A grid-connected inverter parameter identification method and device based on multi-working-condition positive and negative sequence impedance sweep, a storage medium and a product

Abstract

This invention discloses a method, device, storage medium, and product for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweeping. It relates to the field of impedance measurement and parameter identification of grid-connected power electronic devices, and includes the following steps: S1, setting the operating conditions of the grid-connected inverter and constructing a candidate frequency set; S2, obtaining positive and negative sequence impedances; S3, constructing a gray-box sequence impedance model; S4, constructing an objective function across operating conditions and sequence; S5, selecting the identification frequency; S6, obtaining the identification results of the parameters to be identified; S7, evaluating the results of this round based on verification operating conditions, parameter correlation, and consistency of multiple initial values; if the conditions are not met, repeat steps S5 and S6. This invention, employing the above-mentioned method, device, storage medium, and product for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweeping, does not require reading the controller's internal settings and can provide the necessary parameters for impedance model verification and stability analysis.

Description

Technical Field

[0001] This invention relates to the field of impedance measurement and parameter identification of grid-connected power electronic devices, and in particular to a method, device, storage medium and product for identifying parameters of grid-connected inverters based on multi-operating-condition positive and negative sequence impedance frequency sweep. Background Technology

[0002] The sequence impedance of a grid-connected inverter is not solely determined by the filter inductor. Phase-locked loops (PLLs) and current PI controllers exhibit different amplitude and phase characteristics at low and mid-frequency ranges, while sampling delay, voltage feedforward, and sampling filtering further alter the curves at higher frequencies. In engineering projects, only the grid connection point voltage and current are often available; the controller program and settings may not be publicly available. After equipment replacement or firmware upgrades, the parameters in the original model may no longer be applicable. In such cases, it is necessary to deduce the model parameters from the port response.

[0003] When fitting the data using only a single operating point with rated active power and Iq=0, some parameters will compensate for each other. For example, the current loop integral coefficient and the voltage feedforward coefficient may give similar local impedance errors, but the corresponding parameter values ​​are not unique. Frequency sweeping in the project also showed that the frequency shift term near 50 Hz needs to be handled separately, and inconsistencies between the analytical model and simulation data occurred in the 30–70 Hz range. Simply adding uniform frequency points increases the computational load but may not provide sufficient information about the direction of the weakest parameter. On the other hand, even if multi-startpoint optimization converges, the reliability of the parameters cannot be judged solely based on the training error. Summary of the Invention

[0004] The purpose of this invention is to provide a method, device, storage medium and product for identifying grid-connected inverter parameters based on multi-operating-condition positive and negative sequence impedance frequency sweep, to solve the problem of how to reduce mutual compensation between parameters when only port frequency sweep data can be used, as mentioned in the background art, and to determine whether the current frequency set is sufficient to support the joint identification of the selected parameters.

[0005] To achieve the above objectives, this invention provides a method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweep, comprising the following steps: S1. Set at least two operating conditions for grid-connected inverters and construct a candidate frequency set; and divide the data corresponding to different operating conditions into training data and validation data. S2. Apply positive-sequence and negative-sequence perturbations to a subset of frequencies within the candidate frequency set to obtain the grid-connected point voltage phasor and grid-connected current phasor under each operating condition, and obtain the positive-sequence impedance and negative-sequence impedance accordingly. S3. Construct a gray box sequence impedance model. The gray box sequence impedance model takes the parameters to be identified and the operating condition variables as inputs and the predicted positive sequence impedance and the predicted negative sequence impedance as outputs. S4. Based on the deviation between the actual sequencing impedance and the predicted sequence impedance, construct an objective function that spans operating conditions and sequence types; S5. Calculate the sensitivity information of the parameter to be identified in the objective function, and select the identification frequency from the candidate frequency set under at least one of the constraints of frequency band coverage, frequency point spacing and measurement quality. S6. Within the parameter boundary, the objective function is optimized using multiple initial points to obtain the identification results of the parameters to be identified. S7. Based on the verification data and identification results, perform an evaluation. If the evaluation results do not meet the preset conditions, select a supplementary frequency and repeat the sequence impedance acquisition and objective function optimization.

[0006] Therefore, the present invention employs the above-mentioned method, equipment, storage medium, and product for identifying grid-connected inverter parameters based on multi-operating-condition positive and negative sequence impedance frequency sweep, which has the following beneficial effects: (1) The present invention evaluates impedance fitting error and parameter identifiability separately, which can avoid selecting ill-conditioned or highly correlated parameter combinations simply because the fitting error is small, thereby improving the credibility, repeatability and engineering interpretability of the identification results.

[0007] (2) The present invention uses positive and negative sequence impedance data under different operating conditions together for parameter constraints, which can enrich the dynamic information contained in the identification data, reduce the correlation and mutual compensation between different parameters, and improve the stability and cross-operating condition applicability of parameter identification.

[0008] (3) The present invention allows parameters that can be independently determined by spectrum analysis, high-frequency impedance characteristics or DC side data to be fixed in advance, thereby reducing the parameter dimension of joint optimization, improving numerical solution conditions, improving optimization convergence efficiency and consistency of multiple initial value calculations.

[0009] (4) This invention only incorporates the supplementary frequency points that have been actually scanned and passed the data quality check into the identification dataset, which can avoid interference from unmeasured frequency points, interpolation results or model prediction results on the identification data, ensure the traceability of data sources, and improve the reproducibility of the identification process.

[0010] (5) The present invention uses parameter sensitivity to supplement the selection of frequency, instead of using it repeatedly for fitting weight setting. This can avoid the sensitivity information being repeatedly included in the experimental design and parameter estimation stages, reduce the excessive bias of the identification results on a few high-sensitivity frequency points, and improve the balance of frequency coverage and the generalization ability under verification conditions.

