New energy power supply distribution network adaptive distance protection method based on harmonic frequency spectrum model

Abstract

The invention discloses a new energy power supply distribution network adaptive distance protection method based on a harmonic frequency spectrum model in the technical field of power system relay protection, and aims to solve the problem that the nonlinear error of fundamental wave measurement is aggravated due to distance protection measurement impedance distortion caused by harmonic voltage drop. The method comprises the following steps: firstly, extracting each harmonic component in fault current and voltage in real time, and constructing a frequency spectrum model of harmonic and power frequency components by adopting a first-layer recursive least square method; based on the frequency spectrum model, a corrected power frequency impedance measurement value is obtained through a second-layer recursive least square method, the corrected impedance is combined with the fault type, and the fault type is recognized by adopting a support vector machine (SVM) improved peak value measurement method; the three-section distance protection logic is executed, and after the voltage / current abrupt change triggers a fault, fault section judgment and action time limit control are realized by comparing a corrected impedance value with a setting range, so that the protection range narrowing and maloperation risks caused by new energy harmonic waves are eliminated, and the reliability of a power grid is improved.

Description

Technical Field

[0001] This invention relates to an adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model, belonging to the field of power system relay protection technology. Background Technology

[0002] Currently, the number and scale of new energy power plants have both increased significantly. These plants widely employ inverters and converters based on power electronics technology to achieve energy conversion and grid connection. However, their high-frequency switching processes inject a large amount of harmonics into the grid, resulting in a high total harmonic distortion (THD) rate. This harmonic pollution not only degrades power quality but also severely interferes with the measurement and judgment logic of relay protection systems. On the one hand, harmonics cause voltage and current waveform distortion, distorting the extraction of the fundamental frequency component relied upon by traditional protection devices. On the other hand, harmonics alter the system's frequency domain characteristics. The capacitive reactive power compensation devices configured in new energy power plants may form parallel resonance with the system's inductive impedance, amplifying the amplitude of specific harmonics several times, thereby triggering false fault signals. Especially in fault scenarios, harmonic pollution exacerbates the risk of protection misjudgment: distance protection measures impedance deviates from the true power frequency impedance due to harmonic voltage drop, thus increasing the nonlinear error in fundamental frequency measurement.

[0003] Three-stage distance protection is a protection device that achieves selective disconnection of transmission line faults by measuring the impedance distance from the fault point to the protection installation point and setting three operating time limits according to the distance (stage I provides instantaneous protection for 80%-85% of the line, stage II provides a 0.5-second delay to protect the entire line and adjacent sections, and stage III serves as backup protection).

[0004] Recursive Least Squares (RLS) is an adaptive filtering algorithm for online real-time estimation of dynamic system parameters. This algorithm updates parameters iteratively, eliminating the need to store and recalculate all historical data, significantly reducing computational complexity and memory requirements, thus meeting the stringent real-time requirements of relay protection devices. By introducing a forgetting factor, the RLS algorithm assigns higher weights to new data, thereby quickly tracking dynamic changes in the system caused by faults or harmonic characteristic variations. This allows the constructed harmonic spectrum model to reflect the latest state of the system in real time.

[0005] The Support Vector Machine (SVM) improved peak measurement method is an intelligent fault diagnosis strategy that combines traditional signal processing with machine learning. The core innovation of this method lies in its departure from using a single peak indicator as a direct fault criterion. Instead, it systematically extracts multi-dimensional impact-sensitive features from vibration signals, including peak value, peak-to-peak value, peak factor, kurtosis, and envelope spectrum peak value, constructing a feature vector that comprehensively characterizes the equipment's state. Subsequently, leveraging the powerful pattern recognition and classification capabilities of SVM, this set of features is trained and learned, automatically establishing a nonlinear mapping relationship between different fault types and complex feature patterns. This method essentially represents a leap from "single alarm" relying on manual threshold judgment to data-driven "intelligent classification," significantly improving the accuracy of identifying impact faults, its anti-interference capability, and its ability to automatically distinguish fault types. Summary of the Invention

[0006] The purpose of this invention is to propose an adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model to improve the impedance measurement effect caused by harmonics.

[0007] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0008] This invention proposes an adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model, comprising:

[0009] Collect fault electrical quantities and preprocess the fault electrical quantities to obtain three-phase current and three-phase voltage;

[0010] Extract the harmonic components and power frequency components from the three-phase current and three-phase voltage;

[0011] A spectral model of harmonic components and power frequency components is constructed based on the first-level recursive least squares method.

[0012] The characteristic parameters of each harmonic are obtained based on the coefficient matrix output by the spectrum model.

[0013] The initial power frequency impedance is obtained based on the amplitude and phase of the power frequency component, and the harmonic impedance is obtained based on the amplitude and phase of each harmonic component.

[0014] Based on the aforementioned power frequency impedance and the previous nonlinear error term caused by harmonic characteristics. Establish a nonlinear correction model, where, This is a combination vector of the harmonic characteristic parameters. To correct the model weight parameter matrix;

[0015] The weight parameter matrix of the correction model in the nonlinear correction model is updated based on the second-level recursive least squares method, so that the power frequency impedance output by the nonlinear correction model is optimally close to the true impedance. The updated weight parameter matrix of the correction model is then substituted into the nonlinear trimming model to obtain the corrected power frequency impedance estimate.

[0016] Determine whether a fault has occurred in the new energy power grid; if a fault has occurred, obtain the waveform data at the time of the fault.

