DL-MVCNN-based wind power subsynchronous oscillation tracing method and system

Abstract

The invention provides a DL-MVCNN-based wind power subsynchronous oscillation traceability method and system, and relates to the technical field of wind power grid connection and oscillation traceability, and the method comprises the steps: constructing a multi-wind power plant grid connection model, and obtaining instantaneous current and power high-dimension subsynchronous oscillation data; modeling is carried out on instantaneous current and power subsynchronous oscillation intrinsic dynamics based on sampling determination learning, subsynchronous oscillation dynamic characteristics are extracted in a given state space, and a subsynchronous oscillation dynamic diagram is generated; inputting the subsynchronous oscillation kinetic graph into a multi-modal oscillation traceability model, extracting respective kinetic mode characteristics of each wind field, and fusing the kinetic mode characteristics of each wind field through a channel attention mechanism in the multi-modal oscillation traceability model to obtain fused characteristics; different oscillation sources correspond to different determined learning dynamics characteristics, based on fusion characteristics, adjusting a weight giving channel, calculating a corresponding channel weight, and outputting a classification result of wind field traceability.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind power grid connection and oscillation tracing, and in particular to a method and system for tracing subsynchronous oscillations of wind power based on DL-MVCNN. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] At present, most of the oscillation source location methods based on real-time system measurement data use the collected system measurement data to calculate the corresponding discrimination index, and use this as a basis to determine the oscillation source position, thereby obtaining the relationship between the oscillation source and the system measurement. Traditional subsynchronous oscillation source tracing methods include impedance analysis, complex torque coefficient method, eigenvalue analysis method, etc. However, the above methods also have the following common problems: (1) It is necessary to obtain detailed parameters of the actual power grid system and then build an accurate model. However, it is difficult to obtain actual power grid parameters, and data gaps will directly affect the accuracy of analysis; (2) Wind farm systems often operate under time-varying conditions, including different wind speeds and power outputs, which often deviate from the system condition assumptions made by these model-based methods; (3) The power grid system is large in scale, and modeling often requires high-dimensional operations, resulting in poor detection timeliness and only suitable for post-oscillation analysis. Therefore, the above mechanism analysis method has certain limitations and is not suitable for scenarios where real-time and rapid detection of subsynchronous oscillations is required.

[0004] Furthermore, existing energy-based methods have been proposed. These methods can track system-wide energy flows to locate oscillation sources. By calculating the impedance and power of each component's subsynchronous oscillation mode using the voltage and current phasors, they can then link the transient energy flow with the generator's power and damping. This allows the positive / negative damping characteristics of the transient energy flow to be determined, and the subsynchronous oscillations can be traced based on these criteria. However, when multiple wind farms or power electronic equipment simultaneously induce subsynchronous oscillations (SSOs), the oscillation sources may mutually excite or overlap, making it difficult for these methods to distinguish the dominant oscillation source, making it difficult to address the oscillation source tracking problem in wide-area global oscillation scenarios.

[0005] As power system structures and components become increasingly diversified, their dynamic characteristics also exhibit strong nonlinear characteristics, making the application of subsynchronous oscillation source location methods based on models increasingly difficult. With the development of artificial intelligence and big data technologies, smart grid construction is maturing. Data-driven AI-based source tracing methods completely break free from the aforementioned constraints associated with model-based methods and have garnered widespread attention. These methods include traditional machine learning classifiers such as least squares (RLS), K-nearest neighbor (KNN), naive Bayes (NB), and support vector machines (SVM). Furthermore, deep learning algorithms such as long-short memory neural networks (LSTM) and convolutional neural networks (CNN) have also been widely applied in subsynchronous oscillation source tracing. AI-based methods offer significant advantages in data processing and provide new approaches for locating power system oscillation sources. However, the following technical challenges remain: AI-based source tracing methods typically treat wind power systems as black boxes, rely excessively on large amounts of historical operational data, have low model interpretability, and parameter deviations can affect source tracing accuracy, making them difficult to assist in mechanism analysis. In addition, in a scenario where multiple wind farms are connected to the grid, it is difficult to identify which wind farm is experiencing oscillation, and the positioning accuracy is low. Summary of the Invention

[0006] In order to solve the above problems, this paper proposes a wind power subsynchronous oscillation tracing method and system based on DL-MVCNN. Based on the deterministic learning algorithm (DL), the internal dynamics diagram of subsynchronous oscillation is constructed for different oscillation sources. The oscillation source identification model is combined with a multi-perspective convolutional neural network and a channel attention mechanism to obtain the global characteristics of the oscillating wind field, so as to overcome the problem of low oscillation tracing accuracy under a single perspective.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] The wind power subsynchronous oscillation source tracing method based on DL-MVCNN includes

[0009] Build a multi-wind farm grid connection model based on actual grid connection scenarios, obtain instantaneous current and power high-dimensional subsynchronous oscillation data, and preprocess it;

[0010] Based on sampled deterministic learning, the intrinsic dynamics of instantaneous current and power subsynchronous oscillations are modeled. In a given state space, the subsynchronous oscillation dynamic characteristics are extracted and a subsynchronous oscillation dynamics map is generated.

