Subsynchronous oscillation detection and traceability method and system for wind power grid-connected system

Abstract

The invention discloses a subsynchronous oscillation detection and traceability method and system for a wind power grid-connected system, and belongs to the technical field of oscillation traceability of the wind power grid-connected system.The subsynchronous oscillation detection and traceability method for the wind power grid-connected system comprises the steps that measurement data of the wind power grid-connected system is collected, and a subsynchronous oscillation data set of the wind power grid-connected system is constructed; an identification model is constructed based on the subsynchronous oscillation data set, the neural network estimation weight is adjusted, and a dynamic mode library of subsynchronous oscillation is established based on the converged constant neural network approximation model; and establishing a dynamic estimation model based on the dynamic mode library, calculating the residual error of the dynamic estimation model, identifying the operation mode of the wind power grid-connected system based on the residual error, and carrying out oscillation source positioning. According to the method, a dynamic model for describing a subsynchronous oscillation evolution law is established from data such as power grid voltage and current by utilizing deterministic learning, so that rapid detection and traceability of a subsynchronous oscillation mode are realized.

Description

Technical Field

[0001] This invention belongs to the field of oscillation tracing technology for wind power grid-connected systems, specifically relating to a method and system for detecting and tracing subsynchronous oscillations in wind power grid-connected systems. Background Technology

[0002] The statements herein provide only background information in relation to this invention and do not necessarily constitute prior art.

[0003] The frequent occurrence of novel subsynchronous oscillations caused by the interaction between wind turbines and the power grid has become a significant factor affecting the safe and stable operation of wind-powered grid-connected power systems. Current mainstream methods for detecting and tracing subsynchronous oscillations fall into two categories: one involves analyzing the characteristics of subsynchronous oscillations through time-domain or frequency-domain models. However, modeling large-scale wind power grid-connected systems is a massive undertaking, and wind farm / turbine parameters are incomplete and uncertain. Furthermore, the nonlinear dynamic characteristics and complex control strategies of wind farms are difficult to fully capture using traditional models; even if a model can be established, its physical assumptions and simplifications cannot guarantee a perfect match to actual conditions. Therefore, the aforementioned methods are insufficient for the efficient and accurate analysis of subsynchronous oscillations in large-scale wind power systems.

[0004] Secondly, the oscillation source can be located based on measurement data from the perspectives of energy and impedance. However, the energy of the system equipment will change abruptly under different forced oscillation frequencies, which makes it difficult for the energy flow method to accurately locate the oscillation source under different operating conditions.

[0005] Because the subsynchronous oscillation of wind power grid-connected systems exhibits many novel characteristics such as fast time-varying, wide frequency domain, strong nonlinearity, multi-modal coupling, and wide-area propagation, traditional analysis and control theories and methods for low-order, linear systems are insufficient to fully reveal the dynamic characteristics of subsynchronous oscillations and design excellent control strategies. Furthermore, due to the incomplete source tracing methods, it is difficult to accurately determine the minimum control range of new energy units. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for detecting and tracing subsynchronous oscillations in wind power grid-connected systems. It utilizes deterministic learning to establish a dynamic model that comprehensively describes the evolution of subsynchronous oscillations from grid voltage and current data, and achieves rapid detection and accurate tracing of subsynchronous oscillation modes based on dynamic estimation of model residuals.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: On the one hand, the technical solution of the present invention provides a method for detecting and tracing subsynchronous oscillations in a wind power grid-connected system, including: Collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system. An identification model is constructed based on the subsynchronous oscillation dataset, and the estimation weights of the neural network are adjusted. A dynamic pattern library of subsynchronous oscillations is established based on the converged constant neural network approximation model. A dynamic estimation model is established based on a dynamic model library. The residuals of the dynamic estimation model are calculated. Based on the residuals, the operating mode of the wind power grid-connected system is identified and the oscillation source is located.

