Convolutional transient stability assessment method for attention space-time diagram based on power grid structure simplification

Abstract

The invention provides an attention space-time diagram convolution transient stability assessment method based on power grid structure simplification, and the method comprises the steps: taking the voltage amplitudes and phase angles of nodes at three stages before a fault occurs, during the fault and after the fault is removed, and a power grid space structure as the input of a transient stability assessment model; based on spatial structure characteristics and electrical characteristics of the power grid, designing a TOPSIS graph simplification method based on subjective and objective weighting, compressing large-scale graph data information of the power grid, and optimizing input of a model; establishing a time-space diagram convolutional network based on an attention mechanism as a transient stability evaluation model; and inputting the voltage amplitude and the voltage phase angle of the optimized node and the power grid space structure into the transient stability evaluation model, and outputting a transient stability evaluation result of the power system. According to the method, the calculation speed and the evaluation performance of the transient stability evaluation model can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to an attention spatiotemporal graph convolution transient stability assessment method based on simplified power grid structure. Background Art

[0002] With the access of large-scale new energy equipment, power system control is becoming increasingly complex. Maintaining the transient stability of the power system has become a serious technical challenge, and accurate assessment of transient stability is particularly critical. Therefore, it is urgent to establish a reliable and fast transient stability assessment (TSA) model to identify potential transient stability risks in a timely manner to ensure the safe and stable operation of the power grid under the new situation.

[0003] Due to the complexity and uncertainty brought by high-proportion renewable energy systems, traditional transient stability assessment methods, such as time-domain simulation and direct methods, have shown obvious limitations due to their high computational complexity and poor applicability. In recent years, with the in-depth study of the powerful feature aggregation capabilities of deep learning, a large number of scholars have introduced it into transient stability assessment, including stacked autoencoders (SAE), convolutional neural networks (CNN), deep belief networks (DBN), and long short-term memory networks (LSTM). Rong et al. used cascaded convolutional neural networks to capture data features at different time intervals, successfully accelerating the efficiency of model batch evaluation while maintaining the accuracy of transient stability assessment. Wang et al. used deep belief networks to accurately model the nonlinear mapping relationship between input data and stability assessment results, significantly improving the assessment accuracy. Chen et al. integrated attention mechanisms and gated recurrent units to mine the temporal relationship of transient processes and introduced a dual attention mechanism to improve the performance and interpretability of the TSA model. Although the above studies have achieved good stability assessment results with the powerful nonlinear fitting capabilities of deep learning, these methods do not take the spatial characteristics of the power grid into account for transient stability assessment.

[0004] In fact, the topological relationship between power grid nodes and the correlation between the characteristics of adjacent nodes have a significant impact on stability assessment. The emergence of graph deep learning technology has brought a new research approach to transient stability assessment. Huang et al. introduced the topological structure of the power grid into the stability assessment model through graph convolutional neural network, fully considering the impact of topological information on transient stability during power grid operation. JHA et al. fully aggregated the spatial characteristics of the power grid topology in the transient process based on the graph attention neural network, and avoided network degradation by introducing the residual structure, thereby improving the accuracy of transient stability assessment. These studies jointly demonstrate the powerful superiority of graph deep learning technology in transient stability assessment. However, the transient process of the power system is a time series process of transition from a stable state to a new state, in which the physical quantities of the power system, such as generator power angle, voltage, current, etc., show obvious time-varying characteristics. In this process, if we can pay attention to both spatial and temporal features at the same time, and use spatial features to form certain constraints on the mutual influence between different temporal features, the performance of transient stability assessment will be significantly improved. In addition, the modern power grid is large in scale and has many nodes. How to break through the model training time limit under limited resources and conduct efficient and reliable transient stability assessment is still a challenging task. Summary of the invention

[0005] In view of this, the present invention provides an attention spatiotemporal graph convolution transient stability assessment method based on power grid structure simplification.

[0006] The present invention achieves the above technical objectives through the following technical means.