[0011] The subsynchronous oscillation dynamics graph is input into the multimodal oscillation tracing model to extract the dynamic pattern characteristics of each wind field. The dynamic pattern characteristics of each wind field are then fused through the channel attention mechanism in the multimodal oscillation tracing model to obtain the fusion characteristics. Different oscillation sources correspond to different deterministic learning dynamic characteristics. Based on the fusion characteristics, the weights are adjusted to the channels, the corresponding channel weights are calculated, and the classification results of the wind field tracing are output.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions:

[0013] The wind power subsynchronous oscillation source tracing system based on DL-MVCNN includes:

[0014] The data acquisition module is used to build a multi-wind farm grid connection model based on the actual grid connection scenario, obtain instantaneous current and power high-dimensional subsynchronous oscillation data, and pre-process it;

[0015] The dynamic information extraction module is used to model the intrinsic dynamics of instantaneous current and power subsynchronous oscillations based on sampled deterministic learning, extract the subsynchronous oscillation dynamic characteristics in a given state space, and generate a subsynchronous oscillation dynamics map;

[0016] The tracing module is used to input the subsynchronous oscillation dynamics graph into the multimodal oscillation tracing model, extract the dynamic pattern characteristics of each wind field, and fuse the dynamic pattern characteristics of each wind field through the channel attention mechanism in the multimodal oscillation tracing model to obtain the fusion characteristics. Different oscillation sources correspond to different deterministic learning dynamic characteristics. Based on the fusion characteristics, the weights given to the channels are adjusted, the corresponding channel weights are calculated, and the classification results of the wind field tracing are output.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions:

[0018] A computer program product includes a computer program, which, when executed by a processor, implements the wind power subsynchronous oscillation source tracing method based on DL-MVCNN.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the wind power subsynchronous oscillation tracing method based on DL-MVCNN is implemented.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] An electronic device includes: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the wind power subsynchronous oscillation tracing method based on DL-MVCNN.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] This DL-MVCNN-based method for tracing subsynchronous oscillations in wind power utilizes a data-driven approach, employing data collected during wind farm operation to analyze the internal system dynamics of the data to construct oscillation tracing criteria. Based on a deterministic learning algorithm, internal dynamics maps of subsynchronous oscillations are constructed for different oscillation sources. This method differs from end-to-end data processing methods such as deep learning by constructing a weight update rate based on Lyapunov stability theory from control theory, satisfying some continuous excitation conditions and offering greater interpretability than deep learning.

[0025] The wind power subsynchronous oscillation tracing method based on DL-MVCNN disclosed in the present invention proposes a multimodal oscillation tracing model, which includes three parts: feature extraction, feature fusion and prediction of feature recognition expression. The channel attention mechanism is introduced to model the interdependence between channel feature responses, thereby achieving adaptive readjustment of channels. As the data collected by different nodes represents different information from three perspectives, this gives the present invention the opportunity to further optimize feature fusion. Since different oscillation sources correspond to different deterministic learning dynamics characteristics. Therefore, by adjusting the weight given to the channel, you can focus on more important functional areas.

[0026] This paper presents a method for tracing subsynchronous oscillations in wind power, based on DL-MVCNN. This method addresses the low accuracy of localization in scenarios where multiple wind farms are connected to the grid. By combining a multi-perspective convolutional neural network with a channel attention mechanism, the method aims to capture the global characteristics of the oscillating wind farm, overcoming the low accuracy of oscillation tracing from a single perspective. The method achieves an accuracy of 98.05%, an improvement of approximately 10% compared to traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute a part of the present disclosure, are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation to the present disclosure.