[0008] In at least one implementation, measurement data from the wind power grid-connected system are collected to construct a subsynchronous oscillation dataset of the wind power grid-connected system, specifically including: Based on the simulation model of subsynchronous oscillation events of wind power grid connection, after the series compensation circuit is connected and stabilized, the subsynchronous oscillation phenomenon is generated by adjusting the model parameters of different wind farms, and grid data is collected to construct a subsynchronous oscillation dataset; among them, the data before and after the model parameters are changed are marked as normal mode and different oscillation source modes, respectively.

[0009] In at least one implementation, constructing an identification model based on a subsynchronous oscillation dataset and adjusting the neural network estimation weights specifically includes: Spatial trajectories of subsynchronous oscillation modes are formed based on subsynchronous oscillation datasets; Within the trajectory distribution area, construct an RBF neural network identification model; Based on deterministic learning theory, a neural network weight adjustment law based on Lyapunov stability is designed to obtain the estimated values ​​of the neural network weights.

[0010] In at least one implementation, the process of establishing a dynamic pattern library includes: Based on the neural network weight estimation after the transient state, the constant neural network weight is calculated to obtain the dynamic constant neural network approximation model; For subsynchronous oscillation data under different model parameter combinations in the subsynchronous oscillation dataset, constant neural network approximation models of normal mode and different oscillation source modes are calculated, and a dynamic mode library is constructed using the constant neural network approximation models.

[0011] In at least one implementation, the dynamic estimation model is specifically represented as follows:

[0012] In the formula, Indicates different operating modes; Indicates dimension; This indicates different operating scenarios for wind power grid-connected systems; Indicates time; express The measured flow data in dimensions; Indicates the state of the dynamic estimation model; This indicates the dynamic estimation model gain; Indicates the sampling period; This represents the constant neural network weights stored in the dynamic pattern library; This represents the basis functions of a neural network.

[0013] In at least one implementation, the dynamic estimation of model residuals is specifically expressed as follows:

[0014] satisfy:

[0015] In the formula, This represents the residual of the dynamic estimation model; Indicates the state of the dynamic estimation model; This represents the measured data; This indicates the dynamic estimation model gain; This indicates the internal dynamics of the test system.

[0016] In at least one implementation, the operation mode of the wind power grid-connected system is identified based on residuals, and the oscillation source is located, specifically including: The average L1 norm of the residuals in the dynamic estimation model is used as the decision indicator. When the decision index corresponding to the oscillation mode suddenly decreases and falls below the decision index corresponding to the normal mode, it indicates that the wind power grid-connected system has experienced a subsynchronous oscillation. Compare the decision indicators corresponding to all oscillation source modes, and the wind farm corresponding to the mode with the smallest indicator is the oscillation source.

[0017] Secondly, the technical solution of the present invention also provides a subsynchronous oscillation detection and tracing system for wind power grid-connected systems, comprising: The dataset construction module is configured to: collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system. The pattern library construction module is configured to: build an identification model based on the subsynchronous oscillation dataset and adjust the neural network estimation weights; and establish a dynamic pattern library for subsynchronous oscillations based on the converged constant neural network approximation model. The detection and tracing module is configured to: establish a dynamic estimation model based on a dynamic model library, calculate the residuals of the dynamic estimation model, identify the operating mode of the wind power grid-connected system based on the residuals, and locate the oscillation source.

[0018] Thirdly, the technical solution of the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the method for detecting and tracing subsynchronous oscillations in a wind power grid-connected system as described in the first aspect.

[0019] Fourthly, the technical solution of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for detecting and tracing subsynchronous oscillations in a wind power grid-connected system as described in the first aspect.

[0020] The beneficial effects of the above-described technical solution of the present invention are as follows: 1) The method for detecting and tracing the subsynchronous oscillations in a wind power grid-connected system of the present invention can learn the nonlinear system dynamic characteristics behind the subsynchronous oscillation phenomenon from the measurement data of the wind power grid-connected system by deterministic learning, and realize the rapid detection and accurate tracing of the subsynchronous oscillation mode based on the residual of the dynamic estimation model.

