CN117592825A - Voltage sag evaluation method based on graph convolution neural network - Google Patents
Voltage sag evaluation method based on graph convolution neural network Download PDFInfo
- Publication number
- CN117592825A CN117592825A CN202310867591.1A CN202310867591A CN117592825A CN 117592825 A CN117592825 A CN 117592825A CN 202310867591 A CN202310867591 A CN 202310867591A CN 117592825 A CN117592825 A CN 117592825A
- Authority
- CN
- China
- Prior art keywords
- voltage sag
- data
- model
- graph
- sag
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 26
- 238000011156 evaluation Methods 0.000 title claims description 14
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000011176 pooling Methods 0.000 claims abstract description 13
- 230000008859 change Effects 0.000 claims abstract description 9
- 238000007418 data mining Methods 0.000 claims abstract description 6
- 238000005516 engineering process Methods 0.000 claims abstract description 3
- 239000011159 matrix material Substances 0.000 claims description 31
- 230000006870 function Effects 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 23
- 230000004913 activation Effects 0.000 claims description 9
- 238000004088 simulation Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 238000003062 neural network model Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 238000000547 structure data Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 2
- 230000008901 benefit Effects 0.000 abstract description 3
- 239000000284 extract Substances 0.000 abstract 1
- 238000013058 risk prediction model Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- 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
- G06N3/048—Activation functions
-
- 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
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Quality & Reliability (AREA)
- General Health & Medical Sciences (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Power Engineering (AREA)
- Probability & Statistics with Applications (AREA)
- Fuzzy Systems (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a voltage sag assessment method based on a graph roll-up neural network, which adopts the graph roll-up neural network as a data mining means, exerts the advantages of the self-carried network characteristics of the graph roll-up neural network, fully extracts grid structure information and node characteristic information, and can rapidly and accurately predict the voltage sag risk of a power grid; and the pooling technology is added, the characteristic information is compressed into a fixed dimension, the model is prevented from being fitted, meanwhile, the method provided by the invention can predict the voltage sag risk after topology change, and timely send out an early warning signal, so that an electric company can make voltage sag management measures in advance, and the economic loss of users is reduced.
Description
Technical Field
The invention relates to the technical field of voltage sag evaluation, in particular to a voltage sag evaluation method based on a graph convolution neural network.
Background
The traditional voltage sag risk prediction model is mostly a method adopting Monte Carlo fault simulation, and the method is based on the topological structure of a power grid, and the voltage sag characteristic value is obtained by setting faults on each node and each line and calculating the faults. However, as the power grid scale is continuously enlarged and the power equipment is continuously updated, the power grid topological structure is more complex, the fault position, the fault type, the fault probability and the like are more diversified, so that the traditional voltage sag prediction method has long calculation period, low accuracy and low practicability. And after the grid structure or load output is changed, the old model cannot be applied, grid and output information is required to be input again, training is performed again, time consumption is long, efficiency is low, and therefore the traditional voltage sag risk prediction model is further improved.
Disclosure of Invention
Therefore, the present invention aims to provide a voltage sag evaluation method based on a graph convolution neural network, so as to rapidly and accurately predict a voltage sag risk.
In order to achieve the above purpose, the invention adopts the following technical scheme: the voltage sag evaluation method based on the graph convolution neural network comprises the following steps of;
step 1, voltage sag simulation data under different topological structures and different running modes are obtained by utilizing sag random simulation calculation based on a Monte Carlo method;
step 2, preprocessing data, quantizing and standardizing voltage sag data, storing the quantized and standardized voltage sag data into a voltage sag database, taking the data and a grid adjacent matrix as the input of a model, and taking a corresponding residual voltage amplitude as the output of the model;
step 3, after a graph convolutional neural network model is constructed, model training is carried out by utilizing a training set, a predicted result deviation value is calculated by adopting MSE, and model parameters are back propagated to optimize the model parameters; inputting the test set into the model to obtain a prediction error, and indicating that the model is correct within the error range;
and 4, dividing sag early warning levels by utilizing output residual voltage amplitude prediction results and sag tolerance characteristics of different users, and selecting a voltage sag amplitude severity index MSI to represent a voltage sag event risk level so as to finish voltage sag risk early warning.