[0017] Based on the power frequency impedance estimate and the waveform data at the time of the fault, the peak measurement method improved by support vector machine (SVM) is used to identify the fault type.

[0018] Impedance compensation is performed based on the fault type, and a three-stage distance protection logic is executed by setting the impedance value. Accurate fault segment determination and action time limit control are achieved through impedance circle characteristic criterion.

[0019] Furthermore, the process of collecting fault electrical quantities and preprocessing the fault electrical quantities includes:

[0020] At the protection installation point of the new energy power distribution network, voltage transformers and current transformers are used to collect the bus voltage and line short-circuit current at the moment of fault.

[0021] The bus voltage and line short-circuit current at the moment of the fault are low-pass filtered to eliminate high-frequency noise and retain low-frequency signals. At the same time, the DC component of the line short-circuit current is high-pass filtered to reduce DC offset, thus obtaining the three-phase current. and three-phase voltage And ensure that harmonics of a preset number can be effectively captured.

[0022] Furthermore, the extraction of each harmonic component from the three-phase current and three-phase voltage includes:

[0023] For each target harmonic component of order k, where Establish an angular frequency A rotating coordinate system;

[0024] For the three-phase current and three-phase voltage Perform the Clarke transformation to obtain the two-phase stationary coordinate system. Quantity;

[0025] For the two-phase stationary coordinate system The components are subjected to Park transformation to obtain the DC components of each harmonic component k in the rotating coordinate system. , and , ;

[0026] A low-pass filter is applied to the DC component in the rotating coordinate system to filter out the high-frequency components remaining after the transformation, and the voltage amplitude of the kth harmonic component is output. Voltage phase Current amplitude and current phase The voltage amplitude Voltage phase Current amplitude and current phase Obtained through the following formula:

[0027] (1)

[0028] (2)

[0029] (3)

[0030] (4)

[0031] Furthermore, the extraction of the power frequency component from the three-phase current and three-phase voltage includes: obtaining the power frequency voltage amplitude from the three-phase current and three-phase voltage through a fast Fourier transform. Voltage phase Power frequency current amplitude and current phase .

[0032] Furthermore, the spectral models of the harmonic components and the power frequency components are expressed by the following formula:

[0033] (5)

[0034] in, The output vector is represented by the following formula:

[0035] (6)

[0036] The input vector, composed of the amplitude and phase combination of the harmonic components, is expressed by the following formula:

[0037] (7)

[0038] This is the first-level time-varying matrix, and the number of rows corresponds to the output vector. The dimension and number of columns correspond to the input vector. Dimensions Each element in the table represents the influence coefficient of any characteristic of any harmonic component on any power frequency characteristic, where the first row represents the influence weight of harmonic current on power frequency current. The third row shows the weighting of the impact of harmonic voltage on power frequency voltage. ; This represents the modeling error.

[0039] Furthermore, the first layer time-varying matrix Online identification is performed using the first-level recursive least squares method, including:

[0040] Set the initial first-layer time-varying matrix The first-level covariance matrix is ​​initialized as a zero matrix. Initialize the first layer of forgetting factor ;

[0041] Execute at each sampling point t:

[0042] Calculate the estimation error The estimation error This can be expressed by the following formula:

[0043] (8)

[0044] Among them, the estimation error This represents the error in making predictions using the spectral model at time t-1 without using the new data at time t to update the model parameters; Let be the output vector measured at time t; Let be the input vector to the spectral model at time t;

[0045] Calculate the first-level gain matrix at time t. The first layer gain matrix This can be expressed by the following formula:

[0046] (9)

[0047] in, Given all observation data up to time t-1, the time-varying matrix of the first layer... A measure of covariance;

[0048] Update the first-level time-varying matrix :

[0049] (10)

[0050] Update the first-level covariance matrix :

[0051] (11)

[0052] Output the first-level time-varying matrix at the current time. ;

[0053] When the error of several consecutive sampling points exceeds a specified multiple of the rated value, a zero-reset operation is performed on the first-level time-varying matrix.

[0054] Furthermore, obtaining the characteristic parameters of each harmonic based on the coefficient matrix output by the spectrum model includes: from the first layer time-varying matrix Obtaining the influence weight of harmonic current Harmonic voltage influence weight Harmonic amplitude ratio The harmonic amplitude ratio Obtained using the following formula:

[0055] (12).

[0056] Furthermore, the nonlinear correction model is expressed by the following equation:

[0057] (13)

[0058] Among them, the The actual impedance output by the model; This is the nonlinear error term caused by harmonic characteristics; The initial power frequency impedance is obtained based on the amplitude and phase of the power frequency component, using the following formula:

[0059] (14)

[0060] (15)

[0061] (16)

[0062] in, The amplitude of the initial power frequency impedance. The phase of the initial power frequency impedance;

[0063] The To correct the model weight parameter matrix, it is divided into linear and nonlinear terms based on the features. and Therefore, the corrected model structure for the real part is:

[0064] (17)

[0065] The modified model structure for the imaginary part is as follows:

[0066] (18).

