CN115693785A - Method for judging transient power angle stability of power system and related device thereof - Google Patents
Method for judging transient power angle stability of power system and related device thereof Download PDFInfo
- Publication number
- CN115693785A CN115693785A CN202211457405.9A CN202211457405A CN115693785A CN 115693785 A CN115693785 A CN 115693785A CN 202211457405 A CN202211457405 A CN 202211457405A CN 115693785 A CN115693785 A CN 115693785A
- Authority
- CN
- China
- Prior art keywords
- nodes
- graph
- node
- generator
- stability
- 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
Images
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The application discloses a method and a related device for judging the transient power angle stability of a power system, wherein a generator of an original generator graph in the power system is taken as a node, and node clustering is carried out based on edge weights among the nodes to obtain a reduced-dimension graph; mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping; the graph after dimension reduction and the state characteristics thereof are taken as the input of the graph attention force neural network, and the stability judgment result is output; the state characteristics comprise the voltage amplitude of a generator bus, the voltage phase angle, the active power output, the reactive power output, the power angle of a generator, the voltage at the generator terminal of the generator, the alternating current bus voltage of a converter station and the direct current of the converter station, and the technical problems that the dimension disaster is caused by excessive input characteristics and model parameters due to the fact that the system scale is large and the number of related nodes is large, and the input characteristics of all lines and unit input models are extracted, the model training cost is high, the training difficulty is high, the generalization capability is poor, and then the stability judgment result is influenced are solved.
Description
Technical Field
The present application relates to the field of power system technologies, and in particular, to a method and an apparatus for determining transient power angle stability of a power system.
Background
The power grid safety and stability control strategy is to adopt a power cutting machine and load cutting control measure to a system when the power grid is abnormal in operation so as to prevent further expansion of faults. As an important measure for ensuring the stable operation of the power grid, the stability control strategy has the problems of large calculated amount and serious dependence on expert experience in formulation, check and evaluation. Under the background of 'double height', the number of dimensions of a power grid operation mode is increased, the mode is changeable, the difficulty of power grid fault analysis is continuously increased, the scale of an offline analysis and calculation scene is increased explosively, and higher requirements are provided for offline stability control analysis efficiency. One of the key links that restricts the offline stability control analysis efficiency and the degree of automation is stability discrimination. Some stability problems have explicit stability determination rules in terms of physical concepts or engineering calculation specifications, but in many cases, these rules require long-time numerical simulation to provide input information, or do not meet special conditions of analysis requirements (such as loss of synchronization of an insignificant unit far from a fault point), nor can they provide a measure of the degree of stability or instability. The deep learning can reveal key parameter information of the system and the stability degree of the system, can provide high-efficiency automatic stability judgment and stability boundary search functions, and is a key technology development direction for improving the offline stability control analysis efficiency and depth. And establishing a transient stability evaluation deep network model of the stability control strategy for the actual power grid, and facilitating efficient checking and verification of the stability control strategy under high-dimensional and high-order-of-magnitude samples.
In recent years, deep learning has made a significant progress in the field of stable assessment. The deep learning is in a data-driven form, and key information is extracted from a large amount of historical data, so that the stable condition of the system can be predicted quickly and accurately. A deep neural network intelligent stability evaluation model is established through deep learning, and the judgment on the stability can be rapidly output by inputting system parameters and states. Because the space topology of the state quantity of the power grid in the power system can be represented by a graph, and different unit positions in the transient process have obvious influence on the state quantity of the power grid, deep research can be carried out on the stability evaluation method of the power system based on the deep learning technology of the graph. Including Graph Neural Networks (GNNs), graph convolutional neural networks (GCNs), graph attention neural networks (GATs), and the like. The deep graph learning method can be used for extracting the interaction relation among the nodes in a targeted manner, and has higher generalization capability and accuracy in transient stability evaluation. However, for an actual large power grid, due to the large scale of the system and the numerous related nodes, if all lines and units in the network are input into the deep learning model to extract features, dimension disasters are caused by excessive input features and model parameters of deep learning, the model training cost and difficulty are increased, the generalization capability is poor, and the stability judgment result is influenced.
Disclosure of Invention
The application provides a method and a related device for judging the transient power angle stability of a power system, which are used for solving the technical problems that due to the fact that the system scale is large and the related nodes are numerous, all lines and units in a network are input into a deep learning model to extract characteristics, the input characteristics and model parameters are excessive to cause dimension disasters, the model training cost is high, the training difficulty is high, the generalization capability is poor, and further the stability judgment result is influenced.
In view of this, a first aspect of the present application provides a method for determining transient power angle stability of a power system, including:
taking generators in the original generator graph in the power system as nodes, and clustering the nodes based on edge weights among the nodes to obtain a reduced-dimension graph;
mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping, wherein the state characteristics of the nodes in the graph after dimensionality reduction are obtained by calculating the state characteristics of all the nodes in the cluster corresponding to the nodes;
taking the state characteristics of each node in the post-dimensionality reduction graph and the post-dimensionality reduction graph as the input of a graph attention neural network, and outputting the stability judgment result of the power system;
the state characteristics comprise a generator bus voltage amplitude value, a generator bus voltage phase angle, a generator active power output, a generator reactive power output, a generator power angle, a generator terminal voltage, a converter station alternating current bus voltage and a converter station direct current.
