Online parallel fault diagnosis method and system for power grid power equipment

Abstract

The invention relates to the technical field of multivariate data monitoring and diagnosis in the power grid power industry, and provides an online parallel fault diagnosis method and system for power grid power equipment, and the method comprises the steps: receiving power grid feature vector data in real time through an IRichSpout interface, and transmitting the data to a Spout source assembly, thereby forming a to-be-processed data stream; the method comprises the following steps: encapsulating a data stream to be processed into a plurality of Tuple tuples, and generating a unique ID for each Tuple tuple; receiving the Tuple tuple by using a PreBolt component, and preprocessing a data set in the Tuple tuple through a standard fractional method to obtain a standardized sample to be classified; and processing the standardized to-be-classified sample by using a deep classification diagnosis model formed by initialized deep unsupervised training, supervised learning and data flow classification diagnosis deep incremental learning to obtain a fault diagnosis result of the power equipment. According to the method, the classification accuracy is high, and the power grid streaming data can be accurately classified.

Description

Technical Field

[0001] This invention relates to the field of multi-source data monitoring and diagnosis technology in the power grid industry, and in particular to an online parallel fault diagnosis method and system for power grid equipment. Background Technology

[0002] With the development of power systems, power equipment failures have a significant impact on people's lives, thus necessitating continuous monitoring of equipment status. The continuous advancement of sensor and communication technologies has led to an exponential increase in power grid data. This data is characterized by its real-time nature, volatility, and unlimited quantity, making it streaming data requiring continuous monitoring. While the existing platform Hadoop can handle batch data, its real-time performance is poor. Storm, on the other hand, is an open-source distributed real-time computing architecture that can rapidly process massive data streams, compensating for Hadoop's limitations in real-time processing.

[0003] Currently, with the rise of Storm, some applications have emerged in the power industry. A time-based sliding window processing method has been implemented in Storm, and anomaly detection in the power grid data stream is achieved through threshold judgment. Alarm data from power grid equipment is processed quickly, and clustering algorithms are used to process related data streams.

[0004] Deep learning algorithms belong to machine learning algorithms, which can be divided into supervised learning and unsupervised learning. In the real world, most samples are unlabeled, so unsupervised learning is more widely used than supervised learning. Two typical unsupervised deep learning models are Restricted Boltzmann Machines (RBMs) and Autoencoders. They can effectively utilize a large number of unlabeled samples and solve many complex machine learning problems. However, autoencoders have a simple structure but lack probabilistic generative models, while RBMs are too complex. Therefore, classic deep learning algorithms suffer from low classification accuracy and efficiency, and poor noise resistance. Summary of the Invention

[0005] The purpose of this invention is to solve at least one technical problem in the background art and to provide a method and system for online parallel fault diagnosis of power grid equipment.

[0006] To achieve the above objectives, the present invention provides an online parallel fault diagnosis method for power grid equipment, comprising:

[0007] The power grid feature vector data is received in real time through the IRichSpout interface and sent to the Spout source component, forming a data stream to be processed;

[0008] The data stream to be processed is encapsulated into multiple tuples, and a unique ID is generated for each tuple.

[0009] The PreBolt component is used to receive tuples and the dataset in the tuples is preprocessed using the standard score method to obtain standardized samples to be classified.

[0010] By using a deep classification and diagnostic model formed through initial deep unsupervised training, supervised learning, and data stream classification and diagnostic deep incremental learning, standardized samples to be classified are processed to obtain fault diagnosis results for power equipment.

[0011] According to one aspect of the present invention, the deep classification diagnostic model formed by initializing deep unsupervised training, supervised learning, and data stream classification diagnostic deep incremental learning is as follows:

[0012] The DIELMbot processing component is used to initialize deep unsupervised training on standardized samples to be classified, forming the initial training settings for the deep learning model.

[0013] Supervised learning is performed on the deep learning model that has completed initial training settings using labeled samples to form a deep learning model that has completed initialization.

[0014] The standardized samples to be classified are diagnosed by a deep learning model. During the diagnosis process, the deep learning model incrementally learns, adjusts and optimizes the parameters of each layer through deep denoising autoencoder, and forms a deep classification diagnosis model.

[0015] According to one aspect of the invention, the DIELMbott processing component is formed by deploying a deep denoising autoencoder incremental limit learning machine (DEM) within a Bolt component.

[0016] According to one aspect of the present invention, the denoising autoencoder incremental limit learning machine is formed by combining an incremental limit learning machine and a denoising autoencoder, and the deep learning model is formed by a deep denoising autoencoder incremental limit learning machine, wherein its classification layer is classified using an I_ELM classifier.

[0017] According to one aspect of the present invention, the step of performing supervised learning on a deep learning model that has completed initial training settings using labeled samples to form a deep learning model that has completed initialization processing includes:

[0018] Supervised learning is performed using labeled samples by leveraging the IELMBBolt component of the internally deployed I_ELM classifier.

[0019] The IELMBolt component takes the final features extracted by the DIELMolt component as input, maps the feature representation to the target output, and thus completes the initialization process of the deep learning model.

[0020] According to one aspect of the present invention, the deep learning model incrementally learns and adjusts and optimizes the parameters of each layer through deep denoising autoencoder, including:

[0021] Deep learning models learn incrementally through deep denoising autoencoders. Each layer optimizes the parameters using the formulas μ = a1E + a2R and a1 + a2 = 1. A linear combination of E and R is used to evaluate the model's performance. In the formulas, E and R represent the normalized root mean square error and mean absolute error, respectively. μ is the standard for evaluating the model's performance, and a1 and a2 are the weights of E and R, respectively.

