CN118709095B - Partial discharge type identification method and system based on multi-feature extraction and fusion - Google Patents
Partial discharge type identification method and system based on multi-feature extraction and fusion Download PDFInfo
- Publication number
- CN118709095B CN118709095B CN202411194528.7A CN202411194528A CN118709095B CN 118709095 B CN118709095 B CN 118709095B CN 202411194528 A CN202411194528 A CN 202411194528A CN 118709095 B CN118709095 B CN 118709095B
- Authority
- CN
- China
- Prior art keywords
- matrix
- partial discharge
- features
- formula
- feature
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a partial discharge type identification method and a partial discharge type identification system based on multi-feature extraction and fusion, wherein the method comprises the steps of extracting graph structural features of partial discharge signal data to obtain graph structural features corresponding to the partial discharge signal data, fusing periodic features and the graph structural features, and carrying out graph layer normalization processing on the fused fusion features to obtain target fusion features; the method comprises the steps of adding position codes to target fusion features, extracting time sequence features of the added features to be input to obtain final feature vectors, inputting the final feature vectors into a partial discharge recognition model constructed in advance, and outputting partial discharge types corresponding to the final feature vectors by the partial discharge recognition model. The response capability of the model to the input signal change can be enhanced by adaptively adjusting the attention weight, so that the accuracy of feature extraction and the adaptability to dynamic change are improved.
Description
Technical Field
The invention belongs to the technical field of power system monitoring and diagnosis, and particularly relates to a partial discharge type identification method and system based on multi-feature extraction and fusion.
Background
In high voltage cable systems, the partial discharge (PARTIAL DISCHARGE, PD) phenomenon is a localized area electrical discharge due to insulation defects, which is an important precursor to degradation and failure of cable insulation. Thus, identification and diagnosis of partial discharges is critical to ensuring safe operation of the power system.
In the prior art, a machine learning-based method is to utilize a Support Vector Machine (SVM), a decision tree and other traditional machine learning algorithms to classify by extracting the characteristics of partial discharge signals. The deep learning-based method adopts deep learning models such as a Convolutional Neural Network (CNN) and a long and short term memory network (LSTM) to automatically extract the features from partial discharge signals for recognition. The deep learning method is excellent in the aspect of processing complex signals, but needs a large amount of labeling data for training, the interpretation of the model is poor, and the consumption of calculation resources is high.
Disclosure of Invention
The invention provides a partial discharge type identification method and a partial discharge type identification system based on multi-feature extraction and fusion, which are used for solving the technical problems that a large amount of annotation data is required for training, the interpretation of a model is poor, and the consumption of calculation resources is high.
In a first aspect, the present invention provides a partial discharge type recognition method based on multi-feature extraction and fusion, including:
acquiring partial discharge signal data, and extracting periodic characteristics in the partial discharge signal data according to a mode of combining frequency domain analysis and time domain analysis ;
Extracting the graph structural characteristics of the partial discharge signal data to obtain graph structural characteristics corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
Adding position coding to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vector;
The final feature vector is processedInputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
In a second aspect, the present invention provides a partial discharge type recognition system based on multi-feature extraction and fusion, comprising:
An acquisition module configured to acquire partial discharge signal data and extract periodic features in the partial discharge signal data according to a mode combining frequency domain analysis and time domain analysis ;
A first extraction module configured to extract the graph structural features of the partial discharge signal data to obtain the graph structural features corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
A second extraction module configured to add a position code to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vector;
An output module configured to output the final feature vectorInputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
In a third aspect, an electronic device is provided that includes at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the partial discharge type identification method based on multi-feature extraction and fusion of any of the embodiments of the present invention.
In a fourth aspect, the present invention also provides a computer readable storage medium having stored thereon a computer program, which when executed by a processor, causes the processor to perform the steps of the partial discharge type identification method based on multi-feature extraction and fusion of any of the embodiments of the present invention.
The application provides a partial discharge type identification method and a partial discharge type identification system based on multi-feature extraction and fusion, which are used for providing a comprehensive importance measurement method combining node degree and node feature mean value, and calculating the importance of each node by the method. On the basis, the self-adaptive pooling strategy is adopted, the characteristics of important nodes are reserved preferentially, so that the effectiveness and the calculation efficiency of characteristic extraction are improved remarkably, and a dynamic perception attention mechanism is introduced in the time sequence characteristic extraction process. The mechanism can adaptively adjust the attention weight, and enhance the response capability of the model to the change of the input signal, thereby improving the accuracy of feature extraction and the adaptability to dynamic change.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a partial discharge type recognition method based on multi-feature extraction and fusion according to an embodiment of the present invention;
FIG. 2 is a block diagram of a partial discharge type recognition system based on multi-feature extraction and fusion according to an embodiment of the present invention;
Fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. 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.
Referring to fig. 1, a flowchart of a partial discharge type recognition method based on multi-feature extraction and fusion according to the present application is shown.
As shown in fig. 1, the partial discharge type identification method based on multi-feature extraction and fusion specifically includes the following steps:
Step S101, obtaining partial discharge signal data, and extracting periodic characteristics in the partial discharge signal data according to a mode of combining frequency domain analysis and time domain analysis 。
In this step, partial discharge signals are acquired in real time by means of a sensor mounted at the cable joint. The acquired data are time sequence signals, and comprise information such as amplitude, period, phase and the like.
The method comprises the specific steps of carrying out wavelet transformation on the signals to obtain wavelet coefficients of different scales and positions.
Wavelet transformation, namely performing wavelet transformation on the signals to obtain wavelet coefficients of different scales and positions.
And (3) self-adaptive threshold calculation, namely self-adaptively calculating the threshold value of each level of wavelet coefficient according to the statistical characteristics of noise, wherein the formula is as follows:
,
In the formula, Is the standard deviation of the noise, which is the standard deviation of the noise,Is the number of wavelet coefficients. By continuously adaptively adjusting the threshold, dynamic denoising at various scales is possible.
