A method and system for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations

Abstract

This invention discloses a method and system for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations, relating to the field of wind farm cluster oscillation analysis. The method includes: acquiring synchronous measurement data and topological connections of multiple electrical nodes; constructing a network topology model and generating a physical adjacency matrix; extracting the dominant oscillation mode response within the target oscillation frequency band; calculating the edge-level oscillation propagation intensity and edge-level Granger causality intensity between adjacent nodes; constructing node Granger causality scores under candidate source conditions to determine the oscillation source node; identifying the initial propagation channel, and using the initial propagation channel as the link starting point, progressively expanding and determining the dominant propagation link; and finally outputting the identification results. By combining edge-level Granger causality, propagation intensity, and oscillation timing consistency under topological constraints, it achieves the identification of the oscillation source, initial propagation channel, dominant propagation link, and key propagation nodes, enhancing the engineering applicability of the oscillation propagation analysis results.

Description

Technical Field

[0001] This invention relates to the field of wind farm group oscillation analysis, and more specifically to a method and system for tracing the source and identifying the dominant propagation link of wind farm group oscillations. Background Technology

[0002] With the large-scale grid connection of new energy power plants such as wind power, photovoltaics, and energy storage, a large number of grid-connected converters are connected to the power system through collection lines, step-up transformers, and transmission channels. Under weak grid conditions, mismatched control parameters, or external disturbances, subsynchronous oscillations are easily induced in wind farm clusters. Such oscillations may occur not only in a single wind turbine or a single collection branch, but may also propagate along the collection area, step-up substation, transmission channel, and remote main grid nodes, thereby causing wind turbine disconnection, power fluctuations, equipment protection malfunctions, or a decrease in system stability.

[0003] Existing grid-side converter protection for wind turbines typically relies on local protection logic such as overcurrent, overvoltage, undervoltage, and frequency limit violations, enabling rapid response to abnormal conditions of individual devices. However, broadband oscillations propagating in wind farm clusters exhibit significant network coupling characteristics. Relying solely on local measurements and threshold protection of individual converters makes it difficult to determine the node where the oscillation source is located, or to distinguish between the initial propagation path and subsequent relay propagation edges. While existing data-driven analysis methods, such as Granger causality analysis, can reveal directed predictive relationships using time-series data, they tend to focus on oscillation source location and rarely incorporate physical adjacency topology, target frequency band propagation capability, oscillation timing consistency, and dominant propagation links into a unified analysis framework. This results in only providing the location of the oscillation source without revealing the complete diffusion process of the oscillation within the network.

[0004] Existing oscillation mechanism analysis methods based on impedance analysis or small-signal stability typically rely on relatively complete converter control parameters and network models, limiting their online applicability. Purely data-driven methods are sensitive to measurement quality, sample size, and data stationarity, and lack physical interpretability. More importantly, existing methods struggle to uniformly express the side-level propagation relationships, node propagation roles, and dominant propagation links under multi-point measurement conditions in wind farm clusters, failing to generate joint identification results from the oscillation source node to the initial propagation channel and subsequent dominant propagation links. Furthermore, there is a lack of quantitative identification methods for key nodes and propagation hub nodes that play a relay support role in the oscillation propagation process.

[0005] Therefore, how to design a method and system for tracing the source of oscillations in wind farm clusters and identifying the dominant propagation link, so as to enhance the directional interpretability and engineering applicability of oscillation propagation analysis results, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for tracing the source of wind farm cluster oscillations and identifying the dominant propagation link, aiming to solve the problems in existing wind farm cluster oscillation analysis, such as difficulty in locating the oscillation source, difficulty in distinguishing the initial propagation channel from the subsequent relay propagation edge, and inability to form a unified identification result from the source node to the propagation link; the method and system can jointly identify the oscillation source, the initial propagation channel, the dominant propagation link, and key propagation nodes under multi-point measurement and topological constraints, providing a basis for wind farm cluster oscillation monitoring, propagation mechanism analysis, and operational auxiliary decision-making.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for tracing the source of wind farm cluster oscillations and identifying the dominant propagation link, comprising the following steps: S1. Obtain synchronous measurement data and topology connection relationship of multiple electrical nodes in the wind farm group; S2. Based on the aforementioned topological connection relationship, construct a network topology model consisting of a set of nodes and a set of electrical connection edges, and generate a physical adjacency matrix; S3. Preprocess the synchronous measurement data of each node and extract the dominant oscillation mode response within the target oscillation frequency band; S4. Based on the dominant oscillation mode response, construct the input-output transfer matrix of the wind farm group, and calculate the edge-level oscillation propagation intensity between adjacent nodes under the constraint of the physical adjacency matrix. S5. Under the constraints of the physical adjacency matrix, establish a local vector autoregressive model for the dominant oscillation mode response of adjacent nodes and calculate the edge-level Granger causality strength. S6. Based on the relationship between the edge-level Granger causality strength and the topological hierarchy, construct the node Granger causality score under the candidate source condition, and determine the oscillation source node based on the score; S7. Identify the initial propagation channel based on the oscillation source node, the edge-level Granger causality strength, the edge-level oscillation propagation strength, and the consistency of the start-up timing of each node; S8. Taking the initial propagation channel as the starting point of the link, and combining the edge-level comprehensive score and the downstream node Granger causality score, expand and determine the dominant propagation link segment by segment. S9 outputs the oscillation source node, initial propagation channel, dominant propagation link, key propagation edge order, and key propagation node identification results.

[0009] Preferably, S2 includes: The wind turbine grid-side converter, wind power collection node, box-type transformer node, collector line node, substation bus node, and transmission channel node are taken as network nodes, resulting in node set V; The set of edges E is obtained by taking the collector lines, transformer connections, busbar connections, transmission lines and equivalent electrical connections between nodes as electrical connection edges; A physical adjacency matrix is ​​constructed based on whether there are direct electrical connections between nodes. Its elements This indicates that node i and node j have a direct electrical connection. This indicates that there is no direct electrical connection.