[0028] Figure 1 This is a flow chart of a method for tracing the source of subsynchronous oscillations in wind power based on DL-MVCNN according to an embodiment of the present disclosure;

[0029] Figure 2 determining learned oscillatory internal dynamics modeling results for the sampled embodiments of the present disclosure to compare the results;

[0030] in, Figure 2 (a) shows the result of the sampling determination learning algorithm on the internal dynamics of the current model. Figure 2 (b) shows the comparison between the actual current internal dynamics and the modeled dynamics;

[0031] Figure 3 A diagram of subsynchronous oscillation dynamics based on deterministic learning according to an embodiment of the present disclosure;

[0032] in, Figure 3 (a) shows the characteristic diagram of current oscillation dynamics; Figure 3 (b) shows the characteristic diagram of power oscillation dynamics;

[0033] Figure 4 Schematic diagram of DL-MVCNN multi-modal oscillation tracing model training in an embodiment of the present disclosure;

[0034] Figure 5 is the training result of the embodiment of the present disclosure;

[0035] in, Figure 5 (a) in the equation is the loss function of the training process. Figure 5 (b) in the figure is the training accuracy;

[0036] Figure 6 is the test set confusion matrix of the embodiment of the present disclosure;

[0037] Figure 7 Schematic diagram of the structural framework of the DL-MVCNN multimodal oscillation tracing model of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0038] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] Example 1

[0042] At present, most of the oscillation source location methods based on real-time system measurement data use the collected system measurement data to calculate the corresponding discrimination index, and use this as a basis to determine the oscillation source position. Therefore, the following relationship can be obtained between the oscillation source and the system measurement:

[0043] S=g(M)=s(X in )

[0044] Among them, S is the oscillation source label, that is, the wind farm number; M represents the discrimination index; g(·) represents the mapping relationship function between the oscillation source label and the system participation judgment index, and s(·) represents the mapping relationship between the measured data, that is, voltage, current, power and other information and the discrimination index. The above-mentioned traditional subsynchronous oscillation tracing methods include impedance analysis method, complex torque coefficient method, eigenvalue analysis method, etc. However, the above methods also have the following common problems: (1) It is necessary to obtain detailed parameters of the actual power grid system and then build an accurate model. However, it is difficult to obtain the actual power grid parameters, and data gaps will directly affect the accuracy of the analysis; (2) Wind farm systems often operate under time-varying conditions, including different wind speeds and power outputs, which often deviate from the system condition assumptions made by these model-based methods; (3) The power grid system is large in scale, and modeling often requires high-dimensional operations, resulting in poor detection timeliness, and is only suitable for post-oscillation analysis. Therefore, the above-mentioned mechanism analysis method has certain limitations and is not suitable for scenarios where real-time and rapid detection of subsynchronous oscillations is required.

[0045] Therefore, an embodiment of the present disclosure provides a method for tracing the source of subsynchronous oscillations in wind power based on DL-MVCNN, comprising the following steps:

[0046] Step 1: Build a multi-wind farm grid connection model based on the actual grid connection scenario, obtain instantaneous current and power high-dimensional subsynchronous oscillation data, and preprocess it;

[0047] Step 2: Model the intrinsic dynamics of instantaneous current and power subsynchronous oscillations based on sampled deterministic learning, extract the subsynchronous oscillation dynamic characteristics in a given state space, and generate a subsynchronous oscillation dynamics map;

[0048] Step 3: Input the subsynchronous oscillation dynamics graph into the multimodal oscillation tracing model, extract the dynamic pattern characteristics of each wind field, and then fuse the dynamic pattern characteristics of each wind field through the channel attention mechanism in the multimodal oscillation tracing model to obtain the fusion characteristics. Different oscillation sources correspond to different deterministic learning dynamic characteristics. Based on the fusion characteristics, adjust the weights given to the channels, calculate the corresponding channel weights, and output the classification results of the wind field tracing.

[0049] As an embodiment, the specific implementation process of a wind power subsynchronous oscillation tracing method based on DL-MVCNN disclosed in the present invention is as follows:

[0050] Step 1: Construction of oscillation tracing dataset

[0051] Specifically, in actual wind farm grid-connected scenarios, oscillation data is difficult to obtain due to the low frequency of oscillations. Based on this, three wind farm grid-connected models were constructed using Matlab simulation software to simulate actual grid-connected scenarios. Electrical data such as instantaneous current and power were collected to obtain high-dimensional subsynchronous oscillation data and construct an oscillation traceability dataset.

[0052] Preprocessing: Due to the large difference in the order of magnitude of current and power, the acquired instantaneous current and power subsynchronous oscillation data are normalized to ensure the uniformity and comparability of the data.