[0022] Deep denoising incremental learning continuously extracts features from multiple layers of input data, and the parameters of each layer of the deep learning model are continuously adjusted and optimized during the incremental learning process.

[0023] According to one aspect of the present invention, the fault diagnosis results are stored in a database via a DatabaseBolt component, and the fault diagnosis results are stored in a data file via a FileBolt component.

[0024] To achieve the above objectives, the present invention also provides an online parallel fault diagnosis system for power grid equipment, comprising:

[0025] The data stream acquisition module receives power grid feature vector data in real time through the IRichSpout interface and sends it to the Spout source component, forming a data stream to be processed;

[0026] The data stream encapsulation module encapsulates the data stream to be processed into multiple tuples and generates a unique ID for each tuple.

[0027] The module for obtaining standardized samples to be classified uses the PreBolt component to receive tuples and preprocesses the dataset in the tuples using the standard score method to obtain standardized samples to be classified.

[0028] The fault diagnosis result acquisition module uses a deep classification diagnosis model formed by initialization of deep unsupervised training, supervised learning, and data stream classification diagnosis deep incremental learning to process standardized samples to be classified and obtain fault diagnosis results of power equipment.

[0029] To achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the online parallel fault diagnosis method for power grid equipment as described above.

[0030] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the online parallel fault diagnosis method for power grid equipment as described above.

[0031] According to the scheme of the present invention, the classification accuracy of the present invention is high, and it can accurately classify the streaming data of power grid equipment.

[0032] This invention is suitable for streaming data from power equipment because this data is essentially unlabeled, and the invention does not require a large amount of labeled sample data, but only a small amount of labeled sample data for accurate classification.

[0033] This invention enables deep denoising incremental learning during the data stream diagnostic process, continuously optimizing model parameters during diagnosis, and providing stable and reliable diagnostic capabilities for new data.

[0034] This invention improves the efficiency of streaming data processing for power grid equipment by deploying the classification model on the Storm platform and improving the diagnostic efficiency of fault handling through adaptive configuration of the number of processes, meta-components, and the parallelism of processing components.

[0035] According to the above-described scheme of this invention, the present invention can diagnose online faults in power grid equipment in real time. By deploying a deep autoencoder incremental limit learning (DELT) diagnostic model on the Storm platform, efficient classification processing of streaming data from power grid equipment is achieved. This invention combines a denoising autoencoder and a DELT to construct a denoising autoencoder DELT, thus introducing the concept of denoising autoencoders for unsupervised learning. The denoising autoencoder DELT models are then stacked to construct a deep autoencoder DELT. The parameters of the deep diagnostic model are initially determined through unlabeled initialization training and labeled learning, achieving accurate data classification. In subsequent streaming data fault diagnosis processing, the model parameters are continuously updated and adjusted through multi-layer deep denoising incremental learning, continuously optimizing the network's discriminative performance and achieving accurate data stream classification. Furthermore, the monitoring of power equipment fault types by this invention can ensure the safe operation of power equipment, reduce losses to residents' production and lives, and detect various equipment faults early, preventing catastrophic accidents. Attached Figure Description

[0036] Figure 1A flowchart illustrating an online parallel fault diagnosis method for power grid equipment according to an embodiment of the present invention is shown.

[0037] Figure 2 This diagram illustrates a power grid feature vector data processing procedure according to one embodiment of the present invention.

[0038] Figure 3 This diagram illustrates a data access process according to one embodiment of the present invention.

[0039] Figure 4 The flowchart schematically illustrates incremental limit learning according to one embodiment of the present invention. Detailed Implementation

[0040] The invention will now be discussed with reference to exemplary embodiments. It should be understood that the described embodiments are merely intended to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0041] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment".

[0042] Figure 1 This is a schematic flowchart illustrating an online parallel fault diagnosis method for power grid equipment according to an embodiment of the present invention. Figure 1 As shown, in this embodiment, the online parallel fault diagnosis method for power grid equipment includes:

[0043] The power grid feature vector data is received in real time through the IRichSpout interface and sent to the Spout source component, forming a data stream to be processed;

[0044] The data stream to be processed is encapsulated into multiple tuples, and a unique ID is generated for each tuple.

[0045] The PreBolt component is used to receive tuples and the dataset in the tuples is preprocessed using the standard score method to obtain standardized samples to be classified.

[0046] By using a deep classification and diagnostic model formed through initial deep unsupervised training, supervised learning, and data stream classification and diagnostic deep incremental learning, standardized samples to be classified are processed to obtain fault diagnosis results for power equipment.

[0047] Furthermore, Figure 2This diagram schematically illustrates a power grid feature vector data processing procedure according to one embodiment of the present invention. Figure 2 As shown, in this embodiment, the Storm distributed real-time computing platform is introduced to process streaming data. Storm is a distributed real-time data stream processing platform capable of handling massive data streams. The Storm cluster can be dynamically expanded as needed, has high fault tolerance, and can successfully process each piece of data through the ack mechanism.

[0048] The topology consists of a Spout source component and several Bolt logical processing components. Data tuples are emitted by the Spout and flow to subsequent Bolts, forming a directed acyclic graph. The topology designed for online parallel fault diagnosis of power equipment in this invention is as follows: Figure 2 As shown within the dashed box. Figure 2 SendClient in this invention is a client that can be used to send commands to a running Topology to achieve a certain degree of control.