In a preferred embodiment, in step 2, the data is normalized by mean normalization, and the formula is as follows:
wherein X represents the normalized value, X represents the value before normalization, X represents the average value in the sample data, σ represents the variance in the sample data;
after the data processing is finished, storing the data into a voltage sag database; each voltage sag data is stored and input in the form of graph data by taking a fault as a unit, and the data comprises node characteristic data, grid structure data and node labels; and a voltage sag risk prediction database is established to store voltage sag data so as to facilitate the data calling of the model.
In a preferred embodiment, in step 3, the GCN is trained by using the data in the voltage sag database by using the graph roll-up neural network GCN as a data mining means, and the mode between the graph roll-up neural network model layers is as follows:
wherein: h (l) Representing the matrix of voltage sag eigenvectors at layer l, H at all stages (l+1) Representing all the voltage sag feature vector matrixes after one convolution operation;a is an adjacent matrix corresponding to a grid structure, and I is a unit matrix; />Is->Degree matrix of W (l) Initializing weights for random; delta is a certain activation function and the ReLU function is selected as the activation function.
In a preferred embodiment, in step 3, a pooling layer technique is added after stacking the graph, the importance of the nodes is scored by using node voltage sag information through the formula (3), only a fixed number of node information is reserved according to the scoring result through the formulas (4) - (5), the information of the low-level nodes is aggregated to the high-level nodes through the formulas (7) - (9) in the pooling implementation process, the serious information loss is prevented, and finally the processed voltage sag node characteristic X is output out * The method comprises the steps of carrying out a first treatment on the surface of the The pooling operation converts the graph structure from different dimensions to fixed dimensions, so that the training model can cope with the topological-changed power grid structure, and the model overfitting is prevented;
wherein: delta is an activation function, X is H (0) I.e. inputting a voltage sag feature matrix, wherein theta is a node self-attention parameter;
X out =X + ⊙F + (4)
wherein: as indicated by multiplication by element, X + 、F + The feature vector sum score of each node i + of the first k is respectively represented,is an adjacency matrix between the first k nodes i+;
Attention(Q,K,V)=softmax(QK T )V (8)
X out * =X + ⊙F + +MultiHead(X + ,X - ,X - ) (9)
wherein: x is X - Is the characteristic vector of the low-node, Q is the query function, K is the key matrix, V is the value matrix, h is the number of attention heads, E 0 ,Is a parameter matrix which can be learned.
In a preferred embodiment, in step 3, after the graph convolution neural network model is built, node characteristic information and grid structure information are used as input of the model, and node residual voltage is used as output of the model; meanwhile, dividing data into a training set and a testing set, training by using the data of the training set, outputting a residual voltage amplitude predicted value through a multi-layer graph convolutional neural network, carrying out back propagation on model parameters by taking a mean square error MSE as a loss function through calculating the loss function between the predicted value and an actual value, adjusting the model parameters to obtain a minimum value of the loss function, and ending the training when the loss function meets the requirement or reaches the preset iteration times; then inputting the test set into the model to obtain a prediction error, and indicating that the model is correct within the error range;
and because the pooling layer can generate low-dimensional characteristic representation with fixed dimension, the power grid voltage sag prediction after topology change can be predicted, and the node residual voltage amplitude can be directly output by only quantifying and standardizing the node characteristics and inputting a trained model, so that the voltage sag after topology change can be predicted in real time.
In a preferred embodiment, in step 4, the output residual voltage amplitude prediction result and the sag tolerance characteristics of different users are utilized to divide sag early warning levels, and the voltage sag amplitude severity index MSI is selected to represent the risk level of the voltage sag event, so as to complete the voltage sag risk early warning;
wherein:is the per unit value of the residual voltage amplitude, U max And U min The maximum value and the minimum value of the residual voltage amplitude in an uncertain region in a generalized tolerance curve of sensitive equipment are obtained;
after MSI is calculated, the MSI is divided into sections according to the affected condition of the user, and the sag risk is divided into three grades of mild, moderate and severe.
