Broadband oscillation source positioning method and device, electronic equipment, computer readable storage medium and program product

Abstract

The embodiment of the invention provides a broadband oscillation source positioning method and device, electronic equipment, a computer readable storage medium and a program product, and relates to the technical field of electric power. The method adopts a data driving strategy, does not need to depend on accurate physical modeling of a power system, and is suitable for a modern power system with a complex operation state and a changeable topological structure. The causal influence intensity between generator sets is analyzed by introducing transfer entropy, an adjacent matrix is constructed, and reasonable graph structure modeling under the condition of no topological prior is realized. Deep feature extraction is carried out on active power time sequence data of the generator set through the gating circulation unit, dynamic evolution characteristics in the broadband oscillation process are fully captured, and the representation capacity of nodes is enhanced. A causal graph structure and enhanced features are fused in a graph neural network model, space-time joint modeling is realized, the description capability of a disturbance energy propagation path is effectively improved, the spatial positioning precision of a forced oscillation disturbance source is greatly improved, and the method has good engineering applicability and popularization value.

Description

Technical Field

[0001] This invention relates to the field of power technology, and more specifically, to a broadband oscillation source location method, apparatus, electronic device, computer-readable storage medium, and program product. Background Technology

[0002] Against the backdrop of "dual-carbon" goals, high-proportion renewable energy generation, and the widespread application of power electronic equipment, power systems are exhibiting highly electronic characteristics, leading to a significant increase in the frequency of forced oscillations. This has seriously threatened system stability and the safe operation of equipment. Unlike traditional spontaneous oscillations, forced oscillations are typically triggered by periodic disturbances in the control system, equipment failures, or external periodic signals. They have a fixed frequency and stable amplitude, and can propagate over a wider area through the system structure, resulting in degraded power quality and even cascading failures.

[0003] For forced oscillation events with a clearly identifiable disturbance source, quickly and accurately locating the oscillation source is the primary task in suppressing forced oscillations. Therefore, given that only unit signals are collected and there is no direct physical connection between nodes, a method for locating broadband oscillation sources is urgently needed. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a broadband oscillator positioning method, apparatus, electronic device, computer-readable storage medium and program product, which can improve the positioning efficiency and accuracy of broadband oscillator sources.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows: In a first aspect, the present invention provides a method for locating a broadband oscillation source, the method comprising: The active power data of each generator set in the power grid system are normalized to obtain the power time series of each generator set; Calculate the transfer entropy between any two generator sets, and generate an adjacency matrix based on all the transfer entropies; the adjacency matrix is ​​used to characterize the causal relationships between the generator sets in the power system. By using a pre-trained gated recurrent unit to capture the time evolution of the active power corresponding to each generator set based on the power time series of each generator set, the enhanced features of each generator set are obtained. A pre-trained graph neural network model is used to identify broadband oscillation sources based on the adjacency matrix and the enhanced features of all the generator sets, thereby obtaining the probability that each generator set is a broadband oscillation source.

[0006] In an optional implementation, calculating the transfer entropy between any two generator sets includes: Any two generator sets are designated as the first generator set and the second generator set. Calculate the conditional entropy of the first generator set based on the power time series of the first generator set; Calculate the conditional entropy of the second generator set based on the power time series of the second generator set; Calculate the joint conditional entropy of the first generator set and the joint conditional entropy of the second generator set based on the power time series of the first generator set and the power time series of the second generator set. The transfer entropy from the first generator set to the second generator set is determined based on the conditional entropy and joint conditional entropy of the first generator set. The transfer entropy from the second generator set to the first generator set is determined based on the conditional entropy and the joint conditional entropy of the second generator set.

[0007] In an optional implementation, generating the adjacency matrix based on all the transfer entropies includes: The difference between the transfer entropy from the first generator set to the second generator set and the transfer entropy from the second generator set to the first generator set in any two generator sets is defined as the first difference. The difference between the transfer entropy from the second generator set to the first generator set and the transfer entropy from the first generator set to the second generator set in any two generator sets is determined as the second difference. The causal influence coefficients of the first generator set on the second generator set and the second generator set on the first generator set are determined based on the first difference, the second difference, and the preset transfer threshold; all the causal influence coefficients constitute the adjacency matrix.

[0008] In an optional implementation, determining the causal influence coefficient of the first generator set on the second generator set and the causal influence coefficient of the second generator set on the first generator set based on the first difference, the second difference, and a preset transfer threshold includes: If the first difference is not less than the preset transfer threshold, the causal influence coefficient of the first generator set on the second generator set is determined as the first preset value; If the first difference is less than the preset transfer threshold, the causal influence coefficient of the first generator set on the second generator set is determined as the second preset value; If the second difference is not less than the preset transfer threshold, the causal influence coefficient of the second generator set on the first generator set is determined as the first preset value; If the second difference is less than the preset transfer threshold, the causal influence coefficient of the second generator set on the first generator set is determined as the second preset value.

[0009] In an optional implementation, the gated recurrent unit includes multiple hidden layers; the step of using a pre-trained gated recurrent unit to capture the time evolution of the active power corresponding to each generator set based on the power time series of each generator set, and obtaining the enhanced features of each generator set, includes: The power time series of each generator set is input into the gated loop unit; The data in the power time series are processed one by one using the first hidden layer, the hidden state corresponding to the power time series is output, and the hidden state output by the first hidden layer is input into the next hidden layer for processing, until the hidden state output by the last hidden layer is obtained; The hidden state output by the last hidden layer is determined as the enhanced feature of the corresponding generator set.

