An intelligent calculation method for power system wide frequency oscillation damping coefficient

Abstract

The application discloses a kind of power system broadband oscillation damping coefficient intelligent calculation method, comprising the following steps: S1, sample generation;S2, based on PageRank algorithm, the importance of measuring point is sorted;S3, measuring point damping coefficient calculation;S4, system damping coefficient enhancement.The application has high precision: it can effectively improve the precision of system oscillation risk assessment to 98.5%;Measurement requirement is low: system oscillation risk assessment can be realized under the measurement signal of about 40% or so line of system;Fast calculation speed: compared with traditional CPU-based oscillation risk assessment, GPU is called by neural network embedded calculation, and the calculation speed is about 16 times higher.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering, and more particularly to an intelligent calculation method for the broadband oscillation damping coefficient of a power system. Background Technology

[0002] Broadband oscillation risk assessment in power systems involves data collection, system modeling, and oscillation characteristic analysis to identify potential problems, comprehensively consider possible risks, and then implement corresponding risk management measures to ensure the stable operation of the power system. The oscillation damping coefficient is an important parameter reflecting the trend of oscillation changes. It can be used to assess broadband oscillation risk from the perspective of data response, based on measured voltage / current phasor data.

[0003] Existing methods for assessing broadband oscillation risk in systems are mainly based on oscillation damping coefficients and impedance network-based assessment methods. However, these methods have poor accuracy in calculating oscillation damping coefficients, high measurement requirements, and slow calculation speeds. Summary of the Invention

[0004] To address the problems of poor accuracy, high measurement requirements, and slow calculation speed in existing technologies for calculating damping coefficients, this application proposes an intelligent calculation method for broadband oscillation damping coefficients in power systems, thus resolving the aforementioned issues.

[0005] This application discloses an intelligent calculation method for the broadband oscillation damping coefficient of a power system, including the following steps:

[0006] S1, Sample generation: The current data of the power system measurement point under renewable energy access is obtained through time domain simulation. The time series data is divided into multiple data segments according to the oscillation discrimination time requirement, and then data samples are obtained.

[0007] S2, based on the PageRank algorithm, sorts the importance of measurement points and the topology nodes in the power system by importance;

[0008] S3, Calculation of damping coefficient at measuring point, identification of oscillation damping based on measured current data, thereby assessing oscillation risk and providing important information for the subsequent deployment of broadband oscillation suppression devices;

[0009] S4, System Damping Coefficient Enhancement: A broadband oscillation damping coefficient calculation method integrating PageRank and deep learning is proposed. By selecting critical paths and combining deep learning path weights, the system damping coefficient is enhanced.

[0010] Preferably, step S2 specifically includes the following steps:

[0011] S21, Topology Modeling: Using the PageRank algorithm, the power system is treated as an undirected graph, which is represented by a binary matrix A;

[0012] S22, Establish a random agent model: Introduce a random agent that traverses the power system according to the topological connections. When the agent stops at a node during traversal, it considers energy consumed at that node. The probabilistic model is represented by a random browsing matrix P, where P's elements are... ij This represents the probability of jumping from node i to node j;

[0013] S23, Initialize PageRank parameters: Initialize PageRank score v, set hysteresis factor d, maximum number of iterations Imax, convergence threshold ε, and counter n;

[0014] S24, increment the value of counter n, and start updating the PageRank score when n has not reached the maximum number of iterations;

[0015] S25, PageRank score initialization: PageRank scores are treated as a distribution across nodes in the entire power system. Each node is represented by a PageRank score indicating its relative importance. These scores are then treated as a column vector v, v... i This represents the PageRank score of node i;

[0016] S26, PageRank score update: The importance of a node is converted into the weight passed by other nodes through the PageRank algorithm. A node's PageRank score depends on the number and scores of the nodes connected to it. Nodes with higher PageRank scores pass on more weight.

[0017] S27, check if the PageRank value has converged. If the convergence condition is met, the algorithm update ends.

[0018] Preferably, in S21, node A in the binary matrix A... i The following constraints must be met:

[0019]

[0020] In the formula, q is the maximum number of columns in the matrix, and A ij A represents the path coefficient between node i and node j. i The i-th row of the binary matrix represents node A. i Connection status with other nodes, A ii This represents the diagonal of a binary matrix, i.e., the nodes themselves. Since self-loops within nodes are not considered, the coefficient is set to 0.