And (3) thresholding, namely performing soft thresholding on the wavelet coefficients by applying a threshold value, setting coefficients smaller than the threshold value to zero, and subtracting the threshold value from the rest coefficients. Soft thresholdThe processing formula is as follows:
,
Wherein, As a result of the threshold value being set,To adjust the parameters.
Inverse wavelet transform, in which the processed wavelet coefficient is subjected to inverse wavelet transform to reconstruct the denoised signal.
The periodic feature extraction is as follows:
Fast Fourier transform, namely performing fast Fourier transform on the partial discharge signal data, and extracting frequency domain characteristics The frequency domain featuresThe expression of (2) is:
,
In the formula, The signal is processed by denoising;
spectral analysis frequency domain features Each frequency of (3)Calculating each frequency corresponding to an amplitudeAmplitude of (a) of (b)And selecting a frequency with an amplitude greater than a preset threshold as a target frequencyWith the target frequencyCorresponding amplitude valueStored as periodic features, expressed as:
,
,
In the formula, Representing a set of frequencies having an amplitude greater than a preset threshold for a set of primary frequency components,For frequencyThe corresponding absolute value of the spectral amplitude,For the amplitude corresponding to the main frequency component,Is one frequency in the set of frequency components;
empirical Mode Decomposition (EMD) of partial discharge signal data into several intrinsic mode functions and residual errors The characteristic components on different time scales are acquired, and the expression is:
,
In the formula, In order to remove the noise of the signal,As the number of Intrinsic Mode Functions (IMFs),Is the firstA fixed modal function;
And (3) decomposing the self-adaptive signals, namely further decomposing IMFS, and enhancing the precision of period detection. For the first Analyzing the solid mode function, combining the periodic components to form an integral periodic characteristic, wherein the first mode function is used forThe expression for analysis of the individual solid mode functions is:
,
In the formula, The number of components that are the natural mode functions,Is the firstThe j-th component of the intrinsic mode function;
the expression of the overall periodic characteristics is:
,
In the formula, For the period feature set after the adaptive signal decomposition,As a function of the frequency component,For the amplitude value corresponding to the frequency component,For the firstA set of principal frequencies corresponding to the jth component of the natural mode function,For frequencyA corresponding spectral amplitude;
Short-time Fourier transform, namely, carrying out short-time Fourier transform on the partial discharge signal data, analyzing the time-frequency characteristics of the partial discharge signal data, and identifying the change of frequency along with time, wherein the short-time Fourier transform has the expression:
,
In the formula, For time domain features, representing time of daySum frequencyThe spectral values of the lower one of the two,In order to remove the noise of the signal,As a function of the window(s),As a function of the time variable,A kernel function representing a fourier transform, which is a complex exponential function;
,
In the formula, Is the amplitude of the time-frequency characteristic;
,
In the formula, Is a time-frequency characteristic matrix, and is expressed at each momentSum frequencyThe value of the power spectrum in the lower part,Is the firstThe time point at which the time point is the same,Is the firstFrequency points;
autocorrelation function, using the autocorrelation function to analyze partial discharge signal data, the autocorrelation function having the expression:
,
In the formula, As an autocorrelation function, representing the time lag of the signalWhen (1) the correlation is used to determine the correlation,For denoised signal at timeIs used as a reference to the value of (a),For denoised signal at timeIs a value of (2);
period detection, namely, integrating frequency domain analysis and autocorrelation function to detect the period of partial discharge signal data The expression is:
,
,
In the formula, As a major frequency component of the frequency spectrum,Peak position as autocorrelation function;
the obtained cycle characteristics are as follows:
,
In the formula, As a set of periodic features,Is a set of the principal frequency components,For the set of magnitudes corresponding to the dominant frequency component,Is the firstThe function of the individual natural modes,Is the frequency corresponding to the natural mode function,Is the main period of the signal.
Step S102, extracting the graph structural characteristics of the partial discharge signal data to obtain the graph structural characteristics corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics。
In this step, the graph structural features are extracted as:
The signals are represented as a graph structure assuming that the signals in the partial discharge signal data are , wherein,For the number of signal points, each signal point is taken as one node in the graph, and each nodeCorresponding signal point;
Defining dynamic adjacency matricesThe Euclidean distance is adopted to judge whether connection exists, specifically: , wherein, Is a distance threshold;
the degree matrix is a diagonal matrix in which diagonal elements Representing nodesAt the time ofDegree of (i.e. with node)The number of connected nodes and the matrix size isDiagonal elementsThe expression of (2) is:
;
each node Is set directly to the corresponding signal valueInitial feature matrixIs of the size of;
Graph convolution in which a feature matrix is convolved with a dynamic adjacency matrixUsed in combination with the degree matrix to capture the relationship between the signals to obtain the structural features of the graphThe expression is:
,
In the formula, In order to activate the function,Is thatIs used for the spatial convolution operation of (a),For the weight matrix of the initial convolutional layer,Is a degree matrix.
The periodic characteristics are describedSum-picture structural featuresThe expression for fusion is:,
In the formula, As the fusion weights of the structural features of the graph,Is the fusion weight of the periodic features. Specifically, the coefficients are initialized toMultiple experiments were performed on the training set using cross-validation, with the performance of the validation set being used to select the optimalAndValues.
Further, introducing an adaptive pooling method after the graph rolling operation and residual connection, and adaptively selecting a pooling strategy according to the importance of the node comprises the following steps:
Node importance measurement calculation, namely carrying out node importance measurement calculation by adopting a method of combining node degree and node characteristic mean value, wherein the expression is as follows:
,
In the formula, Is a nodeIs a measure of the importance of (1),Is a nodeI.e., the number of nodes connected thereto,Is a nodeIs used for the feature average value of (1),For the weight of the node metric,Weight of node characteristic average valueAndThe initial value is set to be 1, and in the experimental training process, the grid search optimization algorithm is adopted to select the optimalAndValues.