[0007] Attention spatiotemporal graph convolution transient stability assessment method based on simplified power grid structure:

[0008] Establish a spatiotemporal graph convolutional network based on the attention mechanism as a transient stability evaluation model;

[0009] The voltage amplitude and voltage phase angle of the nodes in the three stages before the fault occurs, during the fault and after the fault is removed, as well as the spatial structure of the power grid are used as the input of the transient stability assessment model, and the input is optimized in the following ways: based on the spatial structure characteristics and electrical characteristics of the power grid, a TOPSIS graph simplification method based on subjective and objective weighting is designed to compress the large-scale graph data information of the power grid and optimize the input of the transient stability assessment model;

[0010] The voltage amplitude and voltage phase angle of the optimized nodes and the spatial structure of the power grid are input into the transient stability assessment model, and the transient stability assessment result of the power system is output, that is, whether the power system is stable or unstable.

[0011] Furthermore, a spatiotemporal graph convolution network based on an attention mechanism is established, including constructing a spatiotemporal attention mechanism and a spatiotemporal graph convolution module; the graph convolution formula adopted by the spatiotemporal graph convolution module is to combine the Chebyshev polynomial with the spatiotemporal correlation intensity matrix S':

[0012]

[0013] Among them, g is the convolution kernel, x represents the characteristic vector of the power grid topology, H is the order of the Chebyshev polynomial, θ H is the Chebyshev coefficient, T H () is the H-order Chebyshev polynomial, L is the grid topology, λ max is the largest eigenvalue of L, and E is the unit matrix.

[0014] Furthermore, the TOPSIS graph simplification method based on subjective and objective weighting includes establishing an influence evaluation index system, a TOPSIS graph simplification method and a subjective and objective weighting method. The influence evaluation index system includes node closeness centrality, the rate of change of the number of node spanning trees and the electrical influence index. The TOPSIS graph simplification method uses the TOPSIS method to rank the importance of power grid nodes. The subjective and objective weighting method is used to assign weights to each influence evaluation index.

[0015] Furthermore, the node closeness centrality is:

[0016]

[0017] Where N is the number of nodes in the power grid topology, d ij is the shortest distance from node i to node j.

[0018] Furthermore, the rate of change of the number of node spanning trees is:

[0019]

[0020] Among them, π G is the number of spanning trees in the power grid topology, is the number of spanning trees after removing node i from the power grid topology.

[0021] Furthermore, the electrical influence index is:

[0022]

[0023] Among them, p Loda,i is the load carried by node i, P Gen,i is the installed capacity of node i, (p Loda +P Gen ) maxIndicates the maximum load and installed capacity borne by the node.

[0024] Furthermore, the TOPSIS graph simplification method uses the TOPSIS method to rank the importance of power grid nodes, specifically:

[0025] According to the relative closeness of each node to the ideal solution, they are ranked in order of importance, and some nodes are removed using pruning parameters. Then the topological structure and characteristic matrix of the power system are reconstructed to achieve efficient processing of large-scale graph data of the power grid.

[0026] Furthermore, the TOPSIS graph simplification method first creates a power grid node importance evaluation decision matrix and standardizes it, and then generates a weighted power grid node importance evaluation decision matrix:

[0027] Z=[z im ] N×M =[w m y m ] N×M

[0028] Among them, z im Represents y m The weighted value, y m Indicates I im The standardized value, I im is the index value of the influence evaluation index m of node i, N is the number of nodes in the power grid topology, M is the number of influence evaluation indicators, and w m The weight of the influence evaluation index m is determined by the subjective and objective weighting method, and:

[0029]

[0030] Among them, w cm is the objective weight of the mth influence evaluation index, w am is the subjective weight of the mth influence evaluation indicator.

[0031] A computer device includes a memory and a processor, wherein the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the above-mentioned attention spatiotemporal graph convolution transient stability assessment method based on power grid structure simplification.

[0032] A storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned attention spatiotemporal graph convolution transient stability assessment method based on power grid structure simplification.

[0033] The beneficial effects of the present invention are as follows: the present invention effectively compresses large-scale graph data information of the power grid, optimizes the input of the TSA model, and improves the calculation speed of the transient stability assessment model of the power system through a TOPSIS graph simplification method based on subjective and objective weighting; the present invention simultaneously extracts the spatial and temporal characteristics of the power system through a spatiotemporal graph convolutional network based on an attention mechanism, thereby improving the accuracy of the transient stability assessment of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 This is a structural diagram of the TPS-ASTGCN model described in the present invention;

[0036] FIG2( a ) is a comparison chart of the accuracy of TPS-ASTGCN of the present invention;

[0037] FIG2( b ) is a comparison chart of the time required for training the TPS-ASTGCN of the present invention;