[0010] Preferably, S3 includes: The synchronous measurement signals of each node are sequentially subjected to mean removal, trend removal, target frequency band bandpass filtering, and standardization to obtain the preprocessed signal; The preprocessed signal is constructed using a delayed embedding strategy, and discrete-time eigenvalues ​​and corresponding modes are obtained using the Koopman spectral decomposition algorithm. The discrete-time characteristic values ​​are converted into continuous-time characteristic values, and the oscillation frequency, damping ratio, and modal energy of each mode are calculated. Based on the preset target subsynchronous frequency band constraints and modal energy thresholds, the dominant oscillation mode is selected, and the response components of each node under the dominant oscillation mode are extracted.

[0011] Preferably, S4 includes: Near the steady-state operating point of the wind farm group, a small-disturbance linearization model is established, including the grid-side converter control link, the filtering link, and the network coupling relationship. Based on the state matrix A, input matrix B, output matrix C, and direct transfer matrix D of the small perturbation linearization model, construct the input-output transfer matrix; The frequency domain gain of the equivalent small disturbance input at the grid-connected port at node i to the output response at node j is defined as the oscillation propagation intensity from node i to node j. At the frequency corresponding to the dominant oscillation mode, the oscillation propagation intensity of all adjacent node pairs is calculated to obtain the edge-level oscillation propagation intensity matrix; The edge-level oscillation propagation intensity matrix is ​​normalized to obtain the edge-level propagation intensity.

[0012] Preferably, S5 includes: Based on the physical adjacency matrix, adjacent node pairs with direct electrical connections are selected, and a local vector autoregressive model is established only for the adjacent node pairs. For adjacent nodes i and j, construct a restricted model that includes only the history of node j itself, and an unrestricted model that includes both the history of node j and the history of node i. Compare the residual variances of the restricted model and the unrestricted model, and calculate the initial edge-level Granger causality strength from node i to node j; Insignificant causal edges are removed using significance tests or F-tests, and then normalized to obtain the edge-level Granger causal strength.

[0013] Preferably, S6 includes: Each electrical node is sequentially designated as a candidate source node, and the topological hierarchy of other nodes relative to the candidate source node is calculated based on the shortest path length from the candidate source node to other nodes. Based on the topological hierarchy, the adjacent nodes of each node are divided into deep adjacent nodes, same-level adjacent nodes, and shallow adjacent nodes. Based on deep adjacent nodes, same-layer adjacent nodes, and shallow adjacent nodes, forward output consistency terms, upstream input compensation terms, bidirectional coupling terms, and reverse backflow terms are constructed respectively. Combining the topological hierarchy weights and the above items, calculate the node Granger causality score for each node under the candidate source conditions; The consistency between the candidate source node's own Granger causality score and the score distribution of the entire network is fused to calculate the comprehensive score of each candidate source. The node with the highest comprehensive score among the candidate sources is determined as the oscillation source node.

[0014] Preferably, S7 includes: Hilbert transform is performed on the dominant oscillation mode response of each node to construct the oscillation envelope signal; The oscillation start time of each node is determined based on the moment when the oscillation envelope signal first exceeds the oscillation start threshold. Based on the timing relationship that the upstream node starts oscillating first and the downstream node responds later, an oscillating timing consistency index is constructed for the candidate propagation edge. All outgoing electrical connection edges of the oscillation source node are used as the candidate edge set for the initial propagation channel; Candidate edges are screened based on the oscillation timing consistency index, and the initial propagation channel is determined by comprehensively considering the edge-level Granger causality strength, edge-level oscillation propagation strength, and oscillation timing consistency.

[0015] Preferably, S8 includes: The terminal node of the initial propagation channel is taken as the current link endpoint; Starting from the current link endpoint, candidate propagation edges that meet the conditions of no backtracking in topology hierarchy, consistent oscillation order, and no repeated visits to nodes are selected from its adjacent edges. A link extension score is constructed based on the edge-level comprehensive score of the candidate propagation edge and the node Granger causality score of the downstream node. Select the candidate propagation edge with the highest link expansion score as the next dominant propagation edge, add the corresponding downstream node to the dominant propagation link, and update the current link endpoint; Repeat the above steps until the extended termination condition is met to obtain the complete dominant propagation chain.

[0016] Preferably, in step S9, the identification of key propagation nodes and propagation hub nodes includes: Given the established oscillation source nodes, the effective outward oscillation propagation capability of each node along the main propagation direction and the lateral direction of the same layer is statistically analyzed to obtain the forward oscillation propagation capability of the node. The forward oscillation propagation capability of the nodes is normalized and fused with the normalized node Granger causality score to obtain a key defense score for measuring the node monitoring priority. Statistically count the upstream throughput received by each node from shallow neighboring nodes, and the downstream throughput output to deep and same-layer neighboring nodes; Based on the upstream receiving volume and the downstream distributing volume, a receiving-distributing coordination coefficient is constructed to characterize the relay support capability of the node in the oscillation propagation link; The key defense score is fused with the receiving-distribution coordination coefficient to obtain the propagation center score, and key propagation nodes and propagation center nodes are determined according to preset thresholds.