,
In the formula,For the degree of node i, i.e. the number of nodes connected to it,Is a nodeIn the first placeCharacteristic values of the individual dimensions;
,
In the formula, Representing nodes as contiguous matrix elementsSum nodeWhether a connection exists;
node selection, selecting a node greater than a threshold based on the integrated importance metric Is expressed as:
,
,
In the formula, As a mean value of the current node importance measure,Is the standard deviation of the current node importance measure,For adjusting parameters, controlling the ratio of reserved nodes; Generally in the range of 0.5 to 2, the process is carried out first Initialization is set to 1 by changing the differenceValues are tested. Dividing the data set into a training set and a validation set by comparing the differencesInfluence of values on model performance, selecting an optimumValues.
Feature preservation and pooling, namely carrying out weighted pooling on the selected node features when the feature preservation is carried out, wherein the weighted pooling expression is as follows:
,
Weighted pooled feature weighting the pooled features;
In the formula, For the selected set of important nodes,Is a nodeIs used for the weight coefficient of the (c),Is a nodeIs a feature matrix of (1);
multi-scale feature extraction, namely performing convolution operation on input features by using convolution check of different scales to obtain multi-scale features expressed as:
,
In the formula, As the feature matrix after the fusion,Is the firstFive-fold cross validation is adopted for the weight coefficient of each scale, the weight coefficient is initialized, different weight coefficient combinations are set, five-fold validation is executed for each weight coefficient combination, the performance of each combination on a validation set is calculated, the weight coefficient combination with the optimal performance is selected,Is the firstSelf-adaptive pooling results of individual scales;
and (3) pooling the self-adaptive graph convolution, namely introducing the self-adaptive graph convolution in the pooling process to obtain a characteristic matrix after the self-adaptive graph convolution, wherein the expression is as follows:
,
In the formula, For the feature matrix after the convolution of the adaptive map,AndThe adaptive adjacency matrix and the degree matrix are respectively,In order to activate the function,Is a weight matrix.
In the embodiment, the periodic features and the graph structure features are fused before the self-adaptive graph pooling, compared with the simple process that the periodic features are independently put into the final fusion, the method has the following effects that the depth optimization of feature fusion is achieved, the features with higher hierarchy and stronger recognition force can be extracted through multi-level processing of the fused features, so that the overall performance of the model is improved, the periodic features and the graph structure features can be mutually supplemented and enhanced in the convolution process through the advanced fusion, all the features can be processed under a unified framework through the advanced fusion, the consistency and the uniformity of the model are ensured, the multi-layer graph convolution and residual connection can effectively prevent feature degradation, the stability of the fused features in subsequent processing is ensured, and the robustness and the generalization capability of the model are improved. If the periodic features are kept alone, the lack of such multi-level optimization and enhancement, the robustness and generalization ability of the model may be limited.
Step S103, adding position codes to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vector。
In this step, position coding is added to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vectorComprising the following steps:
Adding adaptive position code into characteristic sequence, using neural network to generate, inputting into time step and signal characteristic, outputting into adaptive position code, finally generating characteristic matrix containing position code Feature matrix comprising position codingThe expression of (2) is:
,
,
,
In the formula, For the purpose of the adaptive position coding,As the raw time-series data is to be obtained,For the neural network to be used to generate the position codes,For use from raw dataA function of the extracted time step information,Is time step information;
before extracting the time sequence features, processing the input features through a multi-scale convolution layer, adding the multi-scale convolution layer, and extracting the multi-scale features through convolution kernels with different sizes, wherein the expression of the multi-scale features is as follows:
,
In the formula, In order for the splicing operation to be performed,In the form of a multi-scale feature matrix,、、For convolution operations of different scales,The characteristic matrix is coded;
dynamic perceived attention mechanisms including generating queries, keys, and values: for multi-scale features Generating a corresponding query matrix, a key matrix and a value matrix, wherein the expression is as follows:
,
,
,
In the formula, In order to query the matrix,In order to query the weight matrix of the matrix,Is a bias term for the key matrix,In the form of a matrix of keys,Is a weight matrix for the key matrix,Is a bias term for the key matrix,In the form of a matrix of values,Is a weight matrix that is a matrix of values,Bias terms that are a matrix of values;
initializing the weight matrix and the bias term by adopting Xaier initialization, wherein the initialization expression is as follows:
,
In the formula, Is any one of a weight matrix of a query matrix, a weight matrix of a key matrix, and a weight matrix of a value matrix,In order to be uniformly distributed,In order to input the dimensions of the device,Is the output dimension;
feature space projection, namely, by linear transformation, input features are projected into different feature spaces, a query matrix is projected into one feature space for providing questions, a key matrix is projected into the other feature space for providing clues of answers, and a value matrix contains actual answer information;
Dynamic perception of input features Performing linear transformation to generate preliminary bias term,Is a weight matrix of a linear transformation,Bias terms for linear transformations;
nonlinear activation is carried out on the result after linear transformation to generate a dynamic bias term ;
Dynamically biasing itemsIs introduced into the calculation of the attention weight and is used for self-adaptive adjustment;
Calculate the attention weight, calculate the dot product of the query matrix Q and key matrix K, and divide by the square root of the dimension of the key to scale: Wherein, Representing the dot product result of the query matrix and the key matrix and dividing by the square root of the dimension of the keyScaling is carried out;
Dynamically biasing items Introduced into the dot product result, the formula is:;
Normalizing the attention score through a softmax function to generate a final attention weight A;
calculating the attention output, wherein the calculated final attention weight A is applied to the value matrix V to obtain the final attention output The attention output calculation formula is:
,
In the formula, In order to be able to take care of the output,Representing nodes as contiguous matrix elementsSum nodeWhether or not there is a connection or not,Is the j-th value vector.
Step S104, the final feature vectorInputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
In this step, the attention output generated by the dynamic perceptual attention mechanism may be passed as input to a multi-layer transducer encoder for further processing and extraction of long-term dependencies. The transducer encoder is a model based on a multi-head attention mechanism, and can effectively capture complex dependency relations in sequence data.
