Method and apparatus for evaluating operational reliability, electronic device and computer storage medium

Abstract

The application provides a method and device for evaluating operation reliability, electronic equipment and computer storage medium; after obtaining a system state to be evaluated, the system state to be evaluated is input into a prediction model to output a minimum load shedding amount of a power system; and the operation reliability evaluation result is determined according to the minimum load shedding amount of the power system. The prediction model is a multi-core collaborative graph convolutional neural network; the extraction capability of the neural network for node features, topological features and physical correlation features is improved at the architecture level, so that the accuracy of the neural network for minimum load shedding calculation is effectively improved, and the adaptability to a complex variable topological scene during operation reliability evaluation is improved.

Description

Technical Field

[0001] This application relates to the field of power systems and their automation technology, and in particular to a method, apparatus, electronic device and computer storage medium for evaluating operational reliability. Background Technology

[0002] Operational reliability assessment is an important tool for assessing and providing early warning of operational risks in power systems. Against the backdrop of the vigorous development of renewable energy worldwide, power systems face significant uncertainties due to the intermittent and random nature of wind and solar power, necessitating real-time operational reliability assessments.

[0003] However, with the collection of reliability data for critical power equipment, both traditional and data-driven methods for assessing operational reliability have faced challenges. On one hand, traditional methods, such as Monte Carlo simulations, require iteratively solving a large number of minimum load shedding optimization problems under numerous system states. However, in real-world large-scale power systems, this computational complexity is high and time-consuming, with the accumulated computational burden hindering online reliability assessment. On the other hand, the collection of extensive data from critical power equipment necessitates simulating more complex and diverse system states. In particular, scenarios involving numerous topological changes will emerge, making neural network training more difficult, and data-driven methods, despite their high computational speed, also face adaptability issues. Summary of the Invention

[0004] In view of this, this application provides an operational reliability assessment method, apparatus, electronic device, and computer storage medium, which effectively improves the accuracy of neural networks in calculating minimum load shedding and enhances the adaptability of operational reliability assessment to complex variable topology scenarios.

[0005] The first aspect of this application provides a method for evaluating operational reliability, including:

[0006] Obtain the system status to be evaluated;

[0007] The system state to be evaluated is input into the prediction model, and the minimum load shedding amount of the power system is output; wherein, the prediction model is a multi-kernel cooperative graph convolutional neural network;

[0008] The operational reliability assessment results are determined based on the minimum load shedding of the power system.

[0009] Optionally, the method for constructing the prediction model includes:

[0010] Construct a training sample set;

[0011] Establish an initial minimum load shear model;

[0012] Construct a graph convolutional layer that is precisely embedded within the initial minimum shear load model;

[0013] A multi-kernel collaborative graph convolutional neural network is established based on the graph convolutional layers described above;

[0014] The multi-kernel collaborative graph convolutional neural network is trained using the training sample set to obtain a prediction model.

[0015] Optionally, the operational reliability assessment method further includes:

[0016] In determining the operational reliability assessment results based on the minimum load shedding of the power system, convergence criteria are used to determine the convergence based on the maximum sample size and the variance of the reliability index.

[0017] Optionally, constructing the training sample set includes:

[0018] Based on typical power system operating scenarios, the power system status is obtained through Monte Carlo sampling.

[0019] For each power system state, the minimum load shedding amount under that power system state is solved using the interior point method;

[0020] A training sample set is constructed based on the minimum load shedding amount under all the described power system states.

[0021] Optionally, constructing the graph convolutional layer precisely embedded in the initial minimum shear load model includes:

[0022] The Gauss-Seidel iteration of the power flow equation is embedded into the graph convolution to construct the graph convolution layer that is precisely embedded in the initial minimum shear load model.

[0023] Optionally, the step of establishing a multi-kernel collaborative graph convolutional neural network based on the graph convolutional layer includes:

[0024] The forward propagation formula is determined based on the node feature convolution and node feature neighborhood aggregation calculation functions.

[0025] The first AGGRE function is determined based on the adjacency matrix;

[0026] The second AGGRE function is determined based on the impedance-based aggregated weight matrix;

[0027] The aggregated weight matrix of the impedance is standardized to obtain the aggregated weight matrix of the target impedance;

[0028] The third AGGRE function is determined based on the aggregated weight matrix of the target impedance;

[0029] Based on the transformed power flow equations, the fourth AGGRE function is determined;

[0030] The first, second, third, and fourth AGGRE functions are dynamically weighted and merged to obtain a multi-kernel collaborative graph convolutional neural network.

[0031] A second aspect of this application provides an operational reliability assessment apparatus, comprising:

[0032] The acquisition unit is used to acquire the state of the system to be evaluated.

[0033] An analysis unit is used to input the system state to be evaluated into a prediction model and output the minimum load shedding amount of the power system; wherein, the prediction model is a multi-kernel cooperative graph convolutional neural network;

[0034] An evaluation unit is used to determine the operational reliability evaluation result based on the minimum load shedding of the power system.

[0035] Optionally, the building blocks of the prediction model include:

[0036] Training sample set construction unit, used to construct the training sample set;

[0037] Minimum load shearing model establishment unit, used to establish the initial minimum load shearing model;

[0038] The graph convolutional layer construction unit is used to construct the graph convolutional layer precisely embedded in the initial minimum shear load model;

[0039] The graph convolutional neural network establishment unit is used to establish a multi-kernel collaborative graph convolutional neural network based on the graph convolutional layer.