[0021] 2) This invention can model and learn the dynamic characteristics of subsynchronous oscillations using power grid data under different operating scenarios. Furthermore, it can expand the model library by learning new oscillation modes, enabling it to handle subsynchronous oscillation detection and tracing tasks in complex real-world scenarios. Compared to traditional parameterized model building, establishing a dynamic subsynchronous oscillation model library is more feasible and has wider applicability.

[0022] 3) In the oscillation modeling process based on deterministic learning theory, the neural network can accurately approximate the internal nonlinear dynamics of the wind power grid-connected system through measurement data, and completely preserve these dynamic laws describing the evolution characteristics of subsynchronous oscillations for subsequent online detection and source tracing tasks, thus enabling a more comprehensive acquisition of the internal nonlinear information of the wind power grid-connected system.

[0023] 4) This invention utilizes a neural network to learn and train different oscillation modes. The learning results not only include state variable information of the wind power grid-connected system but also comprehensively acquire and preserve the internal dynamics of the oscillation modes. In the subsynchronous oscillation detection and source tracing process, the dynamic estimation model designed based on the learning results utilizes the oscillation mode state and dynamic holographic features, enabling more thorough oscillation characteristic analysis and effectively improving the accuracy of detection and source tracing. The dynamic estimation model of this invention can rapidly activate the knowledge stored in the neural network, and the decision indicators are generated in real time with the input of the measured data, effectively ensuring the real-time performance of oscillation detection and source tracing.

[0024] 5) The subsynchronous oscillation detection and source tracing method based on deterministic learning of the present invention uses neural networks for automatic learning within a sampled data framework, and uses the learning results to design a dynamic estimation model to achieve automatic and rapid decision-making in a parallel and real-time manner. It does not require manual setting of various thresholds, has the ability to accurately acquire knowledge and utilize it efficiently, and is conducive to the digitalization and automation of subsynchronous oscillation detection and source tracing. Attached Figure Description

[0025] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0026] Figure 1 This is a schematic diagram of a method for detecting and tracing the subsynchronous oscillations in a wind power grid-connected system disclosed in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the wind power grid connection system model of three wind farms disclosed in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the measurement waveform at the wind farm busbar disclosed in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the spatial trajectory of the measurement signal disclosed in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the neuron node distribution disclosed in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the estimation results of some weights of the RBF neural network disclosed in Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of the subsynchronous oscillation source tracing decision index based on the dynamic estimation model residual disclosed in Embodiment 1 of the present invention. Detailed Implementation

[0027] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] The frequent occurrence of novel subsynchronous oscillations caused by the interaction between wind turbines and the power grid has become a significant factor affecting the safe and stable operation of wind power grid-connected power systems. These oscillations are primarily driven by converter control and involve complex excitation principles, including strong coupling of multiple control links within the equipment, strong interaction between multiple pieces of equipment, and resonance between grid-connected equipment and system series compensation. In recent years, equipment damage, grid disconnection, and even major power outages caused by these oscillations have occurred frequently, seriously threatening the efficient integration of new energy sources and the safe and stable operation of the system. Therefore, in large-scale wind power grid-connected systems, it is generally desirable for the control system to be able to detect subsynchronous oscillations in real time and accurately pinpoint the oscillation-inducing components and factors, thereby providing support for implementing effective suppression measures and ensuring the safe and stable operation of the wind power grid-connected system.

[0029] Methods for analyzing subsynchronous oscillation characteristics using time-domain or frequency-domain models include modal analysis, complex torque coefficient method, and impedance model analysis. These methods typically establish parameterized models based on certain physical assumptions and simplification conditions. They then use model eigenvalues, complex torque coefficients, and the Nyquist criterion to assess system stability. Furthermore, they analyze the influence of different factors on subsynchronous oscillation characteristics by calculating participation factors and damping coefficients, and identify the oscillation-inducing components. However, modeling large-scale wind power grid-connected systems is a massive undertaking, and wind farm / turbine parameters are incomplete and uncertain. Moreover, the nonlinear dynamic characteristics and complex control strategies of wind farms are difficult to fully capture using traditional models. Even if a model can be established, its physical assumptions and simplification conditions cannot guarantee a perfect match to actual conditions. Therefore, the aforementioned methods are insufficient for the efficient and accurate analysis of subsynchronous oscillations in large-scale wind power systems.