Optionally, the method for obtaining the dimensionality reduction graph by using the generator in the original generator graph in the power system as a node and performing node clustering based on the edge weight between the nodes includes:
calculating first-order similarity between nodes by taking a generator in a primary generator graph in a power system as a node and taking generator capacity and an inertia time constant as characteristics;
calculating improved side weights between nodes based on the first-order similarity between the nodes and initial side weights between the nodes in an adjacent matrix of the original generator graph, wherein the initial side weights between the nodes are in negative correlation with the electrical distances between the nodes;
the weight after improvement among the nodes is used as a node similarity index, and node clustering is carried out based on a hierarchical clustering algorithm to obtain a clustering result;
and mapping the nodes belonging to the same cluster into one node based on the clustering result to obtain a reduced-dimension graph.
Optionally, the calculation formula of the first-order similarity between nodes is:
u i ＝(S gi ,T Ji ) T /||(S gi ,T Ji ) T ||；
u j ＝(S gj ,T Jj ) T /||(S gj ,T Jj ) T ||；
in the formula, p ij Is a first order similarity between node i and node j, u i Normalized vector, S, of transient power angle stability characteristic parameter for node i gi Rated capacity, T, of generator for node i Ji Is the inertia time constant, u, of the generator corresponding to node i j Normalized vector of transient power angle stability characteristic parameter of node j, S gj Rated capacity, T, of generator for node j Jj Is the inertia time constant of the generator corresponding to node j.
Optionally, the calculation formula of the improved weights among the nodes is as follows:
in the formula, O i,j As improved back-end weights, ω, between nodes i and j ij Is an initial edge weight, p, between node i and node j ij Is the first order similarity between node i and node j.
Optionally, the outputting the result of judging the stability of the power system by taking the state characteristics of each node in the post-dimensionality reduction graph and the post-dimensionality reduction graph as the input of the graph attention force neural network includes:
taking the state characteristics of each node in the post-dimensionality reduction graph and the post-dimensionality reduction graph as the input of a graph attention neural network, and calculating a similarity coefficient between each node and a neighbor node through the graph attention neural network;
calculating the attention coefficient of each node to the neighbor node according to the similarity coefficient between each node and the neighbor node and the edge weight of each node and the neighbor node in the adjacency matrix of the reduced-dimension graph through the graph attention neural network;
and performing feature fusion on the state features of the neighbor nodes of each node according to the attention coefficients of each node to the neighbor nodes through the graph attention neural network to obtain the output features of each node, and outputting the stability judgment result of the power system according to the output features of each node.
Optionally, the training process of the graph attention neural network is as follows:
acquiring a training sample, and calculating a transient stability index of a generator rotor angle of the training sample;
if the transient stability index of the generator rotor angle of the training sample is larger than 0, setting the label of the training sample as the system transient power angle stability, and if the transient stability index of the generator rotor angle of the training sample is smaller than or equal to 0, setting the label of the training sample as the system transient power angle instability;
training the graph attention neural network through the training sample.
The second aspect of the present application provides an apparatus for determining transient power angle stability of an electric power system, including:
the clustering unit is used for taking the generators in the original generator graph in the power system as nodes and clustering the nodes based on the edge weights among the nodes to obtain a reduced-dimension graph;
the mapping unit is used for mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping, wherein the state characteristics of the nodes in the graph after dimensionality reduction are obtained by calculating the state characteristics of all the nodes in the cluster corresponding to the nodes;
the judging unit is used for taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the graph attention neural network and outputting the stability judging result of the power system;
the state characteristics comprise generator bus voltage amplitude, generator bus voltage phase angle, generator active power output, generator reactive power output, generator power angle, generator terminal voltage, converter station alternating current bus voltage and converter station direct current.
Optionally, the clustering unit is specifically configured to:
taking a generator in an original generator graph in a power system as a node, and calculating first-order similarity between the nodes by taking the capacity of the generator and an inertia time constant as characteristics;
calculating improved side weights between nodes based on the first-order similarity between the nodes and initial side weights between the nodes in an adjacent matrix of the original generator graph, wherein the initial side weights between the nodes are in negative correlation with the electrical distances between the nodes;
the weight after improvement among the nodes is used as a node similarity index, and node clustering is carried out based on a hierarchical clustering algorithm to obtain a clustering result;
and mapping the nodes belonging to the same cluster into one node based on the clustering result to obtain a reduced-dimension graph.
The third aspect of the present application provides a device for judging transient power angle stability of a power system, where the device includes a processor and a memory;
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the method for determining transient power angle stability of a power system according to any one of the first aspect according to instructions in the program code.