[0040] The training unit is used to train the multi-kernel collaborative graph convolutional neural network using the training sample set to obtain a prediction model.

[0041] Optionally, the operational reliability assessment device further includes:

[0042] The convergence judgment unit is used to determine the convergence criterion based on the maximum sample size and the variance of the reliability index during the process of determining the operational reliability assessment result based on the minimum load shedding of the power system.

[0043] Optionally, the training sample set construction unit includes:

[0044] The sampling unit is used to obtain the power system status through Monte Carlo sampling based on typical power system operating scenarios;

[0045] The solution unit is used to solve for the minimum load shedding amount under each power system state using the interior point method;

[0046] A training sample set construction sub-unit is used to construct a training sample set based on the minimum load shedding amount under all the power system states.

[0047] Optionally, the graph convolutional layer construction unit includes:

[0048] The graph convolutional layer construction subunit is used to embed the Gauss-Seidel iteration of the power flow equation into the graph convolution, thereby constructing the graph convolutional layer that is precisely embedded in the initial minimum shear load model.

[0049] Optionally, the graph convolutional neural network building unit includes:

[0050] The first determining unit is used to determine the forward propagation formula based on the node feature convolution and node feature neighborhood aggregation calculation functions.

[0051] The second determining unit is used to determine the first AGGRE function based on the adjacency matrix;

[0052] The third determining unit is used to determine the second AGGRE function based on the impedance-based aggregated weight matrix;

[0053] A normalization unit is used to normalize the aggregated weight matrix of the impedance to obtain the aggregated weight matrix of the target impedance.

[0054] The fourth determining unit is used to determine the third AGGRE function based on the aggregated weight matrix of the target impedance;

[0055] The fifth determining unit is used to determine the fourth AGGRE function based on the transformed power flow equations;

[0056] The dynamic weighted merging unit is used to dynamically weight and merge the first, second, third, and fourth AGGRE functions to obtain a multi-kernel collaborative graph convolutional neural network.

[0057] A third aspect of this application provides an electronic device, comprising:

[0058] One or more processors;

[0059] A storage device on which one or more programs are stored;

[0060] When the one or more programs are executed by the one or more processors, the one or more processors implement the operational reliability assessment method as described in any one of the first aspects.

[0061] The fourth aspect of this application provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the operational reliability assessment method as described in any one of the first aspects.

[0062] As can be seen from the above solutions, this application provides a method, apparatus, electronic device, and computer storage medium for operational reliability assessment. After obtaining the system state to be assessed, the system state is input into a prediction model, and the minimum load shedding amount of the power system is output. Based on the minimum load shedding amount of the power system, the operational reliability assessment result is determined. The prediction model of this application is a multi-core collaborative graph convolutional neural network. At the architectural level, it improves the neural network's ability to extract node features, topological features, and their physical correlation features, thereby effectively improving the accuracy of the neural network in calculating the minimum load shedding amount and enhancing its adaptability to complex variable topology scenarios during operational reliability assessment. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0064] Figure 1 A detailed flowchart of an operational reliability assessment method provided in this application embodiment;

[0065] Figure 2 A flowchart illustrating a method for constructing a prediction model, as provided in another embodiment of this application;

[0066] Figure 3 A flowchart of a method for establishing a multi-kernel collaborative graph convolutional neural network is provided for another embodiment of this application;

[0067] Figure 4 A diagram illustrating a multi-core collaborative graph convolutional neural network architecture is provided in another embodiment of this application.

[0068] Figure 5 Error curves of the training set during training of different neural networks are provided in another embodiment of this application;

[0069] Figure 6 An error curve for a test set during training of different neural networks is provided in another embodiment of this application;

[0070] Figure 7 The intention value of node feature e in a multi-kernel collaborative graph convolutional neural network provided in another embodiment of this application;

[0071] Figure 8 The intention value of node feature f in a multi-kernel collaborative graph convolutional neural network provided in another embodiment of this application;

[0072] Figure 9 This application provides an operational reliability assessment result at different times, as another embodiment of the present application.

[0073] Figure 10 A schematic diagram of an operational reliability assessment device provided in another embodiment of this application;

[0074] Figure 11 This is a schematic diagram of an electronic device that implements an operational reliability assessment method, as provided in another embodiment of this application. Detailed Implementation

[0075] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0076] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0077] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0078] It should be noted that the concepts of "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0079] It should be noted that the terms "a" and "a plurality of" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0080] This application provides a method for evaluating operational reliability, such as... Figure 1 As shown, the specific steps include:

[0081] S101. Obtain the system state to be evaluated.

[0082] In the specific implementation process of this application, the system state can be sampled. The system state is usually sampled according to the load and the probability distribution of renewable energy, as well as the failure probability of important power equipment, and the system state to be evaluated is extracted, which is not limited here.

[0083] Among them, the system state includes but is not limited to equipment state, load level, new energy output, etc., which is not limited here.

[0084] S102. Input the system state to be evaluated into the prediction model, and output the minimum load shedding amount of the power system.

[0085] Among them, the prediction model is a multi-core collaborative graph convolutional neural network.

[0086] Optionally, in another embodiment of this application, an implementation manner of the construction method of the prediction model is as follows Figure 2 shown, including:

[0087] S201. Construct a training sample set.