[0067] Further, the step of updating the correction model weight parameter matrix in the nonlinear correction model based on the second-level recursive least squares method, so that the power frequency impedance output by the nonlinear correction model best approximates the true impedance, and then substituting the updated correction model weight parameter matrix into the nonlinear trimming model to obtain the corrected power frequency impedance estimate, includes:

[0068] Obtain the parameter set of linear term weights updated simultaneously using the second-level recursive least squares method. and nonlinear term weight parameter set A convergent corrected model is obtained;

[0069] At each sampling point t, the feature vector collected based on the spectrum model will be... Input the converged correction model to obtain the real part correction and the imaginary part correction, wherein the real part correction... Obtained using the following formula:

[0070] (19)

[0071] Imaginary part correction Obtained using the following formula:

[0072] (20)

[0073] By combining the real and imaginary correction values, the corrected power frequency impedance estimate is obtained. :

[0074] (twenty one)

[0075] in, The input vector is composed of harmonic characteristic parameters.

[0076] Furthermore, the convergent modified model obtained by simultaneously updating the linear term weight parameter set W(t) and the nonlinear term weight parameter set V(t) based on the second-level recursive least squares method includes:

[0077] Set the initial linear term weight parameter set Initialize the second-level covariance matrix as a zero matrix. Initialize the second-layer forgetting factor ;

[0078] Execute at each sampling point t:

[0079] Calculate the second-layer gain matrix at time t. The second layer gain matrix This can be expressed by the following formula:

[0080] (twenty two)

[0081] Update the linear term weight parameter set :

[0082] (twenty three)

[0083] Update the second-level covariance matrix :

[0084] (twenty four)

[0085] Apply the same second-level recursive least squares method with the same structure to the imaginary part of the true impedance, and update its corresponding linear term weight parameter set. ,and and The updates are independent and uncoupled.

[0086] Furthermore, the step of determining whether a fault has occurred in the new energy power grid, and if so, acquiring waveform data at the time of the fault, includes:

[0087] Determine whether the voltage descent criterion or current surge is met;

[0088] If any criterion is met, a fault is determined to have occurred, and waveform data of the bus voltage and line short-circuit current at the moment of the fault are collected simultaneously.

[0089] Furthermore, the improved peak measurement method for Support Vector Machine (SVM) includes:

[0090] Acquire training sample data, which includes simulation data of various fault types, different fault locations, different harmonic contents, and waveform data when the fault occurs;

[0091] Feature extraction is performed on the training sample data to construct the support vector machine (SVM) feature vector. ;

[0092] The feature vector The data is fed into the standardization module, which uses the mean and variance calculated during offline training to scale each feature component to a standard normal distribution, thereby generating a standardized training set.

[0093] Based on the standardized training set, a radial basis function is used as the kernel function, and a multi-class support vector machine (SVM) model is constructed based on a one-to-one strategy.

[0094] The penalty parameter C and the kernel parameter γ of the multi-class support vector machine (SVM) model are optimized by using grid search and cross-validation techniques to obtain an optimized multi-class support vector machine (SVM) model;

[0095] The optimized multi-class support vector machine (SVM) model is solidified, and the parameters of the solidified model are extracted and written into the memory of the protection device.

[0096] Furthermore, the feature vector is expressed as:

[0097] (25)

[0098] where is the power frequency peak feature, is the symmetrical component peak feature, is the phase difference feature, is the change rate feature, is the impedance amplitude feature.

[0099] 14. The adaptive distance protection method for a new energy power source distribution network based on a harmonic spectrum model according to claim 1, wherein the three-stage distance protection logic is as follows:

[0100] When the impedance compensation meets the setting impedance of segment I, a tripping command is issued within the time duration ;

[0101] When the impedance compensation meets the setting impedance of segment II, a tripping command is issued after a time delay of ;

[0102] When the impedance compensation meets the setting impedance of segment III, a tripping command is issued after a time delay of ;

[0103] where , and the setting impedance of segment I < the setting impedance of segment II < the setting impedance of segment III.

[0104] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0105] First, this invention proposes for the first time a dynamic spectrum modeling method for harmonic pollution from new energy sources. It first collects bus voltage and line current at the moment of a fault using voltage and current transformers, and preprocesses them with low-pass filtering to eliminate high-frequency noise and DC offset, retaining low-frequency harmonic signals. Then, it uses Clarke / Park transform to extract each harmonic component in real time, converting it into DC for measurement and calculation. Based on this, it defines the linear coupling relationship between the harmonic characteristic vector (including harmonic amplitude and phase) and the power frequency state vector (including the amplitude and phase of power frequency voltage and current). It uses recursive least squares to identify the time-varying coupling coefficient matrix online, and iteratively updates the covariance matrix and gain matrix to optimize the model parameters in real time, thereby establishing an accurate harmonic-power frequency spectrum mapping model, providing a data foundation for subsequent impedance correction.

[0106] Second, this invention is the first to construct a real-time nonlinear correction model for power frequency impedance based on harmonic characteristics. The model performs nonlinear correction on power frequency impedance based on harmonic spectral characteristics. By receiving harmonic parameters and initial power frequency impedance measurements, a nonlinear function model is constructed to describe the impedance error caused by harmonics. This model considers the complex relationship between harmonic characteristics and power frequency impedance error. The RLS algorithm is used to dynamically update the model weight parameters to approximate the true impedance in real time. A convergence monitoring and reset mechanism ensures that the model relearns when the system undergoes drastic changes, ultimately outputting a corrected power frequency impedance estimate, effectively eliminating measurement nonlinear errors caused by harmonics. Simultaneously, this invention employs a two-layer RLS to achieve progressively refined error compensation. The first layer of RLS solves the harmonic extraction and modeling problems, quantifying the complex harmonic effects into time-varying coupling coefficients. The second layer of RLS uses the output of the first layer as input and is specifically used to identify and correct the nonlinear measurement error of power frequency impedance caused by harmonics. This clearly defined two-layer structure enables higher accuracy in harmonic correction. When the system's operating state or harmonic characteristics change drastically, the two-layer algorithms can adjust and relearn independently or collaboratively, ensuring that the entire protection system always remains in its optimal operating state.