Compared with the prior art, the invention has the following beneficial effects: the invention provides a voltage sag evaluation method based on a graph convolution neural network, which can be used for realizing the voltage sag risk prediction evaluation of a complex large power grid. The method optimizes the existing risk early warning method: the graph convolution neural network is used as a voltage sag data mining means, grid structure information and node characteristic information can be extracted more effectively, voltage sag risks can be predicted rapidly and accurately, and the voltage sag risk prediction after line operation modes and grid topology changes can be dealt with, so that the voltage sag risks can be predicted in time under various conditions such as line faults, power grid maintenance, power plant startup and shutdown and the like, and therefore voltage sag management can be rapidly unfolded on a power grid side and a user side, and loss caused by the voltage sag is reduced.
Drawings
FIG. 1 is a flow chart of a preferred embodiment of the present invention;
FIG. 2 is a diagram of a graph roll-up neural network in accordance with a preferred embodiment of the present invention;
FIG. 3 is a generalized tolerance curve of a sensitive device according to a preferred embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application; as used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
The invention provides a voltage sag evaluation method based on a graph convolution neural network, which comprises the following steps of:
step 1: and obtaining voltage sag simulation data under different topological structures and different operation modes by utilizing sag random simulation calculation based on a Monte Carlo method.
Step 2: preprocessing data, quantizing and standardizing voltage sag data, storing the quantized and standardized voltage sag data into a voltage sag database, taking the data and a grid adjacent matrix as the input of a model, and taking a corresponding residual voltage amplitude as the output of the model.
Step 3: after constructing a graph convolution neural network model, training the model by utilizing a training set, calculating a predicted result deviation value by adopting MSE, and carrying out back propagation on model parameters to optimize the model parameters; and inputting the test set into the model to obtain a prediction error, and indicating that the model is correct within the error range.
The model can predict the power grid voltage sag after the running mode is changed or the topology is changed in real time, the node residual voltage amplitude can be directly output by only quantifying and standardizing the node characteristics and inputting the node residual voltage amplitude into a trained model, and real-time prediction data is input into a sag database to continuously train and perfect the graph convolutional neural network model.
Step 4: and finally, dividing the sag early warning level by utilizing the output residual voltage amplitude prediction result and the sag tolerance characteristics of different users, selecting a voltage sag amplitude severity index MSI to represent the risk level of a voltage sag event, and finishing the voltage sag risk early warning.
Specifically, in real life, voltage sag monitoring points are not more, the detection period is long, measured data are very few, in order to effectively analyze the association relation between voltage sag influence factors and voltage sag, simulation data obtained by a voltage sag simulation program with large data size are adopted for sag data, actual conditions are simulated by the simulation data, and therefore real-time prediction of voltage sag risk can be achieved after power grid data are accessed in the future.
Analyzing the voltage sag risk prediction influence factors, combining data provided by voltage sag simulation software, screening out data information representing different voltage sag risk influence factors, forming one dimension of input of a voltage sag risk prediction model by each data information, quantifying and standardizing the obtained data, and comprehensively storing to obtain a voltage sag database. The quantitative value table of the dip database is shown in table 1.
Table 1 quantized value table of voltage sag influencing factor data
The voltage sag has discrete data and continuous data as each influencing factor, wherein the dimensions of the continuous data are different, such as the magnitude of the voltage of the tide current and the fault distance. In order to eliminate the variability between data, so that they have the same unit of measure and variation range, the performance of the machine learning algorithm is improved, the algorithm is easier to converge and process, so that the data can be better compared and analyzed, the continuous data is normalized, and the data is normalized by mean normalization, wherein the formula is as follows:
wherein X represents the value after normalization, X represents the value before normalization,represents the average value in the sample data, σ represents the variance in the sample data;
after the data processing is finished, the data is stored in a voltage sag database. Each voltage sag data is stored and input in the form of graph data by taking a fault as a unit, and the data comprises node characteristic data (tide voltage, fault position, fault distance and the like), grid structure data (adjacent matrix) and node labels (actual residual voltage of nodes). And a voltage sag risk prediction database is established to store voltage sag data so as to facilitate the data calling of the model.
