CN118133203A - Fault diagnosis method for electric energy metering detection information - Google Patents
Fault diagnosis method for electric energy metering detection information Download PDFInfo
- Publication number
- CN118133203A CN118133203A CN202410328445.6A CN202410328445A CN118133203A CN 118133203 A CN118133203 A CN 118133203A CN 202410328445 A CN202410328445 A CN 202410328445A CN 118133203 A CN118133203 A CN 118133203A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- unit
- learning
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 135
- 238000000034 method Methods 0.000 title claims abstract description 134
- 238000001514 detection method Methods 0.000 title claims abstract description 72
- 238000012545 processing Methods 0.000 claims abstract description 83
- 230000002787 reinforcement Effects 0.000 claims abstract description 74
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 62
- 238000000605 extraction Methods 0.000 claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 claims abstract description 32
- 230000004927 fusion Effects 0.000 claims abstract description 29
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000013135 deep learning Methods 0.000 claims abstract description 14
- 238000005516 engineering process Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 65
- 238000012549 training Methods 0.000 claims description 45
- 230000009471 action Effects 0.000 claims description 33
- 230000002776 aggregation Effects 0.000 claims description 21
- 238000004220 aggregation Methods 0.000 claims description 21
- 238000009826 distribution Methods 0.000 claims description 20
- 230000007246 mechanism Effects 0.000 claims description 20
- 230000002159 abnormal effect Effects 0.000 claims description 17
- 230000009467 reduction Effects 0.000 claims description 17
- 238000011176 pooling Methods 0.000 claims description 15
- 230000003044 adaptive effect Effects 0.000 claims description 14
- 238000013507 mapping Methods 0.000 claims description 14
- 238000012098 association analyses Methods 0.000 claims description 13
- 238000013461 design Methods 0.000 claims description 13
- 230000010354 integration Effects 0.000 claims description 12
- 238000005065 mining Methods 0.000 claims description 12
- 230000008485 antagonism Effects 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 11
- 238000012300 Sequence Analysis Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 10
- 230000005540 biological transmission Effects 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 10
- 238000010219 correlation analysis Methods 0.000 claims description 10
- 238000004891 communication Methods 0.000 claims description 9
- 230000008713 feedback mechanism Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000003993 interaction Effects 0.000 claims description 8
- 230000005856 abnormality Effects 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 7
- 239000000284 extract Substances 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000003066 decision tree Methods 0.000 claims description 6
- 238000013136 deep learning model Methods 0.000 claims description 6
- 238000003475 lamination Methods 0.000 claims description 6
- 238000012544 monitoring process Methods 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 6
- 238000011426 transformation method Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 claims description 4
- 238000013531 bayesian neural network Methods 0.000 claims description 3
- 238000007621 cluster analysis Methods 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 230000008034 disappearance Effects 0.000 claims description 3
- 238000007499 fusion processing Methods 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 230000007774 longterm Effects 0.000 claims description 3
- 230000014759 maintenance of location Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000007500 overflow downdraw method Methods 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 238000011084 recovery Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000000547 structure data Methods 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 230000001360 synchronised effect Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 3
- 230000004044 response Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0475—Generative networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/092—Reinforcement learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/094—Adversarial learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a fault diagnosis method for electric energy metering detection information, relates to the field of intelligent power grids, and solves the defects of poor data processing instantaneity, dependence on manual experience, lack of self-adaptability and strong limitation in the existing fault diagnosis method; the invention combines the deep learning self-encoder and the graph convolution neural network to realize feature extraction and process complex association relation; self-adaptive learning and real-time adjustment are realized through an anomaly detection algorithm based on the generation of an countermeasure network and reinforcement learning; the real-time performance and the efficiency are improved through an edge computing technology and a distributed processing model; the data association relation is mined through a fusion diagnosis system, so that the accuracy and the comprehensiveness of data diagnosis are improved; model generalization capability is promoted through meta-learning; optimizing the stability of the model by an active learning method; finally, the invention realizes real-time alarm decision and fault processing strategy generation by means of the self-organizing map network model of reinforcement learning, and greatly improves the response speed and accuracy of the existing diagnosis method.
Description
Technical Field
The invention relates to the field of smart grids, and in particular relates to a fault diagnosis method for electric energy metering detection information.
Background
With the continuous development of power systems and the wide application of electric energy metering technologies, the collection and processing of power information becomes more and more important. However, in the past several decades, the power meter detection information fault diagnosis method is still in a relatively late stage. Traditional fault diagnosis methods mainly rely on manual experience and rules, and lack self-adaptability and data processing capability. This results in poor accuracy and efficiency of fault diagnosis, failing to meet the increasing demands of power systems.
In recent years, with the rapid development of artificial intelligence and big data technology, the fault diagnosis method of the electric energy metering detection information is remarkably improved. Based on the development of machine learning, deep learning, graphic neural network and other technologies, the new generation of fault diagnosis method has stronger data processing capability and adaptability. By analyzing and modeling the large-scale electric energy metering data, accurate fault diagnosis and prediction can be realized, and the reliability and safety of the electric power system are improved.
However, the existing fault diagnosis method for the electric energy metering detection information still has some defects, and further application and development of the fault diagnosis method are limited. First, these methods have a problem of low efficiency and poor real-time performance in data processing. Because of the large and complex data volume generated by the electric energy metering system, the traditional method cannot realize real-time diagnosis. Second, existing methods rely on manual experience and rules and lack adaptivity. This makes the fault diagnosis result susceptible to human factors and difficult to accommodate for variations in different environments and system parameters. In addition, the limitations of the existing methods are also apparent. They are often based on specific rules or models, and do not adapt well to complex and varying fault conditions, resulting in insufficiently accurate and reliable diagnostic results.
Therefore, the invention discloses a fault diagnosis method for the electric energy metering detection information.
Disclosure of Invention
Aiming at the defects of the prior art, the invention discloses a fault diagnosis method for electric energy metering detection information, which solves the problems of poor real-time data processing, dependence on manual experience, lack of adaptability and strong limitation in the existing fault diagnosis method for the electric energy metering detection information.
In order to achieve the technical effects, the invention adopts the following technical scheme:
A method of fault diagnosis of electrical energy metering detection information, wherein the method comprises:
As a further technical scheme of the invention, the method comprises the following steps:
step one, data acquisition and pretreatment;
Collecting electric energy metering data through an electric energy meter and sensor equipment, and performing data cleaning, denoising and outlier processing through a signal processing method;
Step two, extracting and selecting characteristics;
extracting data characteristics of the electric energy metering equipment through a deep learning self-encoder, wherein the deep learning self-encoder reduces data dimensionality and retains key information through a back propagation and gradient descent method;
step three, introducing a deep learning model;
complex topological structures and association relations in the electric energy metering system are processed through a graph convolution neural network structure, and the understanding capability of a model to the system state is improved through a mode of node representation learning, information transmission and local global feature processing;
Step four, abnormality detection and self-adaptive learning;
Dynamic detection and self-adaptive learning of potential faults in the electric energy metering system are realized through an anomaly detection algorithm based on generation of an countermeasure network and reinforcement learning;
Fifthly, fusing an edge computing technology;
Real-time data processing and fault diagnosis at the equipment end are realized through a distributed processing model;
step six, multi-source data fusion and association analysis;
The data from different sensors are fused through a fusion diagnosis system, and the fusion diagnosis system excavates the association relation between different data sources through association analysis and a time sequence analysis method;
Seventh, cross-equipment collaborative learning;
Data sharing and model collaborative learning among a plurality of electric energy metering devices are realized through a meta-learning model;
step eight, model optimization and iterative updating;
Introducing an active learning method based on model uncertainty to optimize and iteratively update the model in real time so as to adapt to the dynamic change of the running state of the electric energy metering system;
Step nine, alarming and fault processing in real time;
And realizing real-time alarm decision and fault processing strategy generation through a self-organizing map network model based on reinforcement learning.