[0010] In an optional implementation, the graph neural network model includes a multi-layer graph neural network and a classifier, with the last layer of the graph neural network connected to the classifier; the step of using a pre-trained graph neural network model to identify broadband oscillation sources based on the adjacency matrix and the enhanced features of all the generator sets, and obtaining the probability that each generator set is a broadband oscillation source, includes: The adjacency matrix and the enhanced features of all the generator sets are input into the graph neural network model; The enhanced features of all generator sets are updated using the first layer of the graph neural network based on the adjacency matrix, the weight matrix of the first layer of the graph neural network, and the bias of the first layer of the graph neural network, and the resulting feature update matrix is ​​input into the next layer of the graph neural network. For each layer of the graph neural network after the first layer, the graph neural network updates the feature update matrix output by the previous layer based on the adjacency matrix, the weight matrix of the graph neural network, and the bias of the graph neural network, until the feature update matrix output by the last layer of the graph neural network is obtained. The feature update matrix output from the last layer of the graph neural network is input into the classifier; The classifier is used to identify each generator set as a broadband oscillation source based on the feature update matrix output by the last layer of the graph neural network, thereby obtaining the probability that each generator set is a broadband oscillation source.

[0011] In a second aspect, the present invention provides a broadband oscillation source positioning device, the device comprising: The processing module is used to normalize the active power data of each generator set in the power grid system to obtain the power time series of each generator set. A construction module is used to calculate the transfer entropy between any two generator sets and generate an adjacency matrix based on all the transfer entropies. The adjacency matrix is ​​used to characterize the causal relationship between each generator set in the power system. A pre-trained gated recurrent unit is used to capture the time evolution law of the active power corresponding to each generator set based on the power time series of each generator set, so as to obtain the enhanced features of each generator set. The localization module is used to identify broadband oscillation sources based on the adjacency matrix and the enhanced features of all the generator sets using a pre-trained graph neural network model, and to obtain the probability that each generator set is a broadband oscillation source.

[0012] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing a computer program executable by the processor, the processor executing the computer program to implement the broadband oscillation source localization method described in any of the foregoing embodiments.

[0013] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the broadband oscillation source localization method as described in any of the foregoing embodiments.

[0014] Fifthly, the present invention provides a program product that, when executed by a processor, implements the broadband oscillation source localization method as described in any of the foregoing embodiments.

[0015] Compared to existing technologies, the broadband oscillation source localization method, device, electronic equipment, computer-readable storage medium, and program product provided in this invention employ a data-driven strategy, eliminating the need for precise physical modeling of the power system. This effectively overcomes the poor adaptability of traditional methods in complex system structures and variable operating states, making it suitable for modern power systems with complex operating states and varied topologies. Addressing the challenge of lacking clear physical connections between generator units and the difficulty in constructing graph structures, this invention innovatively introduces transfer entropy analysis to assess the causal influence strength between generator units, constructing an adjacency matrix and achieving reasonable graph structure modeling without topological priors. By using gated cyclic units to extract deep features from the active power time-series data of generator units, the long-term dependence and dynamic evolution characteristics of the broadband oscillation process are fully captured, enhancing the representational ability of nodes. Furthermore, the causal graph structure and enhanced features are integrated into the graph neural network model to achieve spatiotemporal joint modeling, effectively improving the ability to characterize the disturbance energy propagation path and significantly increasing the spatial localization accuracy of forced oscillation disturbance sources. Furthermore, this method relies solely on measurable data from the generator side, which aligns with the current data acquisition practices of wide-area measurement systems, demonstrating good engineering applicability and practical application value.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a broadband oscillation source localization method provided in an embodiment of the present invention is shown.

[0019] Figure 2 This diagram illustrates another flowchart of the broadband oscillation source localization method provided in an embodiment of the present invention.

[0020] Figure 3 A block diagram of a broadband oscillation source positioning device provided in an embodiment of the present invention is shown.

[0021] Figure 4 A block diagram of an electronic device provided in an embodiment of the present invention is shown. Detailed Implementation

[0022] 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] The inventors' research revealed that forced oscillation disturbance source localization methods are generally divided into physical mechanism-based methods and data-driven methods. Physical mechanism-based methods, such as the negative torque coefficient method, state-space method, and impedance method, typically rely on mathematical models or equivalent dynamic characteristic descriptions of the power system, offering good interpretability. However, they have significant limitations in practical engineering: firstly, the complex and frequently changing operating environment of power systems makes it extremely difficult to accurately construct a mathematical model of the entire system; secondly, these methods are highly sensitive to changes in operating points and struggle to adapt to varied disturbances and multi-mode operating conditions, resulting in insufficient stability and generalization ability of the localization results.

[0026] In contrast, data-driven methods do not rely on physical modeling of the system but directly utilize wide-area measurement data for pattern recognition and source localization, making them suitable for complex scenarios with nonlinearity and multiple perturbations. In recent years, deep learning methods such as Long Short-Term Memory (LSTM) networks and convolutional neural networks have been widely used in forced oscillation identification and analysis tasks. However, existing methods mostly focus on temporal feature extraction, neglecting the spatial topological relationships of perturbation propagation in the system, thus limiting the ability to accurately locate the spatial source of the perturbation.

[0027] To overcome the aforementioned problems, Graph Neural Networks (GNNs), due to their ability to integrate structural modeling and feature learning, have been introduced into the study of forced oscillation source localization. They can simulate the propagation process of disturbance energy between nodes based on system topology, theoretically possessing stronger spatial modeling capabilities. However, current GNN methods generally rely on the observability of the entire network's node states, assuming that all bus or load nodes in the system can obtain electrical quantity information. But in actual power systems, limited by factors such as communication network coverage and equipment deployment costs, measurement systems are often only deployed at the generator side, unable to collect data from load nodes or intermediate nodes. Furthermore, generators are not directly connected, lacking directly usable physical connections, making it impossible to directly construct a graph structure, thus limiting the application effectiveness of traditional GNNs in forced oscillation localization tasks.

[0028] Therefore, how to construct a physically meaningful equivalent adjacency matrix to support graph neural network modeling under the premise of only collecting unit signals and no direct physical connection between nodes, while efficiently extracting node temporal features, is a key technical challenge currently facing research on broadband oscillation disturbance source localization.