[0011] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0012] Figure 1This is a flowchart illustrating an embodiment of the present invention: a method, device, storage medium, and product for identifying parameters of a grid-connected inverter based on multi-condition positive and negative sequence impedance frequency sweep. Figure 2 This is a schematic diagram of the multi-condition positive and negative sequence impedance acquisition system, which is an embodiment of the present invention of a grid-connected inverter parameter identification method, device, storage medium and product based on multi-condition positive and negative sequence impedance frequency sweep. Figure 3 This is a schematic diagram of the supplementary frequency selection and iterative identification process based on identifiability evaluation in an embodiment of the present invention, which is a method, device, storage medium and product for identifying parameters of a grid-connected inverter based on multi-operating-condition positive and negative sequence impedance frequency sweeping. Figure 4 This is a schematic diagram of the parameter identification device structure of an embodiment of the present invention, which is based on a multi-condition positive and negative sequence impedance frequency sweep for grid-connected inverters. Figure 5 The residual convergence curves for identifying five parameters from ten initial parameters in an embodiment of the present invention, which is based on a multi-condition positive and negative sequence impedance frequency sweep for grid-connected inverter parameters, including a method, device, storage medium, and product. Figure 6 This is a comparison chart of the amplitude of the measured values ​​of positive and negative sequence impedance under multiple operating conditions and the calculated values ​​of the identification model in an embodiment of the present invention, which is based on a multi-operating condition positive and negative sequence impedance frequency sweep for grid-connected inverter parameters. Figure 7 This is the result of five-parameter identification after fixed sampling frequency, filter inductance and DC voltage in an embodiment of the present invention, which is based on multi-operating-condition positive and negative sequence impedance frequency sweep for grid-connected inverter parameter identification method, equipment, storage medium and product embodiment. Figure 8 This invention provides a method, device, storage medium, and product embodiment for identifying parameters of a grid-connected inverter based on multi-condition positive and negative sequence impedance frequency sweep, showing d-axis and q-axis current tracking waveforms under five operating conditions. Figure 9 This invention provides a comparison of the d / q-axis current response of control parameters and 57 identification parameters under different reactive current conditions in an embodiment of a grid-connected inverter parameter identification method, device, storage medium, and product based on multi-condition positive and negative sequence impedance frequency sweep. Detailed Implementation

[0013] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0014] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0015] Example Please see Figures 1-9 This invention provides a method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweeping. It integrates operating point design, positive and negative sequence frequency sweeping, parameter sensitivity, and independent verification into a single identification process. The frequency ranges requiring additional measurements are determined by existing data, rather than by a pre-defined frequency table applicable to all devices. The specific process is as follows: first, measurements are taken at different operating conditions... , Obtaining positive sequence impedance at given values and negative sequence impedance ,in, The d-axis current at the steady-state operating point of the grid-connected inverter. The q-axis current is the steady-state operating point. The same set of model parameters is then used to simultaneously interpret the data at each operating point. The objective function can be the amplitude-phase residual or the complex relative residual. The normalized parameter Jacobian matrix is ​​calculated for the current solution to check the condition number of the information matrix, parameter correlation, and weakest direction. If the verification condition error is too large or the parameter correlation is too strong, supplementary points are selected from frequencies that have not yet been measured, and the actual frequency sweep is completed before fitting. This method includes the following steps: S1. Set at least two grid-connected inverter operating conditions and construct a candidate frequency set; and divide the data corresponding to different operating conditions into training data and validation data.

[0016] Operating conditions are characterized by a single parameter or any combination of the following: d-axis current setpoint, q-axis current setpoint, active power setpoint, reactive power setpoint, DC side voltage, grid voltage, and short-circuit ratio.

[0017] The at least two grid-connected inverter operating conditions are specifically: using two different active power levels, or configuring at least one non-zero reactive power operating condition, or setting two types of operating conditions simultaneously.

[0018] S2. For some frequencies within the candidate frequency set, apply positive-sequence perturbation and negative-sequence perturbation respectively, obtain the grid connection point voltage phasor and grid connection current phasor under each operating condition, and obtain the positive-sequence impedance and negative-sequence impedance accordingly.

[0019] This step also includes a measurement quality check, where frequencies that fail the check are removed or have their weight reduced. The measurement quality check includes at least one of the following: steady-state check, disturbance current lower limit check, signal-to-noise ratio check, disturbance amplitude linearity check, spectral leakage check, abnormal phase jump check, and fundamental frequency neighborhood model consistency check.

[0020] S3. Construct a gray-box sequence impedance model. The gray-box sequence impedance model takes the parameters to be identified and the operating condition variables as inputs, and outputs the predicted positive sequence impedance and the predicted negative sequence impedance.

[0021] The gray-box sequence impedance model includes: the closed-loop transfer function of the phase-locked loop under positive and negative sequence frequency shift variables, the digital control delay and zero-order hold circuit, the voltage sampling filter circuit and the current sampling filter circuit, and the complex coupling terms of the d-axis operating point current and the q-axis operating point current in the positive and negative sequence domains; among them, when the q-axis operating point current is zero, the complex coupling terms degenerate into coupling terms that are only related to the d-axis operating point current.

[0022] S4. Based on the deviation between the actual sequencing impedance and the predicted sequence impedance, construct an objective function that spans operating conditions and sequence types.

[0023] The objective function includes the weighted sum of squares of the magnitude residual and the phase residual of the sequence impedance, or the weighted sum of squares of the real and imaginary parts of the complex relative residual of the sequence impedance; the weights are determined based on at least one of sequence, operating condition, frequency, measurement quality or parameter prior, and different normalization scales are allowed for the magnitude residual and the phase residual.