As a further technical scheme of the invention, the deep learning self-encoder comprises an encoder module, a decoder module, a feature extraction module and a dimension reduction and reconstruction module; the encoder module dynamically adjusts the learning rate through the adaptive moment estimation optimizer; the adaptive moment estimation optimizer improves model convergence speed and generalization capability through a convolution layer and a pooling layer; the convolution layer realizes feature extraction and nonlinear mapping of input data through convolution operation and an activation function; the pooling layer realizes feature dimension reduction and main information retention through average pooling operation, and reduces the calculated amount at the same time; the decoder module realizes jump connection through a residual connection method, and the residual connection method avoids the problem of gradient disappearance through a transposed convolution layer and a batch normalization layer; the transpose convolution layer realizes the reconstruction and recovery of data through deconvolution operation; the characteristic extraction module is used for effectively extracting and characterizing the data characteristics of the electric energy metering equipment through a convolution self-encoder structure; the convolution self-encoder structure improves the nonlinear fitting capacity of the features through a sparse coding and sparse punishment mechanism; the dimension reduction and reconstruction module realizes sparse representation and dimension reduction compression of data through a sparse self-encoder structure and an L1 regularization method.
As a further technical scheme of the invention, the graph roll-up neural network structure comprises an input layer, a graph roll-up layer, a pooling layer and a full-connection layer; the graph convolution neural network structure represents node characteristic data and an edge connection relation in the electric energy metering system as graph structure input through the input layer, and realizes characteristic extraction and information transmission of the graph structure data through the graph convolution layer; the picture volume lamination comprises a neighbor aggregation unit, a feature conversion unit, an activation function unit and a picture volume lamination unit; the neighbor aggregation unit is used for realizing the aggregation and updating of node characteristics by weighting the characteristic information of adjacent nodes through an adjacent matrix; the feature conversion unit carries out linear transformation on the aggregated neighbor node features through weight parameters so as to obtain higher-level feature representation; the activation function unit carries out nonlinear mapping on the characteristics through a linear rectification function, so that the network learning capacity is enhanced; the graph rolling unit captures the association relation between nodes through graph rolling operation; the graph convolution neural network structure gathers the characteristics of different nodes in the graph through the pooling layer to obtain the characteristic representation of the whole graph, and the pooled graph characteristics are input into the full-connection neural network through the full-connection layer to output a final fault diagnosis result.
As a further technical scheme of the invention, the anomaly detection algorithm based on the generation of the antagonism network and reinforcement learning trains the generation of the antagonism network through generating the objective function of the antagonism network, so that the samples generated by the generator are more similar to the real data distribution, and the discriminator is more good at distinguishing the real data from the generated data; the formula for generating the objective function of the countermeasure network is:
in formula (1), R is a discriminator parameter for discriminating between a real data sample and a dummy sample generated by the generator; y is a real data sample for training the discriminator in training; m is real data distribution, and is used for generating distribution characteristics of a real sample; q is a noise sample, used to generate a false sample; p is the noise distribution, which represents the distribution of noise samples; then, calculating the loss in the classification task through a classification loss function so as to predict a sample label; the formula expression of the classification loss function is as follows:
in the formula (2), z is a real tag, and is used for calculating a tag value of real data; u is a prediction label and is used for calculating a classification predicted value of the model on the data; beta is the number of samples; reconstructing the data sample by the generator to determine a difference between the data sample and the real data; the anomaly detection algorithm calculates the difference between the real data sample and the reconstructed data sample through the reconstructed error loss function; the formula expression of the reconstruction error loss function is as follows:
In the formula (3), c is a real data sample, and is used for calculating the original characteristics of the real data sample; b is a reconstructed data sample generated for evaluating differences between the reconstructed data sample and the real data; θ is a weight adjustment amount for updating model parameters; the change rate of the error to the model parameters is represented and used for guiding the parameter updating direction; discount-accumulating rewards of future time steps through a cumulative reward function based on a reinforcement learning algorithm to evaluate long-term rewards of the current strategy; the formula expression of the jackpot function based on the reinforcement learning algorithm is as follows:
In equation (4), E a is the jackpot at time a, used to measure the sum of the jackpots accumulated by the model at time a during training, and w i is the real-time prize of step i, used to measure the instant prize value obtained by the model at each step; k is a termination time for specifying a termination time of training; finally, calculating the expected return of taking a certain action in the current state by using an action cost function based on a reinforcement learning algorithm, and updating a decision rule by combining the maximum action value at the next moment; the formula expression of the action cost function based on the reinforcement learning algorithm is as follows:
in equation (5), F is taking action in state l For evaluating the action value; s is action taken in state l/>The instant rewards are used for measuring model instant rewards; q is a discount factor for balancing the importance of the current and future rewards.
As a further technical scheme of the invention, the distributed processing model comprises a data receiving module, a feature extraction and selection module, a distributed computing module and a real-time fault diagnosis module; the data receiving module acquires various parameter data of the electric energy metering detection equipment in real time through a data communication interface; the feature extraction and selection module comprises a feature extraction unit and a feature selection unit; the feature extraction unit extracts frequency domain features and time domain feature representations from the preprocessed data through a harmonic analysis and wavelet transformation method; the feature selection unit screens representative and distinguishing feature subsets through correlation analysis and information gain; the distributed computing module comprises edge computing nodes, a model training unit and a model aggregation unit; the edge computing node realizes data processing and model training tasks of the equipment end through a distributed computing and containerization method; the model training unit divides the local data into local data parts and locally trains the model through synchronous gradient descent and asynchronous optimization methods, and then realizes global model updating through the model aggregation unit; the model aggregation unit integrates model parameters obtained by training on each edge computing node through a weighted average and federal learning aggregation method to obtain a global model; the real-time fault diagnosis module comprises a fault diagnosis unit and a fault judgment unit; based on the global model and the real-time data, the fault diagnosis unit judges the fault type and degree through a support vector machine; and according to the fault diagnosis result, the fault judging unit adopts a decision tree and a rule engine to judge whether the equipment has faults or not, and outputs processing advice or alarm information.
As a further technical scheme of the invention, the fusion diagnosis system comprises a data fusion module, a correlation analysis module and a comprehensive diagnosis and decision module; the data fusion module comprises a sensor integration unit and a feature recognition unit; the sensor integration unit performs fusion processing on the data of different sensors through a weighted average and feature fusion method; the feature recognition unit extracts effective feature information from the fused data through a principal component analysis and wavelet transformation method; the association analysis module comprises a time sequence analysis unit and an association mining unit; the time sequence analysis unit is used for excavating time correlation and trend among different data sources through a time sequence model, an autoregressive model and a sliding window; the association mining unit discovers association rules and potential association factors between data through association rule mining, frequent pattern mining and cluster analysis methods; the comprehensive diagnosis and decision module comprises a comprehensive diagnosis unit and a decision support unit; combining the correlation analysis result and the feature extraction information, and the comprehensive diagnosis unit recognizes abnormal conditions in the system through a deep learning model and gives a diagnosis result; based on the comprehensive diagnosis result, the decision support unit provides decision support through a decision tree, an expert system and a fuzzy logic method.
As a further technical scheme of the invention, the meta learning model comprises a model construction module, a multi-device data sharing module, a model collaborative learning module and an adaptive diagnosis module; the model construction module comprises a model structure design unit and a learning strategy design unit; the model structure design unit realizes the rapid learning capacity of the model on a new task through a parameter sharing mechanism in the neural network; the learning strategy design unit performs feature fusion by combining data of a source field and a target field through a field self-adaptive method; the field self-adaptive method realizes the effective utilization of cross-field data and the improvement of model generalization capability through a feature alignment and mapping learning method; the multi-device data sharing module comprises a communication unit and an integration unit; the communication unit encrypts and transmits data through SSL/TLS encryption protocol, and ensures the security of data transmission through an identity authentication mechanism; the integration unit fuses different device data to generate a new sample through an countermeasure generation network; the model collaborative learning module comprises an interaction updating unit and a parameter adjusting unit; the interaction updating unit keeps the synchronism of the model on all devices through a distributed model updating strategy; the distributed model updating strategy realizes updating model parameters on each device through a federal learning method and aggregates the updated parameters; the parameter adjustment unit allows the model to perform partial parameter adjustment after receiving new data through an incremental learning method; the adaptive diagnosis module realizes that the model adjusts parameters according to the feedback information of the equipment in real time through an online learning and feedback mechanism.