Input layer output AttentionOutput of dynamic awareness mechanism is used as input of transducer encoder:;
Transformer encoder layer the Transformer encoder is formed by stacking multiple layers, each layer including a multi-headed attention mechanism and a feed forward neural network (FFN). The specific formula is as follows:
,
Wherein, Is a transducer encoderThe output feature matrix of the layer is used,Is a transducer encoderThe input feature matrix of the layer is,Is of a single layer structure in a transducer encoder, and is used for the first layer of the transducer encoderThe input characteristic matrix of the layer is processed by a multi-head attention mechanism and a feedforward neural network and output to a transducer encoderThe output feature matrix of the layer is used,I.e. the output of the dynamic awareness mechanism as the first layerLayer input.Wherein, Including a multihead attention mechanism and a feedforward neural network,、The weight matrixes of the first layer and the second layer in the feedforward neural network are respectively,、The bias terms of the first layer and the second layer, respectively. By multiple layersLayer L output feature matrix of encoderWhereinIs thatThe L-th layer of the encoder outputs a feature matrix.
The extracted final feature vectorAnd inputting the data into a model for classification recognition.
Classification model and optimization
Feature vector generation by final feature matrixPerforming a flat operation to generate a feature vector z: Through the flat operation, the feature matrix is converted into an input feature vector z for subsequent classification and regression tasks.
Self-attention mechanism, namely weighting the feature vectors by using the self-attention mechanism, enhancing the feature selection capability and generating attention feature vectors。
Self-Attention mechanism, calculate Query (Query, key (Key) and Value):
where z is the input feature vector, Wherein, AndThe first of the query matrix and the key matrix, respectivelyLine and thThe number of rows of the device is,Is the dimension of the key. Applying the Softmax function: Weighted summation: Wherein, For the attention output of the ith query,For the attention score between the ith query and the jth key,For the attention score between the ith query and the kth key,For the attention weight between the ith query and the jth key,Is the first in the value matrixAnd (3) row. Outputting an attention vector:。
Classification model application using fully connected neural networks (Fully Connected Neural Network FCNN) to focus feature vectors And performing classification prediction.
Input layer attention vector。
The hidden layer comprises 3 full-connection layers, a first full-connection layerA second full-connection layerThird full-connection layer。
Output layer-output classification result using softmax activation function:
Challenge training (performed during model training) is introduced to improve the robustness and generalization ability of the model. Challenge training generates challenge samples: Wherein the method comprises the steps of Is a feature vector of the object set,Is the amplitude of the disturbance, representing the magnitude of the disturbance,Is a sign function for determining the direction of the gradient.Is the classification loss function relative to the feature vectorIs a gradient of (a).
Challenge sample training:, And Is the weight and bias of the classification model. Countering loss function: , wherein, Is to combat the loss of the same,Is the value of class c of the real label.
Total loss function: , wherein, Is the weight against the loss of which,Is a regularization term, used to prevent overfitting,Is the regularization strength.
Model training and prediction using total loss functionModel training is carried out, and model parameters are updatedAndTo minimize losses. Applying the trained model to the test data to generate a final classification prediction result。
In this embodiment, the partial discharge type recognition method based on multi-feature extraction and fusion has the following effects:
The recognition accuracy is improved, namely the importance of the nodes is calculated by comprehensively utilizing the node degree and the node characteristic mean value through the self-adaptive image pooling layer, the characteristics of the important nodes are reserved preferentially, the accuracy of characteristic extraction is effectively improved, and the accuracy of partial discharge type recognition is remarkably improved.
The anti-noise capability is enhanced, namely the dynamic perception attention mechanism can flexibly cope with the change of input signals by adaptively adjusting the attention weight, thereby effectively inhibiting environmental noise and electromagnetic interference and improving the anti-noise performance of the recognition system.
The real-time performance is improved, the optimized self-adaptive image pooling layer and the dynamic perception attention mechanism are excellent in the aspects of feature extraction and signal processing calculation efficiency, the requirements of an electric power system on the real-time performance and the high-efficiency performance are met, and the recognition system can be kept to operate efficiently in large-scale real-time data processing.
The adaptability is enhanced, the self-adaptive image pooling layer and the dynamic perception attention mechanism enable the identification system to adapt to different power equipment and complex environments, and the system has good universality and adaptability and ensures stable and reliable operation in various actual scenes.
Dynamic perception capability, namely the dynamic perception attention mechanism is excellent in processing dynamic change signals, and attention weight can be flexibly adjusted according to the change of the signals, so that the processing capability and response capability of complex signals are improved.
The robustness and generalization capability are enhanced, namely an countermeasure training mechanism is introduced in the model training process, so that the robustness and generalization capability of the model are improved, and stable performance in different data sets and practical application is ensured.
The adaptive graph pooling layer performs pooling by adaptively selecting important nodes, so that the effectiveness of feature reservation is ensured, feature dimension is reduced, the computing efficiency is improved, and the system can still operate efficiently in an environment with limited resources.
And the combination of the multi-layer graph convolution network and residual connection ensures that the feature extraction level is deeper, the high-order features in the signals can be captured, and the performance of the identification system is further improved.
In summary, the method of the application provides a comprehensive importance measurement method combining node degree and node characteristic mean value, and the importance of each node is calculated by the method. On the basis, the self-adaptive pooling strategy is adopted, the characteristics of important nodes are reserved preferentially, so that the effectiveness and the calculation efficiency of characteristic extraction are improved remarkably, and a dynamic perception attention mechanism is introduced in the time sequence characteristic extraction process. The mechanism can adaptively adjust the attention weight, and enhance the response capability of the model to the change of the input signal, thereby improving the accuracy of feature extraction and the adaptability to dynamic change.
Referring to fig. 2, a block diagram of a partial discharge type recognition system based on multi-feature extraction and fusion according to the present application is shown.
As shown in fig. 2, the partial discharge type recognition system 200 includes an acquisition module 210, a first extraction module 220, a second extraction module 230, and an output module 240.
Wherein the acquisition module 210 is configured to acquire partial discharge signal data and extract periodic features in the partial discharge signal data according to a combination of frequency domain analysis and time domain analysis;
A first extraction module 220 configured to extract the graph structural features of the partial discharge signal data to obtain the graph structural features corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
A second extraction module 230 configured to add a position code to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vector;
An output module 240 configured to output the final feature vectorInputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
It should be understood that the modules depicted in fig. 2 correspond to the various steps in the method described with reference to fig. 1. Thus, the operations and features described above for the method and the corresponding technical effects are equally applicable to the modules in fig. 2, and are not described here again.