[0029] Based on this, embodiments of the present invention provide a broadband oscillation source localization method, device, electronic device, computer-readable storage medium, and program product. This method, by employing a data-driven strategy, does not rely on precise physical modeling of the power system, effectively overcoming the poor adaptability of traditional methods in real-world scenarios with complex system structures and variable operating states. It is applicable to modern power systems with complex operating states and varied topologies. Addressing the challenge of lacking clear physical connections between generator units and the difficulty in constructing graph structures, this method innovatively introduces transfer entropy analysis to assess the causal influence strength between generator units, constructing an adjacency matrix and achieving reasonable graph structure modeling without topological priors. Deep feature extraction of generator unit active power time-series data is performed using gated cyclic units to fully capture the long-term dependence and dynamic evolution characteristics in the broadband oscillation process, enhancing the representational ability of nodes. Furthermore, the causal graph structure and enhanced features are integrated into the graph neural network model to achieve spatiotemporal joint modeling, effectively improving the ability to characterize disturbance energy propagation paths and significantly increasing the spatial localization accuracy of forced oscillation disturbance sources. In addition, this method relies only on measurable data from the generator side, conforming to the current data acquisition status of wide-area measurement systems, and possesses good engineering applicability and practical promotion value.

[0030] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] Please refer to Figure 1 , Figure 1 A schematic flowchart of a broadband oscillation source localization method provided by an embodiment of the present invention is shown. The method includes the following steps: Step S10: Normalize the active power data of each generator set in the power grid system to obtain the power time series of each generator set.

[0032] In this embodiment of the invention, the active power data of each generator unit in the power system typically have different dimensions, amplitude ranges, and operating baselines. If directly used for subsequent modeling and analysis, the algorithm may become overly sensitive to high-amplitude generator units, affecting the model's convergence and generalization ability. Therefore, the raw data needs to be standardized before oscillation source localization.

[0033] Active power data from each generator unit in the power grid system is collected using a Wide Area Measurement System (WAMS) to obtain time-series information reflecting their dynamic operating status. The collected active power data is then normalized by mapping the raw power values ​​to a uniform numerical range (e.g., [-1,1]) according to a preset scaling rule. This eliminates scale differences between different units caused by factors such as capacity and operating conditions, thus providing a unified and comparable time-series basis for subsequent causal relationship modeling. This power time series serves as the basic data unit reflecting the dynamic behavior of the generator units and constitutes the data input source for the entire method.

[0034] Step S20: Calculate the transfer entropy between any two generator sets, and generate an adjacency matrix based on all transfer entropies; the adjacency matrix is ​​used to characterize the causal relationship between generator sets in the power system.

[0035] In this embodiment of the invention, since power systems typically only collect measurement signals from individual generator sets and cannot acquire data from load nodes or intermediate nodes, and there is no direct physical connection between generator sets, it is difficult to directly construct a graph structure based on physical topology. Under these conditions, constructing an adjacency matrix that reflects the dynamic correlation between generator sets is of great significance for achieving effective graph structure modeling and subsequent oscillation source localization.

[0036] After obtaining the power time series of each generator set, the causal relationship between the generator sets is quantified by calculating the transfer entropy between all generator sets pairwise. The transfer entropy measures the causal impact of the power time series of one generator set on the power time series of another generator set.

[0037] Next, an adjacency matrix is ​​constructed using the transfer entropy between pairs of generator sets. Each element in the adjacency matrix represents whether one generator set has a causal impact on another. The adjacency matrix not only reflects the connectivity in the network topology but also embodies the direction of information flow caused by dynamic power fluctuations, providing a basis for constructing a physically meaningful graph structure.

[0038] Step S30: Using a pre-trained gated recurrent unit, capture the time evolution of the active power of each generator set based on the power time series of each generator set, and obtain the enhanced features of each generator set.

[0039] In this embodiment of the invention, in order to effectively extract dynamic information from the power time series of each generator set and enhance the node feature representation capability of the graph neural network, a pre-trained gated recurrent unit (GRU) is used to process the power time series of each generator set, thereby obtaining enhanced features with time-dependent characteristics.

[0040] It should be understood that the gated recurrent unit (ROU), as a recurrent neural network structure suitable for sequence modeling, possesses a gating mechanism to selectively retain long-term dependency information and suppress irrelevant disturbances. The ROU is trained using historical data from each generator unit in the power grid system, enabling it to learn nonlinear variation patterns in the time dimension, such as periodic fluctuations, sudden disturbance responses, and decay trends, ultimately yielding the trained ROU.

[0041] After the normalized power time series of each generator set is input into a pre-trained gated recurrent unit (GRU), the GRU captures the temporal evolution of the active power corresponding to the generator set and finally outputs a feature vector rich in temporal semantics, i.e., the enhanced feature. Compared with the original power time series, the enhanced feature is more abstract and discriminative, and can be regarded as a highly condensed expression of the dynamic behavior of the generator set, providing support for the node attributes in the subsequent graph neural network model.

[0042] Step S40: Using a pre-trained graph neural network model, broadband oscillation sources are identified based on the adjacency matrix and the enhanced features of all generator sets, and the probability that each generator set is a broadband oscillation source is obtained.

[0043] In this embodiment of the invention, achieving broadband oscillation source localization in a power system requires comprehensive consideration of the dynamic influence relationships between generator units (i.e., the adjacency matrix) and the temporal behavior characteristics of each generator unit (i.e., enhancement features). Specifically, the adjacency matrix and the enhancement features of all generator units are first input into a pre-trained graph neural network model. The adjacency matrix represents the causal relationship structure between generator units, reflecting the potential path of disturbance propagation. The enhancement features of each generator unit are extracted by a gated recurrent unit, containing the deep evolution law of its active power time series. Based on this, the graph neural network model models the entire power grid as a directed graph. In this graph, V represents a graph node, with generator sets serving as the graph nodes. E represents the connection relationships between nodes, defined by an adjacency matrix, thus constructing a physically meaningful equivalent graph structure. Through the information propagation mechanism of this graph structure, the graph neural network model aggregates and updates node features at each layer, integrates the state information of neighboring nodes, gradually extracts higher-order spatial dependencies, and performs discriminant analysis on each generator set based on this, outputting its probability value as a broadband oscillation source. In other words, through the information transmission mechanism in the graph neural network, the online localization of the oscillation source is ultimately achieved.