[0088] In the specific implementation process of this application, it can but is not limited to obtain the power system state through Monte Carlo sampling according to the typical operation scenarios of the power system. Then, for each power system state, solve the minimum load shedding amount under the power system state through the interior point method. Finally, construct a training sample set according to the minimum load shedding amounts under all power system states, which is not limited here.

[0089] S202. Establish an initial minimum load shedding model.

[0090] In the specific implementation process of this application, the establishment of the initial minimum load shedding model can be as follows:

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] in, and These represent the active and reactive load reduction amounts at node i, respectively. The cost coefficient is * (in the formula above) For example, it represents the active power cost coefficient of bus node i). and These represent the active and reactive loads of node i, respectively. and The active and reactive power output of the generator connected to node i; and These are the voltage magnitude and phase angle at node i, respectively; Let be the phase angle difference between the i-th node and the j-th node; and These represent the elements in the i-th row and j-th column of the nodal conductance matrix and the nodal susceptance matrix, respectively. This represents the active power between the i-th node and the j-th node; Indicates the generator status; This represents the active power output of generator i at the previous moment; This represents the maximum gradeability of generator i; , , These are sets of generators, nodes, and branches, respectively. and These represent the upper and lower limits of *, respectively (e.g.: This represents the upper limit of active power between the i-th node and the j-th node. The lower limit of the active power between the i-th node and the j-th node.

[0102] S203. Construct an accurate graph convolutional layer embedded in the initial minimum shear load model.

[0103] It should be noted that in order to achieve accurate embedding of the minimum initial minimum shear load model, it is necessary to maintain physical rules in the graph convolution feature aggregation calculation. Therefore, the derivation based on the power flow model cannot make any assumptions. Thus, in the specific implementation process of this application, Gaussian-Seidel iteration of the power flow equation can be embedded into the graph convolution to construct a graph convolution layer with accurate embedding of the minimum initial minimum shear load model.

[0104] Among these methods, the Gauss-Seidel iteration does not violate the initial power flow equations, and the physical rules are well preserved. More importantly, it can be directly applied to solve the power flow equations.

[0105] The complex form of the power flow model is as follows:

[0106] (1)

[0107] Where Y is the node admittance matrix in complex form; V is the node voltage in complex form; I is the node injected current in complex form; and S is the node injected power in complex form.

[0108] Decompose matrix Y into its diagonal form, denoted as the diagonal element matrix. A matrix of off-diagonal elements is denoted as Equation (1) can be rewritten as Equation (2). Therefore, the i-th node's The Gauss-Seidel iterative function can be rewritten as equation (3).

[0109] (2)

[0110] (3)

[0111] in, and The first Second and third The next iteration ; Let Y be the element in the i-th row and j-th column of the admittance matrix; Inject the i-th element of the power vector S into the node; Let represent the set connected to node i. is the element in the i-th row and i-th column of the admittance matrix Y.

[0112] It can be seen that Equation (3) is a node feature aggregation function using system state (node ​​injected power S) and grid parameters (conductance and susceptance matrices G and B). No assumptions are made during the derivation process, and Equation (3) still satisfies physical laws (such as Kirchhoff's Current Law (KCL) and Kirchhoff's Voltage Law (KVL)). Therefore, this invention uses Equation (3) as a neighborhood feature aggregation function in graph convolution and designs a new graph convolution method. However, all elements in Equation (3) are complex numbers, making it difficult to directly utilize in the forward propagation of neural networks. This invention decouples node features in a Cartesian coordinate system, where... Then the node feature aggregation function is expressed as equation (4), and the expressions for its real part and imaginary part are shown in equation (5).

[0113] (4)

[0114] (5)

[0115] in, and It is the first Features of node i in the next iteration; and Let i represent the i-th row and i-th column elements of the conductance matrix G and the susceptance matrix B, respectively; and The feature representing the clustering of neighboring nodes from node i is represented as:

[0116] (6)

[0117] in, and They are The real and imaginary parts.

[0118] In equations (5) and (6), and Located in the denominator. For different nodes in a given topology It is a non-zero constant. But and This indicates that the output feature of the hidden layer may be zero, leading to errors in the forward propagation calculation of the neural network. To avoid this situation, this invention... A smaller constant was added Finally, the node feature neighborhood aggregation function in graph convolution... and The matrix form can be written as:

[0119] (7)

[0120] (8)

[0121] (9)

[0122] Where P and Q represent the active and reactive power injected into the node, respectively; It is the Hadamard product; and These represent the nodal conductance and susceptance matrices, respectively, without diagonal elements; and These are the diagonal matrices representing the nodal conductance and susceptance matrices, respectively. and Indicates the first Feature vectors of coupled nodes in a layer.

[0123] S204. Establish a multi-kernel collaborative graph convolutional neural network based on the graph convolutional layers.

[0124] Optionally, in another embodiment of this application, one implementation of step S204 is as follows: Figure 3 As shown, it includes:

[0125] S301. Determine the forward propagation formula based on the node feature convolution and node feature neighborhood aggregation calculation functions.

[0126] Understandably, graph convolution calculation differs from traditional matrix or image convolution. It decomposes convolution into two steps: node feature neighborhood aggregation and node feature convolution calculation, thus handling the issue of inconsistent node degrees in a graph. Its forward propagation formula can be expressed as:

[0127] ;

[0128] in, Indicates the first Features of node i in a layered neural network; N(i) represents the set of neighboring nodes of node i and itself; and These are the node feature convolution and node feature neighborhood aggregation calculation functions, respectively.