According to the voltage sag risk prediction model established by the invention, a graph roll neural network (Graph Convolutional Network, GCN) is used as a data mining means, and the GCN neural network is trained by utilizing data in a voltage sag database, so that the connection relation among layers in the voltage sag risk prediction model and the weight relation among neurons are regulated, and the purpose of evaluating the voltage sag risk is achieved. The method for the graph rolling neural network model layer and layer is as follows:
wherein: h (l) Representing the matrix of voltage sag eigenvectors at layer l, H at all stages (l+1) Representing all the voltage sag feature vector matrixes after one convolution operation;a is an adjacent matrix corresponding to a grid structure, and I is a unit matrix; />Is->Degree matrix of W (l) Initializing weights for random; delta is a certain activation function, and the patent selects a ReLU function as the activation function.
In order to cope with the voltage sag risk prediction after topology change, a pooling layer technology is added after the graph is laminated, firstly, node voltage sag information is utilized to score importance of nodes (formula 3), only fixed number of node information is reserved according to the scoring result (formulas 4 and 5), a mutual attention mechanism is introduced in the pooling implementation process, information of low-level nodes is aggregated onto high-level nodes (formulas 7 to 9), serious information loss is prevented, and finally, the processed voltage sag node characteristic X is output out * . The pooling operation can convert the graph structure from different dimensions to fixed dimensions, so that the training model can cope with the topological-changed power grid structure, and the model overfitting can be prevented.
Wherein: delta is an activation function, X is H (0) I.e. the input voltage sag feature matrix, θ is the node self-attention parameter.
Wherein: as indicated by multiplication by element, X + 、F + The feature vector sum score of each node i + of the first k is respectively represented,is the adjacency matrix between the first k nodes i+.
Attention(Q,K,V)=softmax(QK T )V (8)
X out * =X + ⊙F + +MultiHead(X + ,X - ,X - ) (9)
Wherein: x is X - Is the characteristic vector of the low-node, Q is the query function, K is the key matrix, V is the value matrix, h is the number of attention heads, E 0 ,Is a parameter matrix which can be learned.
The structure of the convolutional neural network is shown in fig. 2. After the graph convolution neural network model is established, node characteristic information and grid structure information are used as the input of the model, and node residual voltage is used as the output of the model. Meanwhile, the data are divided into a training set and a testing set, the training is carried out by using the data of the training set, the residual voltage amplitude predicted value is output through the multi-layer graph convolutional neural network, the model parameters are reversely propagated by calculating a loss function (taking a mean square error MSE as the loss function) between the predicted value and the actual value, the model parameters are adjusted to obtain the minimum value of the loss function, and when the loss function meets the requirement or reaches the preset iteration times, the training is ended. And inputting the test set into the model to obtain a prediction error, and indicating that the model is correct within the error range.
And because the pooling layer can generate low-dimensional characteristic representation with fixed dimension, the model can predict the power grid voltage sag after topology change, and the node residual voltage amplitude can be directly output only by quantifying and standardizing the node characteristic and inputting the trained model, so that the voltage sag after topology change can be predicted in real time.
And finally, dividing the sag early warning level by utilizing the output residual voltage amplitude prediction result and the sag tolerance characteristics of different users, selecting a voltage sag amplitude severity index MSI to represent the risk level of a voltage sag event, and finishing the voltage sag risk early warning.
Wherein:is the per unit value of the residual voltage amplitude, U max And U min Maximum and minimum values of residual voltage amplitude in an uncertainty region in a generalized tolerance curve (such as fig. 3) for sensitive equipment.
After MSI is calculated, the MSI is divided into three classes, namely, mild, moderate and severe, according to the affected condition of the user, as shown in table 2.
Table 2: risk level of voltage sag
The invention uses the graph convolution neural network as a data mining means, fully utilizes the advantages of the self-carried network characteristics of the graph convolution neural network, can effectively simulate the characteristics of a power grid, reduces the training time of a voltage sag risk prediction model, and improves the prediction accuracy of the model. The graph convolution neural network has the advantage of strong generalization capability, and can realize voltage sag risk prediction after the change of the power grid topological structure, so that more power grid environments are simulated, and the application range of a voltage sag risk prediction model is enlarged.