[0038] Figure 2(c) is a comparison diagram of the parameters required for training TPS-ASTGCN described in the present invention. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0041] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0042] Example 1

[0043] Reference Figure 1, which is the first embodiment of the present invention, and provides a transient stability assessment method based on power grid structure simplification using attention spatiotemporal graph convolution (TPS-ASTGCN), including:

[0044] (1) The present invention aims to predict and evaluate the transient stability of the power grid by considering the impact of disturbances on the synchronous stability of the power grid. In order to measure this impact, sensors are used to collect the voltage amplitude and voltage phase angle of the nodes in three stages: before the fault occurs, during the fault, and after the fault is removed, and transmitted to the computer as the input features of the transient stability assessment model. The spatial topological characteristics of the power system are the key to transient stability assessment, and the spatial structure of the power grid is included in the input of the transient stability assessment model.

[0045] (2) After determining the input of the transient stability assessment model, a TOPSIS diagram simplification method based on subjective and objective weighting is designed based on the spatial structure characteristics and electrical characteristics of the power grid to compress the large-scale diagram data information of the power grid and optimize the input of the transient stability assessment model, including:

[0046] Design and establish an influence evaluation index system, and construct an influence evaluation index system for various indicators of nodes in the power grid from multiple angles and aspects from the spatial structure characteristics and electrical characteristics of the power grid;

[0047] The TOPSIS graph simplification method (TPS) was designed and established. The importance of power grid nodes was ranked using the TOPSIS method, and nodes with less influence were eliminated.

[0048] A subjective and objective weighting method is designed and established, combining the CRITIC method and the AHP (hierarchy analysis) method to assign weights to various influence evaluation indicators.

[0049] Furthermore, we designed and established an impact evaluation index system, including:

[0050] Node closeness centrality measures the average shortest distance between a node and other nodes. A high centrality means that the node is more central and closer to other nodes. In short, the higher the closeness centrality, the more prominent the node is in the power grid and the shorter the distance to other nodes. Node closeness centrality is defined as:

[0051]

[0052] Where N is the number of nodes in the power grid topology, d ij is the shortest distance from node i to node j.

[0053] In the power system, the failure of a node may cause the spanning tree of the power grid topology to decrease or even lose its connectivity. Therefore, by removing a specific node and its connected transmission line, observing the change rate of the number of spanning trees can be used to evaluate the importance of the node. Therefore, the change rate of the number of spanning trees of a node is defined as:

[0054]

[0055] Among them, π G is the number of spanning trees in the power grid topology G, is the number of spanning trees after removing node i from G.

[0056] In the power system, the importance of a node depends not only on its spatial structure, but also on the load carried by the node or the installed capacity of the generator. The importance of a node increases with the increase of the load carried by the node or the installed capacity of the generator. Therefore, nodes with larger load or installed capacity of generators are considered more critical, and their reliability and stability have a greater impact on the operation of the entire power system. The electrical influence index f is defined as:

[0057]

[0058] Among them, p Loda,i is the load carried by node i, P Gen,i is the installed capacity of node i, (p Loda +P Gen ) max Indicates the maximum load and installed capacity borne by the node.

[0059] Furthermore, a TOPSIS graph simplification method is designed and established, including:

[0060] The TOPSIS method is a multi-attribute decision-making method used to evaluate the similarity of power grid nodes relative to the ideal target and rank them accordingly. In this method, the ideal solution consists of the optimal values ​​of all evaluation indicators, while the anti-ideal solution consists of the worst values ​​of all indicators. The most important node is the node closest to the ideal solution and farthest from the anti-ideal solution, and vice versa, it is the least important node.

[0061] The specific steps of the TOPSIS graph simplification method are as follows:

[0062] a. Create a grid node importance assessment decision matrix and standardize it:

[0063] I=[I1,I2,…,I M ]=[I im ] N×M

[0064]

[0065] Among them, I im is the index value of the influence evaluation index m of node i, M is the number of influence evaluation indexes, and in the present invention, M=3, namely, the change rate of the node closeness centrality, the number of node spanning trees, and the electrical influence index; y m Indicates I im The standardized value, I M Represents the index value of the influence evaluation index M of all nodes.

[0066] b. Generate a weighted grid node importance evaluation decision matrix:

[0067] Z=[z im ] N×M =[w m y m ] N×M

[0068] Among them, w m The weight of the influence evaluation index m determined by the subjective and objective weighting method is specifically determined by the subjective and objective weighting method, which will be described in the next section; im Represents y m Weighted value.

