipment operating state variable y is defined as:

[0019]

[0020] Where K is the voltage amplitude value under the operation of the power plant power equipment; V is the external influence coefficient; L is the current value under the operation of the power plant power equipment; U is the voltage value under the operation of the power plant power equipment; θ ij represents the power factor; i and j represent nodes.

[0021] 2-2) Use computing resources on edge nodes to train and infer models for real-time monitoring and fault diagnosis of power equipment; specifically:

[0022] On the basis of defining the operating status variables of the power plant's power equipment, a deep learning method is introduced to analyze the operating status of the power plant's power equipment;

[0023] In this process, the normal operating state of the relevant power equipment in the power plant is set as A, A={A1,A2,A3,…,A n}, and consider the storage state of the data in A as temporary state;

[0024] A1,A2,A3,…,A n They represent the data status at different times.

[0025] The operation space data of the power equipment is mapped and processed, and the mapped data set is output, which is represented by B. Then B={B1,B2,B3,…,B n}; B1, B2, B3, …, B n They represent the data status at different times.

[0026] The sample data set B is subjected to linear parameter classification, and the classified data set is derived, which is represented as C, where C = {C1, C2, C3, …, C n}, where C is the test set; C1, C2, C3, …, C n They represent the data status at different times.

[0027] The number of nodes in the trust network is selected based on the constraints of the power equipment. In the selection process, data is selected from the sample data sets B and A in the order of 300-500-1000-1500-2000-3000. In this state, the original 2025-dimensional spatial data is mapped to 300-dimensional spatial data, and then the data is tested by IEEE to complete the deep learning of the relevant data.

[0028] 2-3) Convert the power plant power equipment operating status analysis into an NP-hard problem to analyze the power plant power equipment operating status; set the nonlinear conversion objective function to A, then:

[0029]

[0030] In the formula, α ij is the voltage value at the device operation node after nonlinear conversion; β ij is the voltage amplitude value at the equipment operation node after nonlinear conversion; θ ij It is the current under the device operation node after nonlinear transformation.

[0031] Complete the nonlinear transformation of the power plant power equipment operating state. On this basis, set the power plant power equipment operating state analysis expression as ΔP t , then:

[0032] ΔP t =(C t -C t+1)+η d C t+1 +ω d (K ωmax -L ωt+1 )

[0033] Where η d The time when the power plant's power equipment is in operation; ω d The need for decoupling of errors in the operation status analysis of power equipment in power plants; K ωmax is the maximum value of the error in the analysis of the operating status of the power equipment in the power plant; L ωt+1 is the maximum value of the analysis error of the power plant's power equipment operating status on the second day; C t and C t+1 They represent the data status at time t and t+1 in the test collection respectively.

[0034] Furthermore, step 3) specifically includes the following contents:

[0035] 3-1) Adjust the equipment, establish communication connection, record the Harris corner detection processing results when the equipment is operating normally, move the camera smoothly, adjust the focus, and start measuring when the target power equipment clearly appears at the set position;

[0036] 3-2) Read the initial state of the target, collect images, and perform image preprocessing, including image enhancement, image smoothing, and image segmentation;

[0037] 3-3) Obtain target information in the image, process the image using the Harris corner detection method, compare the processed image with the previously recorded image, and determine whether the power equipment is faulty;

[0038] 3-4) If there is a fault, report it and the staff will repair it; if there is no fault, the inspection will end.

[0039] Furthermore, step 4) data fusion specifically includes:

[0040] Let the sensor data be D s ={s1,s2,...,s m}, where s1,s2,...,s m Represents the data of the 1st to mth sensors; the camera data is D c ={c1,c2,...,c f}, where c1, c2, ..., c f Indicates the data of the 1st to fth cameras; the radar data is D r ={r1,r2,...,r b}, where r1, r2, ..., r brepresents the data of the 1st to bth radars; each data source is fused to obtain a comprehensive data set Dcombined = {d1, d2, ..., d s+c+r}, where d1,d2,...,d s+c+r Represents the 1st to s+c+rth data points in the comprehensive data set;

[0041] Use transfer learning to build a classification model. Let the model be M. For a given power equipment data x, the model outputs the corresponding operating status y. state =M(x);