[0030] The method of locating oscillation sources based on measurement data from the perspectives of energy and impedance is essentially a "measure-and-determine" approach. Its key characteristic is the need to construct oscillation source criteria based on measurement data to form an online numerical algorithm. For example, inspired by low-frequency oscillation source tracing methods, researchers have calculated the transient energy flow of subsynchronous oscillations and located the disturbance source based on the sign of the energy flow power at the dominant frequency. The effectiveness of this method has been verified in the forced oscillation source location of doubly-fed induction generator (DFIG) wind turbines. However, at different forced oscillation frequencies, the energy of the system equipment undergoes abrupt changes, making it difficult for the energy flow method to accurately locate the oscillation source under different operating conditions. Another scholar has proposed a wide-area monitoring system for subsynchronous oscillations. This system uses real-time measured voltage and current phasor data to calculate the subsynchronous impedance and active (reactive) power, and locates the subsynchronous oscillation disturbance source based on the equivalent resistance or the sign of the absorbed active power. The effectiveness of this monitoring system has been verified by simulation on electromagnetic transient models of wind power grid-connected systems in Guyuan, North China, and Hami, Xinjiang. However, based on existing research, one challenge of these numerical algorithms lies in balancing the speed and accuracy of online tracing, especially when dealing with subsynchronous oscillation problems with more complex mechanisms. The adaptability of such methods still needs further verification.

[0031] In practical engineering, it is often desirable for control systems to detect oscillations in real time, pinpoint oscillation-inducing components and factors with high precision, and implement effective suppression measures to prevent their development, thereby ensuring the safe and stable operation of the system. However, this is a challenging task, and many key scientific problems remain to be solved.

[0032] Based on this, the purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for detecting and tracing subsynchronous oscillations in wind power grid-connected systems. By utilizing deterministic learning to establish a dynamic model that characterizes the evolution of subsynchronous oscillations from grid voltage and current data, a rapid detection and tracing of subsynchronous oscillation modes can be achieved.

[0033] Example 1 In a typical embodiment of the present invention, such as Figure 1 As shown in the figure, this embodiment discloses a method for detecting and tracing the subsynchronous oscillations in a wind power grid-connected system, including the following steps: S100. Collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system; S200. Construct an identification model based on the subsynchronous oscillation dataset and adjust the neural network estimation weights. Establish a dynamic pattern library of subsynchronous oscillations based on the converged constant neural network approximation model. S300. Establish a dynamic estimation model based on the dynamic model library, calculate the residual of the dynamic estimation model, identify the operation mode of the wind power grid-connected system based on the residual, and locate the oscillation source.

[0034] This method utilizes a deterministic learning dynamic model to learn the evolution law of subsynchronous oscillations from grid voltage and current data. This model contains holographic information about the state and dynamics of the subsynchronous oscillation mode's inherent system, exhibiting good interpretability. Based on this, a recognition mechanism enables rapid detection and accurate tracing of subsynchronous oscillation modes, possessing certain engineering application value and promising to provide a practically significant real-time monitoring method for the safe and stable operation of wind power grid-connected systems. The following detailed description of the above-mentioned method for detecting and tracing subsynchronous oscillations in wind power grid-connected systems, in conjunction with specific implementation methods, provides a detailed explanation.

[0035] S100. Collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system.

[0036] To address the task of tracing the source of subsynchronous oscillations in real-world wind power grid connections, it is necessary to establish a subsynchronous oscillation dataset covering different operating scenarios. Therefore, this step, based on a simulation model of wind power grid-connected subsynchronous oscillation events, generates subsynchronous oscillation phenomena by adjusting model parameters (such as the number of wind turbines, wind speed, and control parameters) for different wind farms after the series compensation circuit is connected and stabilized. Data such as grid current and voltage are collected to construct a subsynchronous oscillation dataset covering different operating scenarios. Data before and after the model parameter changes are labeled as normal mode and different oscillation source modes, respectively. Specifically, data from the point of grid stabilization to before the model parameter changes is recorded as normal mode, while data after the model parameter changes are recorded as different oscillation source modes based on the oscillation source conditions.