[0069] c. Determine the ideal solution to the node importance assessment problem and anti-ideal solution Among them, the ideal solution of the mth influence evaluation index is The anti-ideal solution of the mth influence evaluation index is

[0070] d. Calculate the Euclidean distance between each node of the power grid and the ideal solution and anti-ideal solution:

[0071]

[0072] e. Calculate the relative closeness of each node of the power grid to the ideal solution:

[0073]

[0074] f. Sort the importance of each node according to its relative proximity to the ideal solution and determine the importance of the node. Using the set pruning parameter ξ, selectively retain nodes with higher influence and remove nodes with less influence to ensure the connectivity of the spatial topology of the power grid.

[0075] g. After removing nodes with less impact, the topological structure and characteristic matrix of the power system are reconstructed to achieve efficient processing of large-scale graph data of the power grid.

[0076] Furthermore, a subjective and objective weighting method is designed and established, including:

[0077] The CRITIC method is a decision-making method for objectively assigning weights. Its core idea is to evaluate the comparative strength of influence evaluation indicators by comparing the Gini coefficients between influence evaluation indicators, and to use the Kendall coefficient to measure the degree of conflict between influence evaluation indicators, so as to determine the objective weight of the evaluation indicators.

[0078] Specifically, the Gini coefficient is used to measure the value differences between evaluation indicators. The Gini coefficient of the mth influence evaluation indicator is defined as:

[0079]

[0080] The Kendall coefficient is used to measure the correlation between two evaluation indicators. It measures the consistency between the levels of the two indicators, rather than the specific values. The Kendall coefficient between the mth and tth influence evaluation indicators is defined as:

[0081]

[0082] Among them, Z' m With Z' t are two columns of values ​​obtained by arranging the mth and tth evaluation indicators according to their numerical values, N ws 、N wd Respectively represent Z' m , Z' t The number of equal index values ​​and the number of unequal index values ​​at the same position in the

[0083] The overall Kendall correlation coefficient between the mth influence evaluation index and other influence evaluation indicators is defined as:

[0084]

[0085] Therefore, by adopting the CRITIC method, combining the Gini coefficient and the Kendall coefficient, the objective weight of the mth influence evaluation index is determined as:

[0086]

[0087] The electrical characteristics of nodes (loads carried by nodes or installed capacity of generators) play a key role in power systems. However, it is difficult to fully measure the trade-offs of impact evaluation indicators through objective weighting methods. To make up for this deficiency, the analytic hierarchy process (AHP) is introduced.

[0088] The AHP method is a decision-making method for subjectively assigning weights. It is a multi-criteria decision-making method that combines qualitative and quantitative methods and is used to determine the subjective weights of various influence evaluation indicators.

[0089] Specifically, we first compare different influence evaluation indicators in pairs and construct a judgment matrix A according to a nine-level scale from 1 to 9.

[0090] Secondly, according to the eigenvector method, the maximum eigenvalue λ of the judgment matrix A is obtained max And the corresponding eigenvector w a =[w a1 ,w a2 ,…,w aM ].

[0091] Finally, a consistency test is performed. If CR < 0.10, the consistency test is passed. a As the subjective weight of each influence evaluation index, the subjective weight of the mth influence evaluation index is w am ; Otherwise, reconstruct the judgment matrix A and repeat the above steps. RI is the random consistency index.

[0092] In summary, considering the comprehensive factors of the spatial structure and electrical characteristics of the power grid, the present invention adopts the CRITIC method and the AHP method to construct a subjective and objective weighting method, and defines the comprehensive weight of the mth influence evaluation index as:

[0093]

[0094] (3) Design and establish an attention-based spatiotemporal graph convolutional network (ASTGCN) as a transient stability assessment model; including:

[0095] Design and build a spatiotemporal attention mechanism;

[0096] Design and build a spatiotemporal graph convolution module.

[0097] Furthermore, we design and establish a spatiotemporal attention mechanism, including:

[0098] In order to more accurately model the characteristics of transient processes, the present invention introduces an attention mechanism to capture spatiotemporal correlation characteristics.