[0044] In summary, the broadband oscillation source localization method provided by this invention, by adopting a data-driven strategy, does not rely on precise physical modeling of the power system, effectively overcoming the poor adaptability of traditional methods in real-world scenarios with complex system structures and variable operating states. It is suitable for modern power systems with complex operating states and varied topologies. Addressing the challenge of lacking clear physical connections between generator units and the difficulty in constructing graph structures, this invention innovatively introduces transfer entropy analysis to determine the causal influence strength between generator units, constructing an adjacency matrix and achieving reasonable graph structure modeling without topological priors. By using gated cyclic units to extract deep features from the active power time-series data of generator units, the long-term dependence and dynamic evolution characteristics of the broadband oscillation process are fully captured, enhancing the representational ability of nodes. Furthermore, the causal graph structure and enhanced features are integrated into the graph neural network model to achieve spatiotemporal joint modeling, effectively improving the ability to characterize the disturbance energy propagation path and significantly increasing the spatial localization accuracy of forced oscillation disturbance sources. In addition, this method relies only on measurable data from the generator side, conforming to the current data acquisition status of wide-area measurement systems, and possesses good engineering applicability and practical promotion value.

[0045] It should be understood that accurately characterizing the dynamic causal relationships between generator units in a power system is a key foundation for locating broadband oscillation sources. Traditional correlation analysis struggles to distinguish the direction of influence, while transfer entropy can effectively quantify the degree to which one time series enhances the predictive power of another time series for its future state, thus revealing the directional driving relationship between the two.

[0046] For time series X and Y, the transition entropy is calculated given past data. and right The predictive power is relative to that based solely on The increment of the predictive power of past values. The definition of transfer entropy is as follows:

[0047] in, It is the transfer entropy; It is conditional entropy, which represents the condition that depends only on the past. To predict Uncertainty; It is the joint conditional entropy, representing the entropy when considering the past simultaneously. and Time prediction Uncertainty; It is the t-th element in the time series X. It is the t-th element in the time series Y. It is the (t+1)th element in the time series Y.

[0048] Alternatively, regarding how to determine the transfer entropy between two generator sets, the following is a possible implementation method. Please refer to... Figure 2 , Figure 1 The sub-steps of step S20 may include: Step S200: Determine any two generator sets as the first generator set and the second generator set.

[0049] In this embodiment of the invention, a set of generator sets is selected from all generator sets for causal analysis. This pairing method covers all possible combinations of generator sets, ensuring that the interactions between nodes throughout the power grid are evaluated. For example, generator set I and generator set J are selected to evaluate their interactions, with generator set I designated as the first generator set and generator set J designated as the second generator set.

[0050] Step S210: Calculate the conditional entropy of the first generator set based on the power time series of the first generator set.

[0051] Step S220: Calculate the conditional entropy of the second generator set based on the power time series of the second generator set.

[0052] In this embodiment of the invention, conditional entropy reflects the level of uncertainty when predicting the future state of a generator set solely based on its own historical active power data (i.e., power time series). Assume the power time series of generator set I is... The power time series of generator set J is as follows: The conditional entropy of the first generator unit is based on the power time series of generator unit I. The level of uncertainty when predicting future states. Similarly, the conditional entropy of the second generator unit is based on the power time series of generator unit J. The level of uncertainty when predicting future states.

[0053] Step S230: Calculate the joint conditional entropy of the first generator set and the joint conditional entropy of the second generator set based on the power time series of the first generator set and the power time series of the second generator set.

[0054] In this embodiment of the invention, joint conditional entropy refers to the uncertainty measure of the output of the current generator set at the next moment, under the premise of simultaneously considering the historical states of two generator sets, reflecting the impact of cross-generator set information on prediction accuracy.

[0055] Step S240: Determine the transfer entropy from the first generator set to the second generator set based on the conditional entropy and joint conditional entropy of the first generator set.

[0056] In this embodiment of the invention, the causal influence of the former on the state evolution of the latter is quantified by comparing the difference in prediction uncertainty of the second generator set with or without the introduction of historical information of the first generator set.

[0057] Continuing with power time series as and For example, the formula for calculating the transfer entropy from the first generator set to the second generator set at time t+1 is:

[0058] in, It is the causal effect from the first generator set to the second generator set at time t+1. It is the t-th element in the power time series of the first generator set. It is the t-th element in the power time series of the second generator set. It is the (t+1)th element in the power time series of the second generator set.

[0059] The formula for calculating the transfer entropy from the first generator set to the second generator set for the entire power time series is as follows:

[0060] in, It is the transfer entropy from the first generator set to the second generator set, and N is the number of elements in the power time series.

[0061] Step S250: Determine the transfer entropy from the second generator set to the first generator set based on the conditional entropy and the joint conditional entropy of the second generator set.

[0062] In this embodiment of the invention, the causal influence of the former on the state evolution of the latter is quantified by comparing the difference in prediction uncertainty of the first generator set with or without the introduction of historical information of the second generator set.

[0063] The formula for calculating the transfer entropy from the second generator set to the first generator set at time t+1 is:

[0064] in, It is the causal effect from the second generator set to the first generator set at time t+1. It is the (t+1)th element in the power time series of the first generator set.

[0065] The formula for calculating the transfer entropy from the second generator set to the first generator set for the entire power time series is as follows:

[0066] in, It is the transfer entropy from the second generator set to the first generator set.

[0067] As can be seen, the embodiments of the present invention utilize the difference between conditional entropy and joint conditional entropy to determine the bidirectional transfer entropy of any two generator sets, thereby realizing a quantitative assessment of the direction and intensity of causal influence between the two generator sets. This effectively captures the information flow relationship in the dynamic evolution of active power and provides data support for constructing a causal relationship structure that reflects the internal driving mechanism of the power grid.

[0068] Alternatively, one possible implementation for generating the adjacency matrix is ​​provided below. Please refer to [link / reference]. Figure 2 , Figure 1 The sub-steps of step S20 may include: Step S260: The difference between the transfer entropy from the first generator set to the second generator set and the transfer entropy from the second generator set to the first generator set in any two generator sets is determined as the first difference.