[0037] like Figure 2 As shown, taking a three-wind farm system as an example, the simulation model of the wind power grid-connected system generates 7 seconds of simulation data. A series compensation circuit is connected at 1.5 seconds. After the circuit stabilizes (approximately 3 seconds), the control parameters of the three wind farms are changed at 5 seconds, resulting in the measured waveforms at the wind farm's busbar, as shown below. Figure 3 As shown, from top to bottom, these represent phase A current, phase A voltage, and active power. The collected current, voltage, and power data from the power grid constitute a subsynchronous oscillation dataset. Data from 3 to 5 seconds is labeled as the normal mode, and data after 5 seconds is labeled as its corresponding oscillation source mode.

[0038] Then, the collected data is preprocessed. First, the three-phase voltage / current data is transformed from the stationary ABC coordinate system to the synchronous rotating dq coordinate system using the Park transform. After that, the transformed data is normalized, filtered, and downsampled for use in subsequent steps.

[0039] S200. Construct an identification model based on the subsynchronous oscillation dataset and adjust the neural network estimation weights. Establish a dynamic pattern library of subsynchronous oscillations based on the converged constant neural network approximation model.

[0040] For the data in the normal mode and different oscillation source modes in step S100, this step adopts a deterministic learning framework to identify the nonlinear dynamic system inside the subsynchronous oscillation through measurement data. The specific process is as follows.

[0041] S201. Construct an identification model based on a subsynchronous oscillation dataset.

[0042] In this step, a spatial trajectory is formed based on the preprocessed current signal and active power signal in the subsynchronous oscillation data set, such as... Figure 4 As shown, where it is represented as:

[0043] In the formula, This represents the d-axis current after the Parker transformation; This represents the q-axis current after the Parker transformation; express Active power is always available.

[0044] Then, within the trajectory distribution region, an RBF neural network identification model is constructed, which can be specifically represented as:

[0045] In the formula, This represents the state trajectory corresponding to different modes. Indicates the state of the RBF neural network identification model, superscript Indicates normal mode. This indicates the oscillation modes with oscillation sources being wind farms 1, 2, 3... Subscripts indicate different dimensions of a spatial trajectory; Represents the estimated weights of the neural network; Represents the basis of a neural network. This indicates the number of neurons.

[0046] Taking a three-wind farm system as an example, the RBF neural network identification model constructed within the trajectory distribution area can be represented as:

[0047] In the formula, Indicates the state of the RBF neural network identification model, superscript Indicates normal mode. The oscillation modes represent the oscillation sources as wind farms 1, 2, and 3.

[0048] Selecting the gain of the RBF neural network identification model The sampling period is And a Gaussian function is chosen as the basis for the RBF neural network, that is:

[0049] in, , The center of each neuron Evenly distributed in Within the space. The distribution of neuron nodes along some coordinate axes is as follows: Figure 5 As shown, by constructing multiple RBF neural network identification models with a unified structure, the internal dynamics of different operating modes (i.e., normal mode and different oscillation source modes) can be obtained.

[0050] Then, based on deterministic learning theory, a neural network weight adjustment law based on Lyapunov stability is designed to obtain the estimated values ​​of the neural network weights, specifically expressed as:

[0051] In the formula, the learning gain is , .

[0052] According to deterministic learning theory, the RBF subvectors formed by neurons near the trajectory... Satisfying the continuous activation condition corresponds to the neuron weight sub-vector. The exponential convergence to the ideal true value; for neurons far from the trajectory, the weight sub-vectors Keep it at 0, such as Figure 6 As shown. This means that with Adjustments Nonlinear dynamics that can gradually approximate normal or oscillatory modes within the trajectory neighborhood.

[0053] S202. Establish a dynamic pattern library.