[0053] Step 2: Oscillation dynamics feature extraction, including: modeling the intrinsic dynamics of instantaneous current and power subsynchronous oscillations based on sampling deterministic learning, extracting subsynchronous oscillation dynamics features in a given state space, and generating a subsynchronous oscillation dynamics map.

[0054] By sampling and determining the learning mechanism, the dynamic features in the oscillation signal are further extracted and identified, which are crucial for fault diagnosis. The extracted dynamic features are converted into a dynamic information graph, providing rich features for subsequent model training and prediction.

[0055] Specifically, for the high-dimensional oscillation dynamic data generated in the large-scale wind power grid-connected mode, if all the data is used for deterministic learning modeling, it will lead to high time complexity and dimensionality explosion problems. Based on this, the present disclosure will model the power (P / Q) and three-phase current (IA / IB / IC) separately (because the power grid system is a voltage-stabilized system, the voltage changes little when oscillation occurs, so in this disclosure, only current and power are used for oscillation dynamic modeling). For the pre-processed high-dimensional subsynchronous oscillation data, the inherent dynamics of the subsynchronous oscillation are modeled based on sampling deterministic learning, and a state identifier is constructed using the RBF neural network that models the internal dynamics of the time series. The state identifier is shown in the following formula (1):

[0056]

[0057] Where X=[I A ,I B ] is the instantaneous current of phase A and phase B, I B is the system state that needs to be identified, is the RBF neural network for modeling the internal dynamics of the time series, and τ is the sampling time of the power grid system.

[0058] Furthermore, the weight update rate is constructed based on Lyapunov stability theory, and finally the subsynchronous oscillation unknown nonlinear dynamic model is obtained. According to Lyapunov stability theory, the weight update rate of the RBF neural network is:

[0059]

[0060] Among them, Γ is the weight update gain, e B (k) is the state tracking error.

[0061] Finally, after the neural network converges, the constant neural network weights are obtained, and the current I is obtained using the following formula: B The unknown nonlinear dynamics of :

[0062]

[0063] In addition, the modeling process of power (P / Q) is the same as that of current. Through the above-mentioned deterministic learning dynamic modeling method, the inherent dynamic constant RBF neural network of the time series signal can be obtained along the subsynchronous oscillation trajectory, that is, Since the system state is normalized, the current state trajectory of the system is confined to the space of [-1,1]×[-1,1]. Therefore, although the state trajectories under different oscillation sources are different, they can still be compared in this state space. This method has good robustness. Given the size of the state space, the subsynchronous oscillation dynamics information can form an n×n matrix.

[0064]

[0065] Where n is the length of the time series, and the interval [-1, 1] is evenly divided into n-1 parts. This results in an n×n numerical matrix, which can be considered as a subsynchronous oscillation dynamics map. It is worth noting that the resolution of the dynamics map depends on the size of n. If n is too small, the length of the intercepted time series to be learned is small, which may lead to low accuracy in reproducing the dynamics map. Therefore, this disclosure selects a time series of length 300 to construct a subsynchronous oscillation dynamics map.

[0066] Step 3: Build and train the DL-MVCNN multimodal oscillation tracing model

[0067] After obtaining the subsynchronous oscillation dynamics map, we will further develop a DL-MVCNN multimodal oscillation source tracing model. During training, this model will acquire global dynamic information from multiple wind farms, enabling prediction of multimodal oscillation sources. Furthermore, an attention mechanism is incorporated into the network to focus on regions with high feature weights, enabling more accurate source identification.

[0068] The DL-MVCNN multi-modal oscillation tracing model proposed in this invention is as follows Figure 4 As shown in the figure, the multimodal oscillation tracing model is a deterministic learning-multi-view convolutional neural network model, including a feature extraction module, a feature fusion module and a feature recognition expression prediction module. 1 ,X 2 ,X 3 ) as the input pair of DL-MVCNN, where X N represents the subsynchronous oscillation dynamics diagram of each wind field. Specifically, X N It is a dynamic stack of current data and power data from different wind farms. The basic structure of the feature extractor module adopts a dual-stream structure with 4 layers of convolution blocks, which are respectively from X 1 、X 2 and X 3 Extract the dynamic mode characteristics of wind field 1, wind field 2 and wind field 3. It is worth noting that considering X 1 、X 2 and X 3 The three data streams belong to the same data mode and have similar underlying spatial features. The extractor networks in the network are designed to share weights. and Represented as from X 1 、X 2 and X 3 The features extracted from , where C, H, W represent channel size, height, and width. As shown in the following formula:

[0069] F 1 =f e (X 1 θ e )

[0070] F 2 =f e (X 2 θ e )

[0071] F 3 =f e (X 3 θ e ) (5)

[0072] Among them, f e represents the feature extraction network, θ e is the set of all its parameters.