A fourth aspect of the present application provides a computer-readable storage medium for storing program code, which, when executed by a processor, implements the method for determining transient power angle stability of a power system according to any one of the first aspects.
According to the technical scheme, the method has the following advantages:
the application provides a method for judging the transient power angle stability of a power system, which comprises the following steps: taking generators in an original generator graph in the power system as nodes, and clustering the nodes based on edge weights among the nodes to obtain a reduced-dimension graph; mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping, wherein the state characteristics of the nodes in the graph after dimensionality reduction are obtained by calculating the state characteristics of all the nodes in the cluster corresponding to the nodes; taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the graph attention neural network, and outputting the stability judgment result of the power system; the state characteristics comprise generator bus voltage amplitude, a generator bus voltage phase angle, generator active power output, generator reactive power output, a generator power angle, generator terminal voltage, converter station alternating current bus voltage and converter station direct current.
In the method, a generator in an original generator graph in an electric power system is taken as a node, node clustering is carried out based on edge weights among the nodes, dimension reduction is carried out on the original generator graph in the electric power system, a dimension reduced graph is obtained, state features of the nodes in the original generator graph are mapped into the state features of the nodes in the dimension reduced graph through dimension reduction mapping, and input parameters of a graph attention neural network are reduced through network dimension reduction, so that the technical problems that dimension disasters are caused due to excessive input features and model parameters, the model training cost is high, the training difficulty is high, the generalization capability is poor, and further the stability judgment result is influenced due to the fact that all lines and units in the network input deep learning models extract features due to large system scale and numerous nodes are improved are solved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the description below are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the drawings without inventive labor.
Fig. 1 is a schematic flowchart of a method for determining transient power angle stability of an electric power system according to an embodiment of the present disclosure;
fig. 2 is a schematic structural diagram of a real dc sending system according to an embodiment of the present disclosure;
fig. 3 is a schematic structural diagram of a system obtained by clustering the area A1 in fig. 2 according to an embodiment of the present application;
fig. 4 is a schematic structural diagram of a system after dimension reduction represented by equivalent nodes after clustering of the area A1 in fig. 2 according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of an apparatus for determining transient power angle stability of an electric power system according to an embodiment of the present disclosure.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
For easy understanding, please refer to fig. 1, an embodiment of the present application provides a method for determining transient power angle stability of a power system, including:
and 101, taking generators in an original generator graph in the power system as nodes, and clustering the nodes based on edge weights among the nodes to obtain a reduced-dimension graph.
The method and the device consider that due to the fact that the system is large in scale and numerous in related nodes, if all lines and units in the network are input into the deep learning model to extract features, dimension disasters are caused due to excessive input features and model parameters, model training cost is high, training difficulty is high, generalization capability is poor, and then stability judgment results are affected. In order to improve this problem, the embodiment of the present application considers performing feature dimension reduction processing on a raw generator map in an electric power system.
The feature dimension reduction method mainly comprises feature selection, feature extraction, neural network dimension reduction and the like. The feature selection refers to directly selecting a subset of related features for model construction, and comprises a filtering method, an encapsulation method and an embedding method; feature extraction refers to projecting high-dimensional features to a low-dimensional feature space formed by a few key features, and the feature extraction includes principal component analysis, random forests and the like. The principal component analysis is suitable for the multivariable problem that certain correlation exists among input features, and the applicability is lacked on the features of the power grid topology level. Deep learning models such as convolutional neural networks, encoders and the like belonging to dimension reduction of black boxes together with random forests lack interpretability.
The input feature dimension reduction of the deep learning model for stability control strategy checking needs to meet the following requirements: (1) Embedding the graph topology of the original feature space into the low-dimensional and sparse graph topology while reducing the dimension of the node feature, and taking the graph topology as the topology input of the graph attention network, namely, not only reducing the dimension of the node feature vector, but also reducing the number of nodes and keeping the edge sparsity; (2) Reserving a main research area of a stability control strategy, and reserving state characteristics of a secondary area as much as possible so as to reserve the influence of the secondary area on the stability of the primary area; (3) The dimension reduction process has interpretability, so that when the instability phenomenon is observed, the key influence factors of the original system analysis can be traced back, and a basis is provided for the stability control strategy formulation and check.
The embodiment of the application notes the relation between the dynamic equivalence of the power grid and the dimensionality reduction requirement of the stability assessment model. Obviously, the network after the dynamic equivalence satisfies (1) and (2), however, the goal of the traditional dynamic equivalence is to obtain equivalent dynamic device parameters on the boundary nodes, so that the external network electrical dynamic feedback obtained by the system to be researched from the boundary nodes is similar to that before the equivalence, and an explicit interpretable mapping from the original system state to the boundary node state of the equivalence system is not established, and the dimensionality reduction goal (3) is not satisfied. Based on the above, the method for reducing the dimension of the network is used for reducing the dimension of the power grid (the original generator diagram), and the stability judgment is carried out based on the diagram after the dimension reduction, so that the three characteristic dimension reduction requirements can be realized, the transient state power angle stability judgment of the complex power grid can be efficiently and accurately realized, certain adaptability is provided for the topological change of the power grid, and the automatic development of offline stability control checking is powerfully supported.