[0054] First, for the weight adjustment process of the neural network in step S201, this step selects the time window after the transient state to calculate the weight mean. Specifically, the neural network estimates the weights after the transient state. Calculate the weights of the constant neural network to obtain an approximate model of the dynamic constant neural network. The weights of the constant neural network can be expressed as:

[0055] In the formula, This represents the time interval following the transient state. These constant values ​​are found in RBF neural networks. It can achieve an approximate representation of the internal dynamics within the neighborhood of the subsynchronous oscillation mode trajectory, which means that the learning results contain holographic features of the state and dynamics of the wind power grid-connected system.

[0056] Then, for subsynchronous oscillation data under different model parameter combinations in the subsynchronous oscillation dataset, constant neural network approximation models for normal mode and different oscillation source modes are calculated. A dynamic model library is constructed using these constant neural network approximation models. This dynamic model library can characterize the dynamic evolution of the subsynchronous oscillation system under various operating scenarios. Expanding the model library by performing dynamic modeling on oscillation data under multiple operating scenarios is beneficial for achieving a comprehensive expression of the development characteristics of subsynchronous oscillations.

[0057] In this embodiment, taking three wind farms and eight different combinations of model parameters as examples, the constructed dynamic model library is specifically represented as follows:

[0058] In the formula, This indicates different operating scenarios for wind power grid-connected systems; Indicates different operating modes, Indicates normal mode. The oscillation modes represent the oscillation sources as wind farms 1, 2, and 3.

[0059] Establishing parameterized models for large-scale wind power grid-connected systems is extremely difficult, and even when models are built, their physical assumptions and simplifications cannot guarantee a perfect match to actual conditions. This step, however, utilizes grid data from different operating scenarios to model and learn the dynamic characteristics of subsynchronous oscillations. Furthermore, it expands the dynamic model library by learning new oscillation modes, thus enabling the detection and tracing of subsynchronous oscillations in complex real-world scenarios. Therefore, compared to traditional parameterized model building, establishing a dynamic model library for subsynchronous oscillations in this step is more feasible and has wider applicability.

[0060] S300. Establish a dynamic estimation model based on the dynamic model library, calculate the residual of the dynamic estimation model, identify the operation mode of the wind power grid-connected system based on the residual, and locate the oscillation source.

[0061] S301. Establish a dynamic estimation model.

[0062] Based on the dynamic model library constructed in step S202, dynamic estimation models are built for measured flow data such as grid voltage and current in real-world scenarios. Specifically, this is applied to the measured flow data of the wind power grid-connected system under test. Using dynamic pattern library Construct a dynamic estimation model, specifically expressed as follows:

[0063] In the formula, This represents the state of the dynamic estimation model, and its gain is... These dynamic estimation models incorporate modeling results for normal modes and different oscillation source modes under various operating scenarios. With the measured data The input is used to perform numerical computations on all dynamic estimation models in parallel and in real time. When the measured data... When entering the learning region of the neural network, the dynamic estimation model can quickly activate the constant neural network approximation model. It provides dynamic knowledge of the underlying mechanisms and offers internal dynamic information for the corresponding modes.

[0064] S302. Calculate the residuals of the dynamic estimation model.

[0065] For the measured flow data It can be approximated by the Euler model, specifically expressed as:

[0066] In the formula, Indicates the sampling period; It represents the internal dynamics of the system under test, which is an unknown nonlinear function.

[0067] Define the residuals of the dynamic estimation model as:

[0068] It satisfies:

[0069] If stored in a constant neural network approximation model The dynamic knowledge in the data is effectively activated, along the state trajectory of the wind power grid-connected system under test. , It can accurately reflect the dynamic differences between the tested system's operating mode and various training modes. Using this as a similarity measure between different dynamic modes, similar modes imply smaller estimator residuals. Therefore, the dynamic estimation model residuals can be considered an external manifestation of the similarity between the tested mode and different training modes, and can be used to evaluate the operating status of the wind power grid-connected system.

[0070] S303. Identify the operating mode of the wind power grid-connected system based on residuals and locate the oscillation source.

[0071] In order to perform the online detection and source tracing of subsynchronous oscillations, this step takes the average L1 norm of the residuals of the dynamic estimation model constructed in step S302 as the decision index to make decisions on the detection and source tracing of subsynchronous oscillations.