[0129] S302. Determine the first AGGRE function based on the adjacency matrix.

[0130] To reduce the number of trainable parameters in graph convolutional neural networks (GNNs), most AGGRE functions use only known parameters, making the determination of the AGGRE function a primary task in constructing GNNs. The AGGRE function for the first type of GNN (the first AGGRE function) is constructed based on the adjacency matrix A, and the weights of the features of neighboring nodes are determined using spectral graph theory. Its matrix form mathematical formula can be expressed as:

[0131] ;

[0132] in, It is an input feature with n nodes and k-dimensional features; and Represents the standardized adjacency matrix and node degree matrix. , .

[0133] S303. Determine the second AGGRE function based on the impedance-based aggregated weight matrix.

[0134] In power system applications, graph convolutional neural networks often use physical parameters and models that characterize the influence of neighboring nodes to construct AGGRE functions, such as branch impedance and power flow models. First, aggregate weights for neighborhood features are constructed using branch impedance. This type of AGGRE method employs a Gaussian kernel. ,in This is the scaling factor. It needs to be adjusted in different power systems. To ensure that the path weights are within a reasonable range, preventing paths from being ignored due to excessively small weights, the aggregation function (second AGGRE function) of the second type of graph convolutional neural network can be expressed as:

[0135] (10)

[0136] Where W is a weight matrix calculated using line impedance (an aggregated weight matrix of impedance).

[0137] S304. Standardize the aggregated weight matrix of the impedance to obtain the aggregated weight matrix of the target impedance.

[0138] To further enhance the feature extraction capability of neural networks, the third type of graph convolutional neural network, based on the impedance aggregated weight matrix W, further employs the normalization processing techniques of the first type of graph convolutional neural network to process the W matrix, thereby obtaining the aggregated weight matrix of the target impedance. :

[0139] .

[0140] S305. Determine the third AGGRE function based on the aggregated weight matrix of the target impedance.

[0141] Continuing with the above example, the third AGGRE function can be expressed as:

[0142] (11)

[0143] in, .

[0144] S306. Based on the transformed power flow equations, determine the fourth AGGRE function.

[0145] To further enhance the feature extraction capability of the neural network, the fourth type of graph convolutional neural network also embeds a power flow model in the Cartesian coordinate system. This model will... and Let the features of the coupled nodes in the graph be defined, and update them using a power flow model as a neighborhood aggregation function. Here, it is mainly assumed that the features of the central node i are unknown and updated by its neighborhood, thus including... or The terms are retained on the right side, and other elements are moved to the left side. This is achieved by solving the transformed power flow equations. and The fourth node feature aggregation function (the fourth AGGRE function) is derived as follows:

[0146] ;

[0147] (12)

[0148] in, and It is the first Feature vectors of coupled nodes in the layer; It is a Hadamard product, and:

[0149] ;

[0150] ;

[0151] ;

[0152] ;

[0153] in, and These are nodal conductance and nodal susceptance matrices without diagonal elements; and It is the diagonal matrix of the nodal conductance matrix and the susceptance matrix.

[0154] S307. Dynamically weight and merge the first, second, third, and fourth AGGRE functions to obtain a multi-kernel collaborative graph convolutional neural network.

[0155] like Figure 4 As shown, this invention further uses an attention mechanism to dynamically weight and merge different graph convolution methods to construct a multi-core collaborative graph convolutional neural network architecture.

[0156] In other words, the multi-kernel collaborative graph convolutional neural network in this invention uses multiple neighborhood aggregation calculation methods (e.g., equations (7), (10), (11), and (12)) in the graph convolutional layer to overcome the problem of insufficient feature extraction by a single neighborhood aggregation method. Secondly, it uses a self-attention mechanism to collaboratively process multiple graph convolutional kernels, dynamically adjusting the attention of different node feature neighborhood aggregation methods to extract important features that affect the prediction results. For input to the first If all nodes in the layer have coupling characteristics, then the aggregation characteristic of the s-th type of neighborhood aggregation method is, which can be expressed as:

[0157] ;

[0158] This invention integrates neighborhood aggregation methods (7), (10), (11), and (12) into a multi-kernel collaborative graph convolutional neural network, therefore In equations (10) and (11), the coupling relationship between e and f is not considered, therefore, respectively... and To calculate the aggregation node representation and Since different aggregation methods extract different types of features, concatenating these features can enrich node features and further improve neural network performance. However, combining node representations calculated by different aggregation methods... and Unified concatenation may add redundant or unimportant features, resulting in insufficient attention to important features affecting prediction accuracy. Therefore, this invention uses weights (considered a form of attention) multiplied by aggregated features to scale feature values. If the aggregated feature indicates a significant impact on the neural network output, it is assigned a larger weight; otherwise, a smaller weight is assigned. Unlike traditional attention mechanisms, this invention uses a global pooling layer and a fully connected layer to determine the attention magnitude, expressed as Equation (13). It should be noted that two fully connected layers are used for attention calculations of node features e and f, respectively, expressed as:

[0159] (13)

[0160] in, This represents global pooling computation, and its output dimension is... ; and This represents a fully connected layer that performs two attention calculations, with an output dimension of... .