[0107] Third, this invention is the first to deeply integrate the corrected impedance with Support Vector Machine (SVM) intelligent classification to form adaptive distance protection. It uses the corrected impedance and SVM to improve the peak measurement method to achieve adaptive distance protection. First, fault detection is initiated by voltage drop and current surge criteria, and three-phase voltage and current waveforms are collected simultaneously. The corrected impedance is input, and the fault type is intelligently identified by combining the SVM model. The SVM classifies and outputs fault labels based on offline trained multi-dimensional feature vectors. Impedance compensation is performed according to the fault type, and the set impedance value is set to execute the three-stage distance protection logic. The impedance circle characteristic criterion is used to achieve accurate fault segment determination and action time limit control, thereby improving the reliability of the power grid. Attached Figure Description

[0108] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0109] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use.

[0110] Example 1:

[0111] This embodiment proposes a method for constructing a harmonic spectrum model, including:

[0112] Collect fault electrical quantities and preprocess them to obtain three-phase current and three-phase voltage;

[0113] Extract the harmonic components and power frequency components from the three-phase current and three-phase voltage;

[0114] A spectral model of harmonic components and power frequency components is constructed based on the first-level recursive least squares method.

[0115] In this embodiment, the acquisition of fault electrical quantities and the preprocessing of these quantities include:

[0116] At the protection installation point of the new energy power distribution network, voltage transformers and current transformers are used to collect the bus voltage and line short-circuit current at the moment of fault.

[0117] High-frequency noise ≥2kHz is eliminated by using a low-pass filter, followed by a 0.5Hz high-pass filter to eliminate DC offset, preserving low-frequency signals while ensuring effective capture of harmonics of the 13th order and above, thus obtaining the three-phase current. and three-phase voltage This provides reliable raw data for subsequent harmonic separation.

[0118] In this embodiment, specific harmonics are extracted from the complex mixed signal and converted into DC signals for easy measurement and calculation, including:

[0119] For each target harmonic component of order k, where Establish an angular frequency A rotating coordinate system;

[0120] For three-phase current Perform the Clarke transformation to obtain the two-phase stationary coordinate system. The component, calculated using the following formula:

[0121] (1)

[0122] For two-phase stationary coordinate system The components are subjected to Park transformation to obtain the DC components of each harmonic component k in the rotating coordinate system. and The calculation formula is:

[0123] (2)

[0124] For three-phase voltage Perform the same transformation to obtain , ;

[0125] A 20Hz low-pass filter is applied to the transformed DC component to filter out the residual high-frequency components, and the voltage amplitude of the kth harmonic component is output. Voltage phase Current amplitude and current phase Voltage amplitude Voltage phase Current amplitude and current phase Obtained through the following formula:

[0126] (3)

[0127] (4)

[0128] (5)

[0129] (6)

[0130] In this embodiment, the power frequency voltage amplitude is obtained from the three-phase current and three-phase voltage through Fast Fourier Transform. Voltage phase Power frequency current amplitude and current phase To accurately obtain the amplitude and phase of the power frequency component, this embodiment employs the Fast Fourier Transform (FFT) algorithm. FFT is an efficient calculation method that decomposes a time-domain signal into its frequency-domain representation, revealing the different frequencies, amplitudes, and phases that constitute the signal. First, the preprocessed continuous signal is discretely sampled to form a data sequence of finite length. To improve calculation speed and real-time performance, a data window of 1 to 2 power frequency cycles following the fault is typically selected (e.g., for a 50Hz system, this corresponds to a data length of 20ms to 40ms).

[0131] For the three-phase current in the above data window and three-phase voltage Perform FFT calculations separately. After the calculations are complete, the complex representation of the signal in the frequency domain, i.e., the spectrum, is obtained. This spectrum consists of a series of discrete frequency points, each frequency point corresponding to a complex number result. The magnitude of the complex number represents the amplitude of the frequency component, and the argument of the complex number represents the phase of the frequency component. In the calculated spectrum, locate the spectral line position corresponding to the fundamental power frequency (e.g., 50Hz or 60Hz).

[0132] Amplitude extraction: By directly reading the modulus of the complex number corresponding to the power frequency spectrum line, the power frequency voltage amplitude U1 and the power frequency current amplitude I1 can be obtained.

[0133] Phase extraction: The phase of the power frequency voltage can be obtained by calculating the argument of the complex number corresponding to the power frequency line. Phase with power frequency current This phase is the absolute phase relative to the start time of the data window.

[0134] In this embodiment, the spectral models of harmonic components and power frequency components are represented by the following formula:

[0135] (7)

[0136] in, The output vector represents the measured power frequency "state" after being affected by harmonics, and is expressed by the following formula:

[0137] (8)

[0138] The input vector, composed of the amplitude and phase of the harmonic components, is expressed by the following formula:

[0139] (9)

[0140] This is the first-level time-varying matrix, with the row number corresponding to the output vector. The dimension (4 rows), the number of columns corresponds to the input vector. The dimension (4 × harmonic order). Each element in the table represents the influence coefficient of a certain characteristic of a harmonic (such as the amplitude of the 5th harmonic current) on a certain characteristic of the power frequency (such as the phase of the power frequency voltage). The first row represents the influence weight of the harmonic current on the power frequency current. The third row shows the weighting of the impact of harmonic voltage on power frequency voltage. ; This represents the modeling error.