[0017] Secondly, the present invention provides a system for tracing the source of wind farm cluster oscillations and identifying the dominant propagation link, comprising: Data acquisition module: used to acquire synchronous measurement data and topology connection relationships of multiple electrical nodes in a wind farm cluster; Topology modeling module: used to construct a network topology model consisting of a set of nodes and a set of electrical connection edges based on the topological connection relationship, and to generate a physical adjacency matrix; Dominant mode extraction module: used to preprocess the synchronous measurement data of each node and extract the dominant oscillation mode response within the target oscillation frequency band; Edge propagation intensity calculation module: used to construct the input-output transfer matrix of the wind farm group based on the dominant oscillation mode response, and calculate the edge oscillation propagation intensity between adjacent nodes under the constraint of the physical adjacency matrix; Edge-level Granger causality analysis module: used to establish a local vector autoregressive model for the dominant oscillation mode response of adjacent nodes under the constraints of the physical adjacency matrix, and to calculate the edge-level Granger causality strength; Oscillation source localization module: used to construct node Granger causality scores under candidate source conditions based on the edge-level Granger causality strength and topological hierarchy relationship, and to determine the oscillation source node based on the scores; Initial propagation channel identification module: used to identify the initial propagation channel based on the oscillation source node, the edge-level Granger causality intensity, the edge-level oscillation propagation intensity, and the consistency of the start-up timing of each node; Dominant propagation link determination module: used to expand and determine the dominant propagation link segment by segment, taking the initial propagation channel as the starting point of the link and combining the edge-level comprehensive score and the downstream node Granger causality score; The results output module is used to output the identification results of the oscillation source node, the initial propagation channel, the dominant propagation link, the order of key propagation edges, and the key propagation nodes.

[0018] As can be seen from the above technical solution, compared with the prior art, the technical solution of the present invention has the following beneficial effects: 1. This method calculates the edge-level Granger causality intensity and edge-level oscillation propagation intensity under the constraint of the physical adjacency matrix, and integrates the consistency of the oscillation start-up timing of each node. It can sequentially determine the oscillation source node, the initial propagation channel and the complete dominant propagation link, thereby extending the traditional single-point oscillation tracing result into a propagation path identification result with directionality and topological interpretation.

[0019] 2. By using a network topology model to construct hierarchical relationships between nodes, and under candidate source conditions, Granger causality scores of nodes are constructed through forward output, reverse return and bidirectional coupling terms. This effectively suppresses the interference of synchronous response or common driving factors between non-adjacent nodes on causal analysis, making the oscillation source location and propagation direction determination more consistent with the actual electrical coupling characteristics of wind farms.

[0020] 3. Based on the oscillation source and the dominant propagation link, further statistical analysis is conducted on the forward oscillation propagation capacity, upstream reception capacity, and downstream distribution capacity of each node. Key defense scores and reception-distribution coordination coefficients are constructed to identify the central nodes that play a major relay support role in the oscillation diffusion process. This provides a clear analytical basis for oscillation monitoring, operational risk assessment, and differentiated auxiliary decision-making for wind farm clusters. Attached Figure Description

[0021] 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.

[0022] Figure 1 A flowchart of a method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations provided in an embodiment of the present invention; Figure 2This is a schematic diagram of the topology of wind farm cluster nodes and electrical connection edges provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the propagation intensity matrix of the side-level oscillation provided in an embodiment of the present invention; Figure 4 A schematic diagram of a side-level Granger causality matrix provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the oscillation source identification results provided in an embodiment of the present invention; Figure 6 A schematic diagram illustrating the identification results of the initial propagation channel and the dominant propagation link provided in an embodiment of the present invention; Figure 7 This is a schematic diagram showing the priority defense scoring and ranking of key propagation nodes provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1; like Figure 1 As shown, this embodiment provides a method for tracing the source and identifying the dominant propagation link of wind farm group oscillations. Under physical adjacency topological constraints, it performs oscillation propagation analysis by combining edge-level oscillation propagation intensity, edge-level Granger causality intensity, and oscillation timing consistency. The method includes the following steps: S1. Obtain synchronous measurement data and topology connection relationship of multiple electrical nodes in the wind farm group; S2. Based on the aforementioned topological connection relationship, construct a network topology model consisting of a set of nodes and a set of electrical connection edges, and generate a physical adjacency matrix; S3. Preprocess the synchronous measurement data of each node and extract the dominant oscillation mode response within the target oscillation frequency band; S4. Based on the dominant oscillation mode response, construct the input-output transfer matrix of the wind farm group, and calculate the edge-level oscillation propagation intensity between adjacent nodes under the constraint of the physical adjacency matrix. S5. Under the constraints of the physical adjacency matrix, establish a local vector autoregressive model for the dominant oscillation mode response of adjacent nodes and calculate the edge-level Granger causality strength. S6. Based on the relationship between the edge-level Granger causality strength and the topological hierarchy, construct the node Granger causality score under the candidate source condition, and determine the oscillation source node based on the score; S7. Identify the initial propagation channel based on the oscillation source node, the edge-level Granger causality strength, the edge-level oscillation propagation strength, and the consistency of the start-up timing of each node; S8. Taking the initial propagation channel as the starting point of the link, and combining the edge-level comprehensive score and the downstream node Granger causality score, expand and determine the dominant propagation link segment by segment. S9 outputs the oscillation source node, initial propagation channel, dominant propagation link, key propagation edge order, and key propagation node identification results.

[0025] By integrating the edge-level Granger causality (GC) intensity, edge-level oscillation propagation intensity, and oscillation timing consistency, it sequentially achieves oscillation source node localization, initial propagation channel identification, dominant propagation link determination, and key propagation node quantitative screening under physical topological constraints. This expands the traditional single-point tracing results into complete propagation path identification results with directionality and topological interpretation. At the same time, it uses node Granger causality scores under candidate source conditions to suppress interference between non-adjacent nodes, improving the physical rationality and engineering applicability of the analysis results.

[0026] The following provides further explanation of each step and related features in the above method; In this embodiment, S1, synchronous measurement data and topology connection relationship of multiple electrical nodes in the wind farm group are obtained; The wind farm cluster refers to a power generation cluster consisting of at least two wind turbines and their associated electrical equipment such as box-type transformers, collection lines, collection busbars, step-up substations, and transmission channels. The electrical nodes within the cluster are coupled to each other through physical electrical connections, enabling the collection of synchronous measurement data and the description of topological relationships. The acquired synchronous measurement data includes at least one time-series measurement information of voltage, current, active power, reactive power, frequency, and phase angle of each electrical node, as well as operating status quantities such as DC bus voltage, AC side current, and phase-locked loop output frequency from the converter local controller or wide-area measurement system; the acquired topology connection relationship includes the electrical connection path and connection method between the wind turbine grid-side converter, box-type transformer, collector line, collection bus, step-up substation bus, and transmission channel.