[0161] Finally, the node features based on different neighborhood clustering methods and the forward propagation function of the neural network are compared at the 1st... Each graph convolutional layer can be represented by equations (14) and (15). By further combining pooling layers and fully connected layers, a complete multi-kernel collaborative graph convolutional neural network architecture can be constructed.

[0162] (14)

[0163] (15)

[0164] in, This represents the calculation of splicing different features; This represents the characteristics of coupled nodes that are aggregated using different methods.

[0165] S205. Train the multi-kernel collaborative graph convolutional neural network using the training sample set to obtain the prediction model.

[0166] It is understood that, in the process of training the multi-core collaborative graph convolutional neural network in this application, the training method with the maximum number of iterations can be adopted, or the method with preset convergence conditions can be adopted, and no limitation is made here.

[0167] S103. Determine the operational reliability assessment results based on the minimum load shedding of the power system.

[0168] Understandably, the fundamental indicators of operational reliability are sufficiency indicators, such as the probability of load curtailment (PLC) and the expected demand not supplied (EDNS). After obtaining the minimum load shedding amount under different system states, this invention can calculate the PLC and EDNS using the following formulas:

[0169] ;

[0170] ;

[0171] in, Let be the probability of the i-th system state occurring; S is the set of system states where load reduction occurs. This represents the minimum load shedding amount at node j under system state i.

[0172] In the specific implementation of this application, when determining the operational reliability assessment results based on the minimum load shedding of the power system, it is necessary to determine when to stop the calculation. There are two main convergence criteria: the maximum sample size and the variance of the reliability index. In the specific implementation of this application, convergence criteria are used simultaneously based on the maximum sample size and the variance of the reliability index.

[0173] The present invention will be further described below with reference to specific implementation schemes.

[0174] Example 1:

[0175] In the specific implementation of this application, the operational reliability assessment of a power system typically involves load fluctuations, random renewable energy generation, and outages of critical electrical equipment. For load fluctuations, this embodiment assumes they follow a normal distribution, with a default mean of 0.3 and a standard deviation of 0.3. This embodiment considers two renewable energy sources: wind power and photovoltaics. It is assumed that wind speed and solar irradiance follow Weibull and Beta distributions, respectively, with wind speed following a Weibull distribution. Solar irradiance follows Distribution, among which This embodiment connects multiple wind farms and photovoltaic power plants to different nodes, achieving a renewable energy penetration rate exceeding 20%. In the IEEE 39-node system, four 200MW wind farms and four 200MW photovoltaic power plants are randomly connected to different nodes in the power system. The renewable energy penetration rate is 20.26%. To reduce training data imbalance, this embodiment samples N-1 and N-2 scenarios to cover all training cases. During the reliability assessment phase, this embodiment simulates a 1% failure probability for reliability evaluation. This embodiment generates 20K and 10K samples for training and testing, respectively.

[0176] To demonstrate the effectiveness of the proposed method, this embodiment compares traditional operational reliability assessment methods with five data-driven operational reliability assessment methods. These methods and their descriptions are described below. All neural networks were built and trained using the PyTorch framework on a desktop computer with an Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 16GB RAM, and an NVIDIA GeForce RTX 2080Ti graphics card. The neural networks were trained using the Adam optimizer with a learning rate of 0.001.

[0177] M0: Traditional operational reliability assessment methods use the interior point method to determine the minimum load shedding. This method serves as the benchmark method for data analysis.

[0178] M1: This method uses a typical graph convolutional neural network to calculate the minimum shear load, and its graph convolution kernel is derived from the Laplacian matrix based on spectral theory.

[0179] M2: This method is exactly the same as M1, except for the graph convolution kernel. The graph convolution kernel uses a Gaussian kernel based on line impedance.

[0180] M3: This method is exactly the same as M1, but utilizes the graph convolution kernel embedded in the power flow model.

[0181] M4: This method is the same as M1, but it uses graph convolution kernels that are precisely embedded in the proposed physical model.

[0182] M5: This method is the same as M1, but adopts the multi-core collaborative graph convolutional neural network architecture proposed in this invention.

[0183] This embodiment uses mean absolute error and correlation error to evaluate the performance of different methods. The mathematical formula is as follows:

[0184] ;

[0185] ;

[0186] Wherein, y is the minimum load shedding amount calculated by solving the load reduction model using numerical methods; To predict the minimum load using a neural network; This refers to the number of test data. For operational reliability metrics, the accurate value assessed by M0 is used as a reference.

[0187] like Figure 5 As shown, the error curves of the training set during the training of different neural networks are as follows: Figure 6 As shown, these are the error curves of the test set during the training of different neural networks.

[0188] This embodiment first constructs and trains neural networks using different methods on an IEEE 39-node system. The values ​​in M4 and M5 are set to 0.1. Different neural networks are trained through 2000 iterations. At the completion of different iterations, the mean absolute error of the training and testing data is as follows: Figure 5 and Figure 6 As shown in Table 1, K-fold cross-validation was performed in this embodiment, and the mean absolute error and standard deviation were calculated when K=10. It can be observed that M1 and M2 converged after several iterations, but the validation error was the largest. In M3, the power flow model embedding design can reduce the error to a smaller value. When the neighborhood aggregation method embedded in the power flow model is changed to the neighborhood aggregation method precisely embedded in the physical model, the training and testing errors can be further reduced. Furthermore, when using M5 with a multi-kernel collaborative graph convolutional neural network, the smallest error and standard deviation can be obtained. This verifies the effectiveness of the neural network architecture proposed in this invention.