[0141] In this embodiment, the first layer time-varying matrix Online identification is performed using the first-level recursive least squares method, including:

[0142] Select the harmonic order to be processed (e.g., 3rd, 5th, 7th, etc.), and denote the total number of harmonics as N. Set the initial first-level time-varying matrix. The first-level covariance matrix is ​​initialized as a zero matrix. =Identity matrix × 10000, initialize the first-level forgetting factor ;

[0143] Execute at each sampling point t:

[0144] Calculate the estimation error estimation error This can be expressed by the following formula:

[0145] (10)

[0146] Among them, the estimation error This represents the error in making a prediction using the spectral model at time t-1 without updating the model parameters using the new data at time t. It represents the error in making a prediction using the old (time t-1) model parameters without updating the model parameters using the new data at time t. This error value is crucial because it is used to measure the performance of the current model and as a basis for adjusting and updating the model parameters H(t-1) to make more accurate predictions at the next time step. Let be the output vector measured at time t; Let be the input vector to the spectral model at time t;

[0147] Calculate the first-level gain matrix at time t. First layer gain matrix This can be expressed by the following formula:

[0148] (11)

[0149] in, Given all observation data up to time t-1, the time-varying matrix of the first layer... A measure of covariance; The forgetting factor is a constant between 0 and 1 used to control the rate at which the algorithm "forgets" historical data. This denominator is the normalization factor, which is used to control the numerical stability of the algorithm.

[0150] Update the first-level time-varying matrix :

[0151] (12)

[0152] Update the first-level covariance matrix :

[0153] (13)

[0154] Output the first-level time-varying matrix at the current time. ;

[0155] The convergence status is monitored in real time. When the norm of the covariance matrix P is less than the specified threshold, it is marked as converged. If the error of 10 consecutive sampling points exceeds twice the rated value, H is reset to a zero matrix.

[0156] Example 2:

[0157] This embodiment, based on Embodiment 1, proposes a method for correcting the nonlinearity of power frequency impedance based on harmonic spectrum characteristics, such as... Figure 1 As shown, it includes:

[0158] The characteristic parameters of each harmonic are obtained based on the coefficient matrix output by the spectrum model.

[0159] The initial power frequency impedance is obtained based on the amplitude and phase of the power frequency component, and the harmonic impedance is obtained based on the amplitude and phase of each harmonic component.

[0160] Based on the power frequency impedance and the previous nonlinear error term caused by harmonic characteristics. Establish a nonlinear correction model, where, This is a combination vector of harmonic characteristic parameters. To correct the model weight parameter matrix;

[0161] The weight parameter matrix of the correction model in the nonlinear correction model is updated based on the second-level recursive least squares method, so that the power frequency impedance output by the nonlinear correction model is optimally close to the true impedance. The updated weight parameter matrix of the correction model is then substituted into the nonlinear trimming model to obtain the corrected power frequency impedance estimate.

[0162] In this embodiment, obtaining the characteristic parameters of each harmonic based on the coefficient matrix output by the spectrum model includes: from the first-layer time-varying matrix Obtaining the influence weight of harmonic current Harmonic voltage influence weight Harmonic amplitude ratio Among them, the harmonic amplitude ratio Obtained using the following formula:

[0163] (14)

[0164] In this embodiment, the nonlinear correction model is expressed by the following equation:

[0165] (15)

[0166] in, The actual impedance output by the model; This is the nonlinear error term caused by harmonic characteristics; The initial power frequency impedance is obtained based on the amplitude and phase of the power frequency component, using the following formula:

[0167] (16)

[0168] (17)

[0169] (18)

[0170] in, The amplitude of the initial power frequency impedance. The phase of the initial power frequency impedance;

[0171] To correct the model weight parameter matrix, it is divided into linear and nonlinear terms based on the features. and Therefore, the corrected model structure for the real part is:

[0172] (19)

[0173] The modified model structure for the imaginary part is as follows:

[0174] (20)

[0175] In this embodiment, the weight parameter matrix of the correction model in the nonlinear correction model is updated based on the second-level recursive least squares method, so that the power frequency impedance output by the nonlinear correction model best approximates the true impedance. The updated weight parameter matrix of the correction model is then substituted into the nonlinear trimming model to obtain the corrected power frequency impedance estimate, including:

[0176] Obtain the parameter set of linear term weights updated simultaneously using the second-level recursive least squares method. and nonlinear term weight parameter set A convergent corrected model is obtained;

[0177] At each sampling point t, the feature vector collected based on the spectrum model will be... Inputting a convergent modified model yields real-part and imaginary-part corrections, where the real-part correction is obtained using the following formula:

[0178] (twenty one)

[0179] The imaginary part correction is obtained by the following formula:

[0180] (twenty two)

[0181] By combining the real and imaginary part corrections, the corrected power frequency impedance estimate is obtained. :

[0182] (twenty three)

[0183] in, The input vector is composed of harmonic characteristic parameters, where "1" corresponds to the first item in the set of weight parameters.