[0027] In this embodiment S2, based on the topological connection relationship, a network topology model consisting of a set of nodes and a set of electrical connection edges is constructed, and a physical adjacency matrix is ​​generated; like Figure 2 As shown, the wind farm cluster node and electrical connection edge topology consists of wind farm nodes, collection nodes, switchyard or booster station nodes, transmission lines, and the main grid; the wind farm areas of different colors in the figure are used to distinguish different farms or different collection branches. 、 、 as well as 、 、 The green and blue connecting lines represent the wind turbine units or equivalent wind power units connected to the corresponding power station; the solid circles represent the collection lines or equivalent electrical connection edges inside the power station; the switching station or step-up station nodes are connected to the transmission lines and are connected to the main grid through the transmission channels.

[0028] This step specifically includes: The wind turbine grid-side converter, wind power collection node, box-type transformer node, collector line node, substation busbar node, and transmission channel node are taken as network nodes, resulting in a node set. V ; The set of edges is obtained by treating the collector lines, transformer connections, busbar connections, transmission lines, and equivalent electrical connections between nodes as electrical connection edges. E ; A physical adjacency matrix is ​​constructed based on whether there are direct electrical connections between nodes. Its elements This indicates that node i and node j have a direct electrical connection. This indicates that there is no direct electrical connection; By abstracting a wind farm cluster into a network topology model consisting of a set of nodes and a set of electrical connection edges, and generating a physical adjacency matrix, the direct electrical connection relationships between nodes are clarified, providing topological constraints for subsequent propagation intensity and causal analysis only between adjacent nodes.

[0029] In this embodiment, S3, the synchronization measurement data of each node is preprocessed, and the dominant oscillation mode response within the target oscillation frequency band is extracted; including: The synchronous measurement signals of each node are sequentially subjected to mean removal, trend removal, target frequency band bandpass filtering, and standardization to obtain the preprocessed signal; The preprocessed signal is constructed using a delayed embedding strategy, and discrete-time eigenvalues ​​and corresponding modes are obtained using the Koopman spectral decomposition algorithm. The delayed embedding strategy refers to selecting an embedding dimension and a delay step size d for a one-dimensional time series signal at a single node, and then embedding the signal value at the current time t with its previous d-dimensional values. The signal values ​​at one historical moment are combined into a d-dimensional vector; the constructed up-dimensional observation vector can map the original low-dimensional time series data into a higher-dimensional state space, thereby more fully revealing the nonlinear characteristics and modal information in the oscillatory dynamics. The discrete-time feature values Convert to continuous-time eigenvalues and according to Calculate the oscillation frequency of each mode. Damping ratio and modal energy ;in The sampling interval is... The damping coefficient is... The angular frequency of the oscillation is given by j, where j is the imaginary unit. Based on the preset target subsynchronous frequency band constraints and modal energy threshold The dominant oscillation mode was screened out, and the response components of each node under the dominant oscillation mode were extracted, providing the modally purified analysis signal for subsequent calculation of the side propagation intensity, Granger causality intensity, and oscillation timing identification.

[0030] In this embodiment, S4, based on the dominant oscillation mode response, an input-output transfer matrix of the wind farm group is constructed, and under the constraint of the physical adjacency matrix, the edge-level oscillation propagation intensity between adjacent nodes is calculated; including: Near the steady-state operating point of the wind farm cluster, a small-disturbance linearized model is established, including the grid-side converter control loop, filtering loop, and network coupling relationships. This includes defining the state variable deviation vector of the linearized model as... The perturbation input vector is The measurement output vector is The state-space equation can then be written as:

[0031] In the formula: Let A be the time derivative of the state variable deviation vector, B be the input matrix, C be the output matrix, and D be the direct transfer matrix. These are obtained by small-signal linearization of the original nonlinear differential model at the steady-state operating point of the system. Based on the state matrix A, input matrix B, output matrix C, and direct transfer matrix D of the small perturbation linearization model, construct the input-output transfer matrix;

[0032] In the formula: s is a complex frequency domain variable, and I is the identity matrix. elements This represents the frequency domain transfer relationship between the disturbance input at node i and the output response at node j; The frequency domain gain of the equivalent small disturbance input at the grid-connected port at node i to the output response at node j is defined as the oscillation propagation intensity from node i to node j:

[0033] In the formula: For frequency The response of the output variable of node j to the input disturbance of node i; This represents the input perturbation at node i; The 2-norm of the input perturbation vector; The frequency corresponding to the dominant oscillation mode At this point, calculate the oscillation propagation intensity of all adjacent node pairs to obtain the edge-level oscillation propagation intensity matrix;

[0034] To eliminate dimensional differences for subsequent comprehensive scoring, the edge-level GC intensity and edge-level OPI were normalized respectively:

[0035] In the formula: , Each is a set of connected edges. The maximum and minimum values ​​of the edge-level OPI; Equation (11) maps the edge-level GC intensity and edge-level OPI to the [0,1] interval; like Figure 3 As shown, the above-mentioned side-level oscillation propagation intensity matrix has the source node as the row direction and the target node as the column direction. The colored cells in the matrix represent the normalized frequency domain propagation gain between adjacent nodes from node input disturbance to node output response at the dominant subsynchronous oscillation frequency. In this step, an input-output transfer matrix was constructed based on a small-perturbation linearization model, and the edge-level oscillation propagation intensity between adjacent nodes was calculated under physical adjacency constraints. The normalized frequency domain propagation gain was obtained, providing a physical-level quantitative indicator for distinguishing the direction and intensity of oscillation propagation.