[0189] Table 1

[0190]

[0191] Example 2:

[0192] In another embodiment of this application, operational reliability assessment in a power system typically involves load fluctuations, stochastic renewable energy generation, and outages of critical electrical equipment. For load fluctuations, this embodiment assumes they follow a normal distribution with a default mean of 0.3 and a standard deviation of 0.3. This embodiment considers two renewable energy sources: wind power and solar irradiance. It is assumed that wind speed and solar irradiance follow Weibull and Beta distributions, respectively, with wind speed following a Weibull distribution. Solar irradiance follows Distribution, among which This embodiment connects multiple wind farms and photovoltaic power plants to different nodes, achieving a renewable energy penetration rate exceeding 20%. In the IEEE 39-node system, four 200MW wind farms and four 200MW photovoltaic power plants are randomly connected to different nodes in the power system. The renewable energy penetration rate is 20.26%. To reduce training data imbalance, this embodiment samples N-1 and N-2 scenarios to cover all training cases. During the reliability assessment phase, this embodiment simulates a 1% failure probability for reliability evaluation. This embodiment generates 20K and 10K samples for training and testing, respectively.

[0193] To demonstrate the effectiveness of the proposed method, this embodiment compares traditional operational reliability assessment methods with five data-driven operational reliability assessment methods. These methods and their descriptions are described below. All neural networks were built and trained using the PyTorch framework on a desktop computer with an Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 16GB RAM, and an NVIDIA GeForce RTX 2080Ti graphics card. The neural networks were trained using the Adam optimizer with a learning rate of 0.001.

[0194] M0: Traditional operational reliability assessment methods use the interior point method to determine the minimum load shedding. This method serves as the benchmark method for data analysis.

[0195] M1: This method uses a typical graph convolutional neural network to calculate the minimum shear load, and its graph convolution kernel is derived from the Laplacian matrix based on spectral theory.

[0196] M2: This method is exactly the same as M1, except for the graph convolution kernel. The graph convolution kernel uses a Gaussian kernel based on line impedance.

[0197] M3: This method is exactly the same as M1, but utilizes the graph convolution kernel embedded in the power flow model.

[0198] M4: This method is the same as M1, but it uses graph convolution kernels that are precisely embedded in the proposed physical model.

[0199] M5: This method is the same as M1, but adopts the multi-core collaborative graph convolutional neural network architecture proposed in this invention.

[0200] This embodiment uses mean absolute error and correlation error to evaluate the performance of different methods. The mathematical formula is as follows:

[0201] ;

[0202] ;

[0203] Wherein, y is the minimum load shedding amount calculated by solving the load reduction model using numerical methods; To predict the minimum load using a neural network; This refers to the number of test data. For operational reliability metrics, the accurate value assessed by M0 is used as a reference.

[0204] This embodiment further verifies the effectiveness of the proposed multi-kernel collaborative graph convolutional neural network design. Different methods were implemented on the IEEE 39-node system, and statistical analysis was performed on the attention values ​​of different neighborhood aggregation methods under test samples, such as... Figure 7 and Figure 8 As shown (the columns on each floor are in sequence) , , and It can be observed that the attention values ​​differ in different multi-kernel graph convolutional layers. Neighborhood aggregation methods derived from spectral graph theory (e.g.) and ) in the Layers have large attention values, while aggregation methods for physical model embeddings (e.g.) and The last layer has a larger attention value. This is because the input representations of e and f are constants, and and This allows for more efficient extraction of graph features based on node connections. In deeper hidden layers (e.g., layer 3), features derived from the power flow model... and This approach helps extract key features that influence the minimum shear load in the prediction. Therefore, the multi-kernel design of the graph convolutional layer can extract features more effectively. In graph convolutional layers of different depths, different neighborhood aggregation methods play a dominant role, which can help the neural network extract features that influence the output more accurately and effectively, verifying the effectiveness of the proposed multi-kernel collaborative graph convolutional neural network architecture.

[0205] Example 3:

[0206] This embodiment, based on Embodiment 2, further utilizes a trained neural network for operational reliability assessment. The failure probability of critical equipment is set to 1%. The maximum number of iterations is 30K. The corresponding reliability indices and relative errors for M1-M5 are shown in the table. It can be observed that M1, based on the typical neighborhood aggregation method, has the largest assessment error. As a neighborhood aggregation method embedded within physical variables (e.g., line impedance) and the physical model is progressively constructed, the assessment error gradually decreases, verifying the effectiveness of the proposed neural network architecture.

[0207] As shown in Table 2, when using S1 and S2 for testing, the accuracy of M0-M3 also decreases to some extent, but the accuracy of M4 and M5 can still reach 95%. It is worth noting that in the 2383-node system, the accuracy of VG is around 4%-20%, but the accuracy of M4 and M5 is improved by as much as 90%. Therefore, the local and global nonlinear sample collection method proposed in this invention can effectively improve the generalization ability of neural networks, enabling them to adapt to nonlinear sample testing, thus verifying the effectiveness of the proposed method.