[0072] Taking three wind farms and eight different combinations of model parameters as an example, the decision index is calculated as follows:

[0073] In the formula, This represents the length of the sliding window. The decision indicators calculated based on each dynamic estimation model are as follows: Figure 7 As shown, for Figure 7 Analysis shows that: (1) Before 5s, the decision index corresponding to the normal mode is less than the decision index corresponding to the oscillation mode. This means that the dynamic difference between the actual operation mode and the normal mode is small, and the wind power grid connection system is in normal operation.

[0074] (2) At 5s, wind farm 2, acting as the oscillation source, triggers subsynchronous oscillation. After 5s, the decision index corresponding to the normal mode gradually increases, while the decision index corresponding to the oscillation mode gradually decreases. This means that the dynamic difference between the actual operation mode and the normal mode gradually increases, while the dynamic difference between the actual operation mode and the oscillation mode gradually decreases. When the decision index corresponding to the oscillation mode is less than that corresponding to the normal mode (5.211s), the occurrence of subsynchronous oscillation can be detected.

[0075] (3) After detecting the subsynchronous oscillation phenomenon, the decision indicators corresponding to different oscillation source modes were compared. It was found that the minimum decision indicator corresponds to wind farm 2. Therefore, the oscillation source was located to wind farm 2. This set of measured data was actually caused by the subsynchronous oscillation caused by the parameter change of wind farm 2, which proved that the source tracing result of this method is correct.

[0076] Based on this, the subsynchronous oscillation source detection and tracing mechanism is defined in this step as follows: When the decision index of the oscillation mode decreases and falls below the decision index of the normal mode, it indicates that the wind power grid-connected system has experienced subsynchronous oscillation, thus realizing the detection of subsynchronous oscillation of the wind power grid-connected system. Once subsynchronous oscillation is detected, the decision indicators corresponding to all oscillation source modes are compared. The wind farm corresponding to the mode with the smallest indicator is the oscillation source, thus achieving the source tracing of the oscillation source.

[0077] Example 2 In a typical embodiment of the present invention, this embodiment discloses a subsynchronous oscillation detection and tracing system for a wind power grid-connected system, comprising: The dataset construction module is configured to: collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system. The pattern library construction module is configured to: build an identification model based on the subsynchronous oscillation dataset and adjust the neural network estimation weights; and establish a dynamic pattern library for subsynchronous oscillations based on the converged constant neural network approximation model. The detection and tracing module is configured to: establish a dynamic estimation model based on a dynamic model library, calculate the residuals of the dynamic estimation model, identify the operating mode of the wind power grid-connected system based on the residuals, and locate the oscillation source.

[0078] Example 3 In a typical embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the program implements the steps in the method for detecting and tracing subsynchronous oscillations in a wind power grid-connected system as described in Embodiment 1. The steps include: S100. Collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system; S200. Construct an identification model based on the subsynchronous oscillation dataset and adjust the neural network estimation weights. Establish a dynamic pattern library of subsynchronous oscillations based on the converged constant neural network approximation model. S300. Establish a dynamic estimation model based on the dynamic model library, calculate the residual of the dynamic estimation model, identify the operation mode of the wind power grid-connected system based on the residual, and locate the oscillation source.

[0079] Example 4 In a typical embodiment of the present invention, this embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for detecting and tracing subsynchronous oscillations in a wind power grid-connected system as described in Embodiment 1. These steps include: S100. Collect measurement data from the wind power grid-connected system and construct a subsynchronous oscillation dataset for the wind power grid-connected system; S200. Construct an identification model based on the subsynchronous oscillation dataset and adjust the neural network estimation weights. Establish a dynamic pattern library of subsynchronous oscillations based on the converged constant neural network approximation model. S300. Establish a dynamic estimation model based on the dynamic model library, calculate the residual of the dynamic estimation model, identify the operation mode of the wind power grid-connected system based on the residual, and locate the oscillation source.