[0036] In this embodiment, S5, under the constraint of the physical adjacency matrix, a local vector autoregressive model is established for the dominant oscillation mode response of adjacent nodes, and the edge-level Granger causality strength is calculated; including: Based on the physical adjacency matrix, adjacent node pairs with direct electrical connections are selected, and a local vector autoregressive model is established only for the adjacent node pairs. For adjacent nodes i and j, construct a restricted model that includes only the history of node j itself:

[0037] And an unrestricted model that includes both the history of node j and the history of node i:

[0038] In the formula: p is the lag order of the VAR model; , , These are the coefficients of the VAR model; , For model residuals; , Let k be the dominant oscillation mode response sequence of node i and node j; k is the hysteresis index term. Compare the residual variances of the restricted and unrestricted models, and calculate the initial edge-level Granger causality strength from node i to node j:

[0039] In the formula: and They are respectively and variance Used to characterize the directed statistical driving strength of node i to node j; Insignificant causal edges are removed using a significance test or F-test, and then normalized to obtain the edge-level Granger causality strength:

[0040] In the formula: The normalized boundary-level Granger causality strength; , Each is a set of connected edges. The maximum and minimum values ​​of the GC intensity at the upper level.

[0041] like Figure 4 As shown, the edge-level Granger causality matrix uses the source node as the row direction and the target node as the column direction. Non-zero colored cells in the matrix represent significant directed Granger causality relationships between physically adjacent nodes, and the values ​​within the cells represent the normalized edge-level Granger causality strength. Blank cells indicate that there is no direct physical connection between nodes, the causal relationship has not passed the significance test, or the direction has not been included in the effective propagation edge. This matrix is ​​used to reflect the directed statistical driving relationship between adjacent nodes under the target oscillation mode. Figure 3 and Figure 4 The two types of matrices are used together for subsequent initial propagation channel scoring, dominant propagation link expansion, and key propagation node identification.

[0042] In this embodiment, S6, based on the relationship between the edge-level Granger causality strength and the topological hierarchy, a node Granger causality score is constructed under the candidate source conditions, and the oscillation source node is determined based on the score; including: Each electrical node is sequentially designated as a candidate source node, and the topological hierarchy of other nodes relative to the candidate source node is calculated based on the shortest path length from the candidate source node to other nodes. Based on the topological hierarchy, the adjacent nodes of each node are divided into deep adjacent nodes, same-level adjacent nodes, and shallow adjacent nodes.

[0043] In the formula: Let i be the set of adjacent physical nodes; , and These represent the sets of deep adjacent nodes, the sets of adjacent nodes at the same level, and the sets of shallow adjacent nodes, respectively. Based on deep adjacent nodes, same-layer adjacent nodes, and shallow adjacent nodes, forward output consistency terms, upstream input compensation terms, bidirectional coupling terms, and reverse backflow terms are constructed respectively. Forward output consistency terms:

[0044] In the formula: This is a reduction factor for the same-layer edge, used to reduce the impact of lateral coupling within the same layer on the determination of the main propagation direction. interference; Upstream input compensation items:

[0045] Bidirectional coupling terms:

[0046] Reverse reflux term:

[0047] In the formula: Input the compensation term upstream of node i; This represents the bidirectional GC coupling amount between node i and node j; For node i, the reverse backflow term; This is the reverse backflow suppression coefficient or the reverse causality cancellation coefficient; Combining the topological hierarchy weights and the above items, calculate the node Granger causality score for each node under the candidate source conditions; Topology level weights:

[0048] In the formula: This represents the level penalty coefficient. Node Granger Causality Score:

[0049] In the formula: , , , These are the weight coefficients for the forward output term, upstream input term, bidirectional coupling term, and reverse flow term, respectively. The consistency between the candidate source node's own Granger causality score and the score distribution of the entire network is fused to calculate the comprehensive score of each candidate source.

[0050] In the formula: For global consistency weights; Indicates when node s When used as a candidate oscillation source, its own node GC score; The node with the highest comprehensive score among the candidate sources is determined as the oscillation source node:

[0051] In the formula: To estimate the oscillation source node; Equation (24) comprehensively characterizes the hierarchical position, forward propagation capability and local coupling characteristics of the node under the candidate source assumption, thereby improving the reliability of the oscillation source localization result; like Figure 5 As shown, the oscillation source identification results can be represented by a node score ranking bar chart; the horizontal axis represents the node number, the vertical axis represents the comprehensive score of the candidate source under the candidate source conditions, and the height of the bar represents the probability that the corresponding node is an oscillation source or a key source-side node; different colors are used to distinguish node types such as oscillation source nodes, wind farm nodes, 220kV collection station nodes, and 750kV main grid AC station nodes; the node with the highest score and the strongest consistency with the propagation level of the entire network is identified as the estimated oscillation source node; This step involves setting candidate source nodes, dividing the topology hierarchy, and constructing four scoring criteria: forward output, upstream input, bidirectional coupling, and reverse return flow. This yields the Granger causality score of the node under the candidate source conditions. Finally, the node with the highest comprehensive score is determined as the oscillation source node, thus achieving topologically interpretable oscillation source localization.

[0052] In this embodiment, S7, based on the oscillation source node, the edge-level Granger causality strength, the edge-level oscillation propagation strength, and the consistency of the start-up timing of each node, the initial propagation channel is identified; including: Response of dominant oscillation modes at each node Perform a Hilbert transform to construct an oscillating envelope signal:

[0053] In the formula: For Hilbert transform operators; The oscillation start time of each node is determined based on the moment when the oscillation envelope signal first exceeds the oscillation threshold.