[0049] Storm uses a master-slave architecture, consisting of two types of nodes: Nimbus and Supervisor. There is only one master node (Nimbus) in the entire cluster, while there can be multiple worker Supervisor nodes. Nimbus is responsible for distributing code across the cluster. Worker nodes are those that run the Supervisor daemon.

[0050] Furthermore, Figure 3 This diagram schematically illustrates a data access flowchart according to one embodiment of the present invention. Figure 3 As shown, in this embodiment, the collected power grid feature vector data is accessed to the Spout source component via the IRichSpout interface. The power grid feature vector data is a continuous, uninterrupted data stream, which is continuously sent to the Spout source component, forming a data stream to be processed, i.e., the samples to be diagnosed. These samples are encapsulated into multiple tuples, and the Spout source component sends these tuples to downstream components.

[0051] Furthermore, according to one embodiment of the present invention, the downstream preprocessing component PreBolt receives tuples from the upstream Spout source component. Each tuple encapsulates the sample to be diagnosed, and each tuple contains 1000 feature vector data. The preprocessing component PreBolt preprocesses the feature vector data within the received tuples. In this embodiment, taking transformer fault diagnosis oil chromatography data as an example, the contents of five gases—H2, CH4, C2H6, C2H2, and C2H4—are selected as inputs and preprocessed. This preprocessing uses the standard score method to process these data. This method can reduce sample data skew and improve diagnostic accuracy. The formula is as follows:

[0052]

[0053] Where, x new Here, x represents the standardized gas content value, μ represents the original gas content value, μ represents the average gas content in the training or test set X, and σ represents the standard deviation of the gas content in X.

[0054] After each tuple in the preprocessing component PreBolt has been preprocessed, the downstream component receives the tuple passed from the preprocessing component PreBolt. The data in this tuple is a standardized transformation of the dataset corresponding to the Id number in the tuple passed to the preprocessing component PreBolt. This is called the standardized unclassified sample. These samples participate in the subsequent fault diagnosis process and are finally summarized according to their numbers.

[0055] Furthermore, according to one embodiment of the present invention, the deep classification and diagnostic model formed by initializing deep unsupervised training, supervised learning, and data stream classification and diagnostic deep incremental learning is as follows:

[0056] The DIELMbot processing component is used to initialize deep unsupervised training on standardized samples to be classified, forming the initial training settings for the deep learning model.

[0057] Supervised learning is performed on the deep learning model that has completed initial training settings using labeled samples to form a deep learning model that has completed initialization.

[0058] The standardized samples to be classified are diagnosed by a deep learning model. During the diagnosis process, the deep learning model incrementally learns, adjusts and optimizes the parameters of each layer through deep denoising autoencoder, and forms a deep classification diagnosis model.

[0059] Furthermore, according to one embodiment of the present invention, the DIELMbott processing component is formed by constructing a deep denoising autoencoder incremental limit learning machine (DEM) and deploying it within the Bolt component. The DEM is formed by combining an incremental limit learning machine and a denoising autoencoder, and the deep learning model is formed by the deep DEM, with its classification layer using an I_ELM classifier for classification.

[0060] In this embodiment, after receiving the tuple from the preprocessing component PreBolt, the DIELMbott processing component performs initial deep unsupervised training on the standardized samples to be classified contained in the tuple, thereby realizing the initial training settings of the entire deep learning model.

[0061] During the initial training process of the DIELMbot processing component, it is necessary to set parameters such as the number of deep learning network layers and the number of nodes in each layer, as well as the erasure probability, to build the initial training model and improve classification accuracy. Specifically, this includes:

[0062] (1) Constructing an incremental limit learning machine

[0063] Let the sample size be N, and the sample data be (P, Q). P represents the input matrix, X i This represents the i-th sample data. Q represents the desired output matrix, Y i Let represent the expected output of the i-th sample, n be the number of input layer nodes, L be the number of hidden layer nodes, and m represent the total number of output nodes.

[0064] The input weights ω and biases B between the input layer and the hidden layer are as follows:

[0065] ω=[ω1 ω2 ... ω L ] n×L ;

[0066] B = [b1 b2 ... b] L ] 1×L ;

[0067] ω i b represents the connection weight between the input layer neuron and the i-th neuron in the hidden layer. i This represents the bias of the i-th neuron in the hidden layer.

[0068] Choose a suitable activation function g(x) to map the output layer data from its original space to the feature space of ELM. Let the activation function be as follows:

[0069]

[0070] The output weight matrix β of the hidden layer and the output layer is:

[0071]

[0072] β jk This represents the connection weight between the j-th neuron in the hidden layer and the k-th neuron in the output layer.

[0073] The relationship between the hidden layer and the output layer is as follows:

[0074] H N×L β L×m =Q N×m ;

[0075] Where H is the output matrix of the hidden layer.

[0076] Figure 4 This describes the specific process of the Incremental Extreme Learning Machine (I_ELM). When I_ELM operates, it first determines a neural network containing L0 nodes, and then uses equation H... N×L β L×m =Q N×m Obtain the output weight matrix. To ensure the accuracy of the network, a minimum error function is defined. Because the hidden layer output matrix is ​​not a square matrix, solving it using the traditional least squares method can easily lead to ill-posed problems. Regularization is usually added to ensure numerical stability.

[0077] min||Hβ-Q||+||λIβ||→0;

[0078] Where λ is an adjustable positive parameter, and I represents the value of the regularization term. For equation H N×L β L×m =Q N×m Solving for the weights yields the output weight matrix:

[0079] β=(H T H+λI) -1 H T Q;

[0080] Substituting the calculated output weight matrix into the equation β=(H T H+λI) -1 H T Q is the output matrix predicted by the model.