[0069] In this embodiment of the invention, the first difference reflects whether the first generator set has a significant one-way information guiding effect on the second generator set, and its magnitude reflects the magnitude of the causal influence. The first difference... The calculation formula is:

[0070] Step S270: Determine the second difference as the transfer entropy from the second generator set to the first generator set and the transfer entropy from the first generator set to the second generator set in any two generator sets.

[0071] In this embodiment of the invention, the second difference reflects whether the second generator set has a significant one-way information guiding effect on the first generator set, and its magnitude reflects the magnitude of the causal influence. The first difference... The calculation formula is:

[0072] Step S280: Determine the causal influence coefficient of the first generator set on the second generator set and the causal influence coefficient of the second generator set on the first generator set based on the first difference, the second difference and the preset transfer threshold; all causal influence coefficients constitute an adjacency matrix.

[0073] In this embodiment of the invention, the causal influence coefficient of the first generator set on the second generator set is determined by the difference between the first difference and the preset transfer threshold, and the causal influence coefficient of the second generator set on the first generator set is determined by the difference between the second difference and the preset transfer threshold.

[0074] Finally, the adjacency matrix, composed of all causal influence coefficients, provides a physically meaningful topological input structure for graph neural network modeling, enhancing the graph neural network model's ability to identify oscillation propagation paths.

[0075] As can be seen, the embodiments of the present invention achieve quantitative screening of the dynamic interaction strength and direction between generator sets, enhance the sparsity and physical meaning of the adjacency matrix, and provide a power grid association topology with clear structure and explicit causality for subsequent graph neural network model analysis.

[0076] Alternatively, one possible approach to determining the causal influence coefficient is provided below. Figure 2 The sub-steps of step S280 may include: If the first difference is not less than the preset transfer threshold, the causal influence coefficient of the first generator set on the second generator set is determined as the first preset value; if the first difference is less than the preset transfer threshold, the causal influence coefficient of the first generator set on the second generator set is determined as the second preset value; if the second difference is not less than the preset transfer threshold, the causal influence coefficient of the second generator set on the first generator set is determined as the first preset value; if the second difference is less than the preset transfer threshold, the causal influence coefficient of the second generator set on the first generator set is determined as the second preset value.

[0077] In this embodiment of the invention, a preset transition threshold is introduced in order to filter out significant causal relationships. By setting clear judgment rules, the robustness and structural rationality of the dynamic correlation modeling between power grid units are enhanced. Assuming the first preset value is 1 and the second preset value is 0, the rules for constructing the adjacency matrix are as follows:

[0078] in, It is the causal influence coefficient of generator set s on generator set t. It is the magnitude of the causal influence of generator set s on generator set t, if Therefore, it is assumed that generator set s has a significant causal effect on generator set t, let Conversely, it is 0.

[0079] Optionally, the gated recurrent unit includes multiple hidden layers. One possible implementation for generating the enhanced features is provided below. Figure 1 The sub-steps of step S30 may include: Step S300: Input the power time series of each generator set into the gated loop unit.

[0080] In this embodiment of the invention, the power time series of each generator set is independently entered into the corresponding gated loop unit processing channel to ensure that the feature extraction process of each node maintains independence and integrity in the time dimension.

[0081] Step S310: Process the data in the power time series one by one using the first hidden layer, output the hidden state corresponding to the power time series, and input the hidden state output by the first hidden layer into the next hidden layer for processing, until the hidden state output by the last hidden layer is obtained.

[0082] Step S320: Determine the hidden state output by the last hidden layer as the enhanced feature of the corresponding generator set.

[0083] In this embodiment of the invention, the gated recurrent unit selectively retains long-term historical information and suppresses irrelevant disturbances through its internal reset and update gate mechanisms, thereby effectively capturing periodic, sudden, and decaying fluctuations in the input sequence. The hidden state at each moment integrates a weighted combination of the current input and previous memories, forming a gradual refinement of local dynamic behavior; the multi-layered structure further enhances the nonlinear expressive power, enabling the final output state to reflect deep-seated temporal evolution patterns.

[0084] For the power time series of generator set I Taking the t-th element in the power time series of generator I processed by the first hidden layer of the gated cyclic unit as an example, the recursive process is as follows:

[0085]

[0086]

[0087] in, It is the input of the generator set at time t, that is, the t-th element in the power time series of generator set I; It is the hidden state output at time t. It is the hidden state output at time t-1. It's a door reset. It's an update gate. This represents Hadamard (element-by-element) multiplication. This represents the Sigmoid function; Represents the hyperbolic tangent function. To reset the matrix weights of the gates, To update the matrix weights of the gates, The matrix weights are the output gate weights. To reset the door offset, To update the gate bias, This is the bias value of the output gate.

[0088] Based on the above recursive process, the hidden state corresponding to the power time series of generator set I output by the first hidden layer is obtained. The hidden state output by the first hidden layer is then input into the next hidden layer for processing, until the hidden state corresponding to the power time series of generator set I is output by the last hidden layer, thus obtaining the enhanced feature corresponding to generator set I. , which serve as the node input features in a graph neural network model.

[0089] As can be seen, this embodiment of the invention, by inputting the power time series of each generator set into a gated recurrent unit containing multiple hidden layers and processing it layer by layer, can progressively extract and deepen the dynamic evolution features in the time series. By using each hidden layer to perform progressive modeling of the sequence data, the hidden state output by the last hidden layer fully integrates historical information and nonlinear temporal dependencies, effectively characterizing the variation law of the generator set's active power. Using this hidden state as an enhanced feature improves the expressive power of the node features.

[0090] Optionally, the graph neural network model includes multiple layers of graph neural networks and a classifier, with the last layer of the graph neural network connected to the classifier. One possible implementation for identifying broadband oscillation sources is provided below. Figure 1 The sub-steps of step S40 may include: Step S400: Input the adjacency matrix and the enhanced features of all generator sets into the graph neural network model.

[0091] In this embodiment of the invention, the adjacency matrix representing the connection relationship between generator sets and the enhanced features of each generator set are input into the graph neural network model. The power grid topology represented by the adjacency matrix remains unchanged throughout the forward propagation process. It is provided to all network layers during the model initialization phase and can be accessed by each layer on demand through a shared storage area or transmitted layer by layer, ensuring that each layer of the graph neural network can aggregate and update node features based on a unified graph structure. Finally, the prediction result is output through the classifier.