[0024] The parameters to be identified include at least four of the following: phase-locked loop proportional coefficient, phase-locked loop integral coefficient, current regulation loop proportional coefficient, current regulation loop integral coefficient, feedforward coefficient, switching frequency or sampling frequency, grid-connected equivalent inductance, and DC-side voltage. Among these, parameters that can be obtained through spectral characteristics, high-frequency impedance slope, equipment rating information, or independent testing are fixed, while the remaining parameters are jointly identified through an objective function.

[0025] S5. Calculate the sensitivity information of the parameters to be identified in the objective function. Under at least one of the following constraints—band coverage constraint, frequency spacing constraint, and measurement quality constraint—select the identification frequency from the candidate frequency set. Specifically: based on the preset initial parameters or the identification results of the previous round, calculate the residual Jacobian matrix and the information matrix, and select the identification frequency; for the selected frequencies that have not yet been measured, perform positive and negative frequency sweeps and incorporate the data that have passed the quality check into the training set.

[0026] Sensitivity information includes at least one of the following: normalized parameter Jacobian matrix, information matrix composed of normalized parameter Jacobian matrix, minimum eigenvalue of information matrix, condition number, determinant, covariance matrix, and parameter correlation coefficient; the candidate frequency selection score is formed by a combination of at least two of the following: weak direction sensitivity gain, information gain, target parameter accuracy gain, condition number improvement, and parameter correlation reduction.

[0027] The selection of identification frequencies is specifically as follows: divide the candidate frequencies into multiple frequency bands, and set an upper or lower limit for the number of frequency points in each frequency band; exclude frequency ranges with frequency shift singularities, nonlinear disturbances, or model deviations; and ensure that the distance between the newly selected frequency and the measured frequency is not less than the preset absolute distance and the preset relative distance.

[0028] S6. Within the parameter boundary, the objective function is optimized using multiple initial points to obtain the identification results of the parameters to be identified.

[0029] Multiple initial points include preset initial points and random initial points generated in the logarithmic parameter space; the optimization process employs trust regions, Levenberg-Marquardt algorithms, sequential quadratic programming, swarm intelligence algorithms, or combinations thereof; for unconstrained optimization algorithms, internal optimization variables are mapped to the upper and lower bounds of physical parameters through continuous mapping.

[0030] S7. Evaluate based on verification data and identification results. If the evaluation results do not meet the preset conditions, select a supplementary frequency and repeat the sequence impedance acquisition and objective function optimization. Specifically: evaluate the results of this round based on verification conditions, parameter correlation, and consistency of multiple initial values; if the conditions are not met, repeat S5 supplementary frequency sweep and S6.

[0031] The evaluation results include the root mean square error of the positive-sequence magnitude of the training set, the root mean square error of the positive-sequence phase of the training set, the root mean square error of the negative-sequence magnitude of the training set, the root mean square error of the negative-sequence phase of the training set, the corresponding error of the validation set, the confidence interval of the parameters, the condition number of the information matrix, the maximum absolute correlation coefficient of the parameters, and the consistency of different initial point solutions; the preset conditions are determined by at least one threshold from the above evaluation results.

[0032] The present invention also provides a grid-connected inverter parameter identification device, including a processor, a memory, a data acquisition interface and a communication interface. The memory stores a computer program, and the processor executes the computer program to implement the method of the present invention.

[0033] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method of the present invention.

[0034] The present invention also provides a computer program product, including a computer program, and a method for implementing the present invention when the computer program is executed by a processor.

[0035] Example This embodiment corresponds to the existing PV_VSC simulation model and identification program. Unless otherwise stated, the same symbols in the formulas of this application represent the same or corresponding physical quantities; equivalent transformations, normalization, or the use of expressions with the same physical meaning in the formulas do not affect the technical relationships they represent. The parameter ranges, sweep frequencies, disturbance amplitudes, operating conditions, and algorithm thresholds listed in the embodiments of this application are mainly used to illustrate the implementation process of this technical solution and can be adjusted according to the capacity, control structure, operating conditions, and measurement accuracy of the identified equipment, and do not constitute a limitation on the scope of protection of this application.

[0036] like Figure 1 As shown, this embodiment provides a method for identifying grid-connected inverter parameters based on multi-operating-condition positive and negative sequence impedance frequency sweep, including the following steps: S1. Determine the operating point and the frequency to be measured in this round.

[0037] The existing model directly modifies the two constant blocks PV_VSC / Idr and PV_VSC / Iqr. The per-unit values ​​of Id for the five run points are 0.4, 0.7, 1.0, 0.7, and 0.7, respectively, and the per-unit values ​​of Iq for the last two run points are +0.3 and -0.3, respectively. The last run point is only used for verification and does not participate in parameter solving.

[0038] This embodiment sets the sweep frequency based on the frequency response characteristics of the control loop to be identified. First, 15, 20, and 20 logarithmically spaced candidate frequencies are generated in the 1–20 Hz, 20–200 Hz, and 200–1000 Hz frequency bands, respectively. The generated candidate frequencies are rounded and merged, and duplicate frequencies caused by rounding or overlapping frequency band endpoints are deleted. Considering that the phase-locked loop parameters mainly affect the impedance characteristics near the low frequency and fundamental frequency, this embodiment further adds measurement points at 1 Hz intervals in the 45–49 Hz and 51–55 Hz ranges. 50 Hz is not used as a disturbance frequency to avoid the disturbance component coinciding with the system fundamental frequency. After merging and sorting the frequencies generated by the frequency bands and the encrypted frequencies in the fundamental frequency neighborhood, 57 actual sweep frequencies are obtained in the 1–1000 Hz range. This frequency configuration can cover a wide frequency range and improve the measurement resolution in the low frequency and fundamental frequency neighborhood. The frequency band range, the number of measurement points in each frequency band, and the local encryption interval can all be adjusted according to the rated frequency and control bandwidth of the device under test.