As a further technical scheme of the invention, the active learning method based on model uncertainty comprises the following working steps:
s1, initializing and training a model;
Performing parameter training according to historical data through a neural network model, and establishing an initial fault diagnosis model;
S2, evaluating uncertainty of the model;
based on a Bayesian neural network method, evaluating model parameter uncertainty through posterior distribution of parameters so as to quantify the prediction uncertainty of the model in the current state;
S3, detecting an abnormal sample;
Screening out abnormal samples through local outlier factors according to the prediction uncertainty of the model on training data;
S4, active sample selection;
based on the information quantity and the representative index of the abnormal sample, selecting a sample with important significance for model optimization by an uncertainty sampling method;
s5, sample weighting and model updating;
Updating model parameters in a weighting mode according to the actively selected samples;
s6, evaluating model performance;
testing the updated model through verification set data, evaluating the performance of the model in terms of diagnosis accuracy and stability, and providing feedback and guidance for the next iteration;
S7, a feedback mechanism and iterative updating;
and monitoring the model performance through a feedback mechanism according to the performance evaluation result, adjusting an active learning strategy, and iteratively updating model parameters.
As a further technical scheme of the invention, the self-organizing map network model based on reinforcement learning comprises a data preprocessing module, a self-organizing map network module, a reinforcement learning module and a real-time alarm and fault processing module; the data preprocessing module preprocesses input data through data cleaning, feature extraction and normalization operation; the self-organizing mapping network module maps input data into a neural network with a low-dimensional topological structure through an input layer, a competition layer, a neighborhood function unit and a weight updating unit; the input layer transmits the preprocessed data into the neural network for processing through the forward propagation of the neural network; the competition layer clusters the input through a competition learning mechanism; the competition learning mechanism calculates the matching degree of each unit and input data through Euclidean distance; the neighborhood function unit determines the neighborhood range of each unit in the competition layer through a Gaussian function; according to the input data and the neighborhood function, the weight updating unit updates the weight of the unit in the competition layer through a clustering network rule; the reinforcement learning module realizes real-time alarm decision and fault processing strategy generation through a state unit, an action unit, a reward unit and a strategy unit; the state unit defines a system state through deep reinforcement learning; the action unit defines executable actions through a state value function and an action value function; the rewarding unit adjusts a model learning process through a rewarding function; based on the reward signal and the current state, the strategy unit learns and generates an optimal alarm decision and fault processing strategy through a strategy updating mechanism; and the real-time alarming and fault processing module is used for realizing real-time alarming and fault processing of the electric energy metering detection information through a rule engine according to the alarming decision and fault processing strategy generated by the reinforcement learning module.
Has the positive beneficial effects that:
The invention introduces the deep learning self-encoder to extract and select the characteristics, automatically extracts key information by a back propagation and gradient descent method, and reduces the dependence on artificial experience. Meanwhile, the complex topological structure and the association relation are processed by adopting the graph convolutional neural network structure, so that the understanding capability of the model to the system state is improved, and the limitation of the traditional method in processing the complex association relation is overcome. In addition, the self-adaptive learning is realized based on the anomaly detection algorithm for generating the countermeasure network and the reinforcement learning, potential faults can be detected in real time, the model can be dynamically adjusted, and the adaptability and the instantaneity of the system are improved. By introducing an edge computing technology and a distributed processing model, real-time data processing and fault diagnosis at the equipment end are realized, and the real-time performance and efficiency of the system are improved. The data from different sensors are fused through the fusion diagnosis system, and the association relation between the data is mined by using association analysis and time sequence analysis methods, so that the accuracy and the comprehensiveness of comprehensive diagnosis are improved. And the meta learning model is introduced to realize data sharing among a plurality of devices and model collaborative learning, so that the generalization capability and the overall performance of the model are promoted. The active learning method based on the model uncertainty can optimize and iteratively update the model in real time so as to adapt to the dynamic change of the running state of the system, and improve the stability and reliability of the system. Finally, the self-organizing map network model of reinforcement learning is used for realizing real-time alarm decision and fault processing strategy generation, so that the response speed and accuracy of the system to abnormal conditions are improved.
Drawings
For a clearer description of embodiments of the invention or of solutions in the prior art, the drawings that are necessary for the description of the embodiments or of the prior art will be briefly described, it being apparent that the drawings in the description below are only some embodiments of the invention, from which, without inventive faculty, other drawings can be obtained for a person skilled in the art, in which:
FIG. 1 is a schematic diagram of the steps in the process of the present invention;
FIG. 2 is a step diagram of the operation of the anomaly detection algorithm based on generation of a challenge network and reinforcement learning of the present invention;
FIG. 3 is a schematic block diagram of a meta learning model of the present invention;
FIG. 4 is a schematic diagram of the operation of the reinforcement learning-based self-organizing map network model of the present invention;
fig. 5 is a schematic step diagram of the active learning method based on model uncertainty of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1 to 5, a fault diagnosis method for electric energy metering detection information includes the following steps:
step one, data acquisition and pretreatment;
Collecting electric energy metering data through an electric energy meter and sensor equipment, and performing data cleaning, denoising and outlier processing through a signal processing method;
Step two, extracting and selecting characteristics;
extracting data characteristics of the electric energy metering equipment through a deep learning self-encoder, wherein the deep learning self-encoder reduces data dimensionality and retains key information through a back propagation and gradient descent method;
step three, introducing a deep learning model;
complex topological structures and association relations in the electric energy metering system are processed through a graph convolution neural network structure, and the understanding capability of a model to the system state is improved through a mode of node representation learning, information transmission and local global feature processing;
Step four, abnormality detection and self-adaptive learning;
Dynamic detection and self-adaptive learning of potential faults in the electric energy metering system are realized through an anomaly detection algorithm based on generation of an countermeasure network and reinforcement learning;
Fifthly, fusing an edge computing technology;
Real-time data processing and fault diagnosis at the equipment end are realized through a distributed processing model;
step six, multi-source data fusion and association analysis;
The data from different sensors are fused through a fusion diagnosis system, and the fusion diagnosis system excavates the association relation between different data sources through association analysis and a time sequence analysis method;
Seventh, cross-equipment collaborative learning;
Data sharing and model collaborative learning among a plurality of electric energy metering devices are realized through a meta-learning model;
step eight, model optimization and iterative updating;
Introducing an active learning method based on model uncertainty to optimize and iteratively update the model in real time so as to adapt to the dynamic change of the running state of the electric energy metering system;
Step nine, alarming and fault processing in real time;
And realizing real-time alarm decision and fault processing strategy generation through a self-organizing map network model based on reinforcement learning.
In the above embodiment, the depth learning self-encoder includes an encoder module, a decoder module, a feature extraction module, and a dimension reduction and reconstruction module; the encoder module dynamically adjusts the learning rate through the adaptive moment estimation optimizer; the adaptive moment estimation optimizer improves model convergence speed and generalization capability through a convolution layer and a pooling layer; the convolution layer realizes feature extraction and nonlinear mapping of input data through convolution operation and an activation function; the pooling layer realizes feature dimension reduction and main information retention through average pooling operation, and reduces the calculated amount at the same time; the decoder module realizes jump connection through a residual connection method, and the residual connection method avoids the problem of gradient disappearance through a transposed convolution layer and a batch normalization layer; the transpose convolution layer realizes the reconstruction and recovery of data through deconvolution operation; the characteristic extraction module is used for effectively extracting and characterizing the data characteristics of the electric energy metering equipment through a convolution self-encoder structure; the convolution self-encoder structure improves the nonlinear fitting capacity of the features through a sparse coding and sparse punishment mechanism; the dimension reduction and reconstruction module realizes sparse representation and dimension reduction compression of data through a sparse self-encoder structure and an L1 regularization method.
In a specific embodiment, in the fault diagnosis method of the electric energy metering detection information, the working flow of the deep learning self-encoder is as follows: and inputting original electric energy metering detection data. The data is mapped to the hidden layer representation via the encoder module, extracting high-order features of the data. The decoder module decodes the hidden layer representation as reconstructed data to recover as much of the original input as possible. The feature extraction module learns abstract feature representation of data, and the representation capability of the model is improved. The dimension reduction and reconstruction module realizes dimension reduction processing of data, and simultaneously retains main information and filters noise. Finally, abnormal detection and fault diagnosis are carried out by utilizing the learned characteristic representation, so that the system is helped to find potential problems, manual intervention is reduced, and diagnosis efficiency is improved.
The encoder receives the input data and maps it to a hidden layer representation through a multi-layer neural network. The nodes of each hidden layer are activated using a nonlinear activation function to capture complex features of the data. The decoder receives the output of the encoder and reconstructs the original input data through a back-propagation algorithm. The goal of the decoder is to minimize the reconstruction error so that the reconstructed data is as close as possible to the original input. During training, the self-encoder learns the high-order feature representation of the data by minimizing reconstruction errors. These features enable capturing abstract information of the data, helping to improve the model's ability to characterize the data. The dimension reduction and reconstruction module maps the data into a low-dimension space through the encoder, so that dimension reduction processing of the data is realized. In the decoder module, the low-dimensional representation is reconstructed into the original data, retaining the main information of the data, while filtering out part of the noise.