[0021] Preferably, the method in S26 is as follows:

[0022] Assume the probability that the Agent does not remain on the original node is d∈[0,1]. Each time the PageRank distribution is updated, a random number x is generated, and the update process is as follows:

[0023]

[0024] In the formula: v k ,v k+1 Let I be the PageRank score vector of the nodes in the system obtained after the k-th and k+1-th iterations, respectively, where I is the identity matrix, P is the random browsing matrix, and A is the topological binary matrix.

[0025] Preferably, the formula for calculating the damping coefficient in S3 is:

[0026]

[0027]

[0028] In the formula, x(t) represents the oscillation signal, S is the modal amplitude, f is the oscillation frequency, θ is the modal phase; t0 is the initial time of signal acquisition, t k σ is the current sampling time; m is the length of the data window; Ts is the sampling period; σ represents the damping coefficient; and k is the number of sampling points.

[0029] Preferably, the step of enhancing the system damping coefficient in S4 is as follows:

[0030] Design an enhanced model of the autoencoder structure. The encoder and decoder formulas of the autoencoder are as follows:

[0031]

[0032] x′=g φ (ω);

[0033] In the formula, the encoder output ω is the line weight, satisfying a sum of 1; l is the number of critical lines; σ is the critical line damping coefficient; υ is the critical line importance; x=[σ,υ] is the model input; x′ represents the decoder output; f θ (x) represents the encoder function; g φ (ω) represents the decoder function;

[0034] After obtaining the line weight ω, the system damping coefficient can be obtained as follows:

[0035] The loss function of the augmentation model includes the autoencoder loss and the damping coefficient loss, and its expression is as follows:

[0036]

[0037] In the formula: λ is a constant, This is the loss function.

[0038] The beneficial effects of this invention are:

[0039] 1. High accuracy: The intelligent calculation method for broadband oscillation damping coefficient, which integrates PageRank and deep learning, can effectively improve the accuracy of system oscillation risk assessment to 98.5%.

[0040] 2. Low measurement requirements: System oscillation risk assessment can be achieved by collecting measurement signals from only about 40% of the system's lines.

[0041] 3. High computation speed: Compared with traditional CPU-based oscillation risk assessment, the computation speed is increased by about 16 times by calling the GPU through neural network embedded computation. Attached Figure Description

[0042] Figure 1 This is a flowchart of the intelligent calculation method for the broadband oscillation damping coefficient of a power system according to an embodiment of the present invention;

[0043] Figure 2 This is a flowchart of the PageRank algorithm according to an embodiment of the present invention;

[0044] Figure 3 This is a block diagram of the intelligent enhancement algorithm for the system damping coefficient in an embodiment of the present invention;

[0045] Figure 4 The training curves are those of the intelligent enhancement model in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided with reference to the accompanying drawings and embodiments.

[0047] This application discloses an intelligent calculation method for the broadband oscillation damping coefficient of a power system, such as... Figure 1 As shown, it includes the following steps:

[0048] S1, Sample generation: The current data of the power system measurement point under renewable energy access is obtained through time domain simulation. The time series data is divided into multiple data segments according to the oscillation discrimination time requirement, and then data samples are obtained.

[0049] S2, based on the PageRank algorithm, sorts the importance of measurement points and the topology nodes in the power system by importance;

[0050] The PageRank algorithm is used to evaluate the importance of nodes in an undirected graph structure to measure the relative importance of network systems such as the Internet. Based on graph theory and probability theory, the PageRank algorithm aims to quantify the weights of nodes in a network, which can be represented as various entities. Power systems are typical undirected graph networks; in this embodiment, the topological nodes in the power system are treated as network nodes and ranked by importance.

[0051] like Figure 2 As shown, S2 includes the following steps:

[0052] S21, Topology Modeling: Using the PageRank algorithm, the power system is treated as an undirected graph, which is represented by a binary matrix A. Nodes A in this binary matrix A are... i The following constraints must be met:

[0053]

[0054] In the formula, q is the maximum number of columns in the matrix, and A ij A represents the path coefficient between node i and node j. i The i-th row of the binary matrix represents node A. i Connection status with other nodes, A ii This represents the diagonal of a binary matrix, i.e., the nodes themselves. Since self-loops within nodes are not considered, the coefficient is set to 0.