[0092] Step S410: The enhanced features of all generator sets are updated using the first layer graph neural network based on the adjacency matrix, the weight matrix of the first layer graph neural network, and the bias of the first layer graph neural network, and the obtained feature update matrix is ​​input into the next layer graph neural network.

[0093] Step S420: For each layer of graph neural network after the first layer, the graph neural network updates the feature update matrix output by the previous layer based on the adjacency matrix, the weight matrix of the graph neural network, and the bias of the graph neural network, until the feature update matrix output by the last layer of graph neural network is obtained.

[0094] In this embodiment of the invention, the first layer of the graph neural network updates the enhanced features of each generator set based on the aggregation and nonlinear mapping of the adjacency matrix, and inputs the feature update matrix input from the first layer into the next layer of the graph neural network for processing. Each layer of the graph neural network from the second layer onwards updates the feature update matrix input from the previous layer based on the adjacency matrix, until the feature update matrix output by the last layer of the graph neural network is obtained.

[0095] Taking generator set I as an example, the update formula is as follows:

[0096] in, It represents the features (i.e., feature update matrix) of node i (i.e., generator set I) in the (l+1)th layer of the graph neural network. It is the set of nodes in the adjacency matrix that have a causal relationship with node i. It is the causal influence coefficient from node i to node j in the adjacency matrix. It is the weight matrix of the l-th layer. It is the bias of the l-th layer. It is an activation function.

[0097] The update process of all nodes (i.e., all generator sets) is expressed in matrix form:

[0098] in, It is the feature update matrix of the l-th layer. The first layer of the graph neural network uses a matrix composed of the enhanced features of all generator sets.

[0099] After propagation through multiple layers of the graph neural network, the feature update matrix output by the last layer of the graph neural network is obtained. Each of its rows This represents the final state of node i after integrating its temporal characteristics and causal structure.

[0100] Step S430: Input the feature update matrix output from the last layer of the graph neural network into the classifier.

[0101] Step S440: The classifier is used to identify each generator set as a broadband oscillation source based on the feature update matrix output by the last layer of the graph neural network, so as to obtain the probability that each generator set is a broadband oscillation source.

[0102] In this embodiment of the invention, the feature update matrix output from the last layer of the graph neural network is input into a classifier, which then outputs a predicted probability of whether the generator set is a broadband oscillation source. The probability is calculated using the following formula:

[0103] in, It is the probability that generator set I is a broadband oscillation source. It is the weight matrix of the output layer in the classifier. It is the bias of the output layer in the classifier.

[0104] It should be noted that the weight matrix and bias used in the gated recurrent unit and graph neural network model are updated through training. After training, the weight matrix and bias are known parameters.

[0105] This invention aims to solve the key technical problems in broadband oscillation disturbance source localization, such as modeling difficulties, lack of spatial features, inability to construct structural graphs, and insufficient expression of node features. It proposes a graph neural network broadband oscillation source localization method based on causal graph construction and GRU feature enhancement.

[0106] To address the problem that existing physical mechanism-based positioning methods heavily rely on precise mathematical models of power systems, making them difficult to adapt to complex network structures and frequently changing operating conditions, resulting in poor model generalization ability and unstable positioning results, this invention adopts a data-driven strategy. It directly uses measurable generator active power time series as input to automatically learn the dynamic law of disturbance propagation from actual operating data, thus eliminating the dependence on prior system mechanism models and significantly improving the adaptability and robustness of the method in variable scenarios.

[0107] To overcome the shortcomings of traditional data-driven methods (such as LSTM and CNN) which only focus on extracting features in the time dimension and ignore the spatial correlation of disturbance propagation in the power grid topology, making it difficult to accurately characterize the energy diffusion path of disturbances, this invention introduces a graph neural network (GNN) architecture. It makes full use of the ability of GNN to aggregate spatial information of graph structure data, models the relationship between various units in the power system as a spatial graph structure, and realizes the effective fusion of temporal evolution features and spatial topological features, thereby enhancing the ability to identify the spatial location of disturbance sources.

[0108] Furthermore, addressing the bottleneck problem of "structure not being able to be graphed" caused by the lack of clear physical connection relationships in power grids and the inability to directly construct an effective adjacency matrix in practical applications, this invention innovatively introduces a causal relationship modeling mechanism. It uses the transfer entropy method to quantify the causal influence intensity of active power sequences between different generator sets, thereby constructing a causal adjacency matrix that reflects the direction of disturbance propagation and the degree of coupling. This matrix is ​​then used as the structural input of a graph neural network, achieving "causal graphing" without relying on the real physical topology, thus solving the model input problem under the condition of missing graph structure.

[0109] Furthermore, considering the complex characteristics of disturbance signals during broadband oscillation, such as strong nonlinearity, wide frequency response, and long-term dependence, single static or shallow features are insufficient to fully characterize the dynamic behavior of nodes. Therefore, this invention designs a node feature enhancement module based on gated recurrent units (GRUs) to perform deep time-series modeling on the active power time-series data of each generator set, effectively capturing key time dynamic evolution information before and after the disturbance occurs, generating high-dimensional, discriminative enhanced node features, and significantly improving the sensitivity and classification discriminative power of graph neural networks to node state changes.

[0110] In summary, the embodiments of the present invention construct a broadband oscillation disturbance source localization method that does not rely on system mechanism models, is applicable to scenarios with unknown or incomplete topology, and has strong spatiotemporal awareness through a four-in-one technical system of "data-driven modeling, causal mapping, spatiotemporal feature fusion, and dynamic node representation". This method comprehensively solves the four core challenges existing in the current technology and has good engineering adaptability and significant technical advantages.

[0111] Based on the same inventive concept, the basic principle and technical effects of the broadband oscillation source positioning device provided in this embodiment are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in this embodiment can be referred to the corresponding content in the above embodiments.

[0112] Please refer to Figure 3 , Figure 3 This is a block diagram of a broadband oscillator source positioning device 400 provided in an embodiment of the present invention. The broadband oscillator source positioning device 400 includes a processing module 410, a construction module 420, and a positioning module 430.