[0081] The accuracy of a prediction model can be reflected by the root mean square error and mean absolute error of the network; therefore, a linear combination of the two is chosen as the criterion for evaluating the model's performance.

[0082] μ = a1E + a2R;

[0083] a1 + a2 = 1;

[0084] Where E and R represent the normalized root mean square error and mean absolute error, respectively, μ is the criterion for evaluating the quality of the model, and a1 and a2 are the weights of E and R, respectively.

[0085] If, after a training iteration, the network's absolute error and root mean square error do not meet user requirements or have not reached the preset maximum number of neurons, then nodes are added to the neural network and parameters are updated. Keeping the original input weights ω unchanged, only the input weights ω1 corresponding to the new nodes are generated, resulting in new input weights ω2.

[0086] ω2=[ω ω1] n×(L+l) ;

[0087] Where l represents the number of nodes added.

[0088] Mode The Sigmoid function maps a real number to the interval (0, 1). This function works well when the feature differences of the input matrix are small, so it is chosen as the activation function.

[0089] From this, the hidden layer output matrix H can be obtained. g，ω，b :

[0090]

[0091] After obtaining the new input weights, the model is retrained until the termination condition is met. If the parameters that meet the error condition are not obtained when the number of neurons accumulates to the maximum, then μ is selected during the loop. min The corresponding input weights are used as optimal values ​​to construct the model.

[0092] (2) Constructing a deep denoised autoencoder incremental extreme learning machine

[0093] 1) Constructing a denoised autoencoder incremental extreme learning machine

[0094] Based on I_ELM, the idea of ​​denoising autoencoder is introduced to carry out unsupervised learning, and a denoising autoencoder incremental limit learning machine is constructed.

[0095] By adding a certain type of noise to the input data of I_ELM, the learning objective is to reconstruct the original input data from the noisy input data. The principle is as follows:

[0096] Input sample data is randomly erased using a binomial distribution. The elements are set to have an erasure probability of η. Each element in the input sample data is randomly set to 0 according to probability η, resulting in a noise-added element. Compared to P, P′ loses some features, namely, noise is added to P, X i' represents the i-th sample data with added noise. The denoising autoencoder incremental extreme learning machine uses P′ as the model input. Due to the change in input samples P to P′, the hidden layer output matrix H is transformed into H′. The original data P is reconstructed by minimizing the error, thus achieving the purpose of noise reduction. Its optimization objective function is:

[0097] min||H′β-P||+||λIβ||→0;

[0098] Then the weights between the hidden layer and the output layer of the denoised autoencoder extreme learning machine can be calculated:

[0099] β=(H′TH′+λI) -1 H'P;

[0100] 2) Constructing a deep denoised autoencoder incremental extreme learning machine

[0101] The deep denoising autoencoder incremental limit learning machine (DELM) is composed of multiple stacked DELMs, i.e., it consists of multiple layers of DELMs. The feature extraction process of this deep learning model is as follows:

[0102] Remember R i R is the output of the i-th hidden layer of the deep learning model. i Noise processing was added to obtain H′ i R0 = P, using H′0 as input to train the first denoising autoencoder incremental limit learning machine, obtaining the weight matrix β1 between the hidden layer and the output layer, and then using the connection parameters between the input layer and the hidden layer... Instead, calculate the first hidden layer output R1, then add noise to R1 to obtain H′1. Use H′1 as input to train the second denoised autoencoder incremental extreme learning machine, and then repeat the above steps in sequence. The result can be obtained through calculation using the formula:

[0103] R i =g(H′) i-1 β i T);

[0104] As the training depth increases, each hidden layer of this deep learning model can extract more abstract and easier-to-classify feature representations from the data than the previous layer.

[0105] Furthermore, according to one embodiment of the present invention, supervised learning is performed on a deep learning model that has completed initial training settings using labeled samples to form a deep learning model that has completed initialization processing, including:

[0106] Supervised learning is performed using labeled samples by leveraging the IELMBBolt component of the internally deployed I_ELM classifier.

[0107] The IELMBolt component takes the final features extracted by the DIELMolt component as input, maps the feature representation to the target output, and thus completes the initialization process of the deep learning model.

[0108] Furthermore, according to one embodiment of the present invention, the deep learning model incrementally learns, adjusts, and optimizes the parameters of each layer through deep denoising autoencoder, including:

[0109] Deep learning models learn incrementally through deep denoising autoencoders. Each layer optimizes the parameters using the formulas μ = a1E + a2R and a1 + a2 = 1. A linear combination of E and R is used to evaluate the model's performance. In the formulas, E and R represent the normalized root mean square error and mean absolute error, respectively. μ is the standard for evaluating the model's performance, and a1 and a2 are the weights of E and R, respectively.

[0110] Deep denoising incremental learning continuously extracts features from multiple layers of input data. The parameters of each layer of the deep learning model are continuously adjusted and optimized during the incremental learning process to extract the feature information in the input data to the greatest extent. The parameters are continuously updated and adjusted to maintain the ability to extract features and diagnose new data.

[0111] Furthermore, according to one embodiment of the present invention, the fault diagnosis results are stored in the database through the DatabaseBolt component, and the fault diagnosis results are stored in the data file through the FileBolt component.