[0055] S22, Establish a random agent model: Introduce a random agent that traverses the power system according to the topological connections. When the agent stops at a node during traversal, it considers energy consumed at that node. The probabilistic model is represented by a random browsing matrix P, where P's elements are... ij This represents the probability of jumping from node i to node j;

[0056] S23, Initialize PageRank parameters: Initialize PageRank score v, set hysteresis factor d, maximum number of iterations Imax, convergence threshold ε, and counter n;

[0057] S24, increment the value of counter n, and start updating the PageRank score when n has not reached the maximum number of iterations;

[0058] S25, PageRank score initialization: PageRank scores are treated as the distribution of nodes across the entire power system. Each node is represented by a PageRank score indicating its relative importance. These scores are then treated as a column vector v, v... i This represents the PageRank score of node i;

[0059] S26, PageRank score update: The importance of a node is determined by converting it into weights passed to other nodes through the PageRank algorithm. A node's PageRank score depends on the number and scores of the nodes connected to it. Nodes with higher PageRank scores pass on more weights.

[0060] Assume the probability that the Agent does not remain on the original node is d∈[0,1]. Each time the PageRank distribution is updated, a random number x is generated, and the update process is as follows:

[0061]

[0062] In the formula: v k ,v k+1 Let I be the PageRank score vector of the nodes in the system obtained after the k-th and k+1-th iterations, respectively, where I is the identity matrix, P is the random browsing matrix, and A is the topological binary matrix.

[0063] S27, check if the PageRank value has converged. If the convergence condition is met, the algorithm update ends.

[0064] S3, Calculation of the damping coefficient at the measuring point. The damping coefficient is an important parameter reflecting the trend of oscillation changes, and it can be evaluated from the perspective of data response. The damping coefficient can be used for oscillation discrimination. Oscillation discrimination based on data response does not depend on phasor data and can identify oscillation damping based on measured current data, thereby assessing oscillation risk and providing important information for the subsequent deployment of broadband oscillation suppression devices.

[0065] The method based on the oscillation damping coefficient assumes that the oscillation signal changes in an exponential form. By using the time series formed by the oscillation assignment, the system damping coefficient can be estimated at different times.

[0066] The formula for calculating the damping coefficient is:

[0067]

[0068]

[0069] In the formula, x(t) represents the oscillation signal, S is the modal amplitude, f is the oscillation frequency, θ is the modal phase; t0 is the initial time of signal acquisition, t k σ is the current sampling time; m is the length of the data window; Ts is the sampling period; σ represents the damping coefficient; and k is the number of sampling points.

[0070] S4, the system damping coefficient is enhanced, such as Figure 3As shown, a broadband oscillation damping coefficient calculation method integrating PageRank and deep learning is proposed. By selecting the critical path and combining the deep learning path weights, the system damping coefficient is enhanced.

[0071] In one embodiment, an enhanced model of the autoencoder structure is designed, and the encoder and decoder formulas of the autoencoder are as follows:

[0072]

[0073] x′=g φ (ω);

[0074] In the formula, the encoder output ω is the line weight, satisfying a sum of 1; l is the number of critical lines; σ is the critical line damping coefficient; υ is the critical line importance; x=[σ,υ] is the model input; x′ represents the decoder output; f θ (x) represents the encoder function; g φ (ω) represents the decoder function;

[0075] After obtaining the line weight ω, the system damping coefficient can be obtained as follows:

[0076] The loss function of the augmentation model includes the autoencoder loss and the damping coefficient loss, and its expression is as follows:

[0077]

[0078] In the formula: λ is a constant, The loss function is, for example, the root mean square error (MSE).

[0079] In one specific embodiment, λ is set to 0.1. We selected MSE to test the oscillation discrimination performance under the PageRank algorithm.

[0080] The model training loss versus accuracy curve is as follows: Figure 4 As shown, the test used the IEEE-118 full electromagnetic simulation system, generating subsynchronous oscillation frequency bands of 21.32Hz and 45.67Hz, with a simulation step size of 1e-8 and a total simulation duration of 20s. The neural network was built using Python 3.8 and Tensorflow 2.8. The comparison of the discrimination effect before and after using the PageRank algorithm is shown in Table 1. It can be seen that the oscillation discrimination accuracy was further improved after using the PageRank algorithm.