In other embodiments, embodiments of the present invention further provide a computer readable storage medium having stored thereon a computer program, where the program instructions, when executed by a processor, cause the processor to perform the partial discharge type identification method based on multi-feature extraction and fusion in any of the method embodiments described above;
as one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions configured to:
acquiring partial discharge signal data, and extracting periodic characteristics in the partial discharge signal data according to a mode of combining frequency domain analysis and time domain analysis ;
Extracting the graph structural characteristics of the partial discharge signal data to obtain graph structural characteristics corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
Adding position coding to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vector;
The final feature vector is processedInputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
The computer-readable storage medium may include a storage program area that may store an operating system, an application program required for at least one function, and a storage data area that may store data created according to the use of the partial discharge type recognition system based on multi-feature extraction and fusion, etc. In addition, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer readable storage medium optionally includes memory remotely located with respect to the processor, which may be connected to the partial discharge type recognition system based on multi-feature extraction and fusion via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
Fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and as shown in fig. 3, the device includes a processor 310 and a memory 320. The electronic device may further comprise input means 330 and output means 340. The processor 310, memory 320, input device 330, and output device 340 may be connected by a bus or other means, for example in fig. 3. Memory 320 is the computer-readable storage medium described above. The processor 310 executes various functional applications of the server and data processing by running non-volatile software programs, instructions and modules stored in the memory 320, i.e., implements the partial discharge type recognition method based on multi-feature extraction and fusion of the above-described method embodiments. The input device 330 may receive input numeric or character information and generate key signal inputs related to user settings and function controls of the partial discharge type recognition system based on multi-feature extraction and fusion. The output device 340 may include a display device such as a display screen.
The electronic equipment can execute the method provided by the embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method. Technical details not described in detail in this embodiment may be found in the methods provided in the embodiments of the present invention.
As an implementation manner, the electronic device is applied to a partial discharge type identification system based on multi-feature extraction and fusion, and is used for a client, and comprises at least one processor and a memory in communication connection with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can:
acquiring partial discharge signal data, and extracting periodic characteristics in the partial discharge signal data according to a mode of combining frequency domain analysis and time domain analysis ;
Extracting the graph structural characteristics of the partial discharge signal data to obtain graph structural characteristics corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
Adding position coding to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vector;
The final feature vector is processedInputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on such understanding, the foregoing technical solutions may be embodied essentially or in part in the form of a software product, which may be stored in a computer-readable storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the various embodiments or methods of some parts of the embodiments.
It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present invention, and not for limiting the same, and although the present invention has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present invention.
Claims (9)
1. The partial discharge type identification method based on multi-feature extraction and fusion is characterized by comprising the following steps of:
acquiring partial discharge signal data, and extracting periodic characteristics in the partial discharge signal data according to a mode of combining frequency domain analysis and time domain analysis ;
Extracting the graph structural characteristics of the partial discharge signal data to obtain graph structural characteristics corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
Adding position coding to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vectorWherein the adding of position coding to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vectorComprising the following steps:
Adding adaptive position code into characteristic sequence, using neural network to generate, inputting into time step and signal characteristic, outputting into adaptive position code, finally generating characteristic matrix containing position code Feature matrix comprising position codingThe expression of (2) is:
,
,
,
In the formula, For the purpose of the adaptive position coding,As the raw time-series data is to be obtained,For the neural network to be used to generate the position codes,As a function of the time step information extracted from the raw data,Is time step information;
before extracting the time sequence features, processing the input features through a multi-scale convolution layer, adding the multi-scale convolution layer, and extracting the multi-scale features through convolution kernels with different sizes, wherein the expression of the multi-scale features is as follows:
,
In the formula, In order for the splicing operation to be performed,In the form of a multi-scale feature matrix,、、For convolution operations of different scales,The characteristic matrix is coded;
dynamic perceived attention mechanisms including generating queries, keys, and values: for multi-scale features Generating a corresponding query matrix, a key matrix and a value matrix, wherein the expression is as follows:
,
,
,
In the formula, In order to query the matrix,In order to query the weight matrix of the matrix,Is a bias term for the key matrix,In the form of a matrix of keys,Is a weight matrix for the key matrix,Is a bias term for the key matrix,In the form of a matrix of values,Is a weight matrix that is a matrix of values,Bias terms that are a matrix of values;
initializing the weight matrix and the bias term by adopting Xaier initialization, wherein the initialization expression is as follows:
,
In the formula, Is any one of a weight matrix of a query matrix, a weight matrix of a key matrix, and a weight matrix of a value matrix,In order to be uniformly distributed,In order to input the dimensions of the device,Is the output dimension;
feature space projection, namely, by linear transformation, input features are projected into different feature spaces, a query matrix is projected into one feature space for providing questions, a key matrix is projected into the other feature space for providing clues of answers, and a value matrix contains actual answer information;
Dynamic perception of input features Performing linear transformation to generate preliminary bias term,Is a weight matrix of a linear transformation,Bias terms for linear transformations;
nonlinear activation is carried out on the result after linear transformation to generate a dynamic bias term ;
Dynamically biasing itemsIs introduced into the calculation of the attention weight and is used for self-adaptive adjustment;