Specifically, the process of using the generator in the original generator graph in the power system as a node and performing node clustering based on the edge weight between the nodes to obtain the reduced-dimension graph includes:
s1011, taking the generator in the original generator diagram in the power system as a node, and calculating the first-order similarity between the nodes by taking the generator capacity and the inertia time constant as characteristics;
in the embodiment of the application, a generator in an original generator graph in a power system is used as a node, an initial edge weight between the nodes is determined according to an electrical distance between the nodes, an initial relation graph can be generated, and a first-order similarity between the nodes is calculated by taking the generator capacity and an inertia time constant as characteristics, wherein a specific calculation formula is as follows:
u i ＝(S gi ,T Ji ) T /||(S gi ,T Ji ) T ||；
u j ＝(S gj ,T Jj ) T /||(S gj ,T Jj ) T ||；
in the formula, p ij Is a first order similarity between node i and node j, u i Normalized vector, S, of transient power angle stability characteristic parameter for node i gi Rated capacity, T, of generator for node i Ji Is the inertia time constant, u, of the generator corresponding to node i j Normalized vector of transient power angle stability characteristic parameter of node j, S gj Rated capacity, T, of generator for node j Jj Is the inertia time constant of the generator corresponding to node j.
And calculating first-order similarity through the rated capacity and the inertia time constant of the generator so as to enable the generators with similar rotor motion characteristics to be classified into one cluster in the subsequent clustering process.
S1012, calculating improved side weights among the nodes based on the first-order similarity among the nodes and the initial side weights among the nodes in the adjacent matrix of the original generator graph, wherein the initial side weights among the nodes are in negative correlation with the electrical distances among the nodes;
the calculation formula of the improved weight among the nodes is as follows:
in the formula, O i,j For improved back-end weights between node i and node j, O i,j The weight is larger after improvement, the higher the similarity of the two nodes is, omega ij The initial edge weight between the node i and the node j in the adjacency matrix of the initial relationship graph is negative correlation with the electrical distance between the nodes, and when the electrical distance between the nodes is larger than a certain threshold value, the initial edge weight between the nodes can be made to be zero.
S1013, clustering the nodes based on a hierarchical clustering algorithm by using the improved weights among the nodes as node similarity indexes to obtain clustering results;
in consideration of the fact that an actual power grid has a hierarchical structure based on voltage levels, in order to maintain the structural characteristics in the graph after dimensionality reduction, the embodiment of the application preferably adopts a hierarchical clustering algorithm to perform node clustering. The hierarchical clustering algorithm belongs to the prior art, and specific clustering processes are not described herein again.
And S1014, mapping the nodes belonging to the same cluster into one node based on the clustering result to obtain a dimensionality reduction graph.
After the clustering is completed to obtain a clustering result, mapping the nodes belonging to the same cluster into one node based on the clustering result to obtain a dimensionality-reduced graph, namely, replacing a plurality of nodes belonging to the same cluster in the original generator graph with an equivalent node in the dimensionality-reduced graph.
And 102, mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping.
After the original generator graph is subjected to dimensionality reduction to obtain a dimensionality reduced graph, the state features of the nodes in the original generator graph are mapped to the state features of the nodes in the dimensionality reduced graph through dimensionality reduction mapping, and the state features of the nodes in the dimensionality reduced graph are obtained by calculating the state features of all the nodes in the cluster corresponding to the nodes. Specifically, a plurality of nodes belonging to the same cluster in the original generator graph are replaced by an equivalent node in the reduced-dimension graph, and the state characteristics of the equivalent node are the average value (for voltage amplitude), the sum (for power) or the weighted average value (for phase angle) with the moment of inertia as the weight of the state characteristics of each node in the same cluster, so that the state characteristics of the nodes in the reduced-dimension graph can be explained. In the dynamic equivalence problem of the power system, parameters of the equivalent generator at the equivalent node need to be obtained at the moment, the embodiment of the application aims to obtain a mapping relation between the state characteristics of the equivalent node and the state characteristics of the original node before the equivalent node in the cluster, and the mapping relation between the state characteristics of the node in the graph after dimensionality reduction and the state characteristics of the node in the graph before the dimensionality reduction can refer to table 1, wherein the state characteristics comprise the voltage amplitude of a generator bus, the voltage phase angle of the generator bus, the active power output of the generator, the reactive power output of the generator, the power angle of the generator, the voltage at the generator terminal, the voltage of an alternating current bus of a converter station and the direct current of the converter station.