[0042] Use convolutional neural network to extract spatial features F from camera data camera =CNN(D c ), using a recurrent neural network to learn the temporal information F of radar data radar =RNN(D r ), using convolutional neural networks to extract the spatial features F of sensor data sensor =CNN(D s ), and finally introduce the attention mechanism to dynamically weigh the contribution of each data source F state :

[0043] F state =Softmax(MLP(D c ))·F camera +Softmax(MLP(D r ))·F radar +Softmax(MLP(D s ))·F sensor ;

[0044] Finally, multi-source data are fused to obtain more information about power equipment and improve the accuracy and reliability of system monitoring and diagnosis.

[0045] Furthermore, in step 5), the time series information prediction modeling introduces time series information through continuous data during the operation of the power equipment, and the power equipment model training set is

[0046] D source ={x1,x2,...,x n} and have corresponding labels for their respective states, use the recursive neural network time series model for training, capture the time series dependency of the data, continuously perform training and loss analysis, and finally build a power equipment prediction model to estimate the operating status and fault type of the power equipment; where x1, x2, ..., x n Indicates data at different times.

[0047] The beneficial effects of the present invention are:

[0048] The present invention realizes rapid data processing and analysis through edge computing, and uses transfer learning to realize rapid adaptation and accurate identification of power equipment of different types and sizes. It collects and processes data from multiple data sources such as sensors and cameras in real time, extracts characteristic parameters of power equipment, establishes classification models for power equipment operating status and fault types, and realizes intelligent monitoring and fault warning of power equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is the edge computing graph of the present invention;

[0050] Figure 2 is a diagram of the transfer learning process of the present invention;

[0051] Figure 3 It is a functional diagram of the power feature extraction algorithm of the present invention. DETAILED DESCRIPTION

[0052] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not intended to limit the scope of use of the present invention.

[0053] like Figure 1-3 As shown, the present invention proposes a method for monitoring power equipment faults based on edge computing, which specifically includes the following steps:

[0054] Step 1) Edge computing: Data processing and analysis are performed on edge nodes close to power equipment. Specifically, it includes:

[0055] 1-1) Data collection and data preparation: Install sensors, cameras and other equipment near the power equipment and write corresponding data collection programs. Collect a series of data containing different power equipment scenarios, and annotate the operating status and fault type of the power equipment for each data to serve as training sets and test sets.

[0056] 1-2) Edge computing data processing: Data processing and analysis are performed on edge nodes close to power equipment. Image processing, signal processing and other methods are used to extract characteristic parameters from the data, and data cleaning, normalization and other operations are performed.

[0057] Edge computing is information processing close to the edge of data. It is used to solve the transmission and processing problems of a large amount of device information at the edge of the distribution network. It can provide new application scenarios and solutions for the stable operation and control of the system. The edge computing platform is a new distributed platform that integrates data collection, computing, storage, and analysis at the edge of the network. It can provide edge intelligent services in real time to meet real-time data optimization, information security and other functions. Edge computing migrates part of the computing tasks to be processed by cloud computing to the vicinity of the device layer, greatly reducing the requirements for network bandwidth and computing power for data transmission. Moreover, since its analysis and processing of data is performed locally, it can better reduce the risk of information leakage and improve the security and privacy of information. Power equipment fault detection adopts the concept of edge computing and establishes a four-layer architecture of cloud, pipe, edge, and end. This architecture makes full use of the powerful computing power of the cloud, the high-speed and low-latency communication capabilities of the 5G network, the flexible and reliable layout characteristics of the edge computing layer, and the comprehensive and intelligent data collection and perception capabilities of the device side, and fully coordinates and unifies them, fully corresponding to the characteristics of the distribution network for high real-time requirements, stability and reliability, and flexible and scalable node device connection topology, and establishes support for the coordinated regulation of distribution network cloud-edge resources to improve data security and reliability.