[0113] The processing module 410 is used to normalize the active power data of each generator set in the power grid system to obtain the power time series of each generator set.

[0114] Module 420 is constructed to calculate the transfer entropy between any two generator sets and generate an adjacency matrix based on all transfer entropies. The adjacency matrix is ​​used to characterize the causal relationship between generator sets in the power system. A pre-trained gated recurrent unit is used to capture the time evolution of the active power of each generator set based on the power time series of each generator set, thereby obtaining the enhanced features of each generator set.

[0115] The localization module 430 is used to identify broadband oscillation sources based on the adjacency matrix and enhanced features of all generator sets using a pre-trained graph neural network model, and to obtain the probability that each generator set is a broadband oscillation source.

[0116] In summary, the broadband oscillation source localization device provided in this embodiment of the invention, by adopting a data-driven strategy, effectively overcomes the problem of poor adaptability of traditional methods in real-world scenarios with complex system structures and variable operating states, without relying on precise physical modeling of the power system. It is suitable for modern power systems with complex operating states and variable topologies. Addressing the challenge of lacking clear physical connections between generator units and the difficulty in constructing graph structures, this invention innovatively introduces transfer entropy analysis to determine the causal influence strength between generator units, constructing an adjacency matrix and achieving reasonable graph structure modeling without topological priors. By using gated cyclic units to perform deep feature extraction on the active power time-series data of generator units, the long-term dependence and dynamic evolution characteristics in the broadband oscillation process are fully captured, enhancing the representational ability of nodes. Furthermore, the causal graph structure and enhanced features are integrated into the graph neural network model to achieve spatiotemporal joint modeling, effectively improving the ability to characterize the disturbance energy propagation path and significantly increasing the spatial localization accuracy of forced oscillation disturbance sources. In addition, this method relies only on measurable data from the generator side, conforming to the current data acquisition status of existing wide-area measurement systems, and possesses good engineering applicability and practical promotion value.

[0117] Optionally, module 420 is specifically used to determine any two generator sets as the first generator set and the second generator set; calculate the conditional entropy of the first generator set based on the power time series of the first generator set; calculate the conditional entropy of the second generator set based on the power time series of the second generator set; calculate the joint conditional entropy of the first generator set and the joint conditional entropy of the second generator set based on the power time series of the first generator set and the power time series of the second generator set; determine the transfer entropy from the first generator set to the second generator set based on the conditional entropy of the first generator set and the joint conditional entropy; and determine the transfer entropy from the second generator set to the first generator set based on the conditional entropy of the second generator set and the joint conditional entropy.

[0118] Optionally, the construction module 420 is specifically used to determine the difference between the transfer entropy from the first generator set to the second generator set and the transfer entropy from the second generator set to the first generator set as a first difference; to determine the difference between the transfer entropy from the second generator set to the first generator set and the transfer entropy from the first generator set to the second generator set as a second difference; to determine the causal influence coefficient of the first generator set on the second generator set and the causal influence coefficient of the second generator set on the first generator set based on the first difference, the second difference, and a preset transfer threshold; and to construct an adjacency matrix from all the causal influence coefficients.

[0119] Optionally, the construction module 420 is specifically used to determine the causal influence coefficient of the first generator set on the second generator set as a first preset value if the first difference is not less than a preset transfer threshold; to determine the causal influence coefficient of the first generator set on the second generator set as a second preset value if the first difference is less than the preset transfer threshold; to determine the causal influence coefficient of the second generator set on the first generator set as a first preset value if the second difference is not less than the preset transfer threshold; and to determine the causal influence coefficient of the second generator set on the first generator set as a second preset value if the second difference is less than the preset transfer threshold.

[0120] Optionally, the gated recurrent unit includes multiple hidden layers. The construction module 420 is specifically used to input the power time series of each generator set into the gated recurrent unit; process the data in the power time series one by one using the first hidden layer, output the hidden state corresponding to the power time series, and input the hidden state output by the first hidden layer into the next hidden layer for processing, until the hidden state output by the last hidden layer is obtained; the hidden state output by the last hidden layer is determined as the enhanced feature of the corresponding generator set.

[0121] Optionally, the graph neural network model includes a multi-layer graph neural network and a classifier, with the last layer of the graph neural network connected to the classifier. The localization module 430 is specifically used to input the adjacency matrix and enhanced features of all generator sets into the graph neural network model; The enhanced features of all generator sets are updated using the first-layer graph neural network (Graph Neural Network) based on the adjacency matrix, the weight matrix, and the bias of the first-layer graph neural network. The resulting feature update matrix is ​​then input into the next-layer graph neural network. For each subsequent layer of the graph neural network, the feature update matrix output by the previous layer is updated using the adjacency matrix, the weight matrix, and the bias of the graph neural network, until the feature update matrix output by the last layer of the graph neural network is obtained. The feature update matrix output by the last layer of the graph neural network is then input into a classifier. The classifier uses the feature update matrix output by the last layer of the graph neural network to identify each generator set as a broadband oscillation source, thus obtaining the probability that each generator set is a broadband oscillation source.

[0122] Please refer to Figure 4 This is a block diagram illustrating an electronic device 500 provided in an embodiment of the present invention. The electronic device 500 includes, but is not limited to, a personal computer (PC), a personal digital assistant (PDA), a laptop computer, a tablet computer, and a server. The electronic device 500 includes a memory 510, a processor 520, and a communication module 530. The memory 510, processor 520, and communication module 530 are electrically connected directly or indirectly to each other to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0123] The memory 510 is used to store programs or data. The memory 510 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0124] The processor 520 is used to read / write data or programs stored in the memory 510 and perform corresponding functions. For example, when a computer program stored in the memory 510 is executed by the processor 520, the broadband oscillation source localization method disclosed in the above embodiments can be implemented.

[0125] The communication module 530 is used to establish a communication connection between the electronic device 500 and other communication terminals via a network, and to send and receive data via the network.