Calculate the attention weight, calculate the dot product of the query matrix Q and key matrix K, and divide by the square root of the dimension of the key to scale: Wherein, Representing the dot product result of the query matrix and the key matrix and dividing by the square root of the dimension of the keyScaling is carried out;
Dynamically biasing items Introduced into the dot product result, the formula is:;
Normalizing the attention score through a softmax function to generate a final attention weight A;
calculating the attention output, wherein the calculated final attention weight A is applied to the value matrix V to obtain the final attention output The attention output calculation formula is:
,
In the formula, In order to be able to take care of the output,For adjacency matrix elements, representing nodes and whether there is a connection to a node,Is the j-th value vector;
The final feature vector is processed Inputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
2. The partial discharge type recognition method based on multi-feature extraction and fusion according to claim 1, wherein the extracting periodic features in the partial discharge signal data according to a combination of frequency domain analysis and time domain analysis comprises:
Fast Fourier transform, namely performing fast Fourier transform on the partial discharge signal data, and extracting frequency domain characteristics The frequency domain featuresThe expression of (2) is:
,
In the formula, The signal is processed by denoising;
spectral analysis frequency domain features Each frequency of (3)Calculating each frequency corresponding to an amplitudeAmplitude of (a) of (b)And selecting a frequency with an amplitude greater than a preset threshold as a target frequencyWith the target frequencyCorresponding amplitude valueStored as periodic features, expressed as:
,
,
In the formula, Representing a set of frequencies having an amplitude greater than a preset threshold for a set of primary frequency components,For frequencyThe corresponding absolute value of the spectral amplitude,For the amplitude corresponding to the main frequency component,Is one frequency in the set of frequency components;
empirical Mode Decomposition (EMD) of partial discharge signal data into several intrinsic mode functions and residual errors The characteristic components on different time scales are acquired, and the expression is:
,
In the formula, In order to remove the noise of the signal,As the number of Intrinsic Mode Functions (IMFs),Is the firstA fixed modal function;
adaptive signal decomposition for the first Analyzing the solid mode function, combining the periodic components to form an integral periodic characteristic, wherein the first mode function is used forThe expression for analysis of the individual solid mode functions is:
,
In the formula, The number of components that are the natural mode functions,Is the firstThe j-th component of the intrinsic mode function;
the expression of the overall periodic characteristics is:
,
In the formula, For the period feature set after the adaptive signal decomposition,As a function of the frequency component,For the amplitude value corresponding to the frequency component,For the firstA set of principal frequencies corresponding to the jth component of the natural mode function,For frequencyA corresponding spectral amplitude;
Short-time Fourier transform, namely, carrying out short-time Fourier transform on the partial discharge signal data, analyzing the time-frequency characteristics of the partial discharge signal data, and identifying the change of frequency along with time, wherein the short-time Fourier transform has the expression:
,
In the formula, For time domain features, representing spectral values at time and frequency,The noise-removed signal, the window function,As a function of the time variable,A kernel function representing a fourier transform, which is a complex exponential function;
,
In the formula, Is the amplitude of the time-frequency characteristic;
,
In the formula, Is a time-frequency characteristic matrix, and is expressed at each momentSum frequencyThe value of the power spectrum in the lower part,For the first point in time of the first time,Is the first frequency point;
autocorrelation function, using the autocorrelation function to analyze partial discharge signal data, the autocorrelation function having the expression:
,
In the formula, As an autocorrelation function, representing the time lag of the signalWhen (1) the correlation is used to determine the correlation,For the value of the denoised signal at time,For denoised signal at timeIs a value of (2);
period detection, namely, integrating frequency domain analysis and autocorrelation function to detect the period of partial discharge signal data The expression is:
,
,
In the formula, As a major frequency component of the frequency spectrum,Peak position as autocorrelation function;
the obtained cycle characteristics are as follows:
,
In the formula, As a set of periodic features,Is a set of the principal frequency components,For the set of magnitudes corresponding to the dominant frequency component,Is the firstThe function of the individual natural modes,Is the frequency corresponding to the natural mode function,Is the main period of the signal.
3. The partial discharge type recognition method based on multi-feature extraction and fusion as claimed in claim 1, wherein said performing a graph structural feature extraction on said partial discharge signal data results in a graph structural feature corresponding to said partial discharge signal dataComprising the following steps:
The signals are represented as a graph structure assuming that the signals in the partial discharge signal data are Wherein, for the number of signal points, each signal point is taken as one node in the graph, each nodeCorresponding signal point;
Defining dynamic adjacency matricesThe Euclidean distance is adopted to judge whether connection exists, specifically: , wherein, Is a distance threshold;
the degree matrix is a diagonal matrix in which diagonal elements Representing nodesAt the time ofDegree of (i.e. with node)The number of connected nodes and the matrix size isDiagonal elementsThe expression of (2) is:
;
each node Is set directly to the corresponding signal valueThe initial feature matrix is of the size of;
Graph convolution in which a feature matrix is convolved with a dynamic adjacency matrixUsed in combination with the degree matrix to capture the relationship between the signals to obtain the structural features of the graphThe expression is:
,
In the formula, In order to activate the function,Is thatIs used for the spatial convolution operation of (a),For the weight matrix of the initial convolutional layer,Is a degree matrix.
4. The partial discharge type recognition method based on multi-feature extraction and fusion according to claim 1, wherein the periodic features are selected from the group consisting ofSum-picture structural featuresThe expression for fusion is:
,
In the formula, As the fusion weights of the structural features of the graph,Is the fusion weight of the periodic features.
5. The partial discharge type recognition method based on multi-feature extraction and fusion according to claim 1, wherein the fusion features after fusion are pairedCarrying out layer normalization processing to obtain target fusion characteristicsComprising the following steps:
self-adaptive pooling, namely introducing a self-adaptive pooling method after graph rolling operation and residual connection, and self-adaptively selecting a pooling strategy according to the importance measure of the nodes;
layer normalization, namely carrying out layer normalization processing on the generated feature matrix to obtain target fusion features Wherein the target fuses the featuresThe expression of (2) is:
,
In the formula, For the layer normalization operation,The characteristic matrix is the characteristic matrix after the convolution of the self-adaptive graph.