[0058] Step 2) Deep learning classification: Extract feature information from the data and process noise and outliers in complex scenarios. Use deep learning to establish a classification model for the operating status and fault type of power equipment. Use computing resources on edge nodes to train and infer the model to achieve real-time monitoring and fault diagnosis of power equipment. Specifically include:

[0059] 2-1) Establish a classification model for the operating status and fault types of power equipment; specifically including:

[0060] Define the power plant power equipment operating state variables to ensure that the power plant power equipment operating state is linearly related to the analysis. The present invention defines three groups of variables, including the voltage amplitude value of the power plant power equipment operating, the current value of the power plant power equipment operating, and the voltage value of the power plant power equipment operating;

[0061] The power plant power equipment operating state variable y is defined as:

[0062]

[0063] Where, K is the voltage amplitude value under the operation of the power plant power equipment; V is the external influence coefficient; L is the current value under the operation of the power plant power equipment; U is the voltage value under the operation of the power plant power equipment; θ ij represents the power factor, and i and j represent nodes.

[0064] 2-2) Use computing resources on edge nodes to train and infer models for real-time monitoring and fault diagnosis of power equipment; specifically:

[0065] On the basis of defining the operating status variables of the power plant's power equipment, a deep learning method is introduced to analyze the operating status of the power plant's power equipment;

[0066] In this process, the normal operating state of the relevant power equipment in the power plant is set as A, A={A1,A2,A3,…,A n}, and consider the storage state of the data in A as temporary state;

[0067] A1,A2,A3,…,A n They represent the data status at different times.

[0068] The operation space data of the power equipment is mapped and processed, and the mapped data set is output, which is represented by B. Then B={B1,B2,B3,…,B n}; B1, B2, B3, …, B n They represent the data status at different times.

[0069] The sample data set B is subjected to linear parameter classification, and the classified data set is derived, which is represented as C, where C = {C1, C2, C3, …, C n}, where C is the test set; C1, C2, C3, …, C n They represent the data status at different times.

[0070] The number of nodes in the trust network is selected based on the constraints of the power equipment. In the selection process, data is selected from the sample data sets B and A in the order of 300-500-1000-1500-2000-3000. In this state, the original 2025-dimensional spatial data is mapped to 300-dimensional spatial data, and then the data is tested by IEEE to complete the deep learning of the relevant data.

[0071] 2-3) In order to realize the analysis of the operating status of power plant power equipment, the operating status of power plant power equipment is expressed by deep learning, and the analysis of the operating status of power plant power equipment is converted into an NP-hard problem to analyze the operating status of power plant power equipment; the nonlinear conversion objective function is set to A, then:

[0072]

[0073] In the formula, α ij is the voltage value at the device operation node after nonlinear conversion; β ij is the voltage amplitude value at the equipment operation node after nonlinear conversion; θ ij It is the current value at the device operation node after nonlinear conversion;

[0074] Complete the nonlinear transformation of the power plant power equipment operating state. On this basis, set the power plant power equipment operating state analysis expression as ΔP t , then:

[0075] ΔP t =(C t -C t+1 )+η d C t+1 +ω d (K ωmax -L ωt+1 )

[0076] Where η d The time when the power plant's power equipment is in operation; ω d The need for decoupling of errors in the operation status analysis of power equipment in power plants; K ωmax is the maximum value of the error in the analysis of the operating status of the power equipment in the power plant; L ωt+1 is the maximum value of the analysis error of the power plant's power equipment operating status on the second day; C t and C t+1 They respectively represent the data status at time t and t+1 in the test set; the above formula can be used to analyze the operating status of the power plant's power equipment.

[0077] The present invention introduces edge computing: Since power equipment is widely distributed, transmitting data to the cloud for processing will cause delays, bandwidth consumption and security risks. Therefore, using edge computing to process and analyze data on edge nodes close to power equipment can improve the real-time, efficiency and security of the system.

[0078] The present invention uses deep learning classification: Traditional fault diagnosis methods based on rules or expert systems are difficult to handle complex, nonlinear and high-dimensional data. The classification model based on visual transformers and other deep neural network structures can more effectively extract feature information from data and better handle noise, outliers and other problems in complex scenarios.

[0079] Step 3) Transfer learning: Migrate the existing model to the new power equipment, and adjust the model through the labeled data on the new power equipment to achieve adaptation and recognition of power equipment of different types and sizes.