[0184] In this embodiment, a convergent modified model is obtained by simultaneously updating the linear term weight parameter set W(t) and the nonlinear term weight parameter set V(t) based on the second-level recursive least squares method, including:

[0185] Set the initial linear term weight parameter set Initialize the second-level covariance matrix as a zero matrix. Initialize the second-layer forgetting factor ;

[0186] Execute at each sampling point t:

[0187] Calculate the second-layer gain matrix at time t. Second layer gain matrix This can be expressed by the following formula:

[0188] (twenty four)

[0189] Update the linear term weight parameter set :

[0190] (25)

[0191] Update the second-level covariance matrix :

[0192] (26)

[0193] Apply the same second-level recursive least squares method with the same structure to the imaginary part of the true impedance, and update the corresponding set of linear term weight parameters. ,and and The updates are independent and uncoupled.

[0194] Example 3:

[0195] Based on Example 2, this embodiment proposes an adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model, including:

[0196] Determine whether a fault has occurred in the new energy power grid; if a fault has occurred, obtain the waveform data at the time of the fault.

[0197] Based on the power frequency impedance estimate and waveform data at the time of fault occurrence, a support vector machine (SVM) is used to improve the peak measurement method for fault type identification.

[0198] Impedance compensation is performed based on the fault type, and a three-stage distance protection logic is executed by setting the impedance value. Accurate fault segment determination and action time limit control are achieved through impedance circle characteristic criterion.

[0199] In this embodiment, it is determined whether a fault has occurred in the new energy power grid. If a fault occurs, the system immediately initiates data acquisition and caching, simultaneously recording detailed waveforms of the three-phase bus voltage and line current. This provides a data basis for subsequent harmonic analysis, impedance calculation, and fault identification. The judgment criteria are as follows:

[0200] Voltage sag criterion: The instantaneous rate of change of phase voltage (du / dt) is calculated in real time. If the voltage of any phase drops by more than 30% of its rated value in a very short time (such as 1-2ms), it is judged as abnormal.

[0201] Current surge criterion: Real-time calculation of the instantaneous rate of change (di / dt) of phase current. Any phase current that experiences a sudden increase exceeding 150% of its rated value within a very short period of time is considered abnormal.

[0202] In this embodiment, the improved peak measurement method for Support Vector Machine (SVM) includes:

[0203] Acquire training sample data: Collect a large amount of three-phase voltage and current data under normal conditions for various fault types (such as AG, BG, AB, ABG, ABC, etc.) through simulation software (such as PSCAD, EMTP) or actual power grid fault recorders.

[0204] Feature extraction is performed on the training sample data to construct the support vector machine (SVM) feature vectors. For each segment of fault data, feature vectors are extracted according to the method described in the patent. This vector contains key information such as power frequency peak value, symmetric components, harmonic characteristics, and corrected impedance.

[0205] eigenvectors The data is fed into the standardization module, which uses the mean and variance calculated during offline training to scale each feature component to a standard normal distribution, generating a standardized training set for each set of feature vectors. Label the samples with the corresponding "fault type" tags, such as Normal, AG, BG, etc., and output a large dataset where each row is a feature vector F of a sample, and each row corresponds to a fault type label.

[0206] Based on a standardized training set, a radial basis function is used as the kernel function, and a multi-class support vector machine (SVM) model is constructed based on a one-to-one strategy.

[0207] By using grid search and cross-validation techniques, the penalty parameter C and kernel parameter γ of the multi-class support vector machine (SVM) model are optimized to obtain the optimized multi-class support vector machine (SVM) model.

[0208] The optimized multi-class support vector machine (SVM) model is solidified, and the extracted solidified model parameters are written into the memory of the protection device.

[0209] In this embodiment, the feature vector Represented as:

[0210] (27)

[0211] in, , The peak value characteristic of the power frequency represents the peak value of the power frequency component after harmonic suppression. These are the peak values ​​of the power frequency components of the three-phase current, respectively. This represents the peak value of the power frequency component of the three-phase voltage.

[0212] , Symmetrical component peak characteristics, by The value in is calculated using the symmetric component method, where It is the zero-sequence current. It is a negative sequence current. It is the zero-sequence voltage. It is a negative sequence voltage;

[0213] , The phase difference is a characteristic feature, and the phase difference is represented.

[0214] , The rate of change characteristic represents the maximum value at the initial instant of the fault;

[0215] , The impedance amplitude characteristic is shown in the figure. These are the power frequency impedance amplitudes after parameter correction for phases A, B, and C, respectively.

[0216] In this embodiment, different impedance compensations are selected according to different fault types:

[0217] When the fault type is a single-phase ground fault, the formula for calculating impedance compensation is:

[0218] (28)

[0219] (29)

[0220] When the fault type is a two-phase short circuit, the formula for calculating impedance compensation is:

[0221] (30)

[0222] When the fault type is a two-phase-to-ground short circuit, the formula for calculating impedance compensation is:

[0223] (31)

[0224] When the fault type is a three-phase short circuit, the formula for calculating impedance compensation is:

[0225] (32)

[0226] In this embodiment, three sets of impedance settings are provided:

[0227] The first-stage impedance setting is configured at 80% of the impedance value of the protected line, covering 80% of the protected line's range. The formula is:

[0228] (33)

[0229] in, It describes the total impedance from the installation point of this protection device to the end of the line (i.e., the entire protected line).

[0230] The second-stage impedance setting is configured at 120% of the total impedance value of the protected line, covering the entire length of the protected line and 20% of adjacent lines as backup protection. The formula is as follows:

[0231] (34)

[0232] The third-stage impedance setting is configured according to the requirements of remote backup protection, and is used to cover the protected line and lines in more distant areas. The formula is:

[0233] (35)

[0234] In this embodiment, a three-stage distance protection is adopted, and the action criterion is generated using the impedance circle characteristic. The criterion formula is as follows:

[0235] (36)

[0236] in This indicates the action area; when this formula is satisfied, a protection action is triggered.