[0099] In the time dimension, electrical data in different time periods are correlated, and time-correlation weights can be assigned to the data through time attention, as shown below:

[0100] W=V t ·σ((X T U1)U2(U3X)+b t )

[0101]

[0102]

[0103] in, is the weight matrix of time correlation; is the feature matrix adjusted in the time dimension; V t , b t , U1, U2 and U3 are all learnable parameters; σ is the activation function; t * is the input time series length; X represents the sample feature matrix; W represents the data time correlation weight matrix; W i,j is an element in W; W′ i，j Indicates normalization of W; represents the time series t * elements.

[0104] In the spatial dimension, the features of different nodes in the power grid influence each other, and the weight matrix is ​​constructed for different nodes through spatial attention, as shown below:

[0105]

[0106] Among them, S∈R N×N is the spatiotemporal attention matrix; V s , b s , Q1, Q2 and Q3 are all learnable parameters; the spatiotemporal attention matrix S is obtained by Calculated; S' i,j represents the temporal and spatial correlation strength between node i and node j; S i,j is an element of S.

[0107] Design and build a spatiotemporal graph convolution module, including:

[0108] Traditional CNN is mainly used to process Euclidean spatial data, and is particularly good at extracting features of two-dimensional tensor data, while GCN (graph convolutional neural network) is suitable for processing graph structure data. Therefore, the present invention uses GCN to extract the spatial features of each node in the power grid graph structure.

[0109] Using the normalized Laplace matrix, the grid topology is defined as:

[0110]

[0111] Among them, E is the identity matrix, D is the degree matrix of the power grid topology structure, and A is the adjacency matrix of the power grid topology structure.

[0112] By performing characteristic decomposition on L, we can obtain L = UΛU T, U is a matrix of eigenvectors, and satisfies UU T =E, Λ = [λ1,λ2,...,λ n ] is the matrix composed of the eigenvalues ​​of L, and n is the total number of features.

[0113] Based on U, the graph convolution formula is:

[0114]

[0115] Among them, X is the input of the spatial convolution layer, * is the convolution operation, g is the convolution kernel, is the Hadamard product.

[0116] Using g θ =U T g is used as a learnable convolution kernel, and the graph convolution formula is written as:

[0117] (X*g)=Ug θ U T X

[0118] In order to speed up the calculation of the above formula, the Chebyshev polynomial is used to approximate the convolution kernel, and the convolution kernel can be obtained as follows:

[0119]

[0120] in, λ max is the largest eigenvalue of L, θ H is the Chebyshev coefficient, T H (x) is the H-order Chebyshev polynomial, expressed as:

[0121] T H (x) = 2xT H-1 (x)-T H-2 (x)

[0122] Wherein, T0(x)=1, T1(x)=x, and x represents the characteristic vector of the power grid topology structure.

[0123] In order to dynamically adjust the correlation between adjacent nodes, the Chebyshev polynomial is combined with the spatiotemporal correlation intensity matrix S' (where the components are S i ',j), the final graph convolution formula is:

[0124]

[0125] After the spatiotemporal graph convolution module captures the information on the spatial dimension of the node, temporal convolution is performed by merging the information on adjacent time periods to further update the temporal information of the node:

[0126]

[0127] Among them, φ is the parameter of the convolution kernel in the time dimension, and RELU is the activation function.

[0128] After that, it passes through the fully connected layer Linear and obtains the final output result through linear transformation, as shown in the following formula:

[0129] X″=Linear(RELU(x))

[0130] (4) The voltage amplitude and voltage phase angle of the optimized nodes and the spatial structure of the power grid are input into the transient stability assessment model constructed in (3), and the transient stability assessment result of the power system is output, that is, whether the power system is stable or unstable.

[0131] The steps of the TSA assessment method (i.e. the operation process of the transient stability assessment model) mainly include:

[0132] TOPSIS graph simplification stage: The TOPSIS graph simplification method based on subjective and objective weighting is used to compress the large-scale graph data of the power grid and ensure the connectivity of the spatial topology of the power grid. If the pruning of a node causes the topology of the power grid to no longer be connected, it is necessary to backtrack and undo the last pruning, and then continue to try the pruning of the next node. Finally, the topology and characteristic matrix of the power system are reconstructed.

[0133] Offline training: First, the feature matrix needs to be preprocessed, including normalization and other operations. Then, it is divided into sample training sets and test sets according to a certain ratio. Then, the training set data is input into ASTGCN for training using an appropriate batch size. After each training iteration, the model is tested using the test set data, and the training effect of the network is improved by adjusting the parameters. By continuously optimizing and iterating the network, it aims to achieve better performance evaluation results.