[0112] In this embodiment, the Storm framework itself does not handle the storage of computation results. This invention implements Bolt to handle the storage and aggregation of computation results, such as... Figure 2 The storage component SaveBolt can directly write to a data file or persistently store it in a database. Depending on the needs, the results processing methods of this invention include DatabaseBolt and FileBolt. DatabaseBolt implements the storage operation of model diagnostic results to the database, thereby facilitating the query and retrieval of diagnostic results in the power and related industries; while the FileBolt component stores the diagnostic results in a data file, which can be flexibly copied and migrated.

[0113] According to the above-described scheme of the present invention, the classification accuracy of the present invention is high, and it can accurately classify the streaming data of power grid equipment.

[0114] This invention is suitable for streaming data from power equipment because this data is essentially unlabeled, and the invention does not require a large amount of labeled sample data, but only a small amount of labeled sample data for accurate classification.

[0115] This invention enables deep denoising incremental learning during the data stream diagnostic process, continuously optimizing model parameters during diagnosis, and providing stable and reliable diagnostic capabilities for new data.

[0116] This invention improves the efficiency of streaming data processing for power grid equipment by deploying the classification model on the Storm platform and improving the diagnostic efficiency of fault handling through adaptive configuration of the number of processes, meta-components, and the parallelism of processing components.

[0117] According to the above-described scheme of this invention, the present invention can diagnose online faults in power grid equipment in real time. By deploying a deep autoencoder incremental limit learning (DELT) diagnostic model on the Storm platform, efficient classification processing of streaming data from power grid equipment is achieved. This invention combines a denoising autoencoder and a DELT to construct a denoising autoencoder DELT, thus introducing the concept of denoising autoencoders for unsupervised learning. The denoising autoencoder DELT models are then stacked to construct a deep autoencoder DELT. The parameters of the deep diagnostic model are initially determined through unlabeled initialization training and labeled learning, achieving accurate data classification. In subsequent streaming data fault diagnosis processing, the model parameters are continuously updated and adjusted through multi-layer deep denoising incremental learning, continuously optimizing the network's discriminative performance and achieving accurate data stream classification. Furthermore, the monitoring of power equipment fault types by this invention can ensure the safe operation of power equipment, reduce losses to residents' production and lives, and detect various equipment faults early, preventing catastrophic accidents.

[0118] Furthermore, to achieve the above objectives, the present invention also provides an online parallel fault diagnosis system for power grid equipment, comprising:

[0119] The data stream acquisition module receives power grid feature vector data in real time through the IRichSpout interface and sends it to the Spout source component, forming a data stream to be processed;

[0120] The data stream encapsulation module encapsulates the data stream to be processed into multiple tuples and generates a unique ID for each tuple.

[0121] The module for obtaining standardized samples to be classified uses the PreBolt component to receive tuples and preprocesses the dataset in the tuples using the standard score method to obtain standardized samples to be classified.

[0122] The fault diagnosis result acquisition module uses a deep classification diagnosis model formed by initialization of deep unsupervised training, supervised learning, and data stream classification diagnosis deep incremental learning to process standardized samples to be classified and obtain fault diagnosis results of power equipment.

[0123] Furthermore, such as Figure 2 As shown, in this embodiment, the Storm distributed real-time computing platform is introduced to process streaming data. Storm is a distributed real-time data stream processing platform capable of handling massive data streams. The Storm cluster can be dynamically expanded as needed, has high fault tolerance, and can successfully process each piece of data through the ack mechanism.

[0124] The topology consists of a data source component (Spout) and several logical processing components (Bolt). Data tuples are emitted by the Spout and flow to subsequent Bolts, forming a directed acyclic graph. The topology designed for online parallel fault diagnosis of power equipment in this invention is as follows: Figure 2 As shown within the dashed box. Figure 2 SendClient in this invention is a client that can be used to send commands to a running Topology to achieve a certain degree of control.

[0125] Storm uses a master-slave architecture, consisting of two types of nodes: Nimbus and Supervisor. There is only one master node (Nimbus) in the entire cluster, while there can be multiple worker Supervisor nodes. Nimbus is responsible for distributing code across the cluster. Worker nodes are those that run the Supervisor daemon.

[0126] Furthermore, such as Figure 3 As shown, in this embodiment, the collected power grid feature vector data is accessed to the Spout source component via the IRichSpout interface. The power grid feature vector data is a continuous, uninterrupted data stream, which is continuously sent to the Spout source component, forming a data stream to be processed, i.e., the samples to be diagnosed. These samples are encapsulated into multiple tuples, and the Spout source component sends these tuples to downstream components.

[0127] Furthermore, according to one embodiment of the present invention, the downstream preprocessing component PreBolt receives tuples from the upstream Spout source component. Each tuple encapsulates the sample to be diagnosed, and each tuple contains 1000 feature vector data. The preprocessing component PreBolt preprocesses the feature vector data within the received tuples. In this embodiment, taking transformer fault diagnosis oil chromatography data as an example, the contents of five gases—H2, CH4, C2H6, C2H2, and C2H4—are selected as inputs and preprocessed. This preprocessing uses the standard score method to process these data. This method can reduce sample data skew and improve diagnostic accuracy. The formula is as follows:

[0128]

[0129] Where, x new Here, x represents the standardized gas content value, μ represents the original gas content value, μ represents the average gas content in the training or test set X, and σ represents the standard deviation of the gas content in X.