In the fault diagnosis method of the electric energy metering detection information, the self-encoder can automatically learn abstract feature representation of data, a feature extractor is not required to be designed manually, and the representation capability of a model is improved. In addition, the self-encoder can map the data into a low-dimensional space to realize the dimension reduction processing of the data, and meanwhile, noise in the data can be filtered in the reconstruction process, so that the quality of the data is improved. Secondly, based on the data representation learned from the encoder, anomalies or fault conditions in the data can be effectively detected, helping the diagnostic system to find potential problems. Meanwhile, the self-encoder can learn potential distribution rules of data, has certain generalization capability, and is suitable for different types of data and fault conditions. Finally, through the deep learning self-encoder, the requirement of manual intervention can be reduced, an automatic fault diagnosis process is realized, and the working efficiency and the accuracy are improved.
In the above embodiment, the graph roll-up neural network structure includes an input layer, a graph roll-up layer, a pooling layer, and a full connection layer; the graph convolution neural network structure represents node characteristic data and an edge connection relation in the electric energy metering system as graph structure input through the input layer, and realizes characteristic extraction and information transmission of the graph structure data through the graph convolution layer; the picture volume lamination comprises a neighbor aggregation unit, a feature conversion unit, an activation function unit and a picture volume lamination unit; the neighbor aggregation unit is used for realizing the aggregation and updating of node characteristics by weighting the characteristic information of adjacent nodes through an adjacent matrix; the feature conversion unit carries out linear transformation on the aggregated neighbor node features through weight parameters so as to obtain higher-level feature representation; the activation function unit carries out nonlinear mapping on the characteristics through a linear rectification function, so that the network learning capacity is enhanced; the graph rolling unit captures the association relation between nodes through graph rolling operation; the graph convolution neural network structure gathers the characteristics of different nodes in the graph through the pooling layer to obtain the characteristic representation of the whole graph, and the pooled graph characteristics are input into the full-connection neural network through the full-connection layer to output a final fault diagnosis result.
In particular embodiments, a graph roll-up neural network (GCN) enhances model understanding of system states by way of node representation learning, information propagation, and local global feature processing. In the fault diagnosis method of the electric energy metering detection information, the GCN models elements such as sensor nodes, measuring equipment and the like in an electric energy metering system in a graph mode through graph representation modeling, wherein the nodes represent entities in the system, and edges represent connection relations or interactions among the nodes. The GCN captures the characteristics of the nodes by learning the node's representation, updating each node's representation with the node's neighbor node information. In this way, each node can obtain a rich feature representation by aggregating the information of surrounding nodes. Information transmission is carried out in the graph, and after the characteristics of the nodes are transmitted through multiple rounds of information, richer local and global information can be obtained. This helps to understand the interaction relationships between nodes and the overall system state. In addition, the GCN can simultaneously consider the local neighborhood and the whole graph structure to process the characteristics of the nodes, so that more comprehensive local and global characteristic representations are obtained, and the understanding and analysis of the system state are facilitated.
In the fault diagnosis method of the electric energy metering detection information, an electric energy metering system usually has a complex topological structure, and the GCN can effectively process the complex node association relationship, so that the understanding capability of a model on the system structure is improved. In addition, the GCN can propagate information in the graph structure and comprehensively consider global information, so that the model can better understand the state of the whole system, and is not limited to local information. Secondly, for a dynamic change electric energy metering system, the GCN can perform dynamic feature learning and information propagation according to the updated graph structure and node information, and has certain robustness and adaptability. Meanwhile, the GCN can automatically learn the characteristic representation of the node, so that the requirement on the characteristic representation of the manual design is reduced, and the process of characteristic engineering is simplified.
In the above embodiment, the anomaly detection algorithm based on the generation of the antagonism network and reinforcement learning trains the generation of the antagonism network by generating the objective function of the antagonism network, so that the samples generated by the generator are more similar to the real data distribution, and the discriminator is more good at distinguishing the real data from the generated data; the formula for generating the objective function of the countermeasure network is:
in formula (1), R is a discriminator parameter for discriminating between a real data sample and a dummy sample generated by the generator; y is a real data sample for training the discriminator in training; m is real data distribution, and is used for generating distribution characteristics of a real sample; q is a noise sample, used to generate a false sample; p is the noise distribution, which represents the distribution of noise samples; then, calculating the loss in the classification task through a classification loss function so as to predict a sample label; the formula expression of the classification loss function is as follows:
in the formula (2), z is a real tag, and is used for calculating a tag value of real data; u is a prediction label and is used for calculating a classification predicted value of the model on the data; beta is the number of samples; reconstructing the data sample by the generator to determine a difference between the data sample and the real data; the anomaly detection algorithm calculates the difference between the real data sample and the reconstructed data sample through the reconstructed error loss function; the formula expression of the reconstruction error loss function is as follows:
In the formula (3), c is a real data sample, and is used for calculating the original characteristics of the real data sample; b is a reconstructed data sample generated for evaluating differences between the reconstructed data sample and the real data; θ is a weight adjustment amount for updating model parameters; the change rate of the error to the model parameters is represented and used for guiding the parameter updating direction; discount-accumulating rewards of future time steps through a cumulative reward function based on a reinforcement learning algorithm to evaluate long-term rewards of the current strategy; the formula expression of the jackpot function based on the reinforcement learning algorithm is as follows:
In equation (4), E a is the jackpot at time a, used to measure the sum of the jackpots accumulated by the model at time a during training, and w i is the real-time prize of step i, used to measure the instant prize value obtained by the model at each step; k is a termination time for specifying a termination time of training; finally, calculating the expected return of taking a certain action in the current state by using an action cost function based on a reinforcement learning algorithm, and updating a decision rule by combining the maximum action value at the next moment; the formula expression of the action cost function based on the reinforcement learning algorithm is as follows:
in equation (5), F is taking action in state l For evaluating the action value; s is action taken in state l/>The instant rewards are used for measuring model instant rewards; q is a discount factor for balancing the importance of the current and future rewards.
In particular embodiments, an efficient, dynamic fault detection mechanism may be provided in conjunction with generating an anomaly detection algorithm against the network (GANs) and reinforcement learning. Generating an antagonism network (GANs): GAN consists of two parts, a Generator (Generator) and a discriminator (Discriminator). The goal of the generator is to generate dummy data that is as close as possible to the real data, while the goal of the arbiter is to distinguish whether the input data is real or generated by the generator. Through this countermeasure process, the generator learns to generate more and more realistic data. Reinforcement learning: reinforcement learning is an algorithm that lets a machine learn a specific task by trial and error, which guides the learning process by taking feedback (rewards or penalties) from the environment. In anomaly detection, reinforcement learning can be used to dynamically adjust parameters of the GAN network to optimize anomaly detection performance.
In the fault diagnosis method of the electric energy metering detection information, the implementation process of the abnormality detection algorithm based on the generation of the countermeasure network and reinforcement learning is as follows:
p1, data preparation: and collecting electric energy metering data in a normal running state as training data.
P2, training GAN: and training GAN by using normal data, so that a generator learns to generate data samples in a normal state, and a discriminator learns to distinguish real data from generated data.
P3, reinforcement learning optimization: and dynamically adjusting the parameters of the GAN by using a reinforcement learning algorithm, and optimizing the abnormality detection capability of the discriminator. When the abnormal data are correctly identified by the discriminator, giving positive rewards; otherwise, a penalty is given.
P4, abnormality detection: in practical application, real-time data of the electric energy metering system is input into a trained discriminator, and if the discriminator considers that the data has large difference from normal data in training, the data may indicate that abnormality exists.
In the fault diagnosis method of the electric energy metering detection information, the GAN can generate samples very close to real data through countermeasure training, so that the accuracy of anomaly detection is improved. In addition, reinforcement learning enables the model to continuously adjust strategies according to feedback, and adapt to dynamic changes of the running state of the electric energy metering system. Secondly, through accurate data generation and optimized detection strategies, the possibility of false alarm is reduced.