[0080] Using transfer learning, the existing model is migrated to the new power equipment. A small amount of labeled data on the new power equipment is used to fine-tune the model to achieve rapid adaptation and accurate identification of power equipment of different types and sizes. As a learning platform, deep learning has been widely used in many aspects. However, most of these work under the same underlying distribution, that is, the training data (source domain) and the test data (target domain) have the same feature space. In addition, in practical applications, due to the high cost of data collection and labeling, it is difficult to form a large-scale data set with complete labels, which restricts its further development. When the feature space or feature distribution of the detection data changes, it cannot be used for prediction, and the existing data needs to be reused for reconstruction and retraining, which is costly. In fact, sometimes it is impossible.

[0081] The present invention introduces transfer learning: Due to the wide variety of types and sizes of power equipment, training a general and accurate model requires a large amount of data and computing resources. Transfer learning is a machine learning method that uses existing models or data to improve the learning effect of new tasks or fields. The principle of model-based transfer learning is that there may be some common knowledge or features between different tasks or fields. By transferring this knowledge or features from the original task or field to the target task or field, and using the model parameters or structure of the source field as prior information of the target field to initialize or constrain the model of the target field, the amount of data and training time required for the target task or field can be reduced, while improving the generalization ability and accuracy of the model. By using transfer learning, the existing model can be migrated to the new power equipment, and the knowledge in the existing model can be used to accelerate the training process of the new model and improve the generalization ability of the new model.

[0082] Step 3) specifically includes:

[0083] 3-1) Adjust the equipment, establish communication connection, record the Harris corner detection processing results when the equipment is operating normally, move the camera smoothly, adjust the focus, and start measuring when the target power equipment clearly appears at the set position;

[0084] 3-2) Read the initial state of the target, collect images, and perform image preprocessing, including image enhancement, image smoothing, and image segmentation;

[0085] 3-3) Obtain target information in the image, process the image using the Harris corner detection method, compare the processed image with the previously recorded image, and determine whether the power equipment is faulty;

[0086] 3-4) If there is a fault, report it and the staff will repair it; if there is no fault, the inspection will end.

[0087] Step 4) Data fusion: Fusion of multi-source data including sensors, cameras, radar and sonar to obtain more power equipment information. Specifically including:

[0088] Let the sensor data be D s ={s1,s2,...,s m}, where s1,s2,...,s m Represents the data of the 1st to mth sensors; the camera data is D c ={c1,c2,...,c f}, where c1, c2, ..., c f Indicates the data of the 1st to fth cameras; the radar data is D r ={r1,r2,...,r b}, where r1, r2, ..., r b represents the data of the 1st to bth radars; each data source is fused to obtain a comprehensive data set Dcombined = {d1, d2, ..., d s+c+r}, where d1,d2,...,d s+c+r Represents the 1st to s+c+rth data points in the comprehensive data set;

[0089] Use transfer learning to build a classification model. Let the model be M. For a given power equipment data x, the model outputs the corresponding operating status y. state =M(x);

[0090] Use convolutional neural network to extract spatial features F from camera data camera =CNN(D c ), using a recurrent neural network to learn the temporal information F of radar data radar =RNN(D r ), using convolutional neural networks to extract the spatial features F of sensor data sensor =CNN(D s ), and finally introduce the attention mechanism to dynamically weigh the contribution of each data source F state :

[0091] F state =Softmax(MLP(D c ))·F camera +Softmax(MLP(D r ))·F radar +Softmax(MLP(D s ))·F sensor ;

[0092] Finally, multi-source data are fused to obtain more information about power equipment and improve the accuracy and reliability of system monitoring and diagnosis.

[0093] Step 5) Time series information prediction modeling: Add timestamp information to each data point, use a recursive neural network time series model for training, capture the time series dependency of the data, continuously perform training and loss analysis, build a power equipment prediction model; and use the predictive modeling model to identify the sequence of the data.

[0094] Time series information prediction modeling introduces time series information through continuous data during the operation of power equipment. The power equipment model training set is D source ={x1,x2,...,x n}(where x1, x2, ..., x n Representing data at different moments) and having corresponding labels for their respective states, a recursive neural network time series model is used for training to capture the time series dependency of the data, and training and loss analysis are continuously performed to finally build a power equipment prediction model for estimating the operating status and fault type of power equipment.