6. The partial discharge type recognition method based on multi-feature extraction and fusion according to claim 5, wherein the introducing an adaptive pooling method after a graph rolling operation and residual connection, adaptively selecting a pooling strategy according to importance of nodes, comprises:
Node importance measurement calculation, namely carrying out node importance measurement calculation by adopting a method of combining node degree and node characteristic mean value, wherein the expression is as follows:
,
In the formula, As a measure of the importance of a node,The degree of a node, i.e. the number of nodes connected to it,As a characteristic average value of the node,For the weight of the node metric,The weight of the node characteristic average value;
,
In the formula, For the degree of node i, i.e. the number of nodes connected to it,The characteristic value of the node in the first dimension;
,
In the formula, As adjacency matrix elements, representing nodes and whether there is a connection between the nodes;
node selection, selecting a node greater than a threshold based on the integrated importance metric Is expressed as:
,
,
In the formula, As a mean value of the current node importance measure,Is the standard deviation of the current node importance measure,For adjusting parameters, controlling the ratio of reserved nodes;
Feature preservation and pooling, namely carrying out weighted pooling on the selected node features when the feature preservation is carried out, wherein the weighted pooling expression is as follows:
,
In the formula, For the selected set of important nodes,Is the weighting coefficient of the node and,Is a characteristic matrix of the node;
multi-scale feature extraction, namely performing convolution operation on input features by using convolution check of different scales to obtain multi-scale features expressed as:
,
In the formula, As the feature matrix after the fusion,Is the firstThe weight coefficient of the individual scale is used,Is the firstSelf-adaptive pooling results of individual scales;
and (3) pooling the self-adaptive graph convolution, namely introducing the self-adaptive graph convolution in the pooling process to obtain a characteristic matrix after the self-adaptive graph convolution, wherein the expression is as follows:
,
In the formula, For the feature matrix after the convolution of the adaptive map,AndThe adaptive adjacency matrix and the degree matrix are respectively,In order to activate the function,Is a weight matrix.
7. A partial discharge type recognition system based on multi-feature extraction and fusion, comprising:
An acquisition module configured to acquire partial discharge signal data and extract periodic features in the partial discharge signal data according to a mode combining frequency domain analysis and time domain analysis ;
A first extraction module configured to extract the graph structural features of the partial discharge signal data to obtain the graph structural features corresponding to the partial discharge signal dataAnd characterizing the periodSum-picture structural featuresFusing and combining the fused characteristicsCarrying out layer normalization processing to obtain target fusion characteristics;
A second extraction module configured to add a position code to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vectorWherein the adding of position coding to the target fusion featureAnd for the added features to be inputExtracting time sequence characteristics to obtain final characteristic vectorComprising the following steps:
Adding adaptive position code into characteristic sequence, using neural network to generate, inputting into time step and signal characteristic, outputting into adaptive position code, finally generating characteristic matrix containing position code Feature matrix comprising position codingThe expression of (2) is:
,
,
,
In the formula, For the purpose of the adaptive position coding,As the raw time-series data is to be obtained,For the neural network to be used to generate the position codes,As a function of the time step information extracted from the raw data,Is time step information;
before extracting the time sequence features, processing the input features through a multi-scale convolution layer, adding the multi-scale convolution layer, and extracting the multi-scale features through convolution kernels with different sizes, wherein the expression of the multi-scale features is as follows:
,
In the formula, In order for the splicing operation to be performed,In the form of a multi-scale feature matrix,、、For convolution operations of different scales,The characteristic matrix is coded;
Dynamic perceived attention mechanisms including generating queries, keys, and values: for multi-scale features Generating a corresponding query matrix, a key matrix and a value matrix, wherein the expression is as follows:
,
,
,
In the formula, In order to query the matrix,In order to query the weight matrix of the matrix,Is a bias term for the key matrix,In the form of a matrix of keys,Is a weight matrix for the key matrix,Is a bias term for the key matrix,In the form of a matrix of values,Is a weight matrix that is a matrix of values,Bias terms that are a matrix of values;
initializing the weight matrix and the bias term by adopting Xaier initialization, wherein the initialization expression is as follows:
,
In the formula, Is any one of a weight matrix of a query matrix, a weight matrix of a key matrix, and a weight matrix of a value matrix,In order to be uniformly distributed,In order to input the dimensions of the device,Is the output dimension;
feature space projection, namely, by linear transformation, input features are projected into different feature spaces, a query matrix is projected into one feature space for providing questions, a key matrix is projected into the other feature space for providing clues of answers, and a value matrix contains actual answer information;
Dynamic perception of input features Performing linear transformation to generate preliminary bias term,Is a weight matrix of a linear transformation,Bias terms for linear transformations;
nonlinear activation is carried out on the result after linear transformation to generate a dynamic bias term ;
Dynamically biasing itemsIs introduced into the calculation of the attention weight and is used for self-adaptive adjustment;
Calculate the attention weight, calculate the dot product of the query matrix Q and key matrix K, and divide by the square root of the dimension of the key to scale: Wherein, Representing the dot product result of the query matrix and the key matrix and dividing by the square root of the dimension of the keyScaling is carried out;
Dynamically biasing items Introduced into the dot product result, the formula is:;
Normalizing the attention score through a softmax function to generate a final attention weight A;
calculating the attention output, wherein the calculated final attention weight A is applied to the value matrix V to obtain the final attention output The attention output calculation formula is:
,
In the formula, In order to be able to take care of the output,For adjacency matrix elements, representing nodes and whether there is a connection to a node,Is the j-th value vector;
An output module configured to output the final feature vector Inputting into a pre-constructed partial discharge recognition model, wherein the partial discharge recognition model outputs the final characteristic vectorA corresponding partial discharge type.