In the electric energy meter detection information fault diagnosis method, the data test comparison table based on the anomaly detection algorithm for generating the countermeasure network and reinforcement learning is shown in table 1:
Table 1 data test comparison table
As can be seen from the comparison table, the anomaly detection algorithm based on the generation of the antagonism network and reinforcement learning is superior to the conventional algorithm in terms of accuracy, recall rate, F1 score, accuracy, AUC value, false positive rate, false negative rate and area under the ROC curve, and has a lower false alarm rate. From these metrics, the performance differences of the two algorithms can be more fully evaluated. GAN and reinforcement learning based algorithms represent 0.92 in accuracy, which is higher than 0.85 of the conventional algorithm. This shows that GAN and reinforcement learning based algorithms can more accurately identify true outlier data, reducing the likelihood of false positives. GAN and reinforcement learning based algorithms appear to be 0.89 in recall, slightly higher than 0.78 of the traditional algorithm. This means that the GAN and reinforcement learning based algorithm can better capture the proportion of real anomaly data, reducing the false negative. The F1 score of GAN and reinforcement learning based algorithms is 0.90, significantly higher than 0.81 for the conventional algorithm. This suggests that GAN and reinforcement learning based algorithms achieve a better balance between accuracy and recall, and are a more efficient method of anomaly detection. The accuracy of GAN and reinforcement learning based algorithms is 0.91, which is higher than 0.84 of conventional algorithms. This illustrates that GAN and reinforcement learning based algorithms perform better in overall prediction accuracy. The AUC value for GAN and reinforcement learning based algorithms is 0.93, which is higher than 0.88 for conventional algorithms. AUC values measure the overall performance of the model at different thresholds, and GAN and reinforcement learning based algorithms perform better over the area under the curve. The false positive rate of the algorithm based on GAN and reinforcement learning is 0.08, which is obviously lower than 0.15 of the traditional algorithm. This means that GAN and reinforcement learning based algorithms can label normal data as anomalies less erroneously. The false negative rate of the algorithm based on GAN and reinforcement learning is 0.11, which is improved compared with 0.22 of the traditional algorithm. This shows that GAN and reinforcement learning based algorithms can reduce the false negative, better capture the true outlier. The area under the ROC curve of the GAN and reinforcement learning based algorithm is 0.92, which is higher than 0.87 of the conventional algorithm. This suggests that GAN and reinforcement learning based algorithms have better overall performance at different classification thresholds.
In the above embodiment, the distributed processing model includes a data receiving module, a feature extraction and selection module, a distributed computing module, and a real-time fault diagnosis module; the data receiving module acquires various parameter data of the electric energy metering detection equipment in real time through a data communication interface; the feature extraction and selection module comprises a feature extraction unit and a feature selection unit; the feature extraction unit extracts frequency domain features and time domain feature representations from the preprocessed data through a harmonic analysis and wavelet transformation method; the feature selection unit screens representative and distinguishing feature subsets through correlation analysis and information gain; the distributed computing module comprises edge computing nodes, a model training unit and a model aggregation unit; the edge computing node realizes data processing and model training tasks of the equipment end through a distributed computing and containerization method; the model training unit divides the local data into local data parts and locally trains the model through synchronous gradient descent and asynchronous optimization methods, and then realizes global model updating through the model aggregation unit; the model aggregation unit integrates model parameters obtained by training on each edge computing node through a weighted average and federal learning aggregation method to obtain a global model; the real-time fault diagnosis module comprises a fault diagnosis unit and a fault judgment unit; based on the global model and the real-time data, the fault diagnosis unit judges the fault type and degree through a support vector machine; and according to the fault diagnosis result, the fault judging unit adopts a decision tree and a rule engine to judge whether the equipment has faults or not, and outputs processing advice or alarm information.
In particular embodiments, the distributed processing model collects power metering data from the power meters and the various sensor devices via the data receiving module. The data may include information in multiple dimensions of voltage, current, power, etc. Preprocessing (such as denoising, normalization and the like) is performed on the collected data through a feature extraction and selection module, and key features of the data are extracted through a machine learning algorithm or a deep learning method. The purpose of this step is to identify the most useful information for fault diagnosis and to reduce the amount of data to be processed later. And distributing the data after feature extraction to a network consisting of a plurality of computing nodes through a distributed computing module for parallel processing. Each node may independently process a portion of the data and then aggregate the results to obtain a final diagnostic result. Based on the distributed calculation results, potential faults in the electrical energy metering system are diagnosed in real time by a real-time fault diagnosis module. This may be accomplished by comparing current data to historical data, using a pre-trained machine learning model to identify abnormal patterns, and the like.
In the fault diagnosis method of the electric energy metering detection information, the distributed processing model realizes the parallelization of data processing by dispersing the data processing tasks to a plurality of computing nodes. This means that different nodes can process different parts of data at the same time, thus greatly improving the speed and efficiency of data processing. In addition, the model can dynamically adjust the computing resources according to the needs so as to adapt to different data processing requirements.
In the fault diagnosis method of the electric energy metering detection information, the data processing speed is remarkably improved by the distributed processing model through parallel processing, so that real-time processing of large-scale electric energy metering data is possible. In addition, the distributed architecture increases the redundancy of the system, and even if one node fails, other nodes can still continue to process tasks, thereby improving the overall reliability of the system. Second, the distributed processing model is easy to expand, and computing nodes can be increased or decreased according to data processing requirements, so that the increase of the electric energy metering data quantity or the increase of the complexity are effectively improved. Meanwhile, potential faults in the system can be timely found and diagnosed by the distributed processing model through real-time processing and analysis of the electric energy metering data, and loss and influence caused by the faults are reduced.
In the above embodiment, the fusion diagnosis system includes a data fusion module, a correlation analysis module, and a comprehensive diagnosis and decision module; the data fusion module comprises a sensor integration unit and a feature recognition unit; the sensor integration unit performs fusion processing on the data of different sensors through a weighted average and feature fusion method; the feature recognition unit extracts effective feature information from the fused data through a principal component analysis and wavelet transformation method; the association analysis module comprises a time sequence analysis unit and an association mining unit; the time sequence analysis unit is used for excavating time correlation and trend among different data sources through a time sequence model, an autoregressive model and a sliding window; the association mining unit discovers association rules and potential association factors between data through association rule mining, frequent pattern mining and cluster analysis methods; the comprehensive diagnosis and decision module comprises a comprehensive diagnosis unit and a decision support unit; combining the correlation analysis result and the feature extraction information, and the comprehensive diagnosis unit recognizes abnormal conditions in the system through a deep learning model and gives a diagnosis result; based on the comprehensive diagnosis result, the decision support unit provides decision support through a decision tree, an expert system and a fuzzy logic method.
In particular embodiments, the fusion diagnostic system integrates, fuses data from different sensors, devices, or systems through a data fusion module to provide more comprehensive, accurate data information. Data fusion may include data cleansing, alignment, normalization, etc. processes to ensure consistency and reliability of the data. And the association analysis module discovers potential association between the data by carrying out association analysis on the fused data and exploring the association and rule between each data. This helps identify potential failure modes or anomalies. The comprehensive diagnosis and decision module performs comprehensive diagnosis and decision based on the data fusion and the correlation analysis result. The module can be combined with expert knowledge, a rule base or a machine learning algorithm to comprehensively evaluate the system state and give out corresponding diagnosis results and decision suggestions.
In the fault diagnosis method of the electric energy metering detection information, the fusion diagnosis system can integrate information of different data sources, fully utilize diversified data, improve the comprehensive understanding capability of the system state and help to find potential faults and problems. In addition, through association analysis and comprehensive diagnosis, the system can more accurately judge the health state of the system and possible problems, reduce the false alarm rate and the missing report rate and improve the accuracy of diagnosis. And secondly, the comprehensive diagnosis and decision-making module can quickly make decisions according to the diagnosis results and the rule base and provide the decisions for operators to refer to, so that the operators can quickly take corresponding measures, and the fault processing time is reduced. Meanwhile, the automatic characteristic of the fusion diagnosis system can reduce the need of human intervention, reduce the influence of subjective factors on diagnosis results, and improve the stability and reliability of the system.