[0130] After each tuple in the preprocessing component PreBolt has been preprocessed, the downstream component receives the tuple passed from the preprocessing component PreBolt. The data in this tuple is a standardized transformation of the dataset corresponding to the Id number in the tuple passed to the preprocessing component PreBolt. This is called the standardized unclassified sample. These samples participate in the subsequent fault diagnosis process and are finally summarized according to their numbers.

[0131] Furthermore, according to one embodiment of the present invention, the deep classification and diagnostic model formed by initializing deep unsupervised training, supervised learning, and data stream classification and diagnostic deep incremental learning is as follows:

[0132] The DIELMbot processing component is used to perform initial deep unsupervised training on standardized samples to be classified, forming the initial training settings for the deep learning model;

[0133] Supervised learning is performed on the deep learning model that has completed initial training settings using labeled samples to form a deep learning model that has completed initialization.

[0134] The standardized samples to be classified are diagnosed by a deep learning model. During the diagnosis process, the deep learning model incrementally learns, adjusts and optimizes the parameters of each layer through deep denoising autoencoder, and forms a deep classification diagnosis model.

[0135] Furthermore, according to one embodiment of the present invention, the DIELMbott processing component is formed by constructing a deep denoising autoencoder incremental limit learning machine (DEM) and deploying it within the Bolt component. The DEM is formed by combining an incremental limit learning machine and a denoising autoencoder, and the deep learning model is formed by the deep DEM, with its classification layer using an I_ELM classifier for classification.

[0136] In this embodiment, after receiving the tuple from the preprocessing component PreBolt, the DIELMbott processing component performs initial deep unsupervised training on the standardized samples to be classified contained in the tuple, thereby realizing the initial training settings of the entire deep learning model.

[0137] During the initial training process of the DIELMbot processing component, it is necessary to set parameters such as the number of deep learning network layers and the number of nodes in each layer, as well as the erasure probability, to build the initial training model and improve classification accuracy. Specifically, this includes:

[0138] (1) Constructing an incremental limit learning machine

[0139] Let the sample size be N, and the sample data be (P, Q). P represents the input matrix, X i This represents the i-th sample data. Q represents the desired output matrix, Y i Let represent the expected output of the i-th sample, n be the number of input layer nodes, L be the number of hidden layer nodes, and m represent the total number of output nodes.

[0140] The input weights ω and biases B between the input layer and the hidden layer are as follows:

[0141] ω=[ω1 ω2 ... ω L ] n×L ;

[0142] B = [b1 b2...b] L ] 1×L ;

[0143] ω i b represents the connection weight between the input layer neuron and the i-th neuron in the hidden layer. i This represents the bias of the i-th neuron in the hidden layer.

[0144] Choose a suitable activation function g(x) to map the output layer data from its original space to the feature space of ELM. Let the activation function be as follows:

[0145]

[0146] The output weight matrix β of the hidden layer and the output layer is:

[0147]

[0148] β jk This represents the connection weight between the j-th neuron in the hidden layer and the k-th neuron in the output layer.

[0149] The relationship between the hidden layer and the output layer is as follows:

[0150] H N×L β L×m =Q N×m ;

[0151] Where H is the output matrix of the hidden layer.

[0152] Figure 4 This describes the specific process of the Incremental Extreme Learning Machine (I_ELM). When I_ELM operates, it first determines a neural network containing L0 nodes, and then uses equation H... N×L β L×m =Q N×m Obtain the output weight matrix. To ensure the accuracy of the network, a minimum error function is defined. Because the hidden layer output matrix is ​​not a square matrix, solving it using the traditional least squares method can easily lead to ill-posed problems. Regularization is usually added to ensure numerical stability.

[0153] min||Hβ-Q||+||λIβ||→0;

[0154] Where λ is an adjustable positive parameter, and I represents the value of the regularization term. For equation H N×L β L×m =Q N×m Solving for the weights yields the output weight matrix:

[0155] β=(H T H+λI) -1 H T Q;

[0156] Substituting the calculated output weight matrix into the equation β=(H T H+λI) -1 H T Q is the output matrix predicted by the model.

[0157] The accuracy of a prediction model can be reflected by the root mean square error and mean absolute error of the network; therefore, a linear combination of the two is chosen as the criterion for evaluating the model's performance.

[0158] μ = a1E + a2R;

[0159] a1 + a2 = 1;

[0160] Where E and R represent the normalized root mean square error and mean absolute error, respectively, μ is the criterion for evaluating the quality of the model, and a1 and a2 are the weights of E and R, respectively.

[0161] If, after a training iteration, the network's absolute error and root mean square error do not meet user requirements or have not reached the preset maximum number of neurons, then nodes are added to the neural network and parameters are updated. Keeping the original input weights ω unchanged, only the input weights ω1 corresponding to the new nodes are generated, resulting in new input weights ω2.

[0162] ω2=[ω ω1] n×(L+l) ;

[0163] Where l represents the number of nodes added.

[0164] Mode The Sigmoid function maps a real number to the interval (0, 1). This function works well when the feature differences of the input matrix are small, so it is chosen as the activation function.

[0165] From this, the hidden layer output matrix H can be obtained. g，ω，b :

[0166]

[0167] After obtaining the new input weights, the model is retrained until the termination condition is met. If the parameters that meet the error condition are not obtained when the number of neurons accumulates to the maximum, then μ is selected during the loop. min The corresponding input weights are used as optimal values ​​to construct the model.

[0168] (2) Constructing a deep denoised autoencoder incremental extreme learning machine

[0169] 1) Constructing a denoised autoencoder incremental extreme learning machine

[0170] Based on I_ELM, the idea of ​​denoising autoencoder is introduced to carry out unsupervised learning, and a denoising autoencoder incremental limit learning machine is constructed.