Claims (6)
1. The voltage sag evaluation method based on the graph convolution neural network is characterized by comprising the following steps of;
step 1, voltage sag simulation data under different topological structures and different running modes are obtained by utilizing sag random simulation calculation based on a Monte Carlo method;
step 2, preprocessing data, quantizing and standardizing voltage sag data, storing the quantized and standardized voltage sag data into a voltage sag database, taking the data and a grid adjacent matrix as the input of a model, and taking a corresponding residual voltage amplitude as the output of the model;
step 3, after a graph convolutional neural network model is constructed, model training is carried out by utilizing a training set, a predicted result deviation value is calculated by adopting MSE, and model parameters are back propagated to optimize the model parameters; inputting the test set into the model to obtain a prediction error, and indicating that the model is correct within the error range;
and 4, dividing sag early warning levels by utilizing output residual voltage amplitude prediction results and sag tolerance characteristics of different users, and selecting a voltage sag amplitude severity index MSI to represent a voltage sag event risk level so as to finish voltage sag risk early warning.
2. The voltage sag evaluation method based on a graph roll-up neural network according to claim 1, wherein in step 2, the data is normalized by mean normalization, and the formula is as follows:
wherein X represents the value after normalization, X represents the value before normalization,represents the average value in the sample data, σ represents the variance in the sample data;
after the data processing is finished, storing the data into a voltage sag database; each voltage sag data is stored and input in the form of graph data by taking a fault as a unit, and the data comprises node characteristic data, grid structure data and node labels; and a voltage sag risk prediction database is established to store voltage sag data so as to facilitate the data calling of the model.
3. The voltage dip evaluation method based on a graph roll-up neural network according to claim 1, wherein in step 3, the graph roll-up neural network GCN is used as a data mining means, and the data in the voltage dip database is used to train the GCN neural network, and the mode between the graph roll-up neural network model layers is as follows:
wherein: h (l) Representing the matrix of voltage sag eigenvectors at layer l, H at all stages (l+1) Representing all the voltage sag feature vector matrixes after one convolution operation;a is an adjacent matrix corresponding to a grid structure, and I is a unit matrix; />Is->Degree matrix of W (l) Initializing weights for random; deltaFor a certain activation function, the ReLU function is selected as the activation function.
4. The voltage sag assessment method based on a graph roll-up neural network according to claim 3, wherein in step 3, a graph pooling layer technology is added after graph roll-up, importance of nodes is scored by means of node voltage sag information through formula (3), only fixed number of node information is reserved according to scoring results through formulas (4) - (5), a mutual attention mechanism is introduced in the implementation process of graph pooling through formulas (7) - (9), information of low-level nodes is aggregated to high-level nodes, serious information loss is prevented, and finally the processed voltage sag node characteristics X are output out * The method comprises the steps of carrying out a first treatment on the surface of the The pooling operation converts the graph structure from different dimensions to fixed dimensions, so that the training model can cope with the topological-changed power grid structure, and the model overfitting is prevented;
wherein: delta is an activation function, X is H (0) I.e. inputting a voltage sag feature matrix, wherein theta is a node self-attention parameter;
X out =X + ⊙F + (4)
wherein: as indicated by multiplication by element, X + 、F + The feature vector sum score of each node i + of the first k is respectively represented,is an adjacency matrix between the first k nodes i+;
Attention(Q,K,V)=softmax(QK T )V (8)
X out * =X + ⊙F + +MultiHead(X + ,X - ,X - ) (9)
wherein: x is X - Is the characteristic vector of the low-node, Q is the query function, K is the key matrix, V is the value matrix, h is the number of attention heads, E 0 ,Is a parameter matrix which can be learned.