[0237] The three-stage distance protection logic is as follows:

[0238] When the impedance compensation satisfies the I-section setting impedance, that is, when it satisfies At that time, a trip command is issued within 20ms;

[0239] When the impedance compensation continuously meets the set impedance of section II, that is, when it meets the requirements... At that time, a trip command is issued after a delay of 0.3 to 0.5 seconds;

[0240] When the impedance compensation continuously meets the set impedance of stage III, that is, when the impedance is satisfied... After a delay of 1.0 to 1.5 seconds, a trip command is issued.

[0241] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model, characterized in that, include: Collect fault electrical quantities and preprocess the fault electrical quantities to obtain three-phase current and three-phase voltage; Extract the harmonic components and power frequency components from the three-phase current and three-phase voltage; A spectral model of harmonic components and power frequency components is constructed based on the first-level recursive least squares method. The characteristic parameters of each harmonic are obtained based on the coefficient matrix output by the spectrum model. The initial power frequency impedance is obtained based on the amplitude and phase of the power frequency component, and the harmonic impedance is obtained based on the amplitude and phase of each harmonic component. Based on the aforementioned power frequency impedance and the previous nonlinear error term caused by harmonic characteristics. Establish a nonlinear correction model, where, This is a combination vector of the harmonic characteristic parameters. To correct the model weight parameter matrix; The weight parameter matrix of the correction model in the nonlinear correction model is updated based on the second-level recursive least squares method, so that the power frequency impedance output by the nonlinear correction model is optimally close to the true impedance. The updated weight parameter matrix of the correction model is then substituted into the nonlinear trimming model to obtain the corrected power frequency impedance estimate. Determine whether a fault has occurred in the new energy power grid; if a fault has occurred, obtain the waveform data at the time of the fault. Based on the power frequency impedance estimate and the waveform data at the time of the fault, the peak measurement method improved by support vector machine (SVM) is used to identify the fault type. Impedance compensation is performed based on the fault type, and a three-stage distance protection logic is executed by setting the impedance value. Accurate fault segment determination and action time limit control are achieved through impedance circle characteristic criterion.

2. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 1, characterized in that, The process of collecting fault electrical quantities and preprocessing the fault electrical quantities includes: At the protection installation point of the new energy power distribution network, voltage transformers and current transformers are used to collect the bus voltage and line short-circuit current at the moment of fault. The bus voltage and line short-circuit current at the moment of the fault are low-pass filtered to eliminate high-frequency noise and retain low-frequency signals. At the same time, the DC component of the line short-circuit current is high-pass filtered to reduce DC offset, thus obtaining the three-phase current. and three-phase voltage And ensure that harmonics of a preset number can be effectively captured.

3. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 2, characterized in that, The extraction of harmonic components from the three-phase current and three-phase voltage includes: For each target harmonic component of order k, where Establish an angular frequency A rotating coordinate system; For the three-phase current and three-phase voltage Perform the Clarke transformation to obtain the two-phase stationary coordinate system. Quantity; For the two-phase stationary coordinate system The components are subjected to Park transformation to obtain the DC components of each harmonic component k in the rotating coordinate system. , and , ; A low-pass filter is applied to the DC component in the rotating coordinate system to filter out the high-frequency components remaining after the transformation, and the voltage amplitude of the kth harmonic component is output. Voltage phase Current amplitude and current phase The voltage amplitude Voltage phase Current amplitude and current phase Obtained through the following formula: （1） （2） （3） （4）。 4. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 1, characterized in that, The extraction of the power frequency component from the three-phase current and three-phase voltage includes: obtaining the power frequency voltage amplitude from the three-phase current and three-phase voltage through a fast Fourier transform. Voltage phase Power frequency current amplitude and current phase .

5. The adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model according to claim 3 or 4, characterized in that, The spectral models of the harmonic components and the power frequency components are expressed by the following formula: （5） In the formula, The output vector is represented by the following formula: （6） T is the symbol for matrix transpose. The input vector, composed of the amplitude and phase combination of the harmonic components, is expressed by the following formula: （7） This is the first-level time-varying matrix, and the number of rows corresponds to the output vector. The dimension and number of columns correspond to the input vector. Dimensions Each element in the table represents the influence coefficient of any characteristic of any harmonic component on any power frequency characteristic, where the first row represents the influence weight of harmonic current on power frequency current. The third row shows the weighting of the impact of harmonic voltage on power frequency voltage. ; For modeling error, The amplitude of the third harmonic current. Third harmonic current phase, The third harmonic voltage amplitude, This is the phase of the third harmonic voltage, and the same applies to the following phases; First layer time-varying matrix Online identification is performed using the first-level recursive least squares method, including: Set the initial first-layer time-varying matrix The first-level covariance matrix is ​​initialized as a zero matrix. Initialize the first layer of forgetting factor ; Execute at each sampling point t: Calculate the estimation error The estimation error This can be expressed by the following formula: （8） Among them, the estimation error This represents the error that occurs when using the spectral model at time t-1 to make predictions without updating the model parameters using the new data at time t. Let be the output vector measured at time t; Let be the input vector to the spectral model at time t; Calculate the first-level gain matrix at time t. The first layer gain matrix This can be expressed by the following formula: （9） in, Given all observation data up to time t-1, the time-varying matrix of the first layer... A measure of covariance; Update the first-level time-varying matrix : （10） Update the first-level covariance matrix : （11) Output the first-level time-varying matrix at the current time. ; When the error of several consecutive sampling points exceeds a specified multiple of the rated value, a zero-reset operation is performed on the first-level time-varying matrix.

6. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 5, characterized in that, The process of obtaining the characteristic parameters of each harmonic based on the coefficient matrix output by the spectrum model includes: from the first layer time-varying matrix Obtaining the influence weight of harmonic current Harmonic voltage influence weight Harmonic amplitude ratio The harmonic amplitude ratio Obtained using the following formula: （12)。 7. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 6, characterized in that, The nonlinear correction model is expressed by the following equation: （13） Among them, the The actual impedance output by the model; This is the nonlinear error term caused by harmonic characteristics; The initial power frequency impedance is obtained based on the amplitude and phase of the power frequency component, using the following formula: （14） （15） （16） in, The amplitude of the initial power frequency impedance. The phase of the initial power frequency impedance; The To correct the model weight parameter matrix, it is divided into linear and nonlinear terms based on the features. and Therefore, the corrected model structure for the real part is: (17) The modified model structure for the imaginary part is as follows: (18) Where Re(⋅) and Im(⋅) represent the real and imaginary parts of the complex number, respectively, w0, w1, w2, w3, w4, w5 are the weight coefficients to be identified in the real part correction model, and v0, v1, v2, v3, v4, v5 are the weight coefficients to be identified in the imaginary part correction model.

8. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 7, characterized in that, The step of updating the correction model weight parameter matrix in the nonlinear correction model based on the second-level recursive least squares method, so that the power frequency impedance output by the nonlinear correction model best approximates the true impedance, and then substituting the updated correction model weight parameter matrix into the nonlinear trimming model to obtain the corrected power frequency impedance estimate, includes: Obtain the parameter set of linear term weights updated simultaneously using the second-level recursive least squares method. and nonlinear term weight parameter set A convergent corrected model is obtained; At each sampling point t, the feature vector collected based on the spectrum model will be... Input the converged correction model to obtain the real part correction and the imaginary part correction, wherein the real part correction... Obtained using the following formula: （19） Imaginary part correction Obtained using the following formula: （20） By combining the real and imaginary correction values, the corrected power frequency impedance estimate is obtained. : （21） In the formula, The input vector is composed of harmonic characteristic parameters. Real part correction function Imaginary part correction function yes transpose, yes transpose; The convergent modified model obtained by simultaneously updating the linear term weight parameter set W(t) and the nonlinear term weight parameter set V(t) using the second-level recursive least squares method includes: Set the initial linear term weight parameter set Initialize the second-level covariance matrix as a zero matrix. Initialize the second-layer forgetting factor ; Execute at each sampling point t: Calculate the second-layer gain matrix at time t. The second layer gain matrix This can be expressed by the following formula: （22） Update the linear term weight parameter set : （23） In the formula, Let be the real part of the corrected weight vector at time t-1. For prediction error Update the second-level covariance matrix : （24） In the formula, The covariance matrix at time t-1; Apply the same second-level recursive least squares method with the same structure to the imaginary part of the true impedance, and update its corresponding linear term weight parameter set. ,and and The updates are independent and uncoupled.

9. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 1, characterized in that, The process of determining whether a fault has occurred in the new energy power grid, and if a fault has occurred, acquiring waveform data at the time of the fault, includes: Determine whether the voltage descent criterion or current surge is met; If any criterion is met, a fault is determined to have occurred, and waveform data of the bus voltage and line short-circuit current at the moment of the fault are collected simultaneously.

10. The adaptive distance protection method for new energy power distribution networks based on a harmonic spectrum model according to claim 8 or 9, characterized in that, The improved peak measurement method for Support Vector Machine (SVM) includes: Acquire training sample data, which includes simulation data of various fault types, different fault locations, different harmonic contents, and waveform data when the fault occurs; Feature extraction is performed on the training sample data to construct the support vector machine (SVM) feature vector. ; The feature vector The data is fed into the standardization module, which uses the mean and variance calculated during offline training to scale each feature component to a standard normal distribution, thereby generating a standardized training set. Based on the standardized training set, a radial basis function is used as the kernel function, and a multi-class support vector machine (SVM) model is constructed based on a one-to-one strategy. The penalty parameter C and kernel parameter γ of the multi-class support vector machine (SVM) model are optimized using grid search and cross-validation techniques to obtain the optimized multi-class support vector machine (SVM) model. The optimized multi-class support vector machine (SVM) model is solidified, and the extracted solidified model parameters are written into the memory of the protection device. The feature vector Represented as: （25） In the formula, Characterized by power frequency peak values. It is a symmetrical component peak characteristic. As a phase difference feature, Characteristic of rate of change This represents the impedance amplitude characteristic.

11. The adaptive distance protection method for new energy power distribution networks based on harmonic spectrum model according to claim 1, characterized in that, The three-stage distance protection logic is as follows: When the impedance compensation meets the I-section setting impedance, during the time period The internal circuit breaker trip command was issued. When the impedance compensation meets the set impedance of section II, after a certain period of time... A trip command is issued after a delay; When the impedance compensation meets the set impedance of section III, after a certain period of time... A trip command is issued after a delay; In the formula, , and the setting impedance of section I < the setting impedance of section II < the setting impedance of section III.

Images