TABLE 1
In table 1, subscript i is the node number in the original generator graph that is complete before dimensionality reduction, subscript k is the node number in the graph after dimensionality reduction, and it is assumed that nodes in the region to be studied are protected before and after dimensionality reductionThe memory is unchanged, the outer area is a dimension reduction area, C k A node sequence number set representing the original generator graph contained in the cluster corresponding to the node k, n is C k The number of nodes of the set. As can be seen from table 1, in the post-dimensionality reduction graph corresponding to the external region, the generator bus voltage amplitude of the node k is the average value of the generator bus voltage amplitudes of all nodes of the original generator graph included in the cluster corresponding to the node k, and the direct-current lines are not made to be equivalent, so that the values before and after dimensionality reduction are kept consistent.
Through the process, the state characteristic space of the actual complex large system can be subjected to effective dimension reduction which can reserve the topological relation of the sparse graph, can be interpreted and does not depend on massive training data, so that an intelligent stability judgment model with excellent performance for the stability control strategy checking task of the actual large system can be obtained by matching with a graph attention neural network, the offline stability control analysis efficiency and the automation degree of the electric power system are remarkably improved, and the quick increase of the scale of the stability control checking calculation task under the 'double carbon' background is adapted.
And 103, taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the attention neural network, and outputting the stability judgment result of the power system.
The dimension and the element of the adjacent matrix of the dimension-reduced graph are changed correspondingly, and the edge weight omega 'of each node and the adjacent node of the adjacent matrix of the dimension-reduced graph' kj Satisfies the following conditions:
that is, if the node in the graph after dimensionality reduction is kept as the original node before dimensionality reduction, the edge weight between the nodes keeps the edge weight between the original nodes, and if the node in the graph after dimensionality reduction is an equivalent node obtained by mapping all the original nodes in the same cluster before dimensionality reduction, the edge weight of the node is the sum of the edge weights of all the original nodes in the cluster.
Taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the graph attention neural network, and calculating the phase between each node and the adjacent node through the graph attention neural networkSimilarity coefficient, wherein the similarity coefficient e between the node i and its neighbor node j ij Comprises the following steps:
in the formula, h i Is a status feature of node i, h j The state characteristics of the node j are shown in table 1, the element in the state characteristics represents vector splicing, and the shared parameter W to be learned is used for performing characteristic enhancement on the state characteristics of the node; the attention weight function a (-) is used to map the high-dimensional features onto real numbers as the parameters to be learned.
Because the coupling relation between the generator sets is related to the edge weight of the adjacent matrix in transient stability analysis, when the attention coefficient is calculated, the edge weight in the adjacent matrix of the dimensionality reduced graph is introduced on the basis of the similar coefficient item, and the attention coefficient of each node to the neighboring node is calculated through the graph attention neural network according to the similar coefficient between each node and the neighboring node and the edge weight of each node and the neighboring node in the adjacent matrix of the dimensionality reduced graph, namely:
in the formula, alpha ij Is the attention coefficient, N, of node i to neighbor node j i Is a neighbor node set of node i, ω' ij The edge weight of the node i and the neighbor node j passes through the edge weight omega 'in the adjacency matrix of the dimensionality reduced graph' kj LeakyReLU (-) is obtained as the activation function.
And performing feature fusion on the state features of the neighbor nodes of each node according to the attention coefficients of each node to the neighbor nodes through the graph attention neural network to obtain the output features of each node, and outputting the stability judgment result of the power system according to the output features of each node. The attention mechanism is a mainstream technology of the deep neural network, different attention coefficients are distributed to different neighbor nodes to highlight information which has the largest influence on a target node, and relatively irrelevant information is omitted. Attention coefficients in a graph attention neural network, feature correlations between nodes can be fused into model training.
In the embodiment of the present application, the transient power angle stability line determination of the power system is regarded as a two-class model, that is, the output of the attention neural network in the above graph is a stable result or an unstable result.
Further, the training process of the attention neural network is as follows:
acquiring a training sample, and calculating a transient stability index of a generator rotor angle of the training sample;
if the transient stability index of the generator rotor angle of the training sample is larger than 0, setting the label of the training sample as the system transient power angle stability, and if the transient stability index of the generator rotor angle of the training sample is smaller than or equal to 0, setting the label of the training sample as the system transient power angle instability;
the attention neural network of the graph is trained by training samples.
In the embodiment of the application, when the label is set for the training sample, the transient stability index I of the rotor angle of the generator is adopted TSI As a criterion, the specific calculation process is as follows:
in the formula, Δ δ max For the maximum internal potential phase angle difference between any two generators of interest, when I TSI >And 0, considering the transient power angle of the system to be stable, otherwise, considering the transient power angle of the system to be unstable.
Suppose that the structure of a real direct current sending system obtained according to a certain original generator diagram is shown in fig. 2, a sending end triangular ring area in fig. 2 is set as a research area for stability control strategy checking to be carried out, and a part A1 is an area for dimension reduction. The schematic structural diagrams of the system before and after clustering and dimension reduction of the area A1 are shown in fig. 3 and 4. The original A1 area comprises 879 nodes and 373799 connecting lines; after the dimension reduction method in the embodiment of the application is simplified, the number of the nodes is reduced to 55, and the number of the connecting lines is reduced to 862.