[0171] By adding a certain type of noise to the input data of I_ELM, the learning objective is to reconstruct the original input data from the noisy input data. The principle is as follows:

[0172] Input sample data is randomly erased using a binomial distribution. The elements are set to have an erasure probability of η. Each element in the input sample data is randomly set to 0 according to probability η, resulting in a noise-added element. Compared to P, P′ loses some features, namely, noise is added to P, X i ' represents the i-th sample data with added noise. The denoising autoencoder incremental extreme learning machine uses P′ as the model input. Due to the change in input samples P to P′, the hidden layer output matrix H is transformed into H′. The original data P is reconstructed by minimizing the error, thus achieving the purpose of noise reduction. Its optimization objective function is:

[0173] min=Hβ-P||+||λIβ||→0;

[0174] Then the weights between the hidden layer and the output layer of the denoised autoencoder extreme learning machine can be calculated:

[0175] β=(H′ 1 H′+λI) -1 H'P;

[0176] 2) Constructing a deep denoised autoencoder incremental extreme learning machine

[0177] The deep denoising autoencoder incremental limit learning machine (DELM) is composed of multiple stacked DELMs, i.e., it consists of multiple layers of DELMs. The feature extraction process of this deep learning model is as follows:

[0178] Remember R i R is the output of the i-th hidden layer of the deep learning model. i Noise processing was added to obtain H′ i R0 = P, using H′0 as input to train the first denoising autoencoder incremental limit learning machine, obtaining the weight matrix β1 between the hidden layer and the output layer, and then using the connection parameters between the input layer and the hidden layer... Instead, calculate the first hidden layer output R1, then add noise to R1 to obtain H1′, and use H1′ as input to train the second denoised autoencoder incremental extreme learning machine, and then repeat the above steps in sequence. The result can be obtained through calculation using the formula:

[0179] R i =g(H′) i-1 β i T );

[0180] As the training depth increases, each hidden layer of this deep learning model can extract more abstract and easier-to-classify feature representations from the data than the previous layer.

[0181] Furthermore, according to one embodiment of the present invention, supervised learning is performed on a deep learning model that has completed initial training settings using labeled samples to form a deep learning model that has completed initialization processing, including:

[0182] Supervised learning is performed using labeled samples by leveraging the IELMBBolt component of the internally deployed I_ELM classifier.

[0183] The IELMBolt component takes the final features extracted by the DIELMolt component as input, maps the feature representation to the target output, and thus completes the initialization process of the deep learning model.

[0184] Furthermore, according to one embodiment of the present invention, the deep learning model incrementally learns, adjusts, and optimizes the parameters of each layer through deep denoising autoencoder, including:

[0185] Deep learning models learn incrementally through deep denoising autoencoders. Each layer optimizes the parameters using the formulas μ = a1E + a2R and a1 + a2 = 1. A linear combination of E and R is used to evaluate the model's performance. In the formulas, E and R represent the normalized root mean square error and mean absolute error, respectively. μ is the standard for evaluating the model's performance, and a1 and a2 are the weights of E and R, respectively.

[0186] Deep denoising incremental learning continuously extracts features from multiple layers of input data. The parameters of each layer of the deep learning model are continuously adjusted and optimized during the incremental learning process to extract the feature information in the input data to the greatest extent. The parameters are continuously updated and adjusted to maintain the ability to extract features and diagnose new data.

[0187] Furthermore, according to one embodiment of the present invention, the fault diagnosis results are stored in the database through the DatabaseBolt component, and the fault diagnosis results are stored in the data file through the FileBolt component.

[0188] In this embodiment, the Storm framework itself does not handle the storage of computation results. This invention implements Bolt to handle the storage and aggregation of computation results, such as... Figure 2 The storage component SaveBolt can directly write to a data file or persistently store it in a database. Depending on the needs, the results processing methods of this invention include DatabaseBolt and FileBolt. DatabaseBolt implements the storage operation of model diagnostic results to the database, thereby facilitating the query and retrieval of diagnostic results in the power and related industries; while the FileBolt component stores the diagnostic results in a data file, which can be flexibly copied and migrated.

[0189] According to the above-described scheme of the present invention, the classification accuracy of the present invention is high, and it can accurately classify the streaming data of power grid equipment.

[0190] This invention is suitable for streaming data from power equipment because this data is essentially unlabeled, and the invention does not require a large amount of labeled sample data, but only a small amount of labeled sample data for accurate classification.

[0191] This invention enables deep denoising incremental learning during the data stream diagnostic process, continuously optimizing model parameters during diagnosis, and providing stable and reliable diagnostic capabilities for new data.

[0192] This invention improves the efficiency of streaming data processing for power grid equipment by deploying the classification model on the Storm platform and improving the diagnostic efficiency of fault handling through adaptive configuration of the number of processes, meta-components, and the parallelism of processing components.

[0193] According to the above-described scheme of this invention, the present invention can diagnose online faults in power grid equipment in real time. By deploying a deep autoencoder incremental limit learning (DELT) diagnostic model on the Storm platform, efficient classification processing of streaming data from power grid equipment is achieved. This invention combines a denoising autoencoder and a DELT to construct a denoising autoencoder DELT, thus introducing the concept of denoising autoencoders for unsupervised learning. The denoising autoencoder DELT models are then stacked to construct a deep autoencoder DELT. The parameters of the deep diagnostic model are initially determined through unlabeled initialization training and labeled learning, achieving accurate data classification. In subsequent streaming data fault diagnosis processing, the model parameters are continuously updated and adjusted through multi-layer deep denoising incremental learning, continuously optimizing the network's discriminative performance and achieving accurate data stream classification. Furthermore, the monitoring of power equipment fault types by this invention can ensure the safe operation of power equipment, reduce losses to residents' production and lives, and detect various equipment faults early, preventing catastrophic accidents.