[0208] Table 2

[0209]

[0210] Example 4:

[0211] This embodiment uses a real-world load curve to verify the effectiveness of the proposed neural network architecture in a day-long evaluation. Figure 9 As shown, this embodiment applies the load curve of a certain region from January 1st to January 2nd, 2022 to the IEEE 39-node system. The outage probability of lines and generators is set to 0.01. Subsequently, this embodiment uses M1-M5 to perform operational reliability assessment and plots the error of reliability indicators at different times. All graph convolutional neural networks used consist of three graph convolutional layers and three fully connected layers, with each graph convolutional layer having 16 channels and each fully connected layer having 256 neurons. Then, training samples are generated using the load curve data from January 1st, totaling 48,000 samples (2,000 samples per time point). Then, this embodiment performs operational reliability assessment at different times on January 2nd, and the results are as follows. Figure 6 As shown in the figure, compared with other methods, M5 exhibits lower computational error at various time points, especially between 12:00 and 17:00, where the error rate of other methods increases significantly, while M5 maintains a low computational error throughout this time range. This verifies the effectiveness of the multi-core collaborative graph convolutional neural network operation reliability evaluation method proposed in this invention.

[0212] As can be seen from the above scheme, this application provides an operational reliability assessment method. After obtaining the system state to be assessed, the system state is input into the prediction model, and the minimum load shedding amount of the power system is output. Based on the minimum load shedding amount of the power system, the operational reliability assessment result is determined. The prediction model of this application is a multi-core collaborative graph convolutional neural network. At the architecture level, it improves the neural network's ability to extract node features, topological features, and their physical correlation features, thereby effectively improving the accuracy of the neural network in calculating the minimum load shedding amount and enhancing its adaptability to complex variable topology scenarios during operational reliability assessment.

[0213] Another embodiment of this application provides an operational reliability assessment device, such as... Figure 10 As shown, it specifically includes:

[0214] Acquisition unit 1001 is used to acquire the system state to be evaluated.

[0215] The analysis unit 1002 is used to input the system state to be evaluated into the prediction model and output the minimum load shedding amount of the power system.

[0216] The prediction model is a multi-kernel collaborative graph convolutional neural network.

[0217] Evaluation unit 1003 is used to determine the operational reliability assessment results based on the minimum load shedding of the power system.

[0218] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 1 As shown, it will not be elaborated further here.

[0219] Optionally, in another embodiment of this application, one implementation of the predictive model building unit includes:

[0220] Training sample set construction unit, used to construct the training sample set.

[0221] Minimum load shedding model establishment unit, used to establish the initial minimum load shedding model.

[0222] Graph convolutional layer building units are used to construct graph convolutional layers that are precisely embedded within the initial minimum shear load model.

[0223] The graph convolutional neural network building unit is used to build a multi-kernel collaborative graph convolutional neural network based on the graph convolutional layers.

[0224] The training unit is used to train the multi-kernel collaborative graph convolutional neural network using the training sample set to obtain the prediction model.

[0225] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 2 As shown, it will not be elaborated further here.

[0226] Optionally, in another embodiment of this application, one implementation of the reliability assessment device includes:

[0227] The convergence judgment unit is used to determine the convergence criteria based on the maximum sample size and the variance of the reliability index during the process of determining the operational reliability assessment results based on the minimum load shedding of the power system.

[0228] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, which will not be repeated here.

[0229] Optionally, in another embodiment of this application, one implementation of the training sample set construction unit includes:

[0230] The sampling unit is used to obtain the power system status through Monte Carlo sampling based on typical power system operating scenarios.

[0231] The solution unit is used to solve for the minimum load shedding under each power system state using the interior point method.

[0232] The training sample set construction sub-unit is used to construct the training sample set based on the minimum load shedding amount under all power system states.

[0233] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, which will not be repeated here.

[0234] Optionally, in another embodiment of this application, one implementation of the graph convolutional layer building unit includes:

[0235] The graph convolutional layer construction subunit is used to embed the Gauss-Seidel iteration of the power flow equation into the graph convolution, thus constructing a graph convolutional layer that is precisely embedded in the initial minimum shear load model.

[0236] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, which will not be repeated here.

[0237] Optionally, in another embodiment of this application, one implementation of the graph convolutional neural network building unit includes:

[0238] The first determining unit is used to determine the forward propagation formula based on the node feature convolution and node feature neighborhood aggregation calculation functions.

[0239] The second determining unit is used to determine the first AGGRE function based on the adjacency matrix.

[0240] The third determining unit is used to determine the second AGGRE function based on the impedance-based aggregated weight matrix.

[0241] The normalization unit is used to normalize the aggregated weight matrix of impedance to obtain the aggregated weight matrix of the target impedance.

[0242] The fourth determining unit is used to determine the third AGGRE function based on the aggregated weight matrix of the target impedance.

[0243] The fifth determining unit is used to determine the fourth AGGRE function based on the transformed power flow equations.

[0244] The dynamic weighted merging unit is used to dynamically weight and merge the first, second, third, and fourth AGGRE functions to obtain a multi-kernel collaborative graph convolutional neural network.

[0245] For details on the specific working process of the units disclosed in the above embodiments of this application, please refer to the corresponding method embodiments, such as... Figure 3 As shown, it will not be elaborated further here.

[0246] As can be seen from the above scheme, this application provides an operational reliability assessment device. After the acquisition unit 1001 acquires the system state to be assessed, the analysis unit 1002 inputs the system state to be assessed into the prediction model and outputs the minimum load shedding amount of the power system; the assessment unit 1003 determines the operational reliability assessment result based on the minimum load shedding amount of the power system. The prediction model of this application is a multi-core collaborative graph convolutional neural network; at the architectural level, it improves the neural network's ability to extract node features, topological features, and their physical correlation features, thereby effectively improving the accuracy of the neural network in calculating the minimum load shedding amount and enhancing its adaptability to complex variable topology scenarios during operational reliability assessment.