[0054] In the formula: For nodes i The moment when it first enters an oscillation state; The threshold for determining oscillation; Based on the timing relationship that the upstream node starts oscillating first and the downstream node responds later, an oscillation timing consistency index is constructed for the candidate propagation edges:

[0055] In the formula: This is the timing smoothing coefficient; this index is used to measure the consistency between the side direction and the order of oscillation initiation: when When, explain the node j The oscillation starts no later than the node. i If so, that side should not be considered the initial propagation channel; All outgoing electrical connection edges of the oscillation source node are used as the candidate edge set for the initial propagation channel:

[0056] In the formula: The threshold for oscillation timing consistency is set; an edge is included in the candidate set of initial propagation channels only if the edge direction satisfies the oscillation timing consistency requirement. Candidate edges are screened based on the aforementioned oscillation timing consistency index, and the overall score of the initial propagation channel is calculated by combining the edge-level Granger causality strength, edge-level oscillation propagation strength, and oscillation timing consistency. Determine the initial dissemination channel;

[0057] In the formula: and These are the normalized edge-level GC intensity and edge-level OPI, respectively. , and These are the weighting coefficients for the three indicators; This step determines the oscillation start time of each node based on the Hilbert transform, constructs an oscillation timing consistency index, and integrates the edge-level Granger causality strength and the edge-level oscillation propagation strength. This step screens and identifies the initial propagation channel from the outward edges of the oscillation source node, clarifying the earliest path starting point for the outward diffusion of oscillation energy.

[0058] In this embodiment S8, taking the initial propagation channel as the starting point of the link, and combining the edge-level comprehensive score and the downstream node Granger causality score, the dominant propagation link is expanded segment by segment and determined; including: Suppose that the currently identified dominant propagation path is:

[0059] In the formula: For the first in the propagation chain k One node; L Link length; The downstream node corresponding to the initial propagation channel; the terminal node of the initial propagation channel is taken as the current link endpoint; Starting from the current link endpoint, candidate propagation edges that satisfy the conditions of no backtracking at the topological level, consistent oscillation order, and no repeated visits to nodes are selected from its adjacent edges:

[0060] In the formula: A set of candidate edges that simultaneously satisfies the following requirements: no backtracking of hierarchy, consistent oscillation sequence, and no repeated visits to selected nodes; Based on the edge-level comprehensive score of the candidate propagation edge and the node Granger causality score of the downstream node, a link extension score is constructed:

[0061] In the formula: and These are the weight coefficients for the edge-level comprehensive scoring item and the node GC scoring item, respectively. The normalized GC score for downstream nodes; Select the candidate propagation edge with the highest link expansion score as the next dominant propagation edge, add the corresponding downstream node to the dominant propagation link, and update the current link endpoint; Repeat the above steps until the extended termination condition is met to obtain the complete dominant propagation chain.

[0062] This step takes the initial propagation channel as the starting point of the link, and selects candidate propagation edges that meet the conditions of non-backtracking topology, consistent oscillation sequence, and no repeated node visits by segment. Combined with the link expansion score, it completes the expansion determination of the dominant propagation link, ultimately forming a complete directed propagation path from the oscillation source node along the initial propagation channel to the distant node; for example... Figure 6 As shown, the identification results are output in the form of a node-edge topology graph, where red circles mark the oscillation source nodes, solid arrows represent the initial propagation channels, and dashed or dotted arrows represent the subsequent dominant propagation edges and lateral propagation directions. The source node location, initial propagation channel identification, and dominant propagation link determination results are uniformly mapped onto the physical topology, thereby providing operators with an intuitive panoramic view of oscillation propagation.

[0063] In this embodiment, S9 outputs the oscillation source node, the initial propagation channel, the dominant propagation link, the order of key propagation edges, and the identification results of key propagation nodes.

[0064] The identification of key propagation nodes and central propagation nodes includes: Given a fixed oscillation source node, the effective outward oscillation propagation capability of each node along the main propagation direction and the lateral direction at the same level is statistically analyzed to obtain the forward oscillation propagation capability of the node:

[0065] Normalize the forward oscillation propagation capability of the node to obtain ; and compared it with the normalized node Granger causality score The results are merged to obtain a priority defense score for measuring node monitoring priorities:

[0066] In the formula: This represents the normalized forward oscillation propagation capability. The normalized node GC score; and These are the weight coefficients for the forward propagation capability term and the node GC score term, respectively. Statistically count the upstream throughput received by each node from shallow neighboring nodes, and the downstream throughput output to deep and same-layer neighboring nodes;

[0067] In the formula: Represents a node i The intensity of oscillation propagation received from shallower nodes; Represents a node i The intensity of oscillation propagation distributed to deeper nodes and nodes at the same level; Construct a receiving-distribution coordination coefficient based on the upstream receiving volume and the downstream distribution volume. This is used to characterize the relay support capability of a node in the oscillation propagation link; The key defense score is combined with the receiving-distribution coordination coefficient to obtain the propagation center score. And determine the key propagation nodes and propagation hub nodes based on preset thresholds:

[0068] In the formula: For the set of central nodes of propagation; , and These are the key defense scoring threshold, the receiving-distribution coordination coefficient threshold, and the propagation center scoring threshold, respectively. like Figure 7As shown, the priority defense scores of the key propagation nodes can be output in the form of a bar chart; the horizontal axis of the chart is the node number, the vertical axis is the priority defense score obtained by fusing the node's forward oscillation propagation capability and the node's Granger causality score, and the height of the bars represents the monitoring and defense priority of each node in the oscillation propagation link; different colors are used to distinguish oscillation source nodes, wind farm nodes, 220kV collection station nodes and 750kV main grid AC station nodes; The final oscillation propagation identification result, generated by combining the outputs of each step, includes: oscillation source node, initial propagation channel, dominant propagation link, critical propagation edge ranking, critical propagation nodes, and propagation center node. This identification result can be output through a graphical interface, data file, monitoring platform interface, or online analysis platform on the dispatching side, and can be used for oscillation propagation mechanism analysis, operational risk assessment, and decision support in wind farm clusters.