[0039] S2, Perform positive and negative frequency sweep.

[0040] Each simulation incorporates five perturbation frequencies in parallel, extracts the corresponding voltage and current phasors from the FreqScan records, and calculates: ; ; in, It is a positive sequence impedance. and These are the positive-sequence voltage phasor and positive-sequence current phasor at the disturbance frequency, respectively. It is a negative sequence impedance. and These are the negative sequence voltage phasor and negative sequence current phasor at the disturbance frequency, respectively.

[0041] After the forward sequence is completed, the negative sequence is run. The two sequences do not share the same injection result.

[0042] S3. Write the gray box sequence impedance according to the actual control structure.

[0043] The model retains the SRF-PLL, current PI, one-sampling-cycle delay, zero-order hold, sampling filter, and voltage feedforward. Higher-order components not separately expanded in the model are treated as fixed filters to avoid converting all unmodeled dynamics into parameters to be identified.

[0044] S4. Perform finite differences on the logarithmic parameters at the current parameter point to calculate the incremental information brought by each candidate frequency. During frequency selection, simultaneously check the gain in the weakest direction, parameter correlation, and frequency band distribution; intervals where significant model deviations have been found will not be included in this round of optimization.

[0045] S5. Run bounded nonlinear least squares with ten initial points. The program saves the iteration trajectory, exit flag, and final residual for each starting point. The result with the minimum objective value and parameters not exceeding the bounds is taken as the solution for this round.

[0046] S6. Substitute the parameters from this round into the C5 operating condition (which was not fitted) and calculate the positive and negative sequence amplitudes and phase errors respectively. Simultaneously check the condition number of the information matrix, the confidence interval, and the dispersion of solutions at different starting points. If any check indicates insufficient data, return to step S2 to supplement the frequency measurement.

[0047] For the acquisition and quality control of positive and negative sequence impedance data under multiple operating conditions, such as Figure 2 The image shows the current frequency sweep script. The simulation sampling frequency is 20 kHz, meaning the sampling time is 1 / 20000 s; the default injection amplitude for a single frequency is 15 V. Frequency domain readings start at 1 s, and the analysis window covers 50 fundamental frequency cycles. Injecting 5 frequency points each time is to shorten the runtime; if intermodulation occurs between two frequencies or the current response falls below the extraction lower limit, the script restarts with fewer frequency points.

[0048] Linearity checks are performed using three perturbations: 7.5 V, 15 V, and 22.5 V, with impedance comparisons at 10 Hz, 97 Hz, 109 Hz, and 1000 Hz. Using the 15 V result as a reference, any frequency with a relative deviation exceeding 5% in any sequence is not directly used for identification. This is to prevent altering the original operating point while improving the signal-to-noise ratio. The current gray-box model uses the following parameter order: ; in, For phase-locked loop proportional parameters, These are the integral parameters of the phase-locked loop. This is the proportional parameter for the current loop. These are the integral parameters of the current loop. For the corresponding sampling or switching frequency, For the grid-connected side equivalent inductance, For voltage feedforward coefficients, It is a DC voltage.

[0049] Table 1. Operating Condition Classification of This Embodiment

[0050] Currently, the program does not force all eight parameters to be solved simultaneously. f s First, estimate from the switching spectrum. L f The initial value is given by the high-frequency impedance slope. The measured value was 700 V; subsequently, only four PI parameters were fitted. If these three quantities cannot be obtained independently, then the parameter vector should be expanded.

[0051] Before starting the frequency sweep, set the positive and negative sequence disturbance amplitudes to zero and check whether the d-axis and q-axis currents under the five operating conditions have stabilized. C1 to C3 only change the d-axis current setpoint, corresponding to 0.4 pu, 0.7 pu, and 1.0 pu respectively; the d-axis setpoint of C4 and C5 is 0.7 pu, and the q-axis setpoints are 0.3 pu and -0.3 pu respectively.

[0052] like Figure 8 The current waveforms shown indicate a short transition during startup, after which each operating condition stabilizes near its given value. Statistical analysis of two fundamental cycles within the range of 0.16–0.20 s reveals a maximum deviation of 0.0069 A between the average d-axis and q-axis currents and the given values, approximately 0.033% of the rated current. After a moving average of the current over one fundamental cycle, all five operating conditions enter the 2% error band of the rated current within approximately 21.5 ms. Therefore, subsequent frequency sweeps begin after current stabilization to avoid impacting impedance calculations during startup.

[0053] make s = j 2π f , , , T s =1 / f s According to the order domain definition used in the model, the two frequency shift channels of the phase-locked loop are written as: ; ; in, ω 0 The fundamental angular frequency of the power grid. T s The sampling period is f For the perturbation frequency, and They are respectively in frequency shift variables and The open-loop transfer function of the phase-locked loop is calculated below. Let be the closed-loop coupling coefficient of the phase-locked loop corresponding to the q-th sequential channel. The value represents the steady-state fundamental positive-sequence voltage amplitude in the adopted sequence domain coordinate system. The channel number is the sequence number.

[0054] The program combines a one-sampling-period delay, zero-order hold, and first-order sampling filtering into one... Gv ( s ), Gi ( s In this context, the current syntax is: ; in, Gv ( s Let be the equivalent sampling delay, zero-order hold, and first-order filter transfer function of the voltage channel. Gi ( s ) is the corresponding equivalent transfer function of the current path. This is the first-order filter cutoff angular frequency. Choose a variable for the channel. For voltage channels, Current path.

[0055] Early models only used Id in the PLL coupling term, resulting in significant phase errors in the two non-zero Iq conditions, C4 and C5. The current version uses... , The corresponding positive and negative sequence impedances are: ; ; ; ; ; in, A and B These are the numerator and denominator terms of the positive-sequence impedance model, respectively. C and D These are the numerator and denominator terms of the negative-order impedance model, respectively. This is the feedforward decoupling gain coefficient. and These are the equivalent transfer functions for the current path and the voltage path, respectively. and Phase-locked loop coupling coefficients for positive-sequence and negative-sequence frequency shift channels, respectively. and These are the complex operating point currents used in the positive-sequence and negative-sequence impedance models, respectively.

[0056] The above formulas are consistent with the coordinate transformation, disturbance polarity, and sequence impedance signs in this embodiment. When porting to other inverters, the positive current direction and q-axis definition should be checked first; parameter values ​​cannot be replaced alone. For each measurement record, retain the operating condition c, sequence q, and frequency k. Measured values ​​are denoted as Zc,q,k, and model values ​​are denoted as Ẑc,q,k(θ). The current program uses the following formula to construct the amplitude-phase residual: ; ; in, To normalize the impedance magnitude residual, The impedance calculated for the gray box model. The impedance obtained by frequency sweep. This indicates that the phase difference is reflected back to... , To normalize the phase residual, and These are the scaling factors for the amplitude residual and the phase residual, respectively. For operating condition numbering, Number the frequency points. For the data quality weights corresponding to the impedance points, These are the prior constraint coefficients. For parameter priors or regularization terms.

[0057] Table 2 Optional levels of parameter configuration

[0058] Current =1dB =15°, phase error used The principal phase angle is used to avoid artificial jumps when the curve crosses ±180°. In this round of identification, a uniform value of 1 is assigned; if there are significant differences in phasor quality at the site, weights are then allocated according to the signal-to-noise ratio. The program also retains the complex relative residual form: ; in, The complex relative impedance residual, To prevent the measured impedance amplitude from being too small, resulting in division by zero or amplification of the positive lower limit, The objective function is constructed using complex relative residuals. This represents the data quality weight for the corresponding impedance point.

[0059] When no prior constraints are used, let The frequency selection score only determines whether a data point enters the next round of frequency sweeping and is not multiplied into the least squares weights. This avoids the same sensitivity preference being calculated repeatedly in the frequency selection and fitting stages.

[0060] The retained impedance data are numbered in the order of "Operating Condition - Sequence - Frequency". Each impedance point generates an amplitude residual and a phase residual, and all residuals are arranged as follows: ; in, The magnitude residual at the Nth effective impedance point, The phase residual at the Nth effective impedance point, T For transpose, N It is the number of effective impedance points, therefore r The dimension is 2N.

[0061] This method does not perform polynomial fitting on the impedance curve, but rather recalculates the positive and negative sequence impedances of the gray box model after each parameter update. The parameters to be identified enter the numerator, denominator, and coupling terms of the model through a PLL, current loop, and feedforward circuit. There is no simple linear relationship between the predicted impedance and the parameters. Therefore, the parameter estimation is written as: ; ; in, This represents the parameter that minimizes the objective function. The objective function is a nonlinear least squares function. This is a diagonal weighting matrix. When the measurement quality at a certain impedance point is good, its weight can be appropriately increased; if spectral leakage, excessively small response amplitude, or insufficient linearity occurs, the weight should be decreased or the point should be deleted. If the amplitude residual and phase residual have already been multiplied by the square root of the weights during construction, W Simply use the identity matrix to avoid redundant weighting.

[0062] Parameters such as the PI parameter and voltage feedforward coefficient have well-defined physical ranges. If parameters exceeding these ranges are directly truncated to the boundary during iteration, the objective function will exhibit discontinuous changes near the boundary, and the finite difference results will easily become distorted. Therefore, the program uses internal variables... Indicates the first i There are several parameters to be optimized, and the actual parameters are obtained using the following formula: ; In this way, regardless How it changes, Neither will exceed the set upper or lower bounds. Subsequent finite difference and LM iterations are performed on the internal variable z, and the predicted impedance is then converted to the actual parameters using the above formula.

[0063] If a nonlinear least squares solver that allows direct setting of parameter upper and lower bounds is selected, the solution can also be obtained directly. The above variable mapping will no longer be performed.

[0064] In the t In the next iteration, the current internal variable is denoted as After applying a small increment, the change in the residual is approximately as follows: ; in, Let be the internal variable vector of the t-th iteration. For the increment of the internal variable, In order to be in The residual Jacobian matrix is ​​calculated at point .

[0065] The l-th row and the l-th column of the Jacobian matrix i The column represents the l-th residual with respect to the l-th... i Local rate of change of each internal variable: ; The current program uses central finite difference. When calculating the i-th column, only the following parameters are changed. Keeping all other variables constant, calculate the total impedance residuals in both the positive and negative directions: ; in, Let be the finite difference step size corresponding to the i-th internal variable. For the first i One unit direction. The difference step size is adjusted according to the magnitude of the internal variables: ; in, The relative difference step size coefficient. The value of needs to balance the accuracy of model calculations and numerical rounding errors. If the step size is too large, the finite difference cannot accurately reflect the local changes at the current position; if the step size is too small, the difference between the two impedance calculation results may be covered by floating-point errors.

[0066] The phase residual exhibits periodic jumps around ±180°. The program first reverses the difference between the model phase and the measured phase to... Then, perform finite difference analysis on the interval. Otherwise, even a small parameter change could produce an apparent phase difference of nearly 360°, leading to an abnormally large increase in the corresponding Jacobian element.

[0067] After obtaining the current residuals and Jacobian matrix, the Levenberg-Marquardt algorithm is used to calculate the update amounts of the internal variables: ; in, Damping coefficient. Damping matrix. You can use the identity matrix or the diagonal part of the Jacobian information matrix: ; When using a diagonal damping matrix, the update scale for different parameter directions can vary with their local sensitivity, which can reduce the impact of the order-of-magnitude difference in PI parameters on the iteration.

[0068] get Then, a set of trial variables is formed: ; The program will Convert to actual parameters and recalculate the predicted impedance and objective function for each operating condition. If the objective function at the test point is less than the current value, accept this step. ; The damping coefficient is then reduced to gradually approximate a Gaussian-Newton update in the next step. If the objective function does not decrease, no parameters are written for this trial; the program increases the damping coefficient and recalculates the update amount of the internal variables. This approach allows the algorithm to use a more conservative step size when far from the optimum, while accelerating convergence after entering the stable decreasing region.

[0069] The damping coefficient can also be adjusted based on the ratio of the actual descent to the descent predicted by the local model: ; in, The objective function value at the current iteration point. The actual objective function value was recalculated for the trial point. This represents the predicted value of the objective function at the test point by the local quadratic model. This represents the ratio of the actual decrease to the predicted decrease.

[0070] when A larger value indicates that the Jacobian matrix accurately predicts changes in the current objective function, and can reduce [the impact of these changes]. ;when When it is small or negative, it increases. The specific threshold and adjustment factor are set by the solver program and are not limited to a fixed value.

[0071] This iteration ends when any of the following conditions are met. The first condition is that the relative change in the objective function is already very small: ; in, Let be the objective function value after the (t+1)th iteration following the update. This is the convergence threshold for the relative change of the objective function.

[0072] The second scenario is when the update amount of the internal variable is already less than the set value: ; in, This is the convergence threshold for the relative update amount of the internal variables.

[0073] The maximum number of iterations is used as a safe stopping condition. After the program stops, it records the parameter values, the objective function, the number of iterations, and the reason for stopping, instead of just saving a set of parameters.

[0074] After all initial points have been calculated, first exclude results that have not converged properly or have remained near the parameter boundaries for a long time, and then compare the objective functions of the remaining results. If the parameters obtained from different initial points are similar, the set with the smaller objective function can be used for subsequent verification; if the objective functions are similar but the parameters differ greatly, it indicates that the existing data cannot distinguish some parameters, and further judgment is needed by combining parameter correlation, verification operating condition errors, and supplementary frequency sweep results.

[0075] When solving for the five parameters, the parameter order is [Kp, pll, Ki, pll, Kp, i, Ki, i, Kf]. The current lower bound is [1e-4, 1e-3, 1e-4, 0.05, 1e-5], and the upper bound is [5, 500, 2, 1000, 0.02]. The initial value is fixed at 1.1 times the lower bound, and the remaining initial values ​​are generated in logarithmic space. A total of 10 sets of initial values ​​are compared in each round. When using the Levenberg-Marquardt algorithm, the program uses continuous mappings to restrict the internal variables within the above physical boundaries.

[0076] After the solution is completed, not only the minimum objective value is retained, but also the results of ten initial iterations, the parameter iteration trajectory, the Jacobian matrix, and the predicted impedance for the C5 verification case are saved. The confidence interval is... The generalized inverse approximation is obtained, therefore, when the information matrix is ​​ill-conditioned, it is only used for diagnosis and not as a precise statistical interval. Whether to continue testing is determined by the validation error, condition number, maximum correlation coefficient, and consistency of multiple starting points.

[0077] This embodiment uses the PV_VSC three-phase photovoltaic grid-connected model. C1 to C4 are entered into the objective function, while C5 is only used for verification. fs and Lf are initially given by an independent script, Vdc is fixed at 700 V, and finally, the two PI parameters of the phase loop, the two PI parameters of the current loop, and Kf are calculated. The non-zero Iq condition is entered into the PLL coupling term as a complex operating point current.

[0078] To examine the adaptability of the identification results to the frequency distribution, this round of identification directly used the 57 measured frequencies obtained using the above method, without interpolating the impedance curves. The low-frequency band and the fundamental frequency neighborhood have high frequency resolution, while the mid-to-high frequency band is mainly used to characterize the impact of the current loop, digital delay, and filtering stages on the impedance. Each operating condition includes both positive-sequence and negative-sequence impedances, thus generating a total of 570 sets of complex impedance data across the five operating conditions. This round of calculations uses a fixed sampling frequency. Equivalent inductance on grid-connected side and DC voltage The parameters to be determined are still the two PI parameters of the phase-locked loop, the two PI parameters of the current loop, and the voltage feedforward coefficient.

[0079] like Figure 5 As shown, the residual changes corresponding to ten sets of initial parameters are presented. The starting positions of the ten curves differ significantly, with the fastest decreases occurring in the first few iterations. The initial values ​​of sets 1, 2, 4, 5, and 8 eventually reach the same set of parameters; the remaining initial values ​​stop at higher residual levels, with some parameters simultaneously approaching the search boundary. This indicates that the identification problem has a local solution. In this round, the result corresponding to the first set of initial values ​​is selected as the parameter for subsequent analysis.

[0080] like Figure 6 As shown, the dots represent impedance data obtained from frequency sweep, and the continuous curves represent the calculation results after substituting the identified parameters into the gray box model. Different colors correspond to different operating conditions. The positive and negative sequence amplitude curves are relatively close to the measurement points in the mid-to-high frequency range; the dispersion is greater in the low frequency range and near the fundamental frequency, and the difference in positive sequence phase is also more obvious in this region. The root mean square errors of the positive and negative sequence amplitudes of the training data are 1.6328 dB and 0.6830 dB, respectively, and the root mean square errors of the phases are 6.6445° and 7.0614°, respectively. Under the C5 operating condition without parameter solving, the positive and negative sequence amplitude errors are 1.6277 dB and 0.9691 dB, respectively, and the phase errors are 9.3392° and 12.4266°, respectively. The errors of the training and validation operating conditions are within a similar range, indicating that the current parameters do not only fit the data involved in optimization; at the same time, the low-frequency phase is still a key part that needs to be checked in subsequent model corrections.

[0081] like Figure 7 As shown, the final results for the five parameters are listed: , , , , Compared to the set values ​​in the simulation model, the relative errors of the five parameters are +1.91%, -7.62%, -4.88%, -0.86%, and -3.11%, respectively, with an overall root mean square relative error of approximately 4.38%. The current loop integral coefficient is closest to the set value, while the phase-locked loop integral coefficient shows a relatively larger deviation. This difference is related to... Figure 6 The corresponding phase residuals in the low-to-mid frequency range and near the fundamental frequency also indicate that the problem of uneven parameter sensitivity has not been completely eliminated even after increasing the number of frequency points.

[0082] The controller parameters in the simulation model are known, so the identification results can be compared with the model settings. The identification values ​​and relative errors of the five parameters are shown in Table 3.

[0083] Table 3. Examples of phased five-parameter identification results

[0084] After identifying parameters at 57 frequency points, the five obtained parameters were rewritten into the inverter control model and run under identical conditions. The control group still used the original control parameters of the model. During the comparison process, positive-sequence and negative-sequence frequency sweep injection were turned off, the sampling frequency was kept at 20 kHz, and the main circuit parameters were not adjusted. The resulting current difference mainly comes from the change in control parameters.

[0085] The time-domain comparison uses two operating conditions, C4 and C5. Both have a d-axis current setpoint of 15.001 A and q-axis current setpoints of 6.429 A and -6.429 A, respectively. Using these two operating conditions allows for comparison of the responses to positive and negative reactive currents while keeping the d-axis current constant. C5 is a pre-reserved verification operating condition and was not used in parameter optimization.

[0086] like Figure 9 As shown, the d-axis and q-axis currents corresponding to the original control parameters and identification parameters are presented. The two sets of waveforms are quite similar in terms of startup trend, convergence process, and steady-state position. Taking 0–0.05 s as the startup interval, the root mean square differences of the d-axis and q-axis waveforms in the C4 condition are 0.361 A and 0.152 A, respectively; in the C5 condition, they are 0.255 A and 0.140 A, respectively, with the maximum value accounting for approximately 1.69% of the rated current.

[0087] We then take 0.16–0.20 s as the steady-state interval. The root mean square differences of the d-axis and q-axis waveforms for condition C4 are 0.226 A and 0.112 A, respectively, while for condition C5 they are 0.200 A and 0.124 A, respectively. The maximum difference in the mean steady-state current obtained from the two sets of parameters is 0.0033 A, approximately 0.015% of the rated current. This indicates that although the 57-point identification results are not exactly the same as the model settings, they can still maintain a closed-loop current response close to the original controller after substitution. Similar results were obtained for condition C5, indicating that this set of parameters still has some applicability to negative reactive current conditions that are not involved in optimization.

[0088] like Figure 4 As shown, the device consists of a processor, a memory, and a data acquisition and communication interface. The acquisition interface receives the voltage and current complex phasors and operating point information obtained from the frequency sweep; the memory stores the model, parameter boundaries, and historical frequency sweep data; the processor performs frequency point selection and nonlinear least squares, and outputs the current round parameters, verification error, correlation coefficient, and suggested retest frequencies.

[0089] This embodiment is initially used as an offline tool: after the frequency sweep is completed, the data file is read and the impedance model is updated. If used for online monitoring, the disturbance amplitude, injection period, and allowable frequency should be separately limited to avoid affecting the grid-connected current quality.

[0090] When the equipment has an external impedance scanning device, disturbances can be applied from the grid connection point voltage side; if only a controller simulation model is available, disturbances can also be injected from the current reference side. The two methods cannot directly share the same impedance symbol; they must be rechecked according to the positive directions of voltage and current.

[0091] If the positive and negative orders are not diagonally coupled, the problem cannot be ignored. Z p , Z n The two scalars should be expanded into a 2×2 ordered impedance matrix, and the objective function contains both diagonal and off-diagonal terms; at this point, additional perturbation combinations are needed to separate the matrix elements.

[0092] The number of parameters is determined based on the data conditions. When the equipment parameters are relatively complete, only four PI parameters are identified; when the feedforward coefficients are unknown, a five-parameter form is used; only when both the information matrix and the verification conditions indicate sufficient data are the parameters identified. f s , L f and V dc They were all included in the eight-parameter optimization.

[0093] The above values ​​correspond to this embodiment. The parameter boundaries, frequency sweep range, and model order can be adjusted according to the inverter under test. As long as multi-operating-point positive and negative sequence data are still used, and supplementary measurements are arranged according to the identifiability check, this method can be implemented.

[0094] Therefore, this invention employs the aforementioned method, equipment, storage medium, and product for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweeping. During measurement, the d-axis and q-axis current setpoints are first changed to obtain several operating points. Small positive and negative sequence disturbances are injected at each operating point, and the sequence impedance is calculated from the grid-connected point voltage and current phasors. The calculation model retains components such as the phase-locked loop, current PI converter, digital delay, sampling filtering, and voltage feedforward. When the sampling frequency, filter inductance, or DC voltage can be measured independently, they are not included in the joint optimization; other parameters are solved through multi-condition impedance residuals. The frequency sweep data is divided into fitting data and verification data according to the operating points. Whether a frequency point needs to be remeasured is determined by the Jacobian information matrix, parameter correlation, and verification error. This invention does not require reading the controller's internal settings and can provide the necessary parameters for impedance model verification and stability analysis.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A grid-connected inverter parameter identification method based on multi-working condition positive and negative sequence impedance sweep, characterized in that, Includes the following steps: S1. Set at least two grid-connected inverter operating conditions and construct a candidate frequency set; The data corresponding to different operating conditions are divided into training data and validation data. S2. Apply positive-sequence and negative-sequence perturbations to a subset of frequencies within the candidate frequency set to obtain the grid-connected point voltage phasor and grid-connected current phasor under each operating condition, and obtain the positive-sequence impedance and negative-sequence impedance accordingly. S3. Construct a gray box sequence impedance model. The gray box sequence impedance model takes the parameters to be identified and the operating condition variables as inputs and the predicted positive sequence impedance and the predicted negative sequence impedance as outputs. S4. Based on the deviation between the actual sequencing impedance and the predicted sequence impedance, construct an objective function that spans operating conditions and sequence types; S5. Calculate the sensitivity information of the parameter to be identified in the objective function, and select the identification frequency from the candidate frequency set under at least one of the constraints of frequency band coverage, frequency point spacing and measurement quality. S6. Within the parameter boundary, the objective function is optimized using multiple initial points to obtain the identification results of the parameters to be identified. S7. Evaluate the identification results based on the verification data and identifiability indicators, and select a supplementary frequency for iterative identification when the preset conditions are not met.

2. The method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweep according to claim 1, characterized in that: The operating conditions in S1 are characterized by a single parameter or any combination of the following: d-axis current setpoint, q-axis current setpoint, active power setpoint, reactive power setpoint, DC side voltage, grid voltage, and short-circuit ratio. The at least two grid-connected inverter operating conditions are specifically: using two different active power levels, or configuring at least one non-zero reactive power operating condition, or setting two types of operating conditions simultaneously.

3. The method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweep according to claim 1, characterized in that: Before or after obtaining the positive and negative sequence impedances in S2, a measurement quality check is performed to eliminate or reduce the weight of frequencies that fail the measurement quality check. Measurement quality checks include at least one of the following: steady-state check, disturbance current lower limit check, signal-to-noise ratio check, disturbance amplitude linearity check, spectral leakage check, abnormal phase jump check, and fundamental frequency neighborhood model consistency check.

4. The method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweep according to claim 1, characterized in that: The gray box sequence impedance model in S3 includes the closed-loop transfer function of the phase-locked loop under positive and negative sequence frequency shift variables, the digital control delay and zero-order hold circuit, the voltage sampling filter circuit and the current sampling filter circuit, and the complex coupling terms of the d-axis operating point current and the q-axis operating point current in the positive and negative sequence domains.

5. The method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweep according to claim 1, characterized in that: The objective function in S4 includes the weighted sum of squares of the magnitude residual of the sequence impedance and the phase residual of the sequence impedance, or the weighted sum of squares of the real and imaginary parts of the complex relative residual of the sequence impedance.

6. The method for identifying grid-connected inverter parameters based on multi-condition positive and negative sequence impedance frequency sweep according to claim 1, characterized in that, The specific steps of S5 are as follows: S51. Calculate the sensitivity information of the objective function for the parameters to be identified. The parameters to be identified include at least four of the following: phase-locked loop proportional coefficient, phase-locked loop integral coefficient, current regulation loop proportional coefficient, current regulation loop integral coefficient, feedforward coefficient, switching frequency, sampling frequency, grid-connected equivalent inductance, and DC-side voltage. The sensitivity information includes at least one of the following: normalized parameter Jacobian matrix, information matrix composed of normalized parameter Jacobian matrix, minimum eigenvalue of information matrix, condition number, determinant, covariance matrix, and parameter correlation coefficient. S52. Calculate the selection score of the candidate frequency. The selection score is formed by a combination of at least two of the following: weak direction sensitivity gain, information gain, target parameter accuracy gain, condition number improvement, and parameter correlation reduction. S53. Divide the candidate frequencies into multiple frequency bands and set an upper or lower limit for the number of frequency points in each frequency band; S54. Exclude frequency ranges with frequency shift singularities, perturbation nonlinearities, or model biases; S55. Constrain the distance between the newly selected frequency and the measured frequency to be no less than the preset absolute distance or the preset relative distance. S56. Combining sensitivity information, candidate frequency selection scores, and constraints, the identification frequency is obtained by filtering from the candidate frequency set.

7. The method for identifying grid-connected inverter parameters based on multi-operating-condition positive and negative sequence impedance frequency sweep according to claim 1, characterized in that, The specific steps of S6 are as follows: S61. Configure multiple initial points, including preset initial points and random initial points generated in the logarithmic parameter space; S62. Optimize the objective function using any one or more combinations of trust region algorithm, Levenberg-Marquardt algorithm, sequential quadratic programming, and swarm intelligence algorithm; S63. For unconstrained optimization algorithms, internal optimization variables are mapped to the upper and lower bounds of physical parameters through continuous mapping.

8. A grid-connected inverter parameter identification device, comprising a processor, a memory, a data acquisition interface, and a communication interface, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the grid-connected inverter parameter identification method based on multi-condition positive and negative sequence impedance frequency sweep as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the grid-connected inverter parameter identification method based on multi-condition positive and negative sequence impedance frequency sweep according to any one of claims 1-7.

10. A computer program product, characterized in that: The method includes a computer program that, when executed by a processor, implements the grid-connected inverter parameter identification method based on multi-condition positive and negative sequence impedance frequency sweep as described in any one of claims 1-7.

Images