And setting the head end of the line to generate a three-phase grounding short circuit fault, protecting the misoperation for line removal 0.1s after the fault occurs, and implementing a stable control generator tripping measure 0.3s after the disturbance occurs, wherein the sample data sampling interval is 0.01s. A total of 700 samples were generated. According to the transient power angle stability criterion I TSI And setting a whole graph classification label for the sample set, wherein the number of the unstable samples (namely unstable samples) is 113.
By adopting the method for judging the transient power angle stability of the power system, the comprehensive accuracy rate reaches 96%, is improved by nearly 20% compared with a method for reducing the network dimension (keeping the characteristics of generators with 500kV and 220kV levels as input and neglecting others) which is common in practice, and is also improved by nearly 20% compared with a method for using a classical convolutional neural network model.
In the embodiment of the application, a generator in an original generator graph in an electric power system is taken as a node, node clustering is carried out based on edge weights among the nodes, dimension reduction is carried out on the original generator graph in the electric power system, a dimension reduced graph is obtained, state features of the nodes in the original generator graph are mapped into the state features of the nodes in the dimension reduced graph through dimension reduction mapping, and input parameters of a graph attention neural network are reduced through network dimension reduction, so that the technical problems that dimension disasters are caused due to excessive input features and model parameters, the model training cost is high, the training difficulty is high, the generalization capability is poor, and further the stability judgment result is influenced due to the fact that all lines and units in the network input deep learning models extract features due to large system scale and numerous nodes are involved are solved.
The above is an embodiment of a method for determining transient power angle stability of a power system provided by the present application, and the following is an embodiment of a device for determining transient power angle stability of a power system provided by the present application.
Referring to fig. 5, an apparatus for determining transient power angle stability of an electrical power system according to an embodiment of the present disclosure includes:
the clustering unit is used for clustering nodes based on edge weights among the nodes by taking generators in an original generator graph in the power system as the nodes to obtain a reduced-dimension graph;
the mapping unit is used for mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping, and the state characteristics of the nodes in the graph after dimensionality reduction are obtained by calculating the state characteristics of all the nodes in the cluster corresponding to the nodes;
the judging unit is used for taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the graph attention neural network and outputting the stability judging result of the power system;
the state characteristics comprise generator bus voltage amplitude, a generator bus voltage phase angle, generator active power output, generator reactive power output, a generator power angle, generator terminal voltage, converter station alternating current bus voltage and converter station direct current.
As a further improvement, the clustering unit is specifically configured to:
taking a generator in an original generator graph in a power system as a node, and calculating first-order similarity between the nodes by taking the capacity of the generator and an inertia time constant as characteristics;
calculating improved side weights among the nodes based on the first-order similarity among the nodes and initial side weights among the nodes in an adjacent matrix of the original generator graph, wherein the initial side weights among the nodes are in negative correlation with the electrical distance among the nodes;
the weight after improvement among the nodes is used as a node similarity index, and node clustering is carried out based on a hierarchical clustering algorithm to obtain a clustering result;
and mapping the nodes belonging to the same cluster into one node based on the clustering result to obtain a reduced-dimension graph.
As a further improvement, the determination unit is specifically configured to:
taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the graph attention neural network, and calculating the similarity coefficient between each node and the adjacent node through the graph attention neural network;
calculating the attention coefficient of each node to the neighbor node through the attention neural network of the graph according to the similarity coefficient between each node and the neighbor node and the edge weight of each node and the neighbor node in the adjacent matrix of the graph after dimension reduction;
and performing feature fusion on the state features of the neighbor nodes of each node according to the attention coefficients of each node to the neighbor nodes through the graph attention neural network to obtain the output features of each node, and outputting the stability judgment result of the power system according to the output features of each node.
In the embodiment of the application, a generator in an original generator graph in an electric power system is taken as a node, node clustering is carried out based on edge weights among the nodes, dimension reduction is carried out on the original generator graph in the electric power system, a dimension reduced graph is obtained, state features of the nodes in the original generator graph are mapped into the state features of the nodes in the dimension reduced graph through dimension reduction mapping, and input parameters of a graph attention neural network are reduced through network dimension reduction, so that the technical problems that dimension disasters are caused due to excessive input features and model parameters, the model training cost is high, the training difficulty is high, the generalization capability is poor, and further the stability judgment result is influenced due to the fact that all lines and units in the network input deep learning models extract features due to large system scale and numerous nodes are involved are solved.
The embodiment of the application provides a device for judging the transient power angle stability of a power system, which comprises a processor and a memory;
the memory is used for storing the program codes and transmitting the program codes to the processor;
the processor is configured to execute the method for determining the transient power angle stability of the power system in the foregoing method embodiments according to instructions in the program code.