[0095] Step 6) Strengthen adaptability and self-learning capabilities: Design an adaptive mechanism to monitor the operating status and fault conditions of power equipment in real time. And introduce user feedback, including the confirmation and correction of monitoring results by operation and maintenance personnel, and the subjective evaluation of the current system operation status by users. According to the monitoring results, obtain real-time performance indicators, and summarize and analyze the real-time monitoring information and evaluation, conduct a comprehensive evaluation of the power system performance, and adaptively adjust the system operation status. Enable the power system to optimize and adjust itself according to real-time monitoring results and user feedback. Using reinforcement learning, etc., it can continuously learn and adapt to various power equipment and environmental conditions, thereby continuously improving the accuracy and stability of monitoring and diagnosis. Comprehensively apply the above innovations to create an intelligent monitoring power system with the above-mentioned power equipment fault monitoring method, so that it is more efficient, robust and has rapid adaptability, providing more reliable operating status monitoring and fault diagnosis services for power equipment.

[0096] The present invention uses multi-source data to monitor the operating status of power equipment in real time, realizes accurate diagnosis of power equipment fault types, and realizes rapid adaptation and accurate identification of power equipment of different types and sizes. It can be applied to various types of power equipment, such as transformers, generators, transmission lines, etc. Its high real-time performance, accuracy and universality will greatly improve the safe and stable operation of power equipment and reduce the occurrence rate of faults and maintenance costs.

[0097] The present invention uses artificial intelligence to achieve real-time monitoring and fault diagnosis of power equipment, with high real-time, accuracy and universality. It also uses edge computing nodes deployed near power equipment to collect and process data from multiple data sources such as sensors and cameras, and uses transfer learning technology to achieve rapid adaptation and accurate identification of power equipment of different types and sizes, thus achieving intelligent monitoring and fault warning of power equipment.

[0098] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for monitoring power equipment faults based on edge computing, characterized in that: The method is implemented based on the following steps: 1) Edge computing: Data processing and analysis are performed on edge nodes close to power equipment; 2) Deep learning classification: extracting feature information from data and handling noise and outliers in complex scenarios; 3) Transfer learning: Migrate the existing model to the new power equipment, adjust the model through the labeled data on the new power equipment, and realize the adaptation and recognition of power equipment of different types and sizes; 4) Data fusion: Fusion of multi-source data including sensors, cameras, and radars to obtain more information about power equipment; 5) Time series information prediction modeling: Add timestamp information to each data point, use recursive neural network time series model for training, capture the time series dependency of data, continuously perform training and loss analysis, build a power equipment prediction model; and use the prediction modeling model to identify the sequence of data; 6) Strengthening adaptability and self-learning capabilities: Designing an adaptive mechanism to monitor the operating status and fault conditions of power equipment in real time; obtaining real-time performance indicators based on the monitoring results, and summarizing and analyzing the real-time monitoring information and evaluation, comprehensively evaluating the performance of the power system, and adaptively adjusting the system operating status; Among them, step 1) edge computing specifically includes: 1-1) Data collection and data preparation: Install sensors, cameras and other related equipment near the power equipment, and write corresponding data collection programs; collect a series of data containing different power equipment scenarios, and annotate the operating status and fault type of the power equipment for each data as training sets and test sets; 1-2) Edge computing data processing: Data processing and analysis are performed on edge nodes close to power equipment; methods including image processing and signal processing are used to extract characteristic parameters from the data, and data cleaning and normalization operations are performed; Step 4) Data fusion specifically includes: Let the sensor data be D s ={s1,s2,...,s m }, where s1,s2,...,s m Represents the data of the 1st to mth sensors; the camera data is D c ={c1,c2,...,c f }, where c1, c2, ..., c f Indicates the first The data of f cameras; the radar data is D r ={r1,r2,...,r b }, where r1, r2, ..., r b represents the data of the 1st to bth radars; each data source is fused to obtain a comprehensive data set Dcombined = {d1, d2, ..., d s+c+r }, where d1,d2,...,d s+c+r Represents the 1st to s+c+rth data points in the comprehensive data set; Use transfer learning to build a classification model. Let the model be M. For a given power equipment data x, the model outputs the corresponding operating status y. state =M(x); Use convolutional neural network to extract spatial features F from camera data camera =CNN(D c ), using a recurrent neural network to learn the temporal information F of radar data radar =RNN(D r ), using convolutional neural networks to extract the spatial features F of sensor data sensor =CNN(D s ), and finally introduce the attention mechanism to dynamically weigh the contribution of each data source F state : F state =Softmax(MLP(D c ))·F camera +Softmax(MLP(D r ))·F radar +Softmax(MLP(D s ))·F sensor ; Finally, multi-source data is integrated to obtain more power equipment information and improve the accuracy and reliability of system monitoring and diagnosis; In step 5), the time series information prediction modeling introduces time series information through continuous data during the operation of power equipment. The power equipment model training set is D source ={x1,x2,...,x n } and have corresponding labels for their respective states, use the recursive neural network time series model for training, capture the time series dependency of the data, continuously perform training and loss analysis, and finally build a power equipment prediction model to estimate the operating status and fault type of the power equipment; where x1, x2, ..., x n Indicates data at different times.