[0069] This embodiment constructs a node-edge network topology model of a wind farm cluster and applies physical adjacency constraints. It sequentially calculates the edge-level oscillation propagation intensity and edge-level Granger causality intensity between adjacent nodes, and then uses the node Granger causality score under candidate source conditions to locate the oscillation source node. Based on this, it integrates the consistency of the oscillation timing to identify the initial propagation channel, and uses the initial channel as a starting point to progressively expand and determine the complete dominant propagation link. Finally, through quantitative screening of forward oscillation propagation capability and the receiving-distributing coordination coefficient, it outputs key propagation nodes and propagation hub nodes. The entire process expands the single-point tracing results into a multi-level source-chain-point identification result with directionality and topological interpretation, thus effectively serving the practical engineering needs of wind farm cluster oscillation propagation mechanism analysis, operational risk assessment, and auxiliary decision-making.

[0070] Example 2; A system for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations includes: Data acquisition module: used to acquire synchronous measurement data and topology connection relationships of multiple electrical nodes in a wind farm cluster; Topology modeling module: used to construct a network topology model consisting of a set of nodes and a set of electrical connection edges based on the topological connection relationship, and to generate a physical adjacency matrix; Dominant mode extraction module: used to preprocess the synchronous measurement data of each node and extract the dominant oscillation mode response within the target oscillation frequency band; Edge propagation intensity calculation module: used to construct the input-output transfer matrix of the wind farm group based on the dominant oscillation mode response, and calculate the edge oscillation propagation intensity between adjacent nodes under the constraint of the physical adjacency matrix; Edge-level Granger causality analysis module: used to establish a local vector autoregressive model for the dominant oscillation mode response of adjacent nodes under the constraints of the physical adjacency matrix, and to calculate the edge-level Granger causality strength; Oscillation source localization module: used to construct node Granger causality scores under candidate source conditions based on the edge-level Granger causality strength and topological hierarchy relationship, and to determine the oscillation source node based on the scores; Initial propagation channel identification module: used to identify the initial propagation channel based on the oscillation source node, the edge-level Granger causality intensity, the edge-level oscillation propagation intensity, and the consistency of the start-up timing of each node; Dominant propagation link determination module: used to expand and determine the dominant propagation link segment by segment, taking the initial propagation channel as the starting point of the link and combining the edge-level comprehensive score and the downstream node Granger causality score; The results output module is used to output the identification results of the oscillation source node, the initial propagation channel, the dominant propagation link, the order of key propagation edges, and the key propagation nodes.

[0071] This system can be deployed in wind farm clusters at the site level, in booster station oscillation monitoring terminals, in centralized control center online analysis platforms, or in dispatch-side broadband oscillation online analysis platforms. By accessing data sources such as phasor measurement units, wide-area measurement systems, fault recorders, or converter local controllers, the system acquires real-time synchronous measurement data and topology connections of each electrical node. It can perform online oscillation source location, initial propagation channel identification, dominant propagation link determination, and key propagation node screening. The identification results can be used by operators to trace the causes of oscillation events, analyze propagation paths, and provide early warnings for key equipment. It is also suitable for oscillation propagation analysis in new energy sites such as photovoltaic power plants and energy storage power stations that contain multiple types of power electronic equipment. It achieves fully automated, online oscillation source tracing and propagation link identification from data access to result output, effectively supporting the online monitoring and defense deployment of wind farm cluster oscillations.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations, characterized in that, Includes the following steps: S1. Obtain synchronous measurement data and topology connection relationship of multiple electrical nodes in the wind farm group; S2. Based on the aforementioned topological connection relationship, construct a network topology model consisting of a set of nodes and a set of electrical connection edges, and generate a physical adjacency matrix; S3. Preprocess the synchronous measurement data of each node and extract the dominant oscillation mode response within the target oscillation frequency band; S4. Based on the dominant oscillation mode response, construct the input-output transfer matrix of the wind farm group, and calculate the edge-level oscillation propagation intensity between adjacent nodes under the constraint of the physical adjacency matrix. S5. Under the constraints of the physical adjacency matrix, establish a local vector autoregressive model for the dominant oscillation mode response of adjacent nodes and calculate the edge-level Granger causality strength. S6. Based on the relationship between the edge-level Granger causality strength and the topological hierarchy, construct the node Granger causality score under the candidate source condition, and determine the oscillation source node based on the score; S7. Identify the initial propagation channel based on the oscillation source node, the edge-level Granger causality strength, the edge-level oscillation propagation strength, and the consistency of the start-up timing of each node; S8. Taking the initial propagation channel as the starting point of the link, and combining the edge-level comprehensive score and the downstream node Granger causality score, expand and determine the dominant propagation link segment by segment. S9 outputs the oscillation source node, initial propagation channel, dominant propagation link, key propagation edge order, and key propagation node identification results.

2. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S2 includes: The wind turbine grid-side converter, wind power collection node, box-type transformer node, collector line node, substation bus node, and transmission channel node are taken as network nodes, resulting in node set V; The set of edges E is obtained by taking the collector lines, transformer connections, busbar connections, transmission lines and equivalent electrical connections between nodes as electrical connection edges; A physical adjacency matrix is ​​constructed based on whether there are direct electrical connections between nodes. Its elements This indicates that node i and node j have a direct electrical connection. This indicates that there is no direct electrical connection.

3. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S3 includes: The synchronous measurement signals of each node are sequentially subjected to mean removal, trend removal, target frequency band bandpass filtering, and standardization to obtain the preprocessed signal; The preprocessed signal is constructed using a delayed embedding strategy, and discrete-time eigenvalues ​​and corresponding modes are obtained using the Koopman spectral decomposition algorithm. The discrete-time characteristic values ​​are converted into continuous-time characteristic values, and the oscillation frequency, damping ratio, and modal energy of each mode are calculated. Based on the preset target subsynchronous frequency band constraints and modal energy thresholds, the dominant oscillation mode is selected, and the response components of each node under the dominant oscillation mode are extracted.

4. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S4 includes: Near the steady-state operating point of the wind farm group, a small-disturbance linearization model is established, including the grid-side converter control link, the filtering link, and the network coupling relationship. Based on the state matrix A, input matrix B, output matrix C, and direct transfer matrix D of the small perturbation linearization model, construct the input-output transfer matrix; The frequency domain gain of the equivalent small disturbance input at the grid-connected port at node i to the output response at node j is defined as the oscillation propagation intensity from node i to node j. At the frequency corresponding to the dominant oscillation mode, the oscillation propagation intensity of all adjacent node pairs is calculated to obtain the edge-level oscillation propagation intensity matrix; The edge-level oscillation propagation intensity matrix is ​​normalized to obtain the edge-level propagation intensity.

5. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S5 includes: Based on the physical adjacency matrix, adjacent node pairs with direct electrical connections are selected, and a local vector autoregressive model is established only for the adjacent node pairs. For adjacent nodes i and j, construct a restricted model that includes only the history of node j itself, and an unrestricted model that includes both the history of node j and the history of node i. Compare the residual variances of the restricted model and the unrestricted model, and calculate the initial edge-level Granger causality strength from node i to node j; Insignificant causal edges are removed using significance tests or F-tests, and then normalized to obtain the edge-level Granger causal strength.

6. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S6 includes: Each electrical node is sequentially designated as a candidate source node, and the topological hierarchy of other nodes relative to the candidate source node is calculated based on the shortest path length from the candidate source node to other nodes. Based on the topological hierarchy, the adjacent nodes of each node are divided into deep adjacent nodes, same-level adjacent nodes, and shallow adjacent nodes. Based on deep adjacent nodes, same-layer adjacent nodes, and shallow adjacent nodes, forward output consistency terms, upstream input compensation terms, bidirectional coupling terms, and reverse backflow terms are constructed respectively. Combining the topological hierarchy weights and the above items, calculate the node Granger causality score for each node under the candidate source conditions; The consistency between the candidate source node's own Granger causality score and the score distribution of the entire network is fused to calculate the comprehensive score of each candidate source. The node with the highest comprehensive score among the candidate sources is determined as the oscillation source node.

7. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S7 includes: Hilbert transform is performed on the dominant oscillation mode response of each node to construct the oscillation envelope signal; The oscillation start time of each node is determined based on the moment when the oscillation envelope signal first exceeds the oscillation start threshold. Based on the timing relationship that the upstream node starts oscillating first and the downstream node responds later, an oscillating timing consistency index is constructed for the candidate propagation edge. All outgoing electrical connection edges of the oscillation source node are used as the candidate edge set for the initial propagation channel; Candidate edges are screened based on the oscillation timing consistency index, and the initial propagation channel is determined by comprehensively considering the edge-level Granger causality strength, edge-level oscillation propagation strength, and oscillation timing consistency.

8. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, S8 includes: The terminal node of the initial propagation channel is taken as the current link endpoint; Starting from the current link endpoint, candidate propagation edges that meet the conditions of no backtracking in topology hierarchy, consistent oscillation order, and no repeated visits to nodes are selected from its adjacent edges. A link extension score is constructed based on the edge-level comprehensive score of the candidate propagation edge and the node Granger causality score of the downstream node. Select the candidate propagation edge with the highest link expansion score as the next dominant propagation edge, add the corresponding downstream node to the dominant propagation link, and update the current link endpoint; Repeat the above steps until the extended termination condition is met to obtain the complete dominant propagation chain.

9. The method for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations according to claim 1, characterized in that, In S9, the identification of key propagation nodes and propagation hub nodes includes: Given the established oscillation source nodes, the effective outward oscillation propagation capability of each node along the main propagation direction and the lateral direction of the same layer is statistically analyzed to obtain the forward oscillation propagation capability of the node. The forward oscillation propagation capability of the nodes is normalized and fused with the normalized node Granger causality score to obtain a key defense score for measuring the node monitoring priority. Statistically count the upstream throughput received by each node from shallow neighboring nodes, and the downstream throughput output to deep and same-layer neighboring nodes; Based on the upstream receiving volume and the downstream distributing volume, a receiving-distributing coordination coefficient is constructed to characterize the relay support capability of the node in the oscillation propagation link; The key defense score is fused with the receiving-distribution coordination coefficient to obtain the propagation center score, and key propagation nodes and propagation center nodes are determined according to preset thresholds.

10. A system for tracing the source and identifying the dominant propagation link of wind farm cluster oscillations, characterized in that, include: Data acquisition module: used to acquire synchronous measurement data and topology connection relationships of multiple electrical nodes in a wind farm cluster; Topology modeling module: used to construct a network topology model consisting of a set of nodes and a set of electrical connection edges based on the topological connection relationship, and to generate a physical adjacency matrix; Dominant mode extraction module: used to preprocess the synchronous measurement data of each node and extract the dominant oscillation mode response within the target oscillation frequency band; Edge propagation intensity calculation module: used to construct the input-output transfer matrix of the wind farm group based on the dominant oscillation mode response, and calculate the edge oscillation propagation intensity between adjacent nodes under the constraint of the physical adjacency matrix; Edge-level Granger causality analysis module: used to establish a local vector autoregressive model for the dominant oscillation mode response of adjacent nodes under the constraints of the physical adjacency matrix, and to calculate the edge-level Granger causality strength; Oscillation source localization module: used to construct node Granger causality scores under candidate source conditions based on the edge-level Granger causality strength and topological hierarchy relationship, and to determine the oscillation source node based on the scores; Initial propagation channel identification module: used to identify the initial propagation channel based on the oscillation source node, the edge-level Granger causality intensity, the edge-level oscillation propagation intensity, and the consistency of the start-up timing of each node; Dominant propagation link determination module: used to expand and determine the dominant propagation link segment by segment, taking the initial propagation channel as the starting point of the link and combining the edge-level comprehensive score and the downstream node Granger causality score; The results output module is used to output the identification results of the oscillation source node, the initial propagation channel, the dominant propagation link, the order of key propagation edges, and the key propagation nodes.

Images