[0247] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0248] Another embodiment of this application provides an electronic device, such as... Figure 11 As shown, it includes:

[0249] One or more processors 1101.

[0250] Storage device 1102, on which one or more programs are stored.

[0251] When the one or more programs are executed by the one or more processors 1101, the one or more processors 1101 implement the operational reliability assessment method as described in the above embodiments.

[0252] Another embodiment of this application provides a computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the operational reliability assessment method as described in the above embodiments.

[0253] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0254] It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0255] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0256] Another embodiment of this application provides a computer program product that, when executed, performs the above-described operational reliability assessment method.

[0257] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, it performs the functions defined in the methods of the embodiments of this application.

[0258] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in this application is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely exemplary forms of implementing this application.

[0259] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0260] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described application concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions claimed in this application.

Claims

1. A method of operating reliability assessment, characterized by, The method comprises the following steps: obtaining a system state to be evaluated; inputting the system state to be evaluated into a prediction model to output a minimum load shedding amount of a power system; wherein the prediction model is a multi-core collaborative graph convolutional neural network; determining a running reliability evaluation result according to the minimum load shedding amount of the power system; wherein the construction method of the prediction model comprises: constructing a training sample set; establishing an initial minimum load shedding model; constructing a graph convolutional layer precisely embedded in the initial minimum load shedding model; establishing a multi-core collaborative graph convolutional neural network according to the graph convolutional layer; training the multi-core collaborative graph convolutional neural network using the training sample set to obtain a prediction model; wherein the establishment of the multi-core collaborative graph convolutional neural network according to the graph convolutional layer comprises: determining a forward propagation formula according to a node feature convolution and a node feature neighborhood aggregation calculation function; determining a first AGGRE function based on an adjacency matrix; determining a second AGGRE function based on an impedance aggregation weight matrix; standardizing the impedance aggregation weight matrix to obtain a target impedance aggregation weight matrix; determining a third AGGRE function based on the target impedance aggregation weight matrix; determining a fourth AGGRE function based on a converted power flow equation; dynamically weighting and merging the first AGGRE function, the second AGGRE function, the third AGGRE function and the fourth AGGRE function to obtain a multi-core collaborative graph convolutional neural network.

2. The operational reliability assessment method according to claim 1, characterized in that Further comprising: in the process of determining the running reliability evaluation result according to the minimum load shedding amount of the power system, performing convergence criterion judgment according to the maximum sample number and the variance of the reliability index.

3. The operational reliability assessment method according to claim 1, characterized in that, The construction of the training sample set comprises: obtaining power system states through Monte Carlo sampling according to typical power system operation scenarios; for each power system state, solving the minimum load shedding amount under the power system state by an interior point method; constructing a training sample set according to the minimum load shedding amounts under all the power system states.

4. The method of operational reliability assessment according to claim 1, characterized in that, The construction of the graph convolutional layer precisely embedded in the initial minimum load shedding model comprises: embedding Gauss-Seidel iteration of a power flow equation into graph convolution to construct the graph convolutional layer precisely embedded in the initial minimum load shedding model.

5. A running reliability evaluation device characterized by comprising: The method comprises the following steps: an obtaining unit is configured to obtain a system state to be evaluated; an analysis unit is configured to input the system state to be evaluated into a prediction model to output a minimum load shedding amount of a power system; wherein the prediction model is a multi-core collaborative graph convolutional neural network; an evaluation unit is configured to determine a running reliability evaluation result according to the minimum load shedding amount of the power system; wherein the construction unit of the prediction model comprises: a training sample set construction unit is configured to construct a training sample set; a minimum load shedding model establishment unit is configured to establish an initial minimum load shedding model; a graph convolutional layer construction unit is configured to construct a graph convolutional layer precisely embedded in the initial minimum load shedding model; a graph convolutional neural network establishment unit is configured to establish a multi-core collaborative graph convolutional neural network according to the graph convolutional layer; A training unit is configured to train the multi-core collaborative graph convolutional neural network using the training sample set to obtain a prediction model The graph convolutional neural network establishing unit comprises: A first determining unit is configured to determine a forward propagation formula according to a node feature convolution and a node feature neighborhood aggregation calculation function; A second determining unit is configured to determine a first AGGRE function based on an adjacency matrix; A third determining unit is configured to determine a second AGGRE function based on an impedance aggregation weight matrix; A normalization unit is configured to normalize the impedance aggregation weight matrix to obtain a target impedance aggregation weight matrix; A fourth determining unit is configured to determine a third AGGRE function based on the target impedance aggregation weight matrix; A fifth determining unit is configured to determine a fourth AGGRE function based on the converted power flow equation; A dynamic weighted merging unit is configured to dynamically weight and merge the first AGGRE function, the second AGGRE function, the third AGGRE function and the fourth AGGRE function to obtain the multi-core collaborative graph convolutional neural network.

6. An electronic device, comprising: comprise: one or more processors; a memory device having stored thereon one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the operational reliability evaluation method according to any one of claims 1 to 4.

7. A computer storage medium, characterized in that a computer program is stored thereon, and the computer program is executed by a processor to implement the operational reliability evaluation method according to any one of claims 1 to 4.

Images