2. The power equipment fault monitoring method based on edge computing according to claim 1 is characterized in that: In step 2), deep learning specifically includes: 2-1) Establish a classification model for the operating status and fault types of power equipment; specifically including: Define the power plant power equipment operation state variables to ensure that the power plant power equipment operation state is linearly related to the analysis, including the voltage amplitude value, current value and voltage value of the power plant power equipment operation; The power plant power equipment operating state variable y is defined as: Where K is the voltage amplitude under the operation of the power plant power equipment; V is the external influence coefficient; L is the current value under the operation of the power plant power equipment; U is the voltage value under the operation of the power plant power equipment; θ ij It is expressed as power factor; i and j represent nodes; 2-2) Use computing resources on edge nodes to train and infer models for real-time monitoring and fault diagnosis of power equipment; specifically: On the basis of defining the operating status variables of the power plant's power equipment, a deep learning method is introduced to analyze the operating status of the power plant's power equipment; In this process, the normal operating state of the relevant power equipment in the power plant is set as A, A={A1,A2,A3,…,A n }, and consider the storage state of the data in A as temporary state; A1,A2,A3,…,A n Respectively represent the data status at different times; The operation space data of the power equipment is mapped and processed, and the mapped data set is output, which is represented by B. Then B={B1,B2,B3,…,B n },B1,B2,B3,…,B n Respectively represent the data status at different times; The sample data set B is subjected to linear parameter classification, and the classified data set is derived, which is represented by C, where C = {C1, C2, C3, …, C n }, where C is the test set; C1, C2, C3, …, C n Respectively represent the data status at different times; The number of nodes in the trust network is selected based on the constraints of the power equipment. During the selection process, data is selected from the sample data sets B and A in the order of 300-500-1000-1500-2000-3000. In this state, the original 2025-dimensional spatial data is mapped to 300-dimensional spatial data, and then the data is tested by IEEE to complete the deep learning of the relevant data. 2-3) Convert the power plant power equipment operating status analysis into an NP-hard problem to analyze the power plant power equipment operating status; set the nonlinear conversion objective function to A, then: In the formula, α ij is the voltage value at the device operation node after nonlinear conversion; β ij is the voltage amplitude value at the equipment operation node after nonlinear conversion; θ ij It is the current value at the device operation node after nonlinear conversion; Complete the nonlinear transformation of the power plant power equipment operating state. On this basis, set the power plant power equipment operating state analysis expression as ΔP t , then: ΔP t =(C t -C t+1 )+η d C t+1 +oh d (K ωmax -L ωt+1 ) Where η d The time when the power plant's power equipment is in operation; ω d The need for decoupling of errors in the operation status analysis of power equipment in power plants; K ωmax is the maximum value of the error in the analysis of the operating status of the power equipment in the power plant; L ωt+1 is the maximum value of the analysis error of the power plant's power equipment operating status on the second day; C t and C t+1 They represent the data status at time t and t+1 in the test collection respectively.

Images