[0134] This embodiment also provides a computer device, which is suitable for the case of an attention spatiotemporal graph convolution transient stability assessment method based on simplified power grid structure, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the attention spatiotemporal graph convolution transient stability assessment method based on simplified power grid structure as proposed in the above embodiment.

[0135] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen. The input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0136] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for transient stability assessment based on power grid structure simplification using attention spatiotemporal graph convolution is implemented as proposed in the above embodiment.

[0137] Example 2

[0138] Reference Figure 1 , which is the second embodiment of the present invention, provides a transient stability assessment method based on attention spatiotemporal graph convolution based on simplified power grid structure. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0139] Exemplarily, the evaluation performance test and analysis of TPS-ASTGCN of the present invention can be performed by the following steps: In order to evaluate the superiority of TPS-ASTGCN for TSA, TPS-ASTGCN is compared with GCN, LSTM and SVM (support vector machine) for performance. Specifically, SVM adopts radial basis kernel function and expands sample feature data into a one-dimensional form that meets its requirements. The network structure adopted by LSTM includes 2 LSTM layers, a fully connected layer and a Softmax output layer. The network structure adopted by GCN includes 2 stacked GCN graph convolution layers, a fully connected layer and a Softmax output layer. The input of GCN consists of the adjacency matrix of the power grid graph structure and the feature data of the sample. The above-mentioned deep neural network models (i.e., TPS-ASTGCN, GCN, and LSTM) all adopt cross entropy loss function and Adam optimization algorithm. The pruning parameter ξ in the TOPSIS graph simplification method in TPS-ASTGCN is set to 0.4, which is the result of multiple experiments. Specifically, the performance test results of the four models are shown in Table 1.

[0140] Table 1 Average results of each index of the four models

[0141]

[0142] As can be seen from Table 1, compared with models such as GCN, LSTM and SVM, the TPS-ASTGCN model proposed in the present invention shows better performance in four evaluation indicators, namely, accuracy (Acc), precision (Prec), recall (Rec) and f1-score. As a shallow model, SVM can only capture the shallow features of the data and it is difficult to fully explore the deep features. Therefore, its prediction performance is significantly inferior to that of TPS-ASTGCN, GCN and LSTM models. The GCN model achieved an accuracy of 98.63% by effectively extracting the complex topological structure of the power grid, which is significantly better than the LSTM model. The TPS-ASTGCN model achieves rapid and accurate judgment of the transient stability of the system by simultaneously extracting the spatial structure and time series flow characteristics of the power grid during the transient process, and all indicators are significantly better than GCN. This means that in the learning process, the TPS-ASTGCN model can adaptively learn the importance weights between different nodes and time steps, thereby improving the characterization learning ability of the complex structure and characteristics of the spatiotemporal graph.

[0143] In order to verify the effectiveness of TPS graph simplification, TPS-ASTGCN and ASTGCN were used to predict transient stability respectively. Each model was iterated for 100 rounds. cc ), model training time (Time), and the number of parameters required for model training before and after graph simplification (params). The specific results are shown in Figure 2(a), (b), and (c).

[0144] In the results shown in Figure 2(a), (b), and (c), the TPS-ASTGCN model shows a significant improvement in training efficiency compared to the ASTGCN model. By optimizing the graph structure of the power system (model input), TPS effectively reduces the number of parameters of the ASTGCN model, thereby reducing the burden of computing and storage. Specifically, the TPS-ASTGCN model achieved a 16.67% increase in training speed, significantly accelerating the TSA model training process. At the same time, Figure 2(a) also reveals the potential of TPS technology in improving model prediction accuracy. The prediction accuracy of the TPS-ASTGCN model has increased by 0.22%. This improvement is attributed to the optimization of the power grid graph structure by TPS technology, which removes redundant information and enables the model to focus more on the learning and analysis of key features. By eliminating unnecessary information, the model can more accurately identify the core characteristics and dynamic laws of the power system, thereby enhancing the accuracy of the prediction.

[0145] Comprehensive analysis shows that the model using TPS technology shows obvious advantages in transient stability prediction tasks. TPS not only accelerates the model training process and reduces the demand for computing and storage resources, but also improves the accuracy of predictions, making task execution more efficient. In addition, the application of TPS technology also reduces the amount of node data that needs to be collected during data loading, thereby significantly shortening the data loading time.