[0073] DL-MVCNN obtains the expression of joint representation by concatenation.

[0074]

[0075] In order to fully combine F 1 、F 2 and F 3 The channel attention mechanism is introduced in the feature fusion module based on the channel attention module in image processing. By taking advantage of its features, the mutual dependence between channel feature responses is modeled to achieve adaptive readjustment of the channel. As the data collected by different nodes, F J It represents different information from three perspectives, which gives the present invention an opportunity to further optimize feature fusion. The key point is that different oscillation sources correspond to different deterministic learning dynamics. Therefore, by adjusting the weights given to the channels, we can focus on more important functional areas. The channel attention mechanism essentially uses the global pool and shared MLP network to calculate the corresponding channel weights M. C ∈C×1×1, the formula is as follows:

[0076] M C (F J )=σ(MLP(AvgPool(F J ))+MLP(MaxPool(F J ))) (8)

[0077] Here, σ represents the sigmoid function.

[0078] In the feature recognition expression prediction module, such as Figure 7 As shown, F C represents the final refined output of the channel attention module, and the calculation formula is:

[0079]

[0080] in, Represents element-wise multiplication.

[0081] The specific model parameters of DL-MVCNN are shown in Table 1. DL-MVCNN was trained using the Adam optimizer with an L2 regularization parameter of 0.1. Other network hyperparameters, such as the learning rate, batch size, and epochs, were chosen empirically.

[0082] Table 1 DL-MVCNN model parameter settings

[0083]

[0084] Simulation Results

[0085] Based on the disclosed method, the final accuracy achieved is 98.05%, and the training results are as follows: Figure 5 As shown, Figure 5 (a) in the equation is the loss function of the training process. Figure 5(b) in the figure is the training accuracy. As the number of rounds increases, the training accuracy gradually increases and tends to be stable.

[0086] The theoretical positioning accuracy of current mainstream methods (such as the energy flow method) can be close to 90% in simulation under ideal conditions. The results of the model evaluation using the test set data are as follows: the recognition accuracy is close to 100%. Compared with the traditional method, the recognition accuracy is improved by about 10 percentage points. The confusion matrix of the test set is as follows Figure 6 shown.

[0087] Example 2

[0088] In one embodiment of the present disclosure, a wind power subsynchronous oscillation source tracing system based on DL-MVCNN is provided, comprising:

[0089] The data acquisition module is used to build a multi-wind farm grid connection model based on the actual grid connection scenario, obtain instantaneous current and power high-dimensional subsynchronous oscillation data, and pre-process it;

[0090] The dynamic information extraction module is used to model the intrinsic dynamics of instantaneous current and power subsynchronous oscillations based on sampled deterministic learning, extract the subsynchronous oscillation dynamic characteristics in a given state space, and generate a subsynchronous oscillation dynamics map;

[0091] The tracing module is used to input the subsynchronous oscillation dynamics graph into the multimodal oscillation tracing model, extract the dynamic pattern characteristics of each wind field, and fuse the dynamic pattern characteristics of each wind field through the channel attention mechanism in the multimodal oscillation tracing model to obtain the fusion characteristics. Different oscillation sources correspond to different deterministic learning dynamic characteristics. Based on the fusion characteristics, the weights given to the channels are adjusted, the corresponding channel weights are calculated, and the classification results of the wind field tracing are output.

[0092] Example 3

[0093] In one embodiment of the present disclosure, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method for tracing the source of subsynchronous oscillation of wind power based on DL-MVCNN is implemented.

[0094] Example 4

[0095] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the wind power subsynchronous oscillation tracing method based on DL-MVCNN is implemented.

[0096] Example 5

[0097] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the wind power subsynchronous oscillation tracing method based on DL-MVCNN.

[0098] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0100] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A wind power subsynchronous oscillation source tracing method based on DL-MVCNN is characterized by: include: Build a multi-wind farm grid connection model based on actual grid connection scenarios, obtain instantaneous current and power high-dimensional subsynchronous oscillation data, and preprocess it; Based on sampled deterministic learning, the intrinsic dynamics of instantaneous current and power subsynchronous oscillations are modeled. In a given state space, the subsynchronous oscillation dynamic characteristics are extracted and a subsynchronous oscillation dynamics map is generated. The subsynchronous oscillation dynamics graph is input into the multimodal oscillation tracing model to extract the dynamic pattern characteristics of each wind field. The dynamic pattern characteristics of each wind field are then fused through the channel attention mechanism in the multimodal oscillation tracing model to obtain the fusion characteristics. Different oscillation sources correspond to different deterministic learning dynamic characteristics. Based on the fusion characteristics, the weights are adjusted to the channels, the corresponding channel weights are calculated, and the classification results of the wind field tracing are output.