[0080] 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 detecting and tracing subsynchronous oscillation of a wind power grid-connected system, characterized in that, The method comprises the following steps: Collecting measurement data of a wind power grid-connected system to build a wind power grid-connected system subsynchronous oscillation data set; Based on the subsynchronous oscillation data set, an identification model is built and the neural network estimation weight is adjusted, and a dynamic mode library of subsynchronous oscillation is established based on the converged constant neural network approximation model; Based on the dynamic mode library, a dynamic estimation model is established, the residual error of the dynamic estimation model is calculated, the operating mode of the wind power grid-connected system is identified based on the residual error, and the oscillation source is located.

2. The method of claim 1, wherein the wind power grid-connected system subsynchronous oscillation detection and tracing method is characterized by, The method comprises the following steps: Based on the simulation model of the wind power grid-connected subsynchronous oscillation event, after the series compensation access circuit is connected and stable operation is achieved, the model parameters of different wind farms are adjusted to produce subsynchronous oscillation phenomenon, and grid data is collected to build a subsynchronous oscillation data set; wherein, the data before and after the change of the model parameters are marked as normal mode and different oscillation source mode respectively.

3. The method of claim 1, wherein, Based on the subsynchronous oscillation data set, an identification model is built and the neural network estimation weight is adjusted, and a dynamic mode library of subsynchronous oscillation is established based on the converged constant neural network approximation model; The dynamic mode library establishment process comprises the following steps: Based on the neural network estimation weight after the transient state, the constant neural network weight is calculated to obtain the constant neural network approximation model of dynamics; For the subsynchronous oscillation data in the subsynchronous oscillation data set under different model parameter combinations, the constant neural network approximation model of the normal mode and the different oscillation source mode is calculated, and the dynamic mode library is constructed using the constant neural network approximation model.

4. The method of claim 1, wherein, The dynamic estimation model is specifically represented as: The dynamic estimation model residual error is specifically represented as: Satisfies:

5. The method of claim 1, wherein, Based on the residual error, the operating mode of the wind power grid-connected system is identified and the oscillation source is located, which comprises the following steps: wherein, denotes different operating modes; denotes dimensions; denotes different operating scenarios of a wind power grid-connected system; denotes time; denotes measured flow data of a dimension; denotes a dynamic estimation model state; denotes a dynamic estimation model gain; denotes a sampling period; denotes a constant neural network weight stored in a dynamic mode library; denotes a neural network base function.

6. The method of claim 1, wherein, Taking the average L1 norm of the dynamic estimation model residual error as the decision index; When the decision index corresponding to the oscillation mode suddenly decreases and is lower than the corresponding decision index of the normal mode, it indicates that the wind power grid-connected system has occurred subsynchronous oscillation; wherein represents a dynamic estimation model residual; represents a state of the dynamic estimation model; represents measured data; represents a dynamic estimation model gain; represents an internal dynamics of the test system.

7. The method of claim 1, wherein, Compare the decision indexes corresponding to all oscillation source modes, and the wind farm corresponding to the mode with the smallest index is the oscillation source. The method comprises the following steps: The data set construction module is configured to collect measurement data of a wind power grid-connected system to build a wind power grid-connected system subsynchronous oscillation data set; The mode library construction module is configured to build an identification model based on the subsynchronous oscillation data set and adjust the neural network estimation weight, and establish a dynamic mode library of subsynchronous oscillation based on the converged constant neural network approximation model; 8.A wind power grid-connected system subsynchronous oscillation detection and tracing system, characterized in that, The detection and tracing module is configured to establish a dynamic estimation model based on the dynamic mode library, calculate the residual error of the dynamic estimation model, identify the operating mode of the wind power grid-connected system based on the residual error, and locate the oscillation source. The program is executed by the processor to realize the steps of the wind power grid-connected system subsynchronous oscillation detection and tracing method in any one of claims 1-7. ​ ​ 9. A computer-readable storage medium having stored thereon a computer program, characterized in that, ​ 10. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps in the wind power grid-connected system subsynchronous oscillation detection and tracing method according to any one of claims 1-7 when executing the program.

Images