[0146] The embodiments are preferred implementations of the present invention, but the present invention is not limited to the above-mentioned implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A transient stability assessment method based on attention spatiotemporal graph convolution based on simplified power grid structure, characterized by: Establish a spatiotemporal graph convolutional network based on the attention mechanism as a transient stability evaluation model; The voltage amplitude and voltage phase angle of the nodes in the three stages before the fault occurs, during the fault and after the fault is removed, as well as the spatial structure of the power grid are used as the input of the transient stability assessment model, and the input is optimized in the following ways: based on the spatial structure characteristics and electrical characteristics of the power grid, a TOPSIS graph simplification method based on subjective and objective weighting is designed to compress the large-scale graph data information of the power grid and optimize the input of the transient stability assessment model; The voltage amplitude and voltage phase angle of the optimized nodes and the spatial structure of the power grid are input into the transient stability assessment model, and the transient stability assessment result of the power system is output, that is, whether the power system is stable or unstable.

2. The attention spatiotemporal graph convolution transient stability assessment method according to claim 1, characterized in that: A spatiotemporal graph convolution network based on an attention mechanism is established, including constructing a spatiotemporal attention mechanism and a spatiotemporal graph convolution module; the graph convolution formula adopted by the spatiotemporal graph convolution module is to combine the Chebyshev polynomial with the spatiotemporal correlation intensity matrix S': Among them, g is the convolution kernel, x represents the characteristic vector of the power grid topology, H is the order of the Chebyshev polynomial, θ H is the Chebyshev coefficient, T H () is the H-order Chebyshev polynomial, L is the grid topology, λ max is the largest eigenvalue of L, and E is the unit matrix.

3. The attention spatiotemporal graph convolution transient stability assessment method according to claim 1, characterized in that: The TOPSIS graph simplification method based on subjective and objective weighting includes establishing an influence evaluation index system, a TOPSIS graph simplification method and a subjective and objective weighting method. The influence evaluation index system includes node closeness centrality, the change rate of the number of node spanning trees and an electrical influence index. The TOPSIS graph simplification method uses the TOPSIS method to rank the importance of power grid nodes. The subjective and objective weighting method is used to assign weights to various influence evaluation indicators.

4. The attention spatiotemporal graph convolution transient stability assessment method according to claim 3 is characterized in that: The node closeness centrality is: Where N is the number of nodes in the power grid topology, d ij is the shortest distance from node i to node j.

5. The attention spatiotemporal graph convolution transient stability assessment method according to claim 3, characterized in that: The rate of change of the number of node spanning trees is: Among them, π G is the number of spanning trees in the power grid topology, is the number of spanning trees after removing node i from the power grid topology.

6. The attention spatiotemporal graph convolution transient stability assessment method according to claim 3, characterized in that: The electrical influence index is: Among them, p Loda,i is the load carried by node i, P Gen,i is the installed capacity of node i, (p Loda +P Gen ) max Indicates the maximum load and installed capacity borne by the node.

7. The attention spatiotemporal graph convolution transient stability assessment method according to claim 3, characterized in that: The TOPSIS graph simplification method uses the TOPSIS method to sort the importance of power grid nodes, specifically: According to the relative closeness of each node to the ideal solution, they are ranked in order of importance, and some nodes are removed using pruning parameters. Then the topological structure and characteristic matrix of the power system are reconstructed to achieve efficient processing of large-scale graph data of the power grid.

8. The attention spatiotemporal graph convolution transient stability assessment method according to claim 7, characterized in that: The TOPSIS graph simplification method first creates a power grid node importance evaluation decision matrix and standardizes it, and then generates a weighted power grid node importance evaluation decision matrix: From=[from im ] N×M =[in m and m ] N×M Among them, z im Represents y m The weighted value, y m Indicates I im The standardized value, I im is the index value of the influence evaluation index m of node i, N is the number of nodes in the power grid topology, M is the number of influence evaluation indicators, and w m The weight of the influence evaluation index m is determined by the subjective and objective weighting method, and: Among them, w cm is the objective weight of the mth influence evaluation index, w am is the subjective weight of the mth influence evaluation indicator.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the attention spatiotemporal graph convolution transient stability assessment method based on power grid structure simplification as described in any one of claims 1-8.

10. A storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the attention spatiotemporal graph convolution transient stability assessment method based on power grid structure simplification as described in any one of claims 1 to 8 is implemented.

Images