5. The voltage sag evaluation method based on a graph roll-up neural network according to claim 1, wherein in step 3, after the graph roll-up neural network model is built, node characteristic information and grid structure information are used as inputs of the model, and node residual voltage is used as an output of the model; meanwhile, dividing data into a training set and a testing set, training by using the data of the training set, outputting a residual voltage amplitude predicted value through a multi-layer graph convolutional neural network, carrying out back propagation on model parameters by taking a mean square error MSE as a loss function through calculating the loss function between the predicted value and an actual value, adjusting the model parameters to obtain a minimum value of the loss function, and ending the training when the loss function meets the requirement or reaches the preset iteration times; then inputting the test set into the model to obtain a prediction error, and indicating that the model is correct within the error range;
and because the pooling layer can generate low-dimensional characteristic representation with fixed dimension, the power grid voltage sag prediction after topology change can be predicted, and the node residual voltage amplitude can be directly output by only quantifying and standardizing the node characteristics and inputting a trained model, so that the voltage sag after topology change can be predicted in real time.
6. The voltage sag evaluation method based on the graph roll-up neural network according to claim 1, wherein in the step 4, sag early warning levels are divided by using output residual voltage amplitude prediction results and sag tolerance characteristics of different users, and voltage sag event risk levels are represented by voltage sag amplitude severity indexes MSI to finish voltage sag risk early warning;
wherein:is the per unit value of the residual voltage amplitude, U max And U min The maximum value and the minimum value of the residual voltage amplitude in an uncertain region in a generalized tolerance curve of sensitive equipment are obtained;
after MSI is calculated, the MSI is divided into sections according to the affected condition of the user, and the sag risk is divided into three grades of mild, moderate and severe.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310867591.1A CN117592825A (en) | 2023-07-15 | 2023-07-15 | Voltage sag evaluation method based on graph convolution neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310867591.1A CN117592825A (en) | 2023-07-15 | 2023-07-15 | Voltage sag evaluation method based on graph convolution neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117592825A true CN117592825A (en) | 2024-02-23 |
Family
ID=89910375
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310867591.1A Pending CN117592825A (en) | 2023-07-15 | 2023-07-15 | Voltage sag evaluation method based on graph convolution neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117592825A (en) |
-
2023
- 2023-07-15 CN CN202310867591.1A patent/CN117592825A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105117602B (en) | A kind of metering device running status method for early warning | |
CN106874581B (en) | Building air conditioner energy consumption prediction method based on BP neural network model | |
CN111539515B (en) | Complex equipment maintenance decision method based on fault prediction | |
CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
KR102275126B1 (en) | Electrical facility safety evaluation system using artificial intelligence technique | |
CN114676822B (en) | Multi-attribute fusion air quality forecasting method based on deep learning | |
CN107644297B (en) | Energy-saving calculation and verification method for motor system | |
CN116150897A (en) | Machine tool spindle performance evaluation method and system based on digital twin | |
CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
CN112232561A (en) | Power load probability prediction method based on constrained parallel LSTM quantile regression | |
CN110705859A (en) | PCA-self-organizing neural network-based method for evaluating running state of medium and low voltage distribution network | |
CN115438726A (en) | Device life and fault type prediction method and system based on digital twin technology | |
CN114444660A (en) | Short-term power load prediction method based on attention mechanism and LSTM | |
CN114611372A (en) | Industrial equipment health prediction method based on Internet of things edge calculation | |
CN115829145A (en) | Photovoltaic power generation capacity prediction system and method | |
CN115096357A (en) | Indoor environment quality prediction method based on CEEMDAN-PCA-LSTM | |
CN115358437A (en) | Power supply load prediction method based on convolutional neural network | |
CN113591402A (en) | Digital power transformer health state fuzzy comprehensive evaluation method | |
Niu et al. | Application of AHP and EIE in reliability analysis of complex production lines systems | |
CN117674119A (en) | Power grid operation risk assessment method, device, computer equipment and storage medium | |
CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
CN111738483A (en) | Power grid loss reduction optimization method and system based on clustering and deep belief network | |
CN116776209A (en) | Method, system, equipment and medium for identifying operation state of gateway metering device | |
Bashir et al. | Short-term load forecasting using artificial neural network based on particle swarm optimization algorithm | |
CN111242266A (en) | Operation data management system |
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 |