Wideband Resonant Adaptive Control Method for Multiple Inverters in Parallel
By sampling and extracting current features in a multi-inverter parallel system, the oscillation frequency is identified and tracked, and a frequency-adaptive virtual impedance is constructed to achieve coordinated control and damping distribution of multiple inverters. This solves the problems of resonant frequency offset and uneven damping distribution in multi-inverter parallel systems, and improves resonance suppression effect and system stability.
Patent Information
- Application Number
- CN202611484684.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-22
- Publication Date
- 2026-11-17
AI Technical Summary
In multi-inverter parallel systems, the resonant frequency shifts with the number of parallel units, grid impedance, and operating conditions. Existing methods struggle to accurately pinpoint the dominant oscillation frequency, cannot dynamically reshape the virtual impedance, suffer from uneven damping responsibility among multiple inverters, and cannot optimize control parameters in real time, resulting in poor resonance suppression.
By sampling and extracting the grid connection point current, multiple oscillation frequency components are identified and tracked, a frequency-adaptive virtual impedance is constructed, and combined with multi-inverter coordinated control and damping distribution, the system status is monitored in real time to achieve adaptive updating of control parameters and ensure resonance suppression effect.
It achieves real-time tracking and adaptive locking of multiple oscillation frequencies in a multi-inverter parallel system, dynamically adjusts virtual impedance, and rationally allocates damping tasks to ensure effective suppression of resonance over a wide frequency range, thereby improving system stability and efficiency.
Smart Images

Figure CN123121580A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of power electronics technology, specifically relating to a wideband resonant adaptive control method for multiple inverters in parallel. Background Technology
[0002] With the rapid development and large-scale grid connection of renewable energy generation such as photovoltaic and wind power, parallel operation of multiple inverters has become the norm in new energy power generation systems. As the core equipment for connecting new energy to the grid, the performance of grid-connected inverters has a significant impact on grid connection efficiency and power quality. Currently, LCL filters are widely used in grid-connected inverters due to their advantages such as strong high-frequency harmonic attenuation and high power density. However, LCL filters inherently have resonance peaks, which introduce stability issues in the frequency domain. When multiple LCL inverters are connected to the grid in parallel, the output impedances of each inverter interact and couple through the grid impedance, creating harmonic resonances within the inverters, between inverters, and between the inverters and the grid. Differences in output impedance, control mechanisms, and line impedance between different inverters can trigger harmonic current interactions at the parallel connection points, creating a resonant amplification effect at specific frequencies. This leads to a significant increase in harmonic components, increasing grid losses, shortening equipment lifespan, and even causing large-scale power outages.
[0003] As the number of inverters connected in parallel increases, the system's resonance characteristics become more complex. Studies show that the coupling resonant frequency of a multi-inverter parallel system is negatively correlated with the number of inverters; as the number of inverters increases, the resonant frequency shifts to lower frequencies. Simultaneously, the wide range of grid impedance variations in weak grid environments further exacerbates the coupling between the inverters and the grid, making the resonance problem more prominent. Furthermore, when the inverter parameters are not entirely identical, the resonance characteristics become even more complex. These factors combined result in multi-inverter parallel systems exhibiting significant characteristics such as wide frequency range, multiple modes, and strong time-varying behavior.
[0004] Extensive research has been conducted by scholars both domestically and internationally on the resonance suppression problem in multi-inverter parallel systems. Existing resonance suppression methods can be mainly classified into the following categories:
[0005] (I) Active Damping Methods. Active damping methods effectively increase system damping by introducing additional feedback loops in the control circuit. This approach is flexible and does not increase power consumption. Common solutions include filter capacitor current feedback, inverter-side inductor current feedback, and grid-connected current differential feedback. However, differential feedback is prone to introducing noise interference signals, potentially causing system oscillations; capacitor current feedback requires additional current sensors, increasing system cost. Furthermore, most existing active damping methods are designed for single inverters, and their effectiveness is limited when extended to multi-inverter parallel scenarios.
[0006] (II) Virtual Impedance / Impedance Reshaping Methods. Impedance reshaping methods suppress resonance by altering the inverter's output impedance characteristics. A typical approach is to connect a virtual admittance or virtual resistor in parallel at the point of common coupling (PCC) to achieve resonance suppression at the system level. Some studies have further proposed adaptive virtual impedance methods, which can adjust the virtual resistance value according to changes in system operating conditions. However, most existing adaptive virtual impedance methods only adaptively adjust a single parameter of the virtual resistance, failing to achieve coordinated adaptive adjustment of the parameters of each virtual impedance component (resistance, inductance, capacitance) as the oscillation frequency changes. When the system oscillation frequency shifts significantly, the virtual impedance with a fixed structure struggles to maintain optimal damping across the entire frequency band.
[0007] (III) Global Coordination Control-Based Approaches. Some studies attempt to suppress global resonance at the system level, such as adding a centralized active damping device at the PCC or applying additional damping control to one of the inverters. However, these methods typically rely on additional hardware, increasing system cost and complexity; moreover, they lack a flexible damping task allocation mechanism among multiple inverters, making it difficult to dynamically adjust the sharing of damping responsibilities based on the real-time operating status (such as power margin) of each inverter.
[0008] (iv) Frequency Detection and Tracking-Based Methods. Accurate identification of the resonant frequency is a prerequisite for effective suppression. Existing methods mostly employ Fast Fourier Transform (FFT) for spectrum analysis or Phase-Locked Loop (PLL) for frequency tracking. However, using FFT alone has drawbacks such as high computational cost and poor real-time performance; using PLL alone is difficult to handle situations where multiple frequency components coexist. Existing methods still have significant shortcomings in the parallel detection of multiple oscillation frequency components and adaptive locking of the dominant frequency.
[0009] In summary, existing multi-inverter parallel resonance suppression technologies have the following main shortcomings:
[0010] (1) Insufficient wideband adaptability. Most existing methods are designed to suppress resonance at a single or a few fixed frequency points, which is difficult to cope with the wideband characteristics of resonant frequency shifting with the number of parallel inverters, grid impedance, operating conditions and other factors in a multi-inverter parallel system.
[0011] (2) Limited frequency identification and tracking capabilities. Existing frequency detection methods are unable to simultaneously track multiple oscillation frequency components in real time and adaptively lock onto the dominant oscillation frequency, resulting in the control target of impedance reshaping being unable to accurately target the most threatening oscillation frequency.
[0012] (3) The virtual impedance parameters lack frequency adaptability. The parameters of existing virtual impedance methods are mostly fixed values or only a single parameter can be adjusted. It is difficult to achieve coordinated adaptive adjustment of the parameters of each component of the virtual impedance according to the change of the dominant oscillation frequency, which limits its damping effect over a wide frequency range.
[0013] (4) Lack of a coordinated damping allocation mechanism among multiple inverters. Existing methods rarely consider the differences in power margin of each inverter and cannot reasonably allocate damping tasks according to the real-time operating status of each inverter, which can easily lead to overload of some inverters or insufficient utilization of damping resources.
[0014] (5) The control parameters lack closed-loop adaptive update capability. Existing methods cannot automatically adjust the control parameters to maintain the best resonance suppression effect after the system operating conditions change, and lack a parameter adaptive update mechanism based on real-time evaluation of system state. Summary of the Invention
[0015] This application provides a wideband resonance adaptive control method for multiple inverters in parallel, aiming to solve the problems in the prior art, such as difficulty in accurately locking the dominant oscillation frequency under all operating conditions, the inability of virtual impedance to dynamically reshape with frequency changes, uneven sharing of damping responsibility among multiple inverters, and the inability of control parameters to automatically optimize as the system state deteriorates, resulting in poor wideband resonance suppression effect.
[0016] A broadband resonant adaptive control method for multiple inverters in parallel is proposed, the method comprising:
[0017] S1: Grid-connected point current sampling and feature extraction: Collect the total grid-connected current signal of the multi-inverter parallel system, extract the fundamental angular frequency through coordinate transformation and frequency-locked loop processing, and identify the oscillation frequency components exceeding the amplitude threshold through spectrum analysis to form an oscillation frequency component set. Output the fundamental angular frequency and the oscillation frequency component set as a data stream.
[0018] S2: Oscillation frequency tracking and adaptive locking. The current signal is input to multiple parallel phase-locked loop branches that correspond one-to-one with each component in the oscillation frequency component set. Each branch tracks the frequency with the corresponding oscillation frequency component as the initial center frequency. The frequency signals output by each branch are weighted and fused to obtain the dominant oscillation frequency. The dominant oscillation frequency is output as a data stream.
[0019] S3: Based on frequency adaptive impedance dynamic reshaping, a frequency adaptive virtual impedance is constructed according to the dominant oscillation frequency, the virtual impedance is converted into virtual admittance, and filtered by a fundamental notch filter. The filtered virtual admittance is superimposed on the current control loop output of each inverter to realize dynamic reshaping of the grid-connected inverter output impedance, and the corrected current command is output as a data stream.
[0020] S4: Multi-inverter coordinated control and damping allocation: Collect the output power and operating status of each parallel inverter, calculate the damping allocation coefficient of each inverter based on the rated capacity and current output power of each inverter, allocate the virtual admittance to each inverter according to the damping allocation coefficient, each inverter embeds the allocated virtual admittance into its own current control loop, and outputs the updated output power and damping allocation coefficient of each inverter as a data stream;
[0021] S5: Control parameters are adaptively updated. The total harmonic distortion rate of the grid-connected current, the system damping ratio, and the maximum oscillation amplitude are monitored in real time. A system state evaluation index vector is constructed. When the system state evaluation index vector exceeds the preset stable operation threshold, the reference parameters of the phase-locked loop in S2 and the virtual impedance in S3 are adjusted online according to the current dominant oscillation frequency and the system state evaluation index vector through a fuzzy logic controller or particle swarm optimization algorithm. The updated parameters are sent back to S2 and S3 as a data stream to replace the original parameters.
[0022] S6: Resonance suppression effect evaluation and strategy switching. Calculate the resonance suppression rate. If the resonance suppression rate is greater than the preset suppression rate threshold, maintain the current control parameters and strategy unchanged. If the resonance suppression rate is less than the preset suppression rate threshold, activate the frequency rescanning mechanism in S2