[0126] It should be understood that, Figure 4 The structure shown is only a schematic diagram of the electronic device 500. The electronic device 500 may also include components that are larger than those shown. Figure 4 The more or fewer components shown, or having the same Figure 4 The different configurations shown. Figure 4 The components shown can be implemented using hardware, software, or a combination thereof.

[0127] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor 520, implements the broadband oscillation source localization method disclosed in the above embodiments.

[0128] This invention also provides a program product that, when executed by processor 520, implements the broadband oscillation source localization method disclosed in the above embodiments.

[0129] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0130] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0131] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for locating a broadband oscillation source, characterized in that, The method includes: The active power data of each generator set in the power grid system are normalized to obtain the power time series of each generator set; Calculate the transfer entropy between any two generator sets, and generate an adjacency matrix based on all the transfer entropies; the adjacency matrix is ​​used to characterize the causal relationships between the generator sets in the power system. By using a pre-trained gated recurrent unit to capture the time evolution of the active power corresponding to each generator set based on the power time series of each generator set, the enhanced features of each generator set are obtained. A pre-trained graph neural network model is used to identify broadband oscillation sources based on the adjacency matrix and the enhanced features of all the generator sets, thereby obtaining the probability that each generator set is a broadband oscillation source.

2. The broadband oscillation source localization method according to claim 1, characterized in that, The calculation of the transfer entropy between any two generator sets includes: Any two generator sets are designated as the first generator set and the second generator set. Calculate the conditional entropy of the first generator set based on the power time series of the first generator set; Calculate the conditional entropy of the second generator set based on the power time series of the second generator set; Calculate the joint conditional entropy of the first generator set and the joint conditional entropy of the second generator set based on the power time series of the first generator set and the power time series of the second generator set. The transfer entropy from the first generator set to the second generator set is determined based on the conditional entropy and joint conditional entropy of the first generator set. The transfer entropy from the second generator set to the first generator set is determined based on the conditional entropy and the joint conditional entropy of the second generator set.

3. The broadband oscillation source localization method according to claim 2, characterized in that, The step of generating the adjacency matrix based on all the transfer entropies includes: The difference between the transfer entropy from the first generator set to the second generator set and the transfer entropy from the second generator set to the first generator set in any two generator sets is defined as the first difference. The difference between the transfer entropy from the second generator set to the first generator set and the transfer entropy from the first generator set to the second generator set in any two generator sets is determined as the second difference. The causal influence coefficients of the first generator set on the second generator set and the second generator set on the first generator set are determined based on the first difference, the second difference, and the preset transfer threshold; all the causal influence coefficients constitute the adjacency matrix.

4. The broadband oscillation source localization method according to claim 3, characterized in that, The step of determining the causal influence coefficient of the first generator set on the second generator set and the causal influence coefficient of the second generator set on the first generator set based on the first difference, the second difference, and a preset transfer threshold includes: If the first difference is not less than the preset transfer threshold, the causal influence coefficient of the first generator set on the second generator set is determined as the first preset value; If the first difference is less than the preset transfer threshold, the causal influence coefficient of the first generator set on the second generator set is determined as the second preset value; If the second difference is not less than the preset transfer threshold, the causal influence coefficient of the second generator set on the first generator set is determined as the first preset value; If the second difference is less than the preset transfer threshold, the causal influence coefficient of the second generator set on the first generator set is determined as the second preset value.

5. The broadband oscillation source localization method according to claim 1, characterized in that, The gated recurrent unit includes multiple hidden layers; the method of using a pre-trained gated recurrent unit to capture the time evolution of the active power corresponding to each generator set based on the power time series of each generator set, and obtaining the enhanced features of each generator set, includes: The power time series of each generator set is input into the gated loop unit; The data in the power time series are processed one by one using the first hidden layer, the hidden state corresponding to the power time series is output, and the hidden state output by the first hidden layer is input into the next hidden layer for processing, until the hidden state output by the last hidden layer is obtained; The hidden state output by the last hidden layer is determined as the enhanced feature of the corresponding generator set.

6. The broadband oscillation source localization method according to claim 1, characterized in that, The graph neural network model includes a multi-layer graph neural network and a classifier, with the last layer of the graph neural network connected to the classifier; the step of using the pre-trained graph neural network model to identify broadband oscillation sources based on the adjacency matrix and the enhanced features of all the generator sets, and obtaining the probability that each generator set is a broadband oscillation source, includes: The adjacency matrix and the enhanced features of all the generator sets are input into the graph neural network model; The enhanced features of all generator sets are updated using the first layer of the graph neural network based on the adjacency matrix, the weight matrix of the first layer of the graph neural network, and the bias of the first layer of the graph neural network, and the resulting feature update matrix is ​​input into the next layer of the graph neural network. For each layer of the graph neural network after the first layer, the graph neural network updates the feature update matrix output by the previous layer based on the adjacency matrix, the weight matrix of the graph neural network, and the bias of the graph neural network, until the feature update matrix output by the last layer of the graph neural network is obtained. The feature update matrix output from the last layer of the graph neural network is input into the classifier; The classifier is used to identify each generator set as a broadband oscillation source based on the feature update matrix output by the last layer of the graph neural network, thereby obtaining the probability that each generator set is a broadband oscillation source.

7. A broadband oscillation source positioning device, characterized in that, The device includes: The processing module is used to normalize the active power data of each generator set in the power grid system to obtain the power time series of each generator set. A construction module is used to calculate the transfer entropy between any two generator sets and generate an adjacency matrix based on all the transfer entropies. The adjacency matrix is ​​used to characterize the causal relationship between each generator set in the power system. A pre-trained gated recurrent unit is used to capture the time evolution law of the active power corresponding to each generator set based on the power time series of each generator set, so as to obtain the enhanced features of each generator set. The localization module is used to identify broadband oscillation sources based on the adjacency matrix and the enhanced features of all the generator sets using a pre-trained graph neural network model, and to obtain the probability that each generator set is a broadband oscillation source.

8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program that can be executed by the processor to implement the broadband oscillation source localization method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the broadband oscillation source localization method as described in any one of claims 1-6.

10. A program product, characterized in that, When the program product is executed by the processor, it implements the broadband oscillation source localization method as described in any one of claims 1-6.

Images