[0081] Table 1: Comparison of discrimination results before and after using the PageRank algorithm

[0082]

[0083] The foregoing description illustrates the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A smart calculation method for the broadband oscillation damping coefficient of a power system, characterized in that, Includes the following steps: S1, Sample generation: The current data of the power system measurement point under renewable energy access is obtained through time domain simulation. The time series data is divided into multiple data segments according to the oscillation discrimination time requirement, and then data samples are obtained. S2, ranking the importance of measurement points based on the PageRank algorithm, and ranking the topology nodes in the power system by importance, including the following steps: S21, Topology Modeling: Using the PageRank algorithm, the power system is treated as an undirected graph, which is represented by a binary matrix A; S22, Establish a random agent model: Introduce a random agent that traverses the power system according to the topological connections. When the agent stops at a node during traversal, it considers energy consumed at that node. The probabilistic model is represented by a random browsing matrix P, where P's elements are... ij This represents the probability of jumping from node i to node j; S23, Initialize PageRank parameters: Initialize PageRank score v, set hysteresis factor d, maximum number of iterations Imax, and convergence threshold. , counter n; S24, increment the value of counter n, and start updating the PageRank score when n has not reached the maximum number of iterations; S25, PageRank score initialization: PageRank scores are viewed as the distribution of nodes across the entire power system. Each node is represented by a PageRank score indicating its relative importance. These scores are treated as a column vector v. This represents the PageRank score of node i; S26, PageRank score update: The importance of a node is converted into the weight passed by other nodes through the PageRank algorithm. A node's PageRank score depends on the number and scores of the nodes connected to it. Nodes with higher PageRank scores pass on more weight. S27, check if the PageRank value has converged. If the convergence condition is met, the algorithm update ends. S3, Calculation of damping coefficient at measuring point, identification of oscillation damping based on measured current data, thereby assessing oscillation risk and providing important information for the subsequent deployment of broadband oscillation suppression devices; S4, Enhanced System Damping Coefficient: A broadband oscillation damping coefficient calculation method integrating PageRank and deep learning is proposed. By selecting critical paths and combining deep learning path weights, the system damping coefficient is enhanced. The steps to enhance the system damping coefficient are as follows: Design an enhanced model of the autoencoder structure. The encoder and decoder formulas of the autoencoder are as follows: In the formula, the encoder output The line weights satisfy the condition that the sum is 1. l It is the number of critical paths. It is the damping coefficient of the critical path. It is the importance of the critical path. Input for the model; Indicates the decoder output; Represents the encoder function; Represents the decoder function; Obtain line weights Then, the system damping coefficient can be obtained as follows: ; The loss function of the augmentation model includes the autoencoder loss and the damping coefficient loss, and its expression is as follows: In the formula: It is a constant. This is the loss function.

2. The intelligent calculation method for broadband oscillation damping coefficient of a power system according to claim 1, characterized in that, In S21, node A of the binary matrix A i The following constraints must be met: In the formula, Let A be the maximum number of columns in the matrix. ij This represents the path coefficient between node i and node j. The i-th row of the binary matrix represents the node. Connection status with other nodes, This represents the diagonal of a binary matrix, i.e., the nodes themselves. Since self-loops within nodes are not considered, the coefficient is set to 0.

3. The intelligent calculation method for broadband oscillation damping coefficient of a power system according to claim 2, characterized in that, The method in S26 is as follows: Assume the probability that the Agent does not remain on the original node is A random number is generated each time the PageRank distribution is updated. x The update process is as follows: In the formula: , Let I be the PageRank score vector of the nodes in the system obtained after the k-th and k+1-th iterations, respectively, where I is the identity matrix, P is the random browsing matrix, and A is the topological binary matrix.

4. The intelligent calculation method for broadband oscillation damping coefficient of a power system according to claim 3, characterized in that, The formula for calculating the damping coefficient in S3 is as follows: In the formula, This indicates an oscillation signal, where S is the modal amplitude and f is the oscillation frequency. It is the modal phase; It is the initial moment of signal acquisition. This is the current sampling time; m is the length of the data window; Ts is the sampling period; This represents the damping coefficient, and k is the number of sampling points.

Images