2. The wind power subsynchronous oscillation source tracing method based on DL-MVCNN according to claim 1, characterized in that: The acquired instantaneous current and power high-dimensional subsynchronous oscillation data are respectively subjected to oscillation dynamic modeling. First, the instantaneous current and power high-dimensional subsynchronous oscillation data are normalized to their maximum value. Then, a state identifier is constructed using the RBF neural network that models the internal dynamics of the time series. The weight update rate is constructed based on the Lyapunov stability theory, and finally an unknown nonlinear dynamic model of subsynchronous oscillation is obtained.

3. The wind power subsynchronous oscillation source tracing method based on DL-MVCNN according to claim 1, characterized in that: By determining the learning dynamics modeling method, an inherent dynamic constant RBF neural network of the time series signal is constructed along the subsynchronous oscillation trajectory, that is, the unknown nonlinear dynamic model of the subsynchronous oscillation. The current state trajectory is restricted to the set state space. Under the given state space size, the subsynchronous oscillation dynamic characteristics form a numerical matrix, which is the subsynchronous oscillation dynamics map.

4. The wind power subsynchronous oscillation source tracing method based on DL-MVCNN according to claim 1, characterized in that: The multimodal oscillation tracing model is a deterministic learning-multi-view convolutional neural network model, which includes a feature extraction module, a feature fusion module and a feature recognition expression prediction module. 1 ,X 2 ,X 3 ) as the input pair of the multi-modal oscillation source tracing model, (X 1 ,X 2 ,X 3 ) is the subsynchronous oscillation dynamics diagram of each wind field. The basic structure of the feature extractor module is a dual-stream structure with 4 layers of convolution blocks, which are respectively from X 1 、X 2 and X 3 Extract the dynamic mode characteristics of each wind field.

5. The wind power subsynchronous oscillation source tracing method based on DL-MVCNN according to claim 4, characterized in that: X 1 、X 2 and X 3 The three data streams belong to the same data pattern and have similar underlying spatial features. The feature extraction module is designed to share weights. The feature extraction modules are designed in series to obtain a joint representation expression. The channel attention mechanism is adopted in the feature fusion module to model the interdependence between channel feature responses to achieve adaptive readjustment of the channel. Different oscillation sources correspond to different deterministic learning dynamics characteristics. By adjusting the weights given to the channels, we focus on more important functional areas and use the global pool and shared MLP network to calculate the corresponding channel weights.

6. The wind power subsynchronous oscillation source tracing method based on DL-MVCNN according to claim 1, characterized in that: The feature recognition expression prediction module performs prediction output, which is the classification result of different wind fields.

7. The wind power subsynchronous oscillation source tracing system based on DL-MVCNN is characterized by: include: The data acquisition module is used to build a multi-wind farm grid connection model based on the actual grid connection scenario, obtain instantaneous current and power high-dimensional subsynchronous oscillation data, and pre-process it; The dynamic information extraction module is used to model the intrinsic dynamics of instantaneous current and power subsynchronous oscillations based on sampled deterministic learning, extract the subsynchronous oscillation dynamic characteristics in a given state space, and generate a subsynchronous oscillation dynamics map; The tracing module is used to input the subsynchronous oscillation dynamics graph into the multimodal oscillation tracing model, extract the dynamic pattern characteristics of each wind field, and fuse the dynamic pattern characteristics of each wind field through the channel attention mechanism in the multimodal oscillation tracing model to obtain the fusion characteristics. Different oscillation sources correspond to different deterministic learning dynamic characteristics. Based on the fusion characteristics, the weights given to the channels are adjusted, the corresponding channel weights are calculated, and the classification results of the wind field tracing are output.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the wind power subsynchronous oscillation source tracing method based on DL-MVCNN according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the wind power subsynchronous oscillation tracing method based on DL-MVCNN is implemented according to any one of claims 1 to 6.

10. An electronic device, characterized in that: include: A processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the wind power subsynchronous oscillation tracing method based on DL-MVCNN as described in any one of claims 1 to 6.

Images