In the above embodiment, the meta learning model includes a model construction module, a multi-device data sharing module, a model collaborative learning module, and an adaptive diagnosis module; the model construction module comprises a model structure design unit and a learning strategy design unit; the model structure design unit realizes the rapid learning capacity of the model on a new task through a parameter sharing mechanism in the neural network; the learning strategy design unit performs feature fusion by combining data of a source field and a target field through a field self-adaptive method; the field self-adaptive method realizes the effective utilization of cross-field data and the improvement of model generalization capability through a feature alignment and mapping learning method; the multi-device data sharing module comprises a communication unit and an integration unit; the communication unit encrypts and transmits data through SSL/TLS encryption protocol, and ensures the security of data transmission through an identity authentication mechanism; the integration unit fuses different device data to generate a new sample through an countermeasure generation network; the model collaborative learning module comprises an interaction updating unit and a parameter adjusting unit; the interaction updating unit keeps the synchronism of the model on all devices through a distributed model updating strategy; the distributed model updating strategy realizes updating model parameters on each device through a federal learning method and aggregates the updated parameters; the parameter adjustment unit allows the model to perform partial parameter adjustment after receiving new data through an incremental learning method; the adaptive diagnosis module realizes that the model adjusts parameters according to the feedback information of the equipment in real time through an online learning and feedback mechanism.
In a specific embodiment, the meta learning model constructs a basic structure of the meta learning model through a model construction module, wherein the basic structure comprises a selected algorithm, a neural network architecture and the like. The data sharing among different devices is realized through the multi-device data sharing module, the data of a plurality of devices are integrated into a unified platform, and a more comprehensive data set is provided for training a meta-learning model. The model collaborative learning module can mutually exchange learned knowledge and experience through collaborative learning, so that the overall diagnosis performance is improved. And dynamically adjusting parameters and structures of the meta-learning model by the adaptive diagnosis module according to the data conditions monitored in real time so as to adapt to fault diagnosis requirements under different environments.
In a specific embodiment, the principle of the working mode of the meta-learning model is that the meta-learning model is trained through a large amount of data input, and the model learns the relevance among various fault characteristics in the training process, so that possible faults of the electric energy metering equipment can be accurately identified and diagnosed.
In the fault diagnosis method of the electric energy metering detection information, the meta-learning model improves the accuracy and efficiency of fault diagnosis, and can discover and locate problems more quickly. And the intellectualization and the self-adaptability of the system are enhanced, and fault conditions of different types and complexity can be dealt with. Meanwhile, through multi-device data sharing and model collaborative learning, complementation of knowledge and experience can be realized, and the overall diagnosis performance is improved. In addition, the adaptive diagnosis module can adjust model parameters in time according to actual conditions, and high efficiency and accuracy of diagnosis are maintained.
In the above embodiment, the working steps of the active learning method based on model uncertainty are as follows:
s1, initializing and training a model;
Performing parameter training according to historical data through a neural network model, and establishing an initial fault diagnosis model;
S2, evaluating uncertainty of the model;
based on a Bayesian neural network method, evaluating model parameter uncertainty through posterior distribution of parameters so as to quantify the prediction uncertainty of the model in the current state;
S3, detecting an abnormal sample;
Screening out abnormal samples through local outlier factors according to the prediction uncertainty of the model on training data;
S4, active sample selection;
based on the information quantity and the representative index of the abnormal sample, selecting a sample with important significance for model optimization by an uncertainty sampling method;
s5, sample weighting and model updating;
Updating model parameters in a weighting mode according to the actively selected samples;
s6, evaluating model performance;
testing the updated model through verification set data, evaluating the performance of the model in terms of diagnosis accuracy and stability, and providing feedback and guidance for the next iteration;
S7, a feedback mechanism and iterative updating;
and monitoring the model performance through a feedback mechanism according to the performance evaluation result, adjusting an active learning strategy, and iteratively updating model parameters.
In a specific embodiment, first, an active learning method based on model uncertainty predicts collected electric energy metering data using a current fault diagnosis model, and evaluates the uncertainty of model prediction. Such uncertainty can be measured in a number of ways, such as confidence interval width of the predicted outcome, differences in the predicted outcome of different models, etc. Based on the evaluation of the model uncertainty, the active learning algorithm will select those data points for which the model is least deterministic (i.e., the data points for which the model is most difficult to make accurate predictions) as query objects, requesting manual labeling or expert intervention. Selected data points are labeled by a human or expert to provide accurate fault category or status information. Next, the newly labeled data points are added to the training set, and the fault diagnosis model is retrained or trimmed, thereby updating the model. Finally, repeating the steps, continuously monitoring the uncertainty of the model, continuously selecting new data according to the change of the running state of the system, marking and updating the model, and realizing the real-time optimization and iterative updating of the model.
In the fault diagnosis method of the electric energy metering detection information, the data points which are the most uncertain in the model are focused for learning, and the active learning method can effectively improve the recognition capability of the model to complex or rare fault modes, so that the overall diagnosis accuracy is improved. In addition, compared with the traditional passive learning method (namely randomly selecting data for marking and training), the active learning reduces the requirement for a large amount of marked data through selective marking and learning, and improves the data utilization efficiency. And secondly, the active learning method can continuously adjust learning key points and directions according to the dynamic change of the running state of the electric energy metering system, so that the model always maintains high adaptability to the current system state. Meanwhile, the data with the highest information value is selected to be marked manually, so that the data quantity required by marking is reduced, and the cost and time of manual marking are reduced. Finally, by constantly learning new, unknown or uncertain data, the generalization ability of the model is enhanced, enabling better handling of unseen fault conditions.
In the above embodiment, the self-organizing map network model based on reinforcement learning includes a data preprocessing module, a self-organizing map network module, a reinforcement learning module, and a real-time alarm and fault processing module; the data preprocessing module preprocesses input data through data cleaning, feature extraction and normalization operation; the self-organizing mapping network module maps input data into a neural network with a low-dimensional topological structure through an input layer, a competition layer, a neighborhood function unit and a weight updating unit; the input layer transmits the preprocessed data into the neural network for processing through the forward propagation of the neural network; the competition layer clusters the input through a competition learning mechanism; the competition learning mechanism calculates the matching degree of each unit and input data through Euclidean distance; the neighborhood function unit determines the neighborhood range of each unit in the competition layer through a Gaussian function; according to the input data and the neighborhood function, the weight updating unit updates the weight of the unit in the competition layer through a clustering network rule; the reinforcement learning module realizes real-time alarm decision and fault processing strategy generation through a state unit, an action unit, a reward unit and a strategy unit; the state unit defines a system state through deep reinforcement learning; the action unit defines executable actions through a state value function and an action value function; the rewarding unit adjusts a model learning process through a rewarding function; based on the reward signal and the current state, the strategy unit learns and generates an optimal alarm decision and fault processing strategy through a strategy updating mechanism; and the real-time alarming and fault processing module is used for realizing real-time alarming and fault processing of the electric energy metering detection information through a rule engine according to the alarming decision and fault processing strategy generated by the reinforcement learning module.
In a specific embodiment, the self-organizing map network model based on reinforcement learning performs preprocessing operations such as cleaning, normalization and the like on the original data through a data preprocessing module so as to input the data into the self-organizing map network module for training. The self-organizing map network module is a neural network model with self-learning and self-adapting capabilities for mapping input data to nodes in a two-dimensional space. The method enables similar input data to be mapped to adjacent nodes by adjusting the connection weight between the nodes in the training process. The reinforcement learning module is used for optimizing the performance of the self-organizing map network. The method adopts reinforcement learning algorithm, such as Q-learning or deep reinforcement learning, and continuously updates parameters and strategies of the network by interacting and feeding back with the environment so as to realize better mapping and classification effects. The real-time alarm and fault processing module is used for monitoring the system state and carrying out real-time alarm and fault processing according to the output result of the self-organizing map network model. When an abnormal or fault condition is detected, the module triggers a corresponding alarm mechanism and takes necessary measures to carry out fault diagnosis and treatment.
In the fault diagnosis method of the electric energy metering detection information, the self-organizing map network model based on reinforcement learning can effectively discover the internal structure and characteristic distribution of data through the self-organizing map network module, so that the running state of the system can be better understood. In addition, the reinforcement learning module can perform state evaluation and decision making according to the real-time data, and intelligent monitoring and fault diagnosis of the complex system are realized. The real-time alarm and fault processing module can timely respond to potential fault conditions, reduce loss caused by faults and improve the reliability and stability of the system.
While specific embodiments of the present invention have been described above, it will be understood by those skilled in the art that these specific embodiments are by way of example only, and that various omissions, substitutions, and changes in the form and details of the methods and systems described above may be made by those skilled in the art without departing from the spirit and scope of the invention. For example, it is within the scope of the present invention to combine the above-described method steps to perform substantially the same function in substantially the same way to achieve substantially the same result. Accordingly, the scope of the invention is limited only by the following claims.