The embodiment of the present application provides a computer-readable storage medium, which is used for storing program codes, and when the program codes are executed by a processor, the method for determining the transient power angle stability of a power system in the foregoing method embodiments is implemented.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described system apparatus and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The terms "first," "second," "third," "fourth," and the like in the description of the application and the above-described figures, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that, in this application, "at least one" means one or more, "a plurality" means two or more. "and/or" for describing an association relationship of associated objects, indicating that there may be three relationships, e.g., "a and/or B" may indicate: only A, only B and both A and B are present, wherein A and B may be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" or similar expressions refer to any combination of these items, including any combination of single item(s) or plural items. For example, at least one (one) of a, b, or c, may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b and c may be single or plural.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for executing all or part of the steps of the method described in the embodiments of the present application through a computer device (which may be a personal computer, a server, or a network device). And the aforementioned storage medium includes: a U disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (10)
1. A method for judging the transient power angle stability of a power system is characterized by comprising the following steps:
taking generators in an original generator graph in the power system as nodes, and clustering the nodes based on edge weights among the nodes to obtain a reduced-dimension graph;
mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping, wherein the state characteristics of the nodes in the graph after dimensionality reduction are obtained by calculating the state characteristics of all the nodes in the cluster corresponding to the nodes;
taking the state characteristics of each node in the post-dimensionality reduction graph and the post-dimensionality reduction graph as the input of a graph attention neural network, and outputting the stability judgment result of the power system;
the state characteristics comprise generator bus voltage amplitude, generator bus voltage phase angle, generator active power output, generator reactive power output, generator power angle, generator terminal voltage, converter station alternating current bus voltage and converter station direct current.
2. The method for determining the transient power angle stability of the power system according to claim 1, wherein the step of clustering nodes based on edge weights between the nodes by using the generators in the original generator graph in the power system as the nodes to obtain the dimensionality-reduced graph comprises:
taking a generator in an original generator graph in a power system as a node, and calculating first-order similarity between the nodes by taking the capacity of the generator and an inertia time constant as characteristics;
calculating improved side weights between nodes based on the first-order similarity between the nodes and initial side weights between the nodes in an adjacent matrix of the original generator graph, wherein the initial side weights between the nodes are in negative correlation with the electrical distances between the nodes;
the weight after improvement among the nodes is used as a node similarity index, and node clustering is carried out based on a hierarchical clustering algorithm to obtain a clustering result;
and mapping the nodes belonging to the same cluster into one node based on the clustering result to obtain a reduced-dimension graph.
3. The method according to claim 2, wherein the first-order similarity between nodes is calculated by the following formula:
u i ＝(S gi ,T Ji ) T /||(S gi ,T Ji ) T ||；
u j ＝(S gj ,T Jj ) T /||(S gj ,T Jj ) T ||；
in the formula, p ij Is a first order similarity between node i and node j, u i Normalized vector, S, of transient power angle stability characteristic parameter for node i gi Rated capacity, T, of generator for node i Ji Is the inertia time constant, u, of the generator corresponding to node i j Normalized vector of transient power angle stability characteristic parameter of node j, S gj Rated capacity, T, of generator for node j Jj Is the inertia time constant of the generator corresponding to node j.
4. The method according to claim 2, wherein the formula for calculating the modified weights between nodes is as follows:
in the formula, O i,j As improved back-end weights, ω, between nodes i and j ij Is an initial edge weight, p, between node i and node j ij Is the first order similarity between node i and node j.
5. The method according to claim 1, wherein the outputting the result of determining the stability of the power system with the state characteristics of each node in the post-dimensionality reduction graph and the post-dimensionality reduction graph as the input of a graph attention neural network comprises:
taking the state characteristics of each node in the post-dimensionality reduction graph and the post-dimensionality reduction graph as the input of a graph attention neural network, and calculating a similarity coefficient between each node and a neighbor node through the graph attention neural network;
calculating the attention coefficient of each node to the neighbor node according to the similarity coefficient between each node and the neighbor node and the edge weight of each node and the neighbor node in the adjacency matrix of the reduced-dimension graph through the graph attention neural network;
and performing feature fusion on the state features of the neighbor nodes of each node according to the attention coefficients of each node to the neighbor nodes through the graph attention neural network to obtain the output features of each node, and outputting the stability judgment result of the power system according to the output features of each node.
6. The method according to claim 1, wherein the training process of the graph attention neural network comprises:
acquiring a training sample, and calculating a transient stability index of a generator rotor angle of the training sample;
if the transient stability index of the generator rotor angle of the training sample is larger than 0, setting the label of the training sample as the system transient power angle stability, and if the transient stability index of the generator rotor angle of the training sample is smaller than or equal to 0, setting the label of the training sample as the system transient power angle instability;
training the graph attention neural network through the training sample.