8. An electronic device comprising at least one processor and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411194528.7A CN118709095B (en) | 2024-08-29 | 2024-08-29 | Partial discharge type identification method and system based on multi-feature extraction and fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411194528.7A CN118709095B (en) | 2024-08-29 | 2024-08-29 | Partial discharge type identification method and system based on multi-feature extraction and fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118709095A CN118709095A (en) | 2024-09-27 |
| CN118709095B true CN118709095B (en) | 2025-05-13 |
Family
ID=92822426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411194528.7A Active CN118709095B (en) | 2024-08-29 | 2024-08-29 | Partial discharge type identification method and system based on multi-feature extraction and fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118709095B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119647655B (en) * | 2024-11-15 | 2026-01-06 | 浙江大学 | A method for predicting the remaining life probability of key components of large wind turbine units |
| CN119989198B (en) * | 2025-01-15 | 2025-08-15 | 南开大学 | Partial discharge classification and identification method |
| CN120372407B (en) * | 2025-06-23 | 2025-09-19 | 华东交通大学 | Hand motion recognition method and system |
| CN120541620B (en) * | 2025-07-16 | 2025-11-25 | 广东电网有限责任公司中山供电局 | Short-term power load prediction method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111666912A (en) * | 2020-06-15 | 2020-09-15 | 国网山东省电力公司潍坊供电公司 | Partial discharge fusion feature extraction method considering electrical feature quantity and graphic feature |
| CN116304979A (en) * | 2023-03-02 | 2023-06-23 | 兰州交通大学 | A multi-feature fusion partial discharge type recognition method based on attention mechanism |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100401594B1 (en) * | 2001-07-19 | 2003-10-17 | 엘지전선 주식회사 | Partial discharge measurement system auto-selecting time and frequency domain |
| CN109685138B (en) * | 2018-12-25 | 2023-04-07 | 东南大学 | XLPE power cable partial discharge type identification method |
| CN112686093A (en) * | 2020-12-02 | 2021-04-20 | 重庆邮电大学 | Fusion partial discharge type identification method based on DS evidence theory |
| CN112748317B (en) * | 2021-03-23 | 2022-03-25 | 国网河南省电力公司电力科学研究院 | Switch cabinet partial discharge fault detection method and system based on multiple monitoring data |
| CN113064032B (en) * | 2021-03-26 | 2022-08-02 | 云南电网有限责任公司电力科学研究院 | A Partial Discharge Pattern Recognition Method Based on Spectral Features and Information Fusion |
| CN114943256B (en) * | 2022-05-31 | 2025-05-06 | 常州大学 | A partial discharge identification method and device based on time-frequency characteristics and improved CNN |
| CN115659252A (en) * | 2022-07-25 | 2023-01-31 | 国网安徽省电力有限公司电力科学研究院 | GIS partial discharge mode identification method based on PRPD multi-feature information fusion |
| CN115453286B (en) * | 2022-09-01 | 2023-05-05 | 珠海市伊特高科技有限公司 | GIS partial discharge diagnosis method, model training method, device and system |
| CN118038247A (en) * | 2022-11-03 | 2024-05-14 | 武汉纺织大学 | A method for enhancing the accuracy of partial discharge signal annotation based on multimodal fusion network |
| CN118150945A (en) * | 2023-12-29 | 2024-06-07 | 国网河北省电力有限公司电力科学研究院 | A clock synchronization method and device for overhead line partial discharge detection |
| CN117708760B (en) * | 2024-02-05 | 2024-07-05 | 国网江西省电力有限公司电力科学研究院 | Switchgear multi-source partial discharge pattern recognition method and system based on multi-model fusion |
| CN118070841A (en) * | 2024-02-19 | 2024-05-24 | 三峡大学 | Electric vehicle short-term charging load forecasting method based on LSTM-GNN combined model |
| CN118484666B (en) * | 2024-07-16 | 2024-10-01 | 浙江兴创新能源有限公司 | Energy storage power station evaluation method and system for source network load multi-element application |
-
2024
- 2024-08-29 CN CN202411194528.7A patent/CN118709095B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111666912A (en) * | 2020-06-15 | 2020-09-15 | 国网山东省电力公司潍坊供电公司 | Partial discharge fusion feature extraction method considering electrical feature quantity and graphic feature |
| CN116304979A (en) * | 2023-03-02 | 2023-06-23 | 兰州交通大学 | A multi-feature fusion partial discharge type recognition method based on attention mechanism |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118709095A (en) | 2024-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN118709095B (en) | Partial discharge type identification method and system based on multi-feature extraction and fusion | |
| CN118070107B (en) | Deep learning-oriented network anomaly detection method, device, storage medium and equipment | |
| CN111488904A (en) | Image classification method and system based on confrontation distribution training | |
| CN120448713B (en) | Microservice system time sequence abnormality detection method, device, equipment, medium and product | |
| CN110378319A (en) | Signal detection method and device, computer equipment and storage medium | |
| CN120873557B (en) | Wind power prediction method and system based on time sequence decomposition and multi-model fusion | |
| CN119622318A (en) | A method for fault diagnosis of electric power equipment based on multidimensional features to enhance attention | |
| Fu et al. | A diffusion model-based deep learning approach for denoising acoustic emission signals in concrete | |
| CN111222689A (en) | LSTM load prediction method, medium, and electronic device based on multi-scale temporal features | |
| CN117292301A (en) | Automatic feeding control system and method for poultry farming | |
| CN116523615A (en) | Bank abnormal account detection method, device, system and medium | |
| CN119849548B (en) | Bearing residual life prediction model training method, device, medium and prediction method | |
| CN120910415A (en) | New energy missing data completion method and system based on multi-model fusion | |
| CN119961819A (en) | A method for decomposing high-power electrical load based on long-term and short-term recurrent neural networks | |
| CN119377639A (en) | A natural gas load forecasting method and device | |
| CN118820995A (en) | A photovoltaic power generation system current transformer state analysis and prediction method and system | |
| Qin et al. | Forecasting short-term wind power with multi-view attention mechanism and dual recurrent neural networks | |
| Ren et al. | A new parallel PQDs classification method based on an optimized NLM and an improved DRSN-TCN model | |
| CN114638555A (en) | Power consumption behavior detection method and system based on multilayer regularization extreme learning machine | |
| CN120070944B (en) | Image classification method, training method of image classification model and related equipment | |
| CN120912975B (en) | Soil salinity estimation method and device with self-adaptive characteristics | |
| CN120597112A (en) | A bolt fault diagnosis method, system, device and medium based on acoustic signals | |
| Xu-Dong et al. | Control Chart Recognition Method Based on Transfer Learning | |
| An et al. | AFEformer: An adaptive frequency enhancement transformer for time series prediction | |
| El Madmoune et al. | Improving Face Recognition Robustness: An Innovative Approach Using Tchebichef Hahn Moments and CNNs |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |