A method for coordinated frequency and resonance control of a distributed inverter system

Abstract

This invention relates to the field of inverter control technology, specifically disclosing a method for frequency and resonance coordinated control of a distributed inverter system. The method includes: collecting inverter operating data; evaluating sampling quality based on the operating data; and calculating the operating point variation, resonance peak amplitude, damping ratio, resonance peak frequency, and their confidence level; determining whether wideband impedance identification is triggered based on the sampling quality, operating point variation, and confidence level; if wideband impedance identification is triggered, recalculating the resonance peak frequency, resonance peak amplitude, and damping ratio; calculating the virtual resistor damping voltage; performing centerless consistent inertia fusion and then calculating the angular frequency reference value; and generating control commands for the inverter. This invention predicts and outputs confidence levels online based on a resonance frequency prediction model, triggering wideband impedance sweep frequency measurement as needed, significantly reducing the impact of unnecessary disturbance injection on grid-connected power quality, and simultaneously significantly reducing the average computational load of edge computing units.

Description

Technical Field

[0001] This invention relates to the field of inverter control technology, and in particular to a method for coordinated frequency and resonance control of a distributed inverter system. Background Technology

[0002] With the continuous increase in the penetration rate of new energy power generation, distributed inverters, represented by photovoltaics and energy storage, are being connected to the grid on a large scale. In weak grid scenarios, the equivalent impedance of the grid is relatively large, and resonance can easily occur between the inverter and the grid impedance, leading to grid current distortion, power quality degradation, and even equipment damage. At the same time, the coordinated control of distributed inverter clusters faces challenges such as limited communication bandwidth, high risk of single-point failures, and coupling conflicts among multiple control objectives. How to achieve efficient coordination of multiple inverters while ensuring grid-connected power quality is a key problem that urgently needs to be solved in the field of grid-connected power generation control.

[0003] In impedance identification, existing technologies typically employ a fixed-period injection of wideband disturbance signals followed by full-band discrete Fourier transform analysis to obtain the wideband impedance characteristics of the inverter's grid connection point. However, this method performs a full-band frequency sweep at fixed intervals regardless of system status, continuously injecting disturbances into the grid and affecting grid-connected power quality, while also placing a sustained high computational load on edge computing units. Regarding resonance suppression, traditional virtual resistor schemes apply uniform damping to the output current across the entire frequency band, causing unnecessary fundamental power loss in non-resonant frequency bands, and the damping effect is dispersed, making it difficult to concentrate on the resonant frequency band. In terms of multi-machine collaboration, if each inverter independently adjusts its frequency, circulating currents or oscillations may occur due to asynchronous adjustments; while centralized scheduling at a central node carries the risk of single-point failure and requires a high-bandwidth communication network to transmit full waveform data, making it difficult to deploy in industrial settings. Furthermore, frequency support control and impedance reshaping control are implemented through the same inverter control link. When it is necessary to significantly increase the virtual resistance to suppress severe resonance, the voltage regulation margin will be squeezed. Conversely, when the frequency deviates significantly, strong regulation may cause the modulation depth to approach saturation. There is a coupling conflict between the two, and the existing technology lacks an effective collaborative arbitration mechanism. Summary of the Invention

[0004] The present invention aims to solve the above-mentioned problems. To this end, the present invention provides a method for coordinated frequency and resonance control of a distributed inverter system.

[0005] This invention provides a method for coordinated frequency and resonance control of a distributed inverter system, the technical solution of which includes: S1: Collect the inverter's operating data, evaluate the sampling quality based on the operating data, and calculate the operating point change, resonant peak amplitude, damping ratio, resonant peak frequency, and their confidence level; wherein, the damping ratio, resonant peak amplitude, resonant peak frequency, and their confidence level are calculated by the resonant frequency prediction model based on the operating data; S2: Based on the sampling quality, the change in the running point, and the confidence level, determine whether wideband impedance identification is triggered. If wideband impedance identification is triggered, execute S3; otherwise, execute S4. S3: By using the edge computing unit to superimpose the disturbance signal on the voltage control loop or modulation wave generation stage of the inverter, the resonant peak frequency, resonant peak amplitude and damping ratio are recalculated. S4: Calculate the virtual resistance damping voltage based on the resonance peak amplitude, damping ratio, and resonance peak frequency; S5: Using the running data, perform centerless consistent inertia fusion, and then calculate the angular frequency reference value; S6: Generate control commands for the inverter based on the angular frequency reference value and the virtual resistor damping voltage.

[0006] Furthermore, the process of collecting the inverter's operating data is as follows: The edge computing unit located at the inverter collects the three-phase voltage, three-phase current and DC side voltage at the grid connection point, and calculates the voltage and current in the dq coordinate system, the angular frequency at the grid connection point, the active power and the reactive power.

[0007] Furthermore, the process of evaluating sampling quality based on operational data is as follows: When all four criteria are met, =1, otherwise =0, For sampling quality identification; The four criteria for judgment include: (1) All three-phase voltage and current and DC side voltage were successfully sampled; (2) The time difference between adjacent sampling times and the deviation of the sampling period do not exceed the time deviation threshold; (3) The phase-locked loop is effectively locked and the angular frequency deviation of the grid connection point in adjacent cycles does not exceed the angular frequency jump threshold; (4) The DC side voltage is within the rated range.

[0008] Furthermore, the process of calculating the change in the running point is as follows: Based on active power, reactive power, DC side voltage, and grid connection point angular frequency, a characteristic vector of the operating point is constructed. The weighted Euclidean distance between the current characteristic vector of the operating point and the previous characteristic vector of the operating point is calculated to obtain the change in the operating point.

[0009] Furthermore, the process of calculating the resonant peak frequency and its confidence level is as follows: The resonant frequency prediction model adopts an impedance model in the form of a second-order discrete transfer function, with the d-axis current as the response observation. The resonant peak frequency is obtained by mapping the transfer function to the s-domain through a bilinear transformation and solving for the frequency corresponding to the maximum point of the amplitude-frequency characteristic within the preset frequency range. The formula for calculating the confidence level of the resonant peak frequency is: in, For confidence level, To obtain the maximum value, The model residuals of the resonant frequency prediction model. This represents the upper limit of the normalized residual.

[0010] Furthermore, wideband impedance identification is triggered when the sampling quality indicator is 0, or the change in the running point is greater than the change in the running point threshold, or the confidence level is less than the confidence threshold.

[0011] Furthermore, when broadband impedance identification is not triggered, the virtual resistance damping voltage is calculated using the resonant peak amplitude, damping ratio, and resonant peak frequency obtained from S1. When triggering wideband impedance identification, the virtual resistance damping voltage is calculated using the resonant peak amplitude, damping ratio, and resonant peak frequency obtained from S3 recalculation.

[0012] Furthermore, the presence of a resonance event is determined based on the resonance peak amplitude and damping ratio; When a resonance event is determined to exist. in, This is the virtual resistance after the transition. To restrict the numerical range of functions, This is the resistance conversion factor. The amplitude of the resonance peak. For the damping ratio, To minimize damping protection value and prevent division by zero, This represents the minimum virtual resistance. This represents the maximum virtual resistance.

[0013] When it is determined that no resonance event exists , As a reference virtual resistor; The formula for calculating the virtual resistance damping voltage is: in, The virtual resistance damping voltage along the d-axis. For d-axis current, This represents the d-axis current component near the resonant peak frequency.

[0014] Furthermore, the process of using operational data to perform centerless consistent inertia fusion and then calculating the angular frequency reference value is as follows: Calculate the local inertia adjustment using operational data; The local inertia regulation is fused multiple times using the inertia regulation of adjacent inverters to achieve centerless consistent inertia fusion and obtain the fused inertia regulation. The angular frequency reference value is calculated based on the fused inertia adjustment.

[0015] Furthermore, the process of generating control commands for the inverter based on the angular frequency reference value and the virtual resistor damping voltage is as follows: The control quantity is synthesized from the angular frequency reference value and the virtual resistor damping voltage. The control quantity is limited; the inverter control command is generated based on the angular frequency reference value and the limited control quantity.

[0016] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: 1. This invention maintains a computationally insignificant resonant frequency prediction model locally on each inverter, replacing frequent wideband impedance sweep measurements with online model prediction to continuously track the changing trends of system impedance characteristics. A confidence level mechanism is introduced to quantitatively evaluate the reliability of the model predictions; costly wideband impedance identification is triggered only when the confidence level is too low, the operating point changes significantly, or sampling is abnormal. This on-demand triggering mode significantly reduces the impact of unnecessary disturbance injections on grid-connected power quality, while also significantly reducing the average computational load on edge computing units.

[0017] 2. This invention improves the virtual resistance scheme to a discrete jump-change mode: the virtual resistance jumps from a reference value to a suppression resistance value calculated based on the severity of the resonance only when a resonance event that needs to be suppressed is determined to exist; in non-resonant frequency bands and when no resonance occurs, only a very small reference virtual resistance is maintained. Simultaneously, the current component in the resonant frequency band is extracted through bandpass filtering, and the incremental resistance only acts on the resonant frequency band, remaining unaffected in the non-resonant frequency band. This invention minimizes power loss when no resonance occurs, and the damping energy is concentrated in the resonant frequency band, resulting in a more precise suppression effect.

[0018] 3. This invention employs a decentralized consensus algorithm for multi-machine inertia fusion. Each edge computing unit only exchanges a highly concise state summary (inertia adjustment amount, confidence level, and timestamp) with its neighboring units. A few iterations are sufficient to bring the inertia adjustment amounts of each unit towards consistency. The decentralized architecture of this invention eliminates the risk of system-wide failure due to a central node failure; the amount of communication data per cycle is only about tens of bytes, which can be implemented on low-bandwidth industrial buses such as CAN and RS485, resulting in low deployment costs. Simultaneously, the confidence level is dynamically adjusted by incorporating adjacency weights. When communication is interrupted or the confidence level of a neighboring unit is too low, the weights are automatically adjusted, further enhancing system robustness.

[0019] 4. This invention combines the voltage regulation value generated by converting the angular frequency reference value with the virtual resistor damping voltage into a unified control quantity, effectively avoiding the problem of resonance suppression squeezing the voltage regulation margin or strong frequency regulation causing modulation depth saturation, ensuring that the inverter operates within the safety boundary under various operating conditions.

[0020] 5. The confidence level of the resonant frequency prediction model of this invention simultaneously drives the adjacency weight adjustment of the gating decision and the centerless consensus algorithm, organically coupling the various technical aspects: high confidence levels avoid triggering precise identification, reducing disturbances and computational overhead; adjacent units with low confidence levels automatically have their weights reduced during consensus fusion, preventing unreliable data from contaminating the global coordination results. This closed-loop architecture achieves synergistic efficiency among the control aspects, with overall performance superior to the simple superposition of the various aspects.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the method provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but should not be used to limit the scope of this invention.

[0025] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0026] The following is combined with Figure 1 This invention provides a more detailed description of a distributed inverter system frequency and resonance coordinated control method, applicable to a grid-connected power generation system comprising multiple parallel distributed inverters and an equivalent impedance to a weak grid. Each inverter is configured with an edge computing unit, and the edge computing units are connected to each other via low-bandwidth communication links to form a decentralized communication topology.

[0027] In this embodiment, as Figure 1 As shown, a method for coordinated frequency and resonance control of a distributed inverter system is provided, comprising the following steps: S1: Collect the inverter's operating data, evaluate the sampling quality based on the operating data, and calculate the operating point change, resonant peak amplitude, damping ratio, resonant peak frequency, and their confidence level; wherein, the damping ratio, resonant peak amplitude, resonant peak frequency, and their confidence level are calculated by the resonant frequency prediction model based on the operating data.

[0028] In this embodiment, a computationally inexpensive resonant frequency prediction model is maintained locally on each inverter. This model's online prediction replaces frequent wideband impedance sweep measurements. The model continuously tracks the changing trends of the system's impedance characteristics, initiating precise but costly wideband impedance identification only when the confidence level is too low.

[0029] S1.1: Collect the inverter's operating data, evaluate the sampling quality, and generate a sampling quality identifier.

[0030] In the The sampling period, the first Each inverter edge computing unit collects three-phase voltage, three-phase current, DC-side voltage, and switch status at the grid connection point, all with timestamps and edge computing unit numbers. The voltage and current in dq coordinate system are obtained through Clark / Park transformation; the angular frequency at the grid connection point is obtained through a phase-locked loop; and active and reactive power are calculated.

[0031] Based on the collected data, a sampling quality identifier for this inverter sampling is generated. , The criteria for judgment are as follows: (1) all three-phase voltage and current and DC side voltage are successfully sampled; (2) the time difference between adjacent sampling times and the deviation of the sampling period does not exceed the time deviation threshold (e.g., 0.1 times the sampling period); (3) the phase-locked loop is effectively locked and the angular frequency deviation of the grid connection point between adjacent periods does not exceed the angular frequency jump threshold (e.g., 2π·1Hz); (4) the DC side voltage is within the rated range (85%-115% of the rated value). When all four criteria are met, =1, otherwise The value is 0. By using sampling quality indicators, the system can quickly identify sensor faults, communication packet loss, and power grid anomalies, providing a unified judgment entry point for subsequent gating decisions and anomaly degradation. A value of 0 indicates that the sampling quality of the current running data does not meet the requirements (not meeting all four judgment criteria at the same time).

[0032] S1.2: Calculate the change in the running point based on the running data.

[0033] Based on active power, reactive power, DC side voltage, and grid connection point angular frequency, a characteristic vector of the operating point is constructed. The weighted Euclidean distance between the current operating point characteristic vector and the previous valid operating point characteristic vector is calculated to obtain the change in the operating point.

[0034] In this embodiment, the running point feature vector is defined as follows: , in, For the first The operating point feature vector of an inverter For the first The active power of each inverter For the first The reactive power of each inverter For the first DC side voltage of the inverter For the first The grid connection point angular frequency of the inverter This is the matrix transpose.

[0035] The weighted Euclidean distance between the current running point feature vector and the previous valid running point feature vector is calculated as follows: Change in running point In this embodiment, the weights of active power, reactive power, DC-side voltage, and grid connection point angular frequency are set to 1.0, 1.0, 0.5, and 2.0, respectively; the grid connection point angular frequency is given the highest weight, making the detection more sensitive to frequency-related changes. When the value exceeds 0.05, the operating point is considered to have changed significantly.

[0036] S1.3: The resonant frequency prediction model calculates the resonant peak frequency and its confidence level based on the operating data (d-axis current).

[0037] In this embodiment, the resonant frequency prediction model adopts an impedance model in the form of a second-order discrete transfer function, which characterizes the local impedance characteristics from inverter output voltage disturbance to output current response. The mathematical expression is as follows: in, This is the impedance estimate. These are the second-order molecular coefficients. These are the first-order molecular coefficients. The zeroth-order molecular coefficient, The coefficients of the first-order denominator are... The zeroth order denominator coefficient, The delay operator is a unit.

[0038] definition , The model parameter vector is updated online using a recursive least squares method with a forgetting factor (forgetting factor is 0.98), and the d-axis current (calculated from the current in the dq coordinate system) is used as the response observation.

[0039] Model residuals Defined as the average absolute error between the predicted value (d-axis current estimate) of the resonant frequency prediction model within the sliding window and the actual current measurement. In this embodiment, the window length is 20.

[0040] The confidence level of the resonant peak frequency is: in, For confidence level, To obtain the maximum value, The model residuals of the resonant frequency prediction model. The upper limit of the normalized residual, This is the sampling period index. In this embodiment, the upper limit of the normalized residual is... Take the 95th percentile value of the residuals for the first 500 periods after model initialization. If the initialization data is insufficient... Take 0.05 times the rated output current amplitude of the inverter.

[0041] The confidence level ranges from [0,1], with a higher value indicating a more reliable prediction of the resonant frequency. The confidence level resets to zero when the model residuals exceed the normalization upper limit.

[0042] The resonant peak frequency calculated by the resonant frequency prediction model This is achieved by mapping the transfer function to the s-domain via a bilinear transformation and solving for the frequencies corresponding to the maximum amplitude-frequency response points within a preset frequency range. In this embodiment, the preset frequency range is... .

[0043] Using the d-axis current as the response observation, a standard second-order system is constructed by mapping it to the s-domain through a bilinear transformation based on the second-order discrete impedance transfer function, and then the resonance peak amplitude and damping ratio are analytically solved.

[0044] The resonant frequency prediction model requires only a small number of recursive parameter updates per cycle, with a computational load far less than that of full-band DFT analysis (which requires processing thousands of sampling points). The confidence mechanism enables the system to quantitatively evaluate the reliability of the model's predictions, providing a basis for decision-making regarding whether to initiate precise identification.

[0045] S2: Based on the sampling quality, the change in the running point, and the confidence level, determine whether wideband impedance identification is triggered. If wideband impedance identification is triggered, execute S3; otherwise, execute S4.

[0046] Traditional methods perform a full-band frequency sweep at fixed intervals, injecting wideband disturbances regardless of the system state. This method improves upon this by implementing an on-demand mode, initiating precise full-band identification only when the confidence level is too low, sampling is abnormal, or the operating point changes significantly. This greatly reduces unnecessary disturbance injection and computational overhead.

[0047] Wideband impedance identification is triggered when the sampling quality indicator is 0, or the change in the running point is greater than the change in the running point threshold, or the confidence level is less than the confidence threshold.

[0048] Wideband impedance identification trigger mark Determined by the following logic: in, The confidence threshold. This is the threshold for the change in the running point.

[0049] In this embodiment, The initial value is 0.6, and it is adaptively fine-tuned based on the sampling noise level over nearly 100 cycles. , The standard deviation of the sampling noise is normalized to the rated current. In the first The confidence threshold for each period.

[0050] when (If not triggered) directly Send to S5, and simultaneously force the disturbance injection amplitude for this cycle to be zero.

[0051] when When triggered, enter S3 to perform wideband impedance identification.

[0052] This gating mechanism can reduce the proportion of wideband impedance identification triggering, reduce the impact of disturbances on normal system operation, and at the same time reduce the average computing load of edge computing units.

[0053] S3: When triggering wideband impedance identification, the local edge computing unit of the inverter is used to superimpose the disturbance signal in the voltage control loop or modulation wave generation stage of the inverter as a software variable to recalculate the resonant peak frequency, resonant peak amplitude and damping ratio.

[0054] When the S2 gating mechanism triggers wideband impedance identification, this step generates and superimposes a wideband disturbance signal as a variable within the inverter control software. The principle is to utilize the inverter's own PWM modulation capability to superimpose a small pseudo-random disturbance onto the voltage reference waveform, and then deduce the system impedance characteristics by analyzing the output current response. The disturbance amplitude is strictly controlled within 5% of the rated voltage, and multiple safety constraints are set to ensure that equipment safety is not affected.

[0055] S3.1: Disturbance signal generation and injection.

[0056] The perturbation signal uses a pseudo-random binary sequence (PRBS) with a clock frequency of 5 kHz and a sequence length of [missing information]. for Disturbance amplitude It is 12V (approximately 3.86% of the rated voltage). The disturbance is added to the d-axis voltage reference value as a software variable.

[0057] S3.2: Calculation of broadband impedance spectrum.

[0058] During the injected disturbance, the d-axis voltage disturbance component and d-axis current response component of the inverter are simultaneously acquired (obtained by subtracting the fundamental component under normal operating conditions, which is extracted by moving average filtering). A total of 2047 sampling points are collected. The acquired data undergoes Discrete Fourier Transform (DFT) in the local edge computing unit, and the broadband impedance spectrum is calculated at the angular frequency sampling points. The resonant peak frequency (taking the local maximum point of the amplitude-frequency characteristic), resonant peak amplitude, and damping ratio are extracted from the impedance spectrum. The damping ratio is calculated using the half-power bandwidth method; when the half-power bandwidth boundary cannot be reliably identified, it is marked as uncalcifiable.

[0059] It should be noted that the following conditions should be checked before each disturbance injection, and the injection should be stopped immediately if any of the conditions are met: (1) the output current amplitude reaches 95% of the maximum allowable value, (2) the DC bus voltage exceeds the range of 0.9-1.1 times the rated value, (3) the modulation depth reaches 0.95, (4) the instantaneous value of any phase of the grid current is close to 95% of the hardware protection limit.

[0060] Timing processing during identification: PRBS identification requires continuous acquisition of 2047 sampling points, during which each control cycle continues to execute normally. In the identification process where data acquisition is complete but DFT analysis is not yet finished, ... S5 remains at 1 (to prevent the gating logic from exiting before identification is complete), and continues to use the resonance peak parameters from the previous valid identification cycle. The DFT analysis is performed in the... Each cycle completes and updates the resonance peak parameters in one go. This represents the total number of sampling points. =2047. If the identification process is interrupted due to a disturbance caused by safety constraints, the identification will be invalid. Reset to 0 and reassess the gating conditions in the next cycle.

[0061] S4: Calculate the virtual resistance damping voltage based on the resonance peak amplitude, damping ratio, and resonance peak frequency.

[0062] Traditional virtual resistor schemes apply uniform damping to the current across the entire frequency band, causing unnecessary fundamental power loss in non-resonant frequency bands, and the damping effect is diffuse. This method changes the operation of the virtual resistor to a discrete mode: only at the identified specific resonant frequency, the virtual resistor jumps from a reference value to a suppression resistance value calculated based on the severity of the resonance; while in non-resonant frequency bands and when no resonance occurs, it maintains only the reference value. Frequency selectivity is achieved by extracting the current component in the resonant frequency band through bandpass filtering; the virtual resistor only acts on this frequency component, and the non-resonant frequency band remains unaffected.

[0063] The presence of a resonant event requiring suppression is determined based on the resonant peak amplitude and damping ratio. The criterion is: a resonant event requiring suppression is considered to exist when the resonant peak amplitude exceeds 0.3 pu and the damping ratio is below 0.05. When broadband impedance identification is not triggered in the current cycle, the historical values ​​of the resonant peak amplitude and damping ratio from the previous effective identification cycle are used.

[0064] When a resonance event is determined to exist, the virtual resistance after the transition is calculated using the following formula: in, This is the virtual resistance after the transition. To restrict the numerical range of functions, This is the resistance conversion factor. , The amplitude of the resonance peak. For the damping ratio, To minimize damping protection value and prevent division by zero, , This represents the minimum virtual resistance. This represents the maximum virtual resistance.

[0065] When it is determined that no resonance event exists , The reference virtual resistor.

[0066] The output current is bandpass filtered to extract the current component near the resonant peak frequency. In this embodiment, a second-order bandpass filter is used, with the center frequency being the resonant peak frequency and the quality factor being [missing information]. Then, the virtual resistor after the frequency band is applied to obtain the virtual resistor damping voltage: in, The virtual resistance damping voltage along the d-axis. For d-axis current, This represents the d-axis current component near the resonant peak frequency. Incremental resistance acting on current across the entire frequency band It only applies to the current in the resonant frequency band. When resonance does not occur, it maintains only a minimal reference damping (reference virtual resistance) to minimize power loss. The virtual resistance damping voltage on the q-axis is calculated in the same way. When it is determined that there is no resonant event that needs to be suppressed for 10 consecutive cycles, the virtual resistance returns to the reference value to prevent frequent switching due to instantaneous fluctuations.

[0067] S5: Using the running data, perform centerless consistent inertia fusion, and then calculate the angular frequency reference value.

[0068] In scenarios where multiple inverters operate in parallel, frequency support requires coordination among all inverters. If each inverter adjusts independently, circulating currents or oscillations may occur due to asynchronous adjustments. This method employs a centerless consensus algorithm, where each edge computing unit only exchanges a highly concise state summary (inertia adjustment, confidence level, and timestamp) with its neighboring units. Through multiple iterations, the inertia adjustment of each edge computing unit tends to be consistent, without the need to exchange full waveform data, resulting in extremely low communication bandwidth requirements.

[0069] S5.1: Based on a virtual synchronizer, inertia adjustment is calculated using runtime data.

[0070] In the Each edge computing unit calculates the angular frequency deviation, the rate of change of angular frequency deviation (approximated by differential), and the active power deviation based on operating data. It then calculates the local inertia adjustment using the inertia-damping equation of the virtual synchronous machine (VSG). Simultaneously, it calculates the differential delay of active power.

[0071] S5.2: The inertia adjustment amount is obtained by using adjacent inverters to perform a centerless consistency iterative fusion of the inertia adjustment amount.

[0072] For the Each inverter, in each cycle, utilizes the inertia adjustment values ​​of adjacent inverters to repeatedly fuse the local inertia adjustment value. Specifically, within each cycle, it executes... In the nth consistency iteration, the th... Each edge computing unit performs a consensus iteration with its neighboring edge computing units, exchanging only state summary data during the iteration: The formula for the r-th iteration is: in, Let be the inertia adjustment amount for the (r+1)th iteration of the g-th edge computing unit. Let be the inertia adjustment amount for the g-th edge computing unit in the r-th iteration. For consistent step size, , Let be the inertia adjustment amount for the j-th edge computing unit in the r-th iteration. Let g be the adjacency weight between the g-th edge computing unit and its adjacent j-th edge computing unit. This is a set of adjacent cells. Initial value. This is the local inertia adjustment amount.

[0073] Adjacency weight The dynamic adjustment rules are as follows: the base weights use Metropolis weights; if the arrival time of the latest summary of the adjacent edge computing unit j exceeds the timeout threshold (5 times the sampling period), the adjacent weight is reset to zero (communication interruption); if the confidence level of edge computing unit j... If the value is less than 0.4, the adjacent weight is halved; finally, row normalization is performed. When communication between all adjacent units is interrupted, the system operates independently using only local adjustment values.

[0074] In this embodiment, The consensus iteration basically converges within 5 rounds, and the communication data volume is only about tens of bytes per cycle, which can be implemented on low-bandwidth industrial buses such as CAN / RS485. The decentralized architecture eliminates the risk of single point of failure.

[0075] S5.3: Update the no-load angular frequency and droop coefficient based on the fused inertia adjustment, and generate an angular frequency reference value.

[0076] The no-load angular frequency is recovered without error based on the integral of the frequency deviation (limiting). The active frequency droop coefficient is adaptively adjusted based on the frequency deviation and delay power difference (limiting). This allows the frequency support control quantity to gradually reduce the steady-state frequency deviation within the limiting range.

[0077] S6: Generate control commands for the inverter based on the angular frequency reference value and the virtual resistor damping voltage.

[0078] Frequency support and impedance reshaping are achieved through the same inverter control link, which may lead to coupling conflicts. For example, suppressing severe resonance requires a significant increase in virtual resistance, which will squeeze the margin of VSG voltage regulation; conversely, strong regulation during large frequency shifts may cause the modulation depth to approach saturation. This step combines the two into a unified control quantity and then performs three-level limiting and conflict arbitration to ensure that safety constraints always take precedence.

[0079] The angular frequency reference value is converted into a voltage regulation value, which is then combined with the virtual resistor damping voltage to form a control quantity. in, This is the d-axis control quantity. VSG control based on angular frequency reference value The generated d-axis voltage.

[0080] in, This is the q-axis control variable. This refers to the q-axis voltage controlled by the VSG. The virtual resistance damping voltage along the q-axis.

[0081] The control quantity is subjected to current limiting, modulation depth limiting, and frequency deviation limiting in sequence to obtain the limited control quantity.

[0082] Based on the angular frequency reference value and the limited control quantity, control commands that the inverter voltage loop / current loop / modulation loop can receive are generated, including: d-axis voltage reference value, q-axis voltage reference value, and angular frequency reference value.

[0083] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for frequency and resonance coordination control of a distributed inverter system, characterized in that, include: S1: Collect the inverter's operating data, evaluate the sampling quality based on the operating data, and calculate the operating point change, resonant peak amplitude, damping ratio, resonant peak frequency, and their confidence level; wherein, the damping ratio, resonant peak amplitude, resonant peak frequency, and their confidence level are calculated by the resonant frequency prediction model based on the operating data; S2: Based on the sampling quality, the change in the running point, and the confidence level, determine whether wideband impedance identification is triggered. If wideband impedance identification is triggered, execute S3; otherwise, execute S4. S3: By using the edge computing unit to superimpose the disturbance signal on the voltage control loop or modulation wave generation stage of the inverter, the resonant peak frequency, resonant peak amplitude and damping ratio are recalculated. S4: Calculate the virtual resistance damping voltage based on the resonance peak amplitude, damping ratio, and resonance peak frequency; S5: Using the running data, perform centerless consistent inertia fusion, and then calculate the angular frequency reference value; S6: Generate control commands for the inverter based on the angular frequency reference value and the virtual resistor damping voltage.

2. The method of claim 1, wherein, The process of collecting inverter operating data is as follows: The edge computing unit located at the inverter collects the three-phase voltage, three-phase current and DC side voltage at the grid connection point, and calculates the voltage and current in the dq coordinate system, the angular frequency at the grid connection point, the active power and the reactive power.

3. The method of claim 2, wherein, The process of evaluating sampling quality based on operational data is as follows: All four criteria are met, is 1, otherwise is 0, is a sampling quality identifier; The four criteria for judgment include: (1) All three-phase voltage and current and DC side voltage were successfully sampled; (2) The time difference between adjacent sampling times and the deviation of the sampling period do not exceed the time deviation threshold; (3) The phase-locked loop is effectively locked and the angular frequency deviation of the grid connection point in adjacent cycles does not exceed the angular frequency jump threshold; (4) The DC side voltage is within the rated range.

4. The method of claim 2, wherein, The process of calculating the change in running point is as follows: Based on active power, reactive power, DC side voltage, and grid connection point angular frequency, a characteristic vector of the operating point is constructed. The weighted Euclidean distance between the current characteristic vector of the operating point and the previous characteristic vector of the operating point is calculated to obtain the change in the operating point.

5. The frequency and resonance coordinated control method for a distributed inverter system as described in claim 1, characterized in that, The process of calculating the resonant peak frequency and its confidence level is as follows: The resonant frequency prediction model adopts an impedance model in the form of a second-order discrete transfer function, with the d-axis current as the response observation. The resonant peak frequency is obtained by mapping the transfer function to the s-domain through a bilinear transformation and solving for the frequency corresponding to the maximum point of the amplitude-frequency characteristic within the preset frequency range. The formula for calculating confidence level is: in, For confidence level, To obtain the maximum value, The model residuals of the resonant frequency prediction model. This represents the upper limit of the normalized residual.

6. The frequency and resonance coordinated control method for a distributed inverter system as described in claim 3, characterized in that, Wideband impedance identification is triggered when the sampling quality indicator is 0, or the change in the running point is greater than the change in the running point threshold, or the confidence level is less than the confidence threshold.

7. The frequency and resonance coordinated control method for a distributed inverter system as described in claim 1, characterized in that, When broadband impedance identification is not triggered, the virtual resistance damping voltage is calculated using the resonant peak amplitude, damping ratio, and resonant peak frequency obtained from S1. When triggering wideband impedance identification, the virtual resistance damping voltage is calculated using the resonant peak amplitude, damping ratio, and resonant peak frequency obtained from S3 recalculation.

8. The frequency and resonance coordinated control method for a distributed inverter system as described in claim 1, characterized in that, Determine whether a resonance event exists based on the resonance peak amplitude and damping ratio; When a resonance event is determined to exist. in, This is the virtual resistance after the transition. To restrict the numerical range of functions, This is the resistance conversion factor. The amplitude of the resonance peak. For the damping ratio, To minimize damping protection value and prevent division by zero, This represents the minimum virtual resistance. This represents the maximum virtual resistance. When it is determined that no resonance event exists , As a reference virtual resistor; The formula for calculating the virtual resistance damping voltage is: in, The virtual resistance damping voltage along the d-axis. For d-axis current, This represents the d-axis current component near the resonant peak frequency.

9. The frequency and resonance coordinated control method for a distributed inverter system as described in claim 1, characterized in that, The process of using runtime data to perform centerless consistent inertia fusion and then calculating the angular frequency reference value is as follows: Calculate the local inertia adjustment using operational data; The local inertia regulation is fused multiple times using the inertia regulation of adjacent inverters to achieve centerless consistent inertia fusion and obtain the fused inertia regulation. The angular frequency reference value is calculated based on the fused inertia adjustment.

10. The frequency and resonance coordinated control method for a distributed inverter system as described in claim 1, characterized in that, The process of generating inverter control commands based on angular frequency reference values ​​and virtual resistor damping voltage is as follows: The control quantity is synthesized from the angular frequency reference value and the virtual resistor damping voltage. The control quantity is limited; the inverter control command is generated based on the angular frequency reference value and the limited control quantity.

Images