Claims (9)
1. A fault diagnosis method for electric energy metering detection information is characterized in that: the method comprises the following steps:
step one, data acquisition and pretreatment;
Collecting electric energy metering data through an electric energy meter and sensor equipment, and performing data cleaning, denoising and outlier processing through a signal processing method;
Step two, extracting and selecting characteristics;
extracting data characteristics of the electric energy metering equipment through a deep learning self-encoder, wherein the deep learning self-encoder reduces data dimensionality and retains key information through a back propagation and gradient descent method;
step three, introducing a deep learning model;
complex topological structures and association relations in the electric energy metering system are processed through a graph convolution neural network structure, and the understanding capability of a model to the system state is improved through a mode of node representation learning, information transmission and local global feature processing;
Step four, abnormality detection and self-adaptive learning;
Dynamic detection and self-adaptive learning of potential faults in the electric energy metering system are realized through an anomaly detection algorithm based on generation of an countermeasure network and reinforcement learning;
Fifthly, fusing an edge computing technology;
Real-time data processing and fault diagnosis at the equipment end are realized through a distributed processing model;
step six, multi-source data fusion and association analysis;
The data from different sensors are fused through a fusion diagnosis system, and the fusion diagnosis system excavates the association relation between different data sources through association analysis and a time sequence analysis method;
Seventh, cross-equipment collaborative learning;
Data sharing and model collaborative learning among a plurality of electric energy metering devices are realized through a meta-learning model;
step eight, model optimization and iterative updating;
Introducing an active learning method based on model uncertainty to optimize and iteratively update the model in real time so as to adapt to the dynamic change of the running state of the electric energy metering system;
Step nine, alarming and fault processing in real time;
And realizing real-time alarm decision and fault processing strategy generation through a self-organizing map network model based on reinforcement learning.
2. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the deep learning self-encoder comprises an encoder module, a decoder module, a feature extraction module and a dimension reduction and reconstruction module; the encoder module dynamically adjusts the learning rate through the adaptive moment estimation optimizer; the adaptive moment estimation optimizer improves model convergence speed and generalization capability through a convolution layer and a pooling layer; the convolution layer realizes feature extraction and nonlinear mapping of input data through convolution operation and an activation function; the pooling layer realizes feature dimension reduction and main information retention through average pooling operation, and reduces the calculated amount at the same time; the decoder module realizes jump connection through a residual connection method, and the residual connection method avoids the problem of gradient disappearance through a transposed convolution layer and a batch normalization layer; the transpose convolution layer realizes the reconstruction and recovery of data through deconvolution operation; the characteristic extraction module is used for effectively extracting and characterizing the data characteristics of the electric energy metering equipment through a convolution self-encoder structure; the convolution self-encoder structure improves the nonlinear fitting capacity of the features through a sparse coding and sparse punishment mechanism; the dimension reduction and reconstruction module realizes sparse representation and dimension reduction compression of data through a sparse self-encoder structure and an L1 regularization method.
3. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the graph roll-up neural network structure comprises an input layer, a graph roll-up layer, a pooling layer and a full-connection layer; the graph convolution neural network structure represents node characteristic data and an edge connection relation in the electric energy metering system as graph structure input through the input layer, and realizes characteristic extraction and information transmission of the graph structure data through the graph convolution layer; the picture volume lamination comprises a neighbor aggregation unit, a feature conversion unit, an activation function unit and a picture volume lamination unit; the neighbor aggregation unit is used for realizing the aggregation and updating of node characteristics by weighting the characteristic information of adjacent nodes through an adjacent matrix; the feature conversion unit carries out linear transformation on the aggregated neighbor node features through weight parameters so as to obtain higher-level feature representation; the activation function unit carries out nonlinear mapping on the characteristics through a linear rectification function, so that the network learning capacity is enhanced; the graph rolling unit captures the association relation between nodes through graph rolling operation; the graph convolution neural network structure gathers the characteristics of different nodes in the graph through the pooling layer to obtain the characteristic representation of the whole graph, and the pooled graph characteristics are input into the full-connection neural network through the full-connection layer to output a final fault diagnosis result.
4. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the anomaly detection algorithm based on the generation of the antagonism network and reinforcement learning trains the generation of the antagonism network through generating an objective function of the antagonism network, so that samples generated by the generator are more similar to real data distribution, and meanwhile, a discriminator is more good at distinguishing real data from generated data; the formula for generating the objective function of the countermeasure network is:
in formula (1), R is a discriminator parameter for discriminating between a real data sample and a dummy sample generated by the generator; y is a real data sample for training the discriminator in training; m is real data distribution, and is used for generating distribution characteristics of a real sample; q is a noise sample, used to generate a false sample; p is the noise distribution, which represents the distribution of noise samples; then, calculating the loss in the classification task through a classification loss function so as to predict a sample label; the formula expression of the classification loss function is as follows:
in the formula (2), z is a real tag, and is used for calculating a tag value of real data; u is a prediction label and is used for calculating a classification predicted value of the model on the data; beta is the number of samples; reconstructing the data sample by the generator to determine a difference between the data sample and the real data; the anomaly detection algorithm calculates the difference between the real data sample and the reconstructed data sample through the reconstructed error loss function; the formula expression of the reconstruction error loss function is as follows:
In the formula (3), c is a real data sample, and is used for calculating the original characteristics of the real data sample; b is a reconstructed data sample generated for evaluating differences between the reconstructed data sample and the real data; θ is a weight adjustment amount for updating model parameters; the change rate of the error to the model parameters is represented and used for guiding the parameter updating direction; discount-accumulating rewards of future time steps through a cumulative reward function based on a reinforcement learning algorithm to evaluate long-term rewards of the current strategy; the formula expression of the jackpot function based on the reinforcement learning algorithm is as follows:
In equation (4), E a is the jackpot at time a, used to measure the sum of the jackpots accumulated by the model at time a during training, and w i is the real-time prize of step i, used to measure the instant prize value obtained by the model at each step; k is a termination time for specifying a termination time of training; finally, calculating the expected return of taking a certain action in the current state by using an action cost function based on a reinforcement learning algorithm, and updating a decision rule by combining the maximum action value at the next moment; the formula expression of the action cost function based on the reinforcement learning algorithm is as follows:
in equation (5), F is taking action in state l For evaluating the action value; s is action taken in state l/>The instant rewards are used for measuring model instant rewards; q is a discount factor for balancing the importance of the current and future rewards.
5. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the distributed processing model comprises a data receiving module, a feature extraction and selection module, a distributed computing module and a real-time fault diagnosis module; the data receiving module acquires various parameter data of the electric energy metering detection equipment in real time through a data communication interface; the feature extraction and selection module comprises a feature extraction unit and a feature selection unit; the feature extraction unit extracts frequency domain features and time domain feature representations from the preprocessed data through a harmonic analysis and wavelet transformation method; the feature selection unit screens representative and distinguishing feature subsets through correlation analysis and information gain; the distributed computing module comprises edge computing nodes, a model training unit and a model aggregation unit; the edge computing node realizes data processing and model training tasks of the equipment end through a distributed computing and containerization method; the model training unit divides the local data into local data parts and locally trains the model through synchronous gradient descent and asynchronous optimization methods, and then realizes global model updating through the model aggregation unit; the model aggregation unit integrates model parameters obtained by training on each edge computing node through a weighted average and federal learning aggregation method to obtain a global model; the real-time fault diagnosis module comprises a fault diagnosis unit and a fault judgment unit; based on the global model and the real-time data, the fault diagnosis unit judges the fault type and degree through a support vector machine; and according to the fault diagnosis result, the fault judging unit adopts a decision tree and a rule engine to judge whether the equipment has faults or not, and outputs processing advice or alarm information.
6. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the fusion diagnosis system comprises a data fusion module, a correlation analysis module and a comprehensive diagnosis and decision module; the data fusion module comprises a sensor integration unit and a feature recognition unit; the sensor integration unit performs fusion processing on the data of different sensors through a weighted average and feature fusion method; the feature recognition unit extracts effective feature information from the fused data through a principal component analysis and wavelet transformation method; the association analysis module comprises a time sequence analysis unit and an association mining unit; the time sequence analysis unit is used for excavating time correlation and trend among different data sources through a time sequence model, an autoregressive model and a sliding window; the association mining unit discovers association rules and potential association factors between data through association rule mining, frequent pattern mining and cluster analysis methods; the comprehensive diagnosis and decision module comprises a comprehensive diagnosis unit and a decision support unit; combining the correlation analysis result and the feature extraction information, and the comprehensive diagnosis unit recognizes abnormal conditions in the system through a deep learning model and gives a diagnosis result; based on the comprehensive diagnosis result, the decision support unit provides decision support through a decision tree, an expert system and a fuzzy logic method.
7. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the meta learning model comprises a model construction module, a multi-device data sharing module, a model collaborative learning module and an adaptive diagnosis module; the model construction module comprises a model structure design unit and a learning strategy design unit; the model structure design unit realizes the rapid learning capacity of the model on a new task through a parameter sharing mechanism in the neural network; the learning strategy design unit performs feature fusion by combining data of a source field and a target field through a field self-adaptive method; the field self-adaptive method realizes the effective utilization of cross-field data and the improvement of model generalization capability through a feature alignment and mapping learning method; the multi-device data sharing module comprises a communication unit and an integration unit; the communication unit encrypts and transmits data through SSL/TLS encryption protocol, and ensures the security of data transmission through an identity authentication mechanism; the integration unit fuses different device data to generate a new sample through an countermeasure generation network; the model collaborative learning module comprises an interaction updating unit and a parameter adjusting unit; the interaction updating unit keeps the synchronism of the model on all devices through a distributed model updating strategy; the distributed model updating strategy realizes updating model parameters on each device through a federal learning method and aggregates the updated parameters; the parameter adjustment unit allows the model to perform partial parameter adjustment after receiving new data through an incremental learning method; the adaptive diagnosis module realizes that the model adjusts parameters according to the feedback information of the equipment in real time through an online learning and feedback mechanism.
8. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the active learning method based on the model uncertainty comprises the following working steps:
s1, initializing and training a model;
Performing parameter training according to historical data through a neural network model, and establishing an initial fault diagnosis model;
S2, evaluating uncertainty of the model;
based on a Bayesian neural network method, evaluating model parameter uncertainty through posterior distribution of parameters so as to quantify the prediction uncertainty of the model in the current state;
S3, detecting an abnormal sample;
Screening out abnormal samples through local outlier factors according to the prediction uncertainty of the model on training data;
S4, active sample selection;
based on the information quantity and the representative index of the abnormal sample, selecting a sample with important significance for model optimization by an uncertainty sampling method;
s5, sample weighting and model updating;
Updating model parameters in a weighting mode according to the actively selected samples;
s6, evaluating model performance;
testing the updated model through verification set data, evaluating the performance of the model in terms of diagnosis accuracy and stability, and providing feedback and guidance for the next iteration;
S7, a feedback mechanism and iterative updating;
and monitoring the model performance through a feedback mechanism according to the performance evaluation result, adjusting an active learning strategy, and iteratively updating model parameters.
9. The power metering detection information fault diagnosis method according to claim 1, characterized in that: the self-organizing map network model based on reinforcement learning comprises a data preprocessing module, a self-organizing map network module, a reinforcement learning module and a real-time alarm and fault processing module; the data preprocessing module preprocesses input data through data cleaning, feature extraction and normalization operation; the self-organizing mapping network module maps input data into a neural network with a low-dimensional topological structure through an input layer, a competition layer, a neighborhood function unit and a weight updating unit; the input layer transmits the preprocessed data into the neural network for processing through the forward propagation of the neural network; the competition layer clusters the input through a competition learning mechanism; the competition learning mechanism calculates the matching degree of each unit and input data through Euclidean distance; the neighborhood function unit determines the neighborhood range of each unit in the competition layer through a Gaussian function; according to the input data and the neighborhood function, the weight updating unit updates the weight of the unit in the competition layer through a clustering network rule; the reinforcement learning module realizes real-time alarm decision and fault processing strategy generation through a state unit, an action unit, a reward unit and a strategy unit; the state unit defines a system state through deep reinforcement learning; the action unit defines executable actions through a state value function and an action value function; the rewarding unit adjusts a model learning process through a rewarding function; based on the reward signal and the current state, the strategy unit learns and generates an optimal alarm decision and fault processing strategy through a strategy updating mechanism; and the real-time alarming and fault processing module is used for realizing real-time alarming and fault processing of the electric energy metering detection information through a rule engine according to the alarming decision and fault processing strategy generated by the reinforcement learning module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410328445.6A CN118133203A (en) | 2024-03-21 | 2024-03-21 | Fault diagnosis method for electric energy metering detection information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410328445.6A CN118133203A (en) | 2024-03-21 | 2024-03-21 | Fault diagnosis method for electric energy metering detection information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118133203A true CN118133203A (en) | 2024-06-04 |
Family
ID=91231463
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410328445.6A Pending CN118133203A (en) | 2024-03-21 | 2024-03-21 | Fault diagnosis method for electric energy metering detection information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118133203A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118332291A (en) * | 2024-06-12 | 2024-07-12 | 中国人民解放军海军航空大学 | Multi-sensor data fault prediction method for aircraft |
| CN118400852A (en) * | 2024-06-28 | 2024-07-26 | 勤源(江苏)科技有限公司 | Remote monitoring and fault detection method and system for integrated energy-saving lamp |
| CN118536576A (en) * | 2024-07-25 | 2024-08-23 | 成都携恩科技有限公司 | Deep learning model countermeasure training method based on abnormal perception |
-
2024
- 2024-03-21 CN CN202410328445.6A patent/CN118133203A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118332291A (en) * | 2024-06-12 | 2024-07-12 | 中国人民解放军海军航空大学 | Multi-sensor data fault prediction method for aircraft |
| CN118400852A (en) * | 2024-06-28 | 2024-07-26 | 勤源(江苏)科技有限公司 | Remote monitoring and fault detection method and system for integrated energy-saving lamp |
| CN118536576A (en) * | 2024-07-25 | 2024-08-23 | 成都携恩科技有限公司 | Deep learning model countermeasure training method based on abnormal perception |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jiang et al. | Spatio-temporal meta-graph learning for traffic forecasting | |
| CN111914873B (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN118133203A (en) | Fault diagnosis method for electric energy metering detection information | |
| CN112966714B (en) | Edge time sequence data anomaly detection and network programmable control method | |
| CN113792754B (en) | Converter transformer DGA online monitoring data processing method for firstly removing abnormal state and then repairing | |
| CN109389325B (en) | Method for evaluating state of electronic transformer of transformer substation based on wavelet neural network | |
| CN114048546B (en) | Method for predicting residual service life of aeroengine based on graph convolution network and unsupervised domain self-adaption | |
| CN117035424B (en) | Mine safety situation collaborative sensing system and method based on anti-deduction learning | |
| CN117349595A (en) | Industrial scene-oriented extremely few sample fault diagnosis method | |
| CN115470850A (en) | Water quality abnormal event recognition early warning method based on pipe network water quality time-space data | |
| CN116485029A (en) | Method, device, equipment and medium for detecting accuracy of dynamic data in oilfield development | |
| CN118003961B (en) | Intelligent charging pile group control system and method | |
| CN116520806A (en) | Intelligent fault diagnosis system and method for industrial system | |
| CN113884807B (en) | Power distribution network fault prediction method based on random forest and multi-layer architecture clustering | |
| CN114996110A (en) | Deep inspection optimization method and system based on micro-service architecture | |
| CN114581699A (en) | Transformer state evaluation method based on deep learning model in consideration of multi-source information | |
| CN112613494B (en) | Power line monitoring abnormality identification method and system based on deep countermeasure network | |
| CN117909881A (en) | Fault diagnosis method and device for multi-source data fusion pumping unit | |
| CN117454266A (en) | Multielement time sequence anomaly detection model | |
| CN111405605A (en) | Wireless network interruption detection method based on self-organizing mapping | |
| CN115459982A (en) | Power network false data injection attack detection method | |
| Yang et al. | Lightweight Fault Prediction Method for Edge Networks | |
| CN114861792A (en) | Complex power grid key node identification method based on deep reinforcement learning | |
| CN118473910B (en) | Electric power Internet of things fault detection method and system based on edge cloud cooperation | |
| Liao et al. | Research on line loss analysis and intelligent diagnosis of abnormal causes in distribution networks: artificial intelligence based method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |