Knowledge-data dual-driven small-signal stability online evaluation method for multi-infeed system
By employing a knowledge-data dual-driven approach, and utilizing a BP neural network model and admittance dataset, a small-signal model for a new energy multi-feed system was constructed. This solved the problem of online assessment of small-signal instability risk in new energy systems, achieving efficient and reliable stability assessment and improving assessment accuracy and real-time performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies make it difficult to conduct accurate online assessments of small-signal instability risks in multi-infeed systems of new energy sources in advance and over a wide bandwidth. Furthermore, the "black box" nature of new energy units and their time-varying operating conditions make it difficult to establish accurate analytical models.
A knowledge-data dual-driven approach is adopted. By collecting the dq-axis admittance matrix of new energy units, an admittance dataset is constructed. A BP neural network model is trained and optimized, and a small-signal model of a new energy multi-infeed system is constructed in combination with the line topology to evaluate the system stability in real time.
It enables efficient and reliable online assessment of small-signal stability of multi-infeed systems for new energy sources, improves the prediction accuracy and real-time assessment of broadband instability risks, avoids dependence on the internal structure and control parameters of the unit, and has stronger engineering applicability and robustness.
Smart Images
Figure CN122371289A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of stability analysis technology for multi-converter grid-connected systems, specifically to a knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems. Background Technology
[0002] Currently, a large number of renewable energy sources are being distributed and connected to the grid in a wide area using converters, and multi-infeed systems for renewable energy are gradually taking shape. However, converters have wide-bandwidth control dynamic characteristics, and impedance mismatches between converters and between converters and the grid can easily lead to small-signal instability, threatening the safe and stable operation of the grid. Therefore, conducting online assessments of the small-signal stability of multi-infeed systems for renewable energy and accurately identifying instability risks is an important requirement for ensuring the stable operation of the system.
[0003] Currently, online stability assessment of multi-infeed systems for renewable energy is mainly divided into direct monitoring and model-based analytical methods. Direct monitoring, based on sensors such as synchronous phasor measurement devices, collects real-time system data, analyzes and extracts this data to obtain information such as oscillation frequency, amplitude, and damping ratio, enabling online assessment of small-signal stability. However, current global oscillation monitoring primarily focuses on the subsynchronous / supersynchronous frequency band, and these methods can only perform post-event oscillation analysis, making it difficult to achieve efficient and timely pre-event assessment of system instability risks. Model-based analytical methods clarify the risk and mechanism of small-signal instability by establishing an analytical model of the system. However, the large-scale integration of commercial "black box" renewable energy units and the random fluctuations in renewable energy output cause time-varying operating conditions for these units, making it difficult to analyze the impedance / admittance characteristics of multi-infeed systems for renewable energy. Therefore, stability assessment methods based on analytical models are not applicable. Summary of the Invention
[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a knowledge-data dual-driven online assessment method for small-signal stability of multi-infeed systems. This method solves the problems of existing methods being unable to achieve accurate online assessment of small-signal instability risks in new energy multi-infeed systems in advance and over a wide bandwidth, and the difficulty in establishing effective analytical models due to the influence of black-box units and time-varying operating conditions. To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A knowledge- and data-driven online evaluation method for small-signal stability of multi-feed systems includes the following steps: The dq-axis admittance matrices of new energy generating units under different operating points and different disturbance frequencies were collected. The operating point parameters and disturbance frequencies were used as input data and the dq-axis admittance matrices were used as output data. After normalization processing, a new energy generating unit admittance dataset was constructed. The BP neural network model is trained using the admittance dataset of new energy units to learn the nonlinear relationship between input and output data. The total loss function composed of mean square error and L2 regularization term is introduced to optimize the model parameters, and finally a trained BP neural network model is generated. The operating point parameters at the port of each new energy unit are acquired in real time under different disturbance frequencies. The operating point parameters and the corresponding disturbance frequencies are input into the trained BP neural network model to obtain the dq axis admittance matrix of each new energy unit. The voltage phase at the port of each new energy unit is obtained. After global coordinate transformation of the dq axis admittance matrix of each new energy unit, and combined with the line topology, a small-signal model of the new energy multi-infeed system containing active and passive subsystems is constructed, and the output response of the new energy multi-infeed system is obtained. Based on the output response, the eigenvalues of the open-loop transfer function of the new energy multi-infeed system are calculated, and the small-signal stability of the new energy multi-infeed system is evaluated online using the Nyquist criterion.
[0005] The present invention has the following beneficial effects: The knowledge-data dual-driven online assessment method for small-signal stability of multi-infeed systems proposed in this invention effectively solves the technical difficulties of existing technologies in achieving accurate online assessment of small-signal instability risks in multi-infeed systems of new energy sources in advance and across a wide bandwidth, as well as the difficulty in establishing accurate analytical models due to the influence of "black box" new energy units and time-varying operating conditions. At the same time, it can achieve efficient and reliable online assessment of small-signal stability of multi-infeed power systems containing "black box" new energy units under dynamic operating conditions, significantly improving the prediction accuracy and real-time performance of wide bandwidth instability risks, while avoiding dependence on the detailed internal structure and control parameters of new energy units, thus exhibiting stronger engineering applicability and robustness. Attached Figure Description
[0006] Figure 1 This is a flowchart illustrating the online evaluation method for small-signal stability of multi-feed systems driven by both knowledge and data proposed in this invention. Figure 2 This is a schematic diagram of the BP neural network model in the embodiment; Figure 3 This is a schematic diagram of the time-domain simulation model of the new energy unit in the embodiment; Figure 4 This is a schematic diagram illustrating the error between the admittance identification result output by the BP neural network model in the embodiment and the actual admittance value. Figure 5 This is a schematic diagram of the structure of the new energy multi-infeed system in the embodiment; Figure 6 This is a schematic diagram of the online stability assessment results for a multi-infeed system for new energy sources. Detailed Implementation
[0007] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0008] The specific embodiments adopted in this invention are as follows: like Figure 1 As shown, the knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems includes the following steps: Step 1: Using a series perturbation power supply and a single perturbation frequency injection method, collect the dq-axis admittance matrix of the new energy unit at different operating points and different perturbation frequencies. Use the operating point parameters and perturbation frequency as input data and the dq-axis admittance matrix as output data. After normalization processing, construct the admittance dataset of the new energy unit.
[0009] In this step, the perturbation frequency The frequency of the injected small signal is used to excite the new energy units to generate dynamic responses under different disturbance frequencies, so as to facilitate the extraction of the admittance characteristics of each unit; the operating point parameters include the d-axis voltage steady-state value. q-axis voltage steady-state value d-axis current steady-state value q-axis current steady-state value The two are used to characterize the steady-state operating conditions of new energy units; together they serve as inputs to the neural network model, thereby enabling online prediction of the admittance of new energy units under different operating conditions and frequencies.
[0010] Specifically, the formula for calculating the dq-axis admittance matrix of a new energy unit is as follows:
[0011] In the formula, For the dq-axis admittance matrix of the new energy unit; , , , These are the d-axis self-admittance component, dq-axis mutual admittance component, qd-axis mutual admittance component, and q-axis self-admittance component of the dq-axis admittance matrix of the new energy unit, respectively. , These are the current response components on the d-axis and q-axis respectively during the first perturbation frequency injection; , These are the current response components on the d-axis and q-axis respectively during the second perturbation frequency injection; , These are the voltage response components on the d-axis and q-axis, respectively, during the first perturbation frequency injection. , These are the voltage response components on the d-axis and q-axis respectively during the second perturbation frequency injection.
[0012] Specifically, during normalization, the normalization result of the input data is represented as follows:
[0013] In the formula, For the first Normalized results of the input data; , These are the upper and lower limits of the normalized interval for all input data, respectively. For the first The original values of the group input data; , These are the minimum and maximum values among all input data, respectively.
[0014] Specifically, during normalization, the normalized result of the output data is represented as follows:
[0015] in, For the first Normalized results of group output data; , These are the upper and lower limits of the normalized interval for all output data, respectively. For the first The raw values of the group output data; , These are the minimum and maximum values among all output data, respectively.
[0016] The purpose of normalizing all the acquired data in this step is to prevent variables with large numerical ranges from dominating the training process, which helps the model fit the data better.
[0017] Step 2: Train the BP neural network model using the admittance dataset of new energy units, learn the nonlinear relationship between input and output data, and optimize the model parameters by introducing a total loss function composed of mean square error and L2 regularization term, and finally generate a trained BP neural network model.
[0018] In this step, the structure and connection relationships of the BP neural network model are as follows: Figure 2 As shown, it includes an input layer, a hidden layer, and an output layer. The hidden layer consists of several neurons. The input layer receives five dimensions of data, including operating point parameters and perturbation frequencies, and is represented as follows: The hidden layer extracts features layer by layer through a combination of weight matrices and activation functions to fit the nonlinear relationship between the input and output data. The output layer outputs the real and imaginary parts of all elements in the dq-axis admittance matrix of the new energy unit, totaling 8 dimensions, represented as follows: , , , , , , , , , , , These are the real parts of the d-axis self-admittance component, dq-axis mutual admittance component, qd-axis mutual admittance component, and q-axis self-admittance component of the dq-axis admittance matrix of the new energy unit, respectively. , , , These are the imaginary parts of the d-axis self-admittance component, dq-axis mutual admittance component, qd-axis mutual admittance component, and q-axis self-admittance component of the dq-axis admittance matrix of the new energy unit, respectively.
[0019] Specifically, the formula for calculating the total loss function is as follows:
[0020] In the formula, This represents the total loss function value. This represents the number of training samples in the dataset. The dimension of the output layer of the model; For the first The training sample at the th ... The true value in each output dimension; For the first The training sample at the th ... Model identification values in each output dimension; The regularization coefficient is used. This represents the total number of layers in the model. For the first The number of neurons in a layer; For the first The number of neurons in a layer; For the first In the layer, connect the first Layer The first neuron and the second The first in the layer The weight parameters of each neuron.
[0021] In this step, based on the total loss function composed of mean squared error and L2 regularization, the weights and biases within the model are iteratively optimized to minimize the total loss function value, thereby driving the model output (admittance identification result) to approximate the true admittance sample value (true value). Specifically, mean squared error is used to measure the deviation between the model's output admittance identification result and the true value; mean squared error, on the other hand, applies a penalty to the weight parameters to constrain model complexity, thereby preventing overfitting and improving the model's generalization ability. Furthermore, the iteration terminates when the total loss function converges to a preset threshold or the maximum number of training epochs is reached, at which point training stops, outputting a high-performance BP neural network model, which is then used as the admittance identification model.
[0022] Step 3: Real-time acquisition of operating point parameters at the port of each new energy unit under different disturbance frequencies, inputting the operating point parameters and corresponding disturbance frequencies into the trained BP neural network model to obtain the dq-axis admittance matrix of each new energy unit.
[0023] In this step, during system operation, power flow calculation is first used to obtain in real time the steady-state voltage and current data (i.e., operating point parameters) at the port of each new energy unit under different disturbance frequencies, and these operating point parameters include the first... steady-state value of d-axis voltage at the port of a new energy unit , No. steady-state value of q-axis voltage at the port of a new energy unit , No. steady-state value of d-axis current at the port of a new energy unit , No. steady-state value of q-axis current at the port of a new energy unit Then, the operating point parameters and corresponding disturbance frequencies at the ports of each new energy unit are input into the trained BP neural network model to obtain the dq-axis admittance matrix of each new energy unit.
[0024] Step 4: Obtain the voltage phase at the port of each new energy unit. After performing global coordinate transformation on the dq-axis admittance matrix of each new energy unit, and combining it with the line topology, construct a small-signal model of the new energy multi-infeed system containing active and passive subsystems, and obtain the output response of the new energy multi-infeed system, specifically: Based on the voltage phase at the port of each new energy unit The dq-axis admittance matrix of each new energy unit is transformed globally to generate the dq-axis admittance matrix of each new energy unit in the global coordinate system, i.e.:
[0025]
[0026] In the formula, For the first Voltage phase at the port of each new energy unit; For the first Admittance matrix of each new energy unit along the dq axis in the global coordinate system; , These are the sine function and the cosine function, respectively. For the first The angle difference between the dq coordinate system of each new energy unit and the global coordinate system; For the first The dq-axis admittance matrix of each new energy unit; For the first Voltage phase at the port of each new energy unit; This is the phase angle in the global coordinate system.
[0027] The purpose of performing a global coordinate transformation on the dq-axis admittance matrix of each new energy unit in this step is to unify the coordinate system of all unit variable calculations in order to reduce the complexity of the model.
[0028] Based on the dq admittance matrix in the global coordinate system and combined with the line topology, a small-signal model of a new energy multi-infeed system containing source and passive subsystems is constructed to obtain the output response of the new energy multi-infeed system, i.e.:
[0029] In the formula, For the output response of a new energy multi-infeed system; for An identity matrix of order 1; For a small-signal model of an active subsystem, it is represented as: , This represents the equivalent admittance of the first new energy unit. Indicates the first Equivalent admittance of each new energy unit Here is the impedance matrix of power grid 1. For power grid The impedance matrix; For a passive subsystem, a small-signal model is used. The equivalent power vector of the new energy multi-infeed system; This is the open-loop transfer function of the new energy multi-infeed system.
[0030] In this step, the purpose of constructing a small-signal model of a new energy multi-infeed system containing source and passive subsystems and obtaining the output response of the new energy multi-infeed system is to establish an open-loop transfer function for stability analysis and provide a mathematical model for eigenvalue calculation.
[0031] Step 5: Based on the output response, calculate the eigenvalues of the open-loop transfer function of the new energy multi-infeed system, and then use the Nyquist criterion to evaluate the small-signal stability of the new energy multi-infeed system online.
[0032] Specifically, the formula for calculating the eigenvalues of the open-loop transfer function of a new energy multi-infeed system is as follows:
[0033]
[0034] In the formula, It is a matrix determinant operator; For the open-loop transfer function of the new energy multi-infeed system; It is the eigenvalue matrix; For a small-signal model of an active subsystem; For a passive subsystem, a small-signal model is used. for An identity matrix of order 1.
[0035] Specifically, the process of using the Nyquist criterion to evaluate the small-signal stability of a new energy multi-infeed system online is as follows: Plot curves on the complex plane using the real parts of the eigenvalues as the real axis and the imaginary parts of the eigenvalues as the imaginary axis, generating Nyquist curves for the eigenvalues. Determine whether the Nyquist curves for all eigenvalues do not enclose any curve on the complex plane. point, The imaginary unit, This is an operation where the ordinate in the complex plane is set to 0. If so, the new energy multi-infeed system is stable; otherwise, the new energy multi-infeed system is unstable.
[0036] In this step, the Nyquist curve of the eigenvalues refers to the curve drawn in the complex plane with the real part of the eigenvalues as the real axis and the imaginary part as the imaginary axis, where the point is... This refers to a point on the complex plane with an x-coordinate of -1 and a y-coordinate of 0. Specifically, on the complex plane, the x-coordinate represents the real part, and the y-coordinate represents the imaginary part. Generally, when representing coordinates on the complex plane, the imaginary unit 'j' is added before the y-coordinate parameter to distinguish it from the real coordinate system. 0 indicates that the ordinate in the complex plane is 0.
[0037] Based on this, by performing the above operations using the calculated eigenvalue data, the stability of the system can be determined, and the execution time is independent of the system's stability. Furthermore, when the system is stable, a stability evaluation index can be extracted from the intersection of the eigenvalue trajectory and the unit circle: the system's... eigenvalues The difference between the phase angle of the eigenvector at the intersection with the unit circle and -180° is the phase margin. The smallest phase margin is the phase margin of the multi-feed system. Furthermore, the potential oscillation frequency at the minimum phase margin is the system's potential oscillation frequency. f ocs Phase margin is one of the indicators for measuring the stability margin of a system. The purpose of calculating the phase margin is to select the minimum phase margin and the potential oscillation frequency of the system. The purpose of obtaining the minimum phase margin qP and the potential oscillation frequency is to provide an indicator for measuring the stability level of the system.
[0038] To verify the effectiveness of the knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems proposed in this invention, the following experiments were conducted: Build such in Matbac / Simulink Figure 3 The time-domain simulation model of the new energy unit (showing the circuit and control structures of the new energy unit) is illustrated in Table 1. Table 1 Parameter Table for New Energy Generating Units
[0039] exist Figure 3 In this context, PCC is the common coupling point of the new energy grid-connected system; For the first The filter inductor of each new energy unit; SVPWM is sinusoidal pulse width modulation technology; dq / abc is the conversion from the dq synchronous rotating coordinate system to the three-phase rotating coordinate system, and abc / dq is the conversion from the three-phase coordinate system to the dq synchronous rotating coordinate system; For the first The voltage at the grid connection point of each new energy unit, and , , The first The voltage components on the d-axis and q-axis at the grid connection point of each new energy unit; For the first The current response components at the grid connection point of each new energy unit, and , , The first The current response components on the d-axis and q-axis at the grid connection point of a new energy unit; , The first The current reference values of the d-axis and q-axis at the grid connection point of each new energy unit; For the first Current loop proportional integrator for a new energy unit; For the first A phase-locked loop proportional integrator for a new energy unit; It is an integral operator.
[0040] Admittance data of the new energy unit at different operating points were obtained using a frequency sweep method. The design of the operating conditions is shown in Table 2, totaling 196,020 data sets. Furthermore, due to the d-axis voltage phase-locked loop at the grid connection point of the new energy unit, the q-axis voltage at the unit's port remains constant at 0.
[0041] Table 2. Scheme for Generating Admittance Datasets for New Energy Units
[0042] The acquired admittance datasets of new energy generating units under different operating conditions were normalized, with the input data normalized to the range [0, 1] and the output data normalized to the range [-1, 1]. The datasets were then divided into training and test sets in a 4:1 ratio.
[0043] Construct a BP neural network model, the model structure is as follows: Figure 2 As shown: the input layer is The model consists of 5 dimensions. The hidden layer extracts features layer by layer through a combination of weight matrices and activation functions to fit the nonlinear relationship between input and output variables. The output layer contains the dq admittance data of the new energy unit, which has 8 dimensions. The new energy unit admittance dataset is input into the BP neural network model. The internal weights and biases are adjusted through iterative optimization to minimize the loss function, thereby driving the neural network output to approximate the true admittance sample value.
[0044] Based on this Figure 4 Showing the test set , , fixed, Admittance error under varying operating conditions , These are the voltage response components on the d-axis and q-axis, respectively. , These are the current response components along the d-axis and q-axis, respectively. And from... Figure 4 It can be seen that the admittance identification results output by the BP neural network model have small errors compared with the actual admittance values, and the admittance identification results show a high degree of consistency with the actual admittance values. The neural network admittance identification model can effectively capture the physical mapping relationship between the operating point and the admittance of the new energy unit.
[0045] at the same time, Figure 5 A new energy multi-infeed system comprising four new energy generating units and two grid power sources was demonstrated. The new energy generating units are of the same model, and the system parameters are shown in Table 3. Table 3 System Parameter Table
[0046] exist Figure 5 middle, , , , These are the equivalent current sources of new energy unit 1, new energy unit 2, new energy unit 4, and new energy unit 4, respectively. , , , , , , , All of these represent the impedance of harness 1, line 2, line 3, line 4, line 5, line 6, line 7, and line 8; This refers to the grid voltage. , The resistors are those of power grid 1 and power grid 2, respectively. , , , The equivalent admittances for Unit 1, Unit 2, Unit 3, and Unit 4 are respectively. , , , , , These are the steady-state current values for lines 1 through 6, respectively. , , , , , These are the steady-state voltage values for each grid-connected node 1 to 6.
[0047] Based on the aforementioned multi-infeed system structure, different operating conditions were designed, and the method proposed in this invention was used to verify the accuracy of online stability assessment of the multi-infeed system under dynamic operating conditions. The online assessment results are as follows: Figure 6 As shown. Among them, Figure 6 Figure (a) shows the waveforms of the operating point changes for each new energy unit; Figure 6 Figure (b) shows the results of the online stability assessment; Figure 6 Figure (c) shows the voltage and current waveforms output by the new energy unit 1; Figure 6 Figure (d) shows the voltage FFT analysis results of new energy unit 1. The specific analysis is as follows: (1) Operating condition 1: Within 0.50~1.00s, each new energy unit The instruction values are all 0.67 pu. The instruction value is 0.00 pu. Combined with... Figure 4An online stability assessment was conducted on the new energy multi-infeed system. The assessment results showed a phase margin of 9.96°, indicating system stability. Furthermore, time-domain simulation analysis revealed stable voltage and current waveforms for new energy unit 1, with a voltage harmonic content of 2.47%, also indicating system stability. The online stability assessment results were consistent with the simulation results.
[0048] (2) Operating Condition 2: When the simulation reaches 1.00s, the new energy units The command values were increased to 0.80pu, 0.72pu, 0.78pu, and 0.80pu respectively to simulate fluctuations in the output of new energy sources. At this point, the online stability assessment results were: a phase margin of -0.52°, system instability, and an oscillation frequency of 144.00Hz. In the time-domain simulation results, the voltage harmonic content of new energy unit 1 was 28.30%, and the oscillation frequency was (93+193) / 2=143.00Hz, with a relative error of 0.70% compared to the online assessment.
[0049] (3) Operating Condition 3: When the simulation reaches 1.50s, the current command values of new energy units 2 and 3 change as follows: , , , , , These are the steady-state values of the d-axis and q-axis currents of the new energy unit 2, respectively. , The d-axis and q-axis steady-state current values of new energy unit 3 are used to simulate the active power output and reactive power regulation changes of some units. At this time, the online stability assessment results are: phase margin of -3.50°, increased system instability, and oscillation frequency of 138.00Hz. Meanwhile, in the time-domain simulation, the harmonic content of the voltage of new energy unit 1 increases to 34.30%, and the oscillation frequency becomes (89+189) / 2=139.00Hz, with a relative error of 0.72% compared to the online assessment.
[0050] Therefore, through the analysis of the above-mentioned online stability assessment results and simulation results, it can be seen that the oscillation frequency error between the online stability assessment results and the actual simulation calculation results of the new energy multi-infeed system is <1%, and the instability and oscillation degree are consistent. This indicates that the knowledge-data dual-driven online stability assessment method for multi-infeed systems proposed in this invention can accurately reflect the stability level and oscillation frequency, verifying the correctness and effectiveness of the method.
[0051] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
[0052] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.
Claims
1. A knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems, characterized in that, Includes the following steps: The dq-axis admittance matrices of new energy generating units under different operating points and different disturbance frequencies were collected. The operating point parameters and disturbance frequencies were used as input data and the dq-axis admittance matrices were used as output data. After normalization processing, a new energy generating unit admittance dataset was constructed. The BP neural network model is trained using the admittance dataset of new energy units to learn the nonlinear relationship between input and output data. The total loss function composed of mean square error and L2 regularization term is introduced to optimize the model parameters, and finally a trained BP neural network model is generated. The operating point parameters at the port of each new energy unit are acquired in real time under different disturbance frequencies. The operating point parameters and the corresponding disturbance frequencies are input into the trained BP neural network model to obtain the dq axis admittance matrix of each new energy unit. The voltage phase at the port of each new energy unit is obtained. After global coordinate transformation of the dq axis admittance matrix of each new energy unit, and combined with the line topology, a small-signal model of the new energy multi-infeed system containing active and passive subsystems is constructed, and the output response of the new energy multi-infeed system is obtained. Based on the output response, the eigenvalues of the open-loop transfer function of the new energy multi-infeed system are calculated, and the small-signal stability of the new energy multi-infeed system is evaluated online using the Nyquist criterion.
2. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, Operating point parameters include the steady-state value of the d-axis voltage. q-axis voltage steady-state value d-axis current steady-state value q-axis current steady-state value .
3. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, The formula for calculating the dq-axis admittance matrix of a new energy power unit is: in, For the dq-axis admittance matrix of the new energy unit; , , , These are the d-axis self-admittance component, dq-axis mutual admittance component, qd-axis mutual admittance component, and q-axis self-admittance component of the dq-axis admittance matrix of the new energy unit, respectively. , These are the current response components on the d-axis and q-axis respectively during the first perturbation frequency injection; , These are the current response components on the d-axis and q-axis respectively during the second perturbation frequency injection; , These are the voltage response components on the d-axis and q-axis, respectively, during the first perturbation frequency injection. , These are the voltage response components on the d-axis and q-axis respectively during the second perturbation frequency injection.
4. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, During normalization, the normalization result of the input data is represented as: in, For the first Normalized results of the input data; , These are the upper and lower limits of the normalized interval for all input data, respectively. For the first The original values of the group input data; , These are the minimum and maximum values among all input data, respectively.
5. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, During normalization, the normalized result of the output data is represented as follows: in, For the first Normalized results of group output data; , These are the upper and lower limits of the normalized interval for all output data, respectively. For the first The raw values of the group output data; , These are the minimum and maximum values among all output data, respectively.
6. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, The formula for calculating the total loss function is: in, This represents the total loss function value. This represents the number of training samples in the dataset. The dimension of the output layer of the model; For the first The training sample at the th ... The true value in each output dimension; For the first The training sample at the th ... Model identification values in each output dimension; The regularization coefficient is used. This represents the total number of layers in the model. For the first The number of neurons in a layer; For the first The number of neurons in a layer; For the first In the layer, connect the first Layer The first neuron and the second The first in the layer The weight parameters of each neuron.
7. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, The process of obtaining the voltage phase at the port of each new energy unit, performing global coordinate transformation on the dq-axis admittance matrix of each new energy unit, and constructing a small-signal model of the new energy multi-infeed system containing active and passive subsystems based on the line topology, and obtaining the output response of the new energy multi-infeed system is as follows: Based on the voltage phase at the port of each new energy unit, the dq-axis admittance matrix of each new energy unit is transformed globally to generate the dq admittance matrix of each new energy unit in the global coordinate system, i.e.: in, For the first Admittance matrix of each new energy unit along the dq axis in the global coordinate system; , These are the sine function and the cosine function, respectively. For the first The angle difference between the dq coordinate system of each new energy unit and the global coordinate system; For the first The dq-axis admittance matrix of each new energy unit; For the first Voltage phase at the port of each new energy unit; Phase angle in global coordinate system; Based on the dq admittance matrix in the global coordinate system and combined with the line topology, a small-signal model of a new energy multi-infeed system containing source subsystems and passive subsystems is constructed to obtain the output response of the new energy multi-infeed system.
8. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 7, characterized in that, The output response of a new energy multi-infeed system is expressed as: in, For the output response of a new energy multi-infeed system; for An identity matrix of order 1; For a small-signal model of an active subsystem; For a passive subsystem, a small-signal model is used. The equivalent power vector of the new energy multi-infeed system; This is the open-loop transfer function of the new energy multi-infeed system.
9. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, The formula for calculating the eigenvalues of the open-loop transfer function of a new energy multi-infeed system is as follows: in, It is a matrix determinant operator; For the open-loop transfer function of the new energy multi-infeed system; For a small-signal model of an active subsystem; For a passive subsystem, a small-signal model is used. It is the eigenvalue matrix; for An identity matrix of order 1.
10. The knowledge-data dual-driven online evaluation method for small-signal stability of multi-feed systems according to claim 1, characterized in that, The process of online evaluation of the small-signal stability of a new energy multi-infeed system using the Nyquist criterion is as follows: Plot curves on the complex plane using the real parts of the eigenvalues as the real axis and the imaginary parts of the eigenvalues as the imaginary axis, generating Nyquist curves for the eigenvalues. Determine whether the Nyquist curves for all eigenvalues do not enclose any curve on the complex plane. point, The imaginary unit, This is an operation where the ordinate in the complex plane is set to 0. If so, the new energy multi-infeed system is stable; otherwise, the new energy multi-infeed system is unstable.