[0194] Furthermore, to achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the online parallel fault diagnosis method for power grid equipment as described above.

[0195] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the online parallel fault diagnosis method for power grid equipment as described above.

[0196] Those skilled in the art will recognize that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0197] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and equipment can be referred to the corresponding process in the foregoing method implementation, and will not be repeated here.

[0198] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0199] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the objectives of the embodiments of the present invention, depending on actual needs.

[0200] In addition, the functional modules in the embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0201] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the energy-saving signal transmission / reception methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0202] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this application.

[0203] It should be understood that the sequence number of each step in the invention and its embodiments does not absolutely imply 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 embodiments of the present invention.

Claims

1. A method for online parallel fault diagnosis of power grid power equipment, characterized in that: include: Receive power grid feature vector data in real time through the IRichSpout interface and send it to the Spout source component, forming a data stream to be processed; Encapsulate the data stream to be processed into multiple Tuple tuples and generate a unique ID for each Tuple tuple; Use the PreBolt component to receive the Tuple tuple and preprocess the data set in the Tuple tuple using the standard score method to obtain standardized samples to be classified; The deep classification diagnosis model formed by initializing deep unsupervised training, supervised learning and deep incremental learning of data stream classification diagnosis is used to process standardized samples to be classified and obtain the fault diagnosis results of power equipment.

2. The online parallel fault diagnosis method for power grid power equipment according to claim 1, characterized in that: The deep classification diagnosis model formed by initializing deep unsupervised training, supervised learning and data stream classification diagnosis deep incremental learning is: Use the DIELMBolt processing component to initialize deep unsupervised training on standardized samples to be classified, forming the initial training settings of the deep learning model; Using labeled samples to perform supervised learning on the deep learning model that has completed the initialization training setting, to form a deep learning model that has completed the initialization processing; The standardized samples to be classified are diagnosed through a deep learning model. During the diagnosis process, the deep learning model adjusts and optimizes the parameters of each layer through deep denoising autoencoding incremental learning to form a deep classification diagnosis model.

3. The online parallel fault diagnosis method for power grid power equipment according to claim 2 is characterized in that: The DIELMBolt processing component is formed by constructing a multi-layer denoising autoencoder incremental extreme learning machine and deploying a deep denoising autoencoder incremental extreme learning machine in a Bolt component.

4. The online parallel fault diagnosis method for power grid power equipment according to claim 3 is characterized in that: The denoising autoencoder incremental extreme learning machine is formed by combining an incremental extreme learning machine and a denoising autoencoder, and the deep learning model is formed by a deep denoising autoencoder incremental extreme learning machine, and its classification layer uses an I_ELM classifier for classification.

5. The online parallel fault diagnosis method for power grid power equipment according to claim 2, characterized in that: The method of using labeled samples to perform supervised learning on the deep learning model that has completed the initialization training setting to form a deep learning model that has completed the initialization processing includes: The IELMBolt component uses the internally deployed I_ELM classifier to perform supervised learning using labeled samples; The IELMBolt component takes the final features extracted by the DIELMBolt component as input and maps the feature representation to the target output, thus completing the initialization process of the deep learning model.

6. The method for online parallel fault diagnosis of power grid power equipment according to claim 2, characterized in that: The deep learning model uses deep denoising autoencoder incremental learning to adjust and optimize the parameters of each layer, including The deep learning model uses deep denoising autoencoder incremental learning. Each layer optimizes the parameters through the formula μ=a1E+a2R and the formula a1+a2=1. The linear combination of E and R is used to evaluate the quality of the model. In the formula, E and R represent the normalized root mean square error and mean absolute error, μ is the standard for evaluating the quality of the model, and a1 and a2 are the weights of E and R respectively. Deep denoising incremental learning continuously extracts features from multi-layer input data, and the parameters of each layer of the deep learning model are continuously adjusted and optimized during the incremental learning process.

7. The method for online parallel fault diagnosis of power grid power equipment according to any one of claims 1 to 6, characterized in that: The fault diagnosis result is stored in a database through the DatabaseBolt component, and the fault diagnosis result is stored in a data file through the FileBolt component.

8. Online parallel fault diagnosis system for power grid power equipment, characterized in that: include: The data stream acquisition module receives the power grid feature vector data in real time through the IRichSpout interface and sends it to the Spout source component, forming a data stream to be processed; The data stream encapsulation module encapsulates the data stream to be processed into multiple Tuple tuples and generates a unique ID for each Tuple tuple; The standardized sample acquisition module uses the PreBolt component to receive the Tuple tuple and preprocesses the data set in the Tuple tuple using the standard score method to obtain the standardized sample to be classified; The fault diagnosis result acquisition module uses the deep classification diagnosis model formed by initializing deep unsupervised training, supervised learning and data stream classification diagnosis deep incremental learning to process standardized samples to be classified and obtain the fault diagnosis results of the power equipment.

9. An electronic device, characterized in that The method comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the method for online parallel fault diagnosis of power equipment in a power grid as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for online parallel fault diagnosis of power equipment in a power grid according to any one of claims 1 to 7 is implemented.

Images