7. An apparatus for determining transient power angle stability of an electric power system, comprising:
the clustering unit is used for taking the generators in the original generator graph in the power system as nodes and clustering the nodes based on the edge weights among the nodes to obtain a reduced-dimension graph;
the mapping unit is used for mapping the state characteristics of the nodes in the original generator graph into the state characteristics of the nodes in the graph after dimensionality reduction through dimensionality reduction mapping, wherein the state characteristics of the nodes in the graph after dimensionality reduction are obtained by calculating the state characteristics of all the nodes in the cluster corresponding to the nodes;
the judging unit is used for taking the state characteristics of each node in the dimensionality reduced graph and the dimensionality reduced graph as the input of the graph attention neural network and outputting the stability judging result of the power system;
the state characteristics comprise generator bus voltage amplitude, generator bus voltage phase angle, generator active power output, generator reactive power output, generator power angle, generator terminal voltage, converter station alternating current bus voltage and converter station direct current.
8. The apparatus according to claim 7, wherein the clustering unit is specifically configured to:
taking a generator in an original generator graph in a power system as a node, and calculating first-order similarity between the nodes by taking the capacity of the generator and an inertia time constant as characteristics;
calculating improved side weights between nodes based on the first-order similarity between the nodes and initial side weights between the nodes in an adjacent matrix of the original generator graph, wherein the initial side weights between the nodes are in negative correlation with the electrical distances between the nodes;
the weight after improvement among the nodes is used as a node similarity index, and node clustering is carried out based on a hierarchical clustering algorithm to obtain a clustering result;
and mapping the nodes belonging to the same cluster into a node based on the clustering result to obtain a graph after dimension reduction.
9. The device for judging the transient power angle stability of the power system is characterized by comprising a processor and a memory;
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the power system transient power angle stability determination method according to any one of claims 1 to 6 according to instructions in the program code.
10. A computer-readable storage medium for storing program code, which when executed by a processor implements the power system transient power angle stability determination method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211457405.9A CN115693785A (en) | 2022-11-21 | 2022-11-21 | Method for judging transient power angle stability of power system and related device thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211457405.9A CN115693785A (en) | 2022-11-21 | 2022-11-21 | Method for judging transient power angle stability of power system and related device thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115693785A true CN115693785A (en) | 2023-02-03 |
Family
ID=85054822
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211457405.9A Pending CN115693785A (en) | 2022-11-21 | 2022-11-21 | Method for judging transient power angle stability of power system and related device thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115693785A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116776131A (en) * | 2023-08-23 | 2023-09-19 | 腾讯科技（深圳）有限公司 | Feature dimension reduction method based on graph, intention rating method and related equipment |
-
2022
- 2022-11-21 CN CN202211457405.9A patent/CN115693785A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116776131A (en) * | 2023-08-23 | 2023-09-19 | 腾讯科技（深圳）有限公司 | Feature dimension reduction method based on graph, intention rating method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Thukaram et al. | Artificial neural network and support vector machine approach for locating faults in radial distribution systems | |
| Do Coutto Filho et al. | Forecasting-aided state estimation—Part I: Panorama | |
| Krishnan et al. | Efficient database generation for decision tree based power system security assessment | |
| Al Karim et al. | Dynamic event detection using a distributed feature selection based machine learning approach in a self-healing microgrid | |
| CN112383052B (en) | Power grid fault repairing method and device based on power internet of things | |
| Wu et al. | A genetic-algorithm support vector machine and DS evidence theory based fault diagnostic model for transmission line | |
| Gholami et al. | Static security assessment of power systems: A review | |
| CN102074955A (en) | Method based on knowledge discovery technology for stability assessment and control of electric system | |
| Fang et al. | A statistical approach to estimate imbalance-induced energy losses for data-scarce low voltage networks | |
| Zhang et al. | Intelligent systems for power system dynamic security assessment: Review and classification | |
| CN115693785A (en) | Method for judging transient power angle stability of power system and related device thereof | |
| CN113902946A (en) | Power system fault direction judging method and device, terminal equipment and storage medium | |
| Jamali et al. | Identification of faulted line section in microgrids using data mining method based on feature discretisation | |
| Huang et al. | An intelligent system for advanced dynamic security assessment | |
| CN115618249A (en) | Low-voltage power distribution station area phase identification method based on LargeVis dimension reduction and DBSCAN clustering | |
| CN112564107A (en) | Transient stability assessment method for power system | |
| Bhuiyan et al. | A deep learning through DBN enabled transmission line fault transient classification framework for multimachine microgrid systems | |
| CN116581890B (en) | Intelligent monitoring method and system for operation stability of power grid | |
| Azim et al. | A comparative analysis of intelligent classifiers for passive islanding detection in microgrids | |
| CN109932617B (en) | Self-adaptive power grid fault diagnosis method based on deep learning | |
| CN112595918A (en) | Low-voltage meter reading fault detection method and device | |
| Moreira et al. | Evaluation of pattern recognition algorithms for applications on power factor compensation | |
| Guo et al. | Identification of power system dynamic signature using hierarchical clustering | |
| Olaru et al. | Data mining tools and application in power system engineering | |
| CN116204771A (en) | Power system transient stability key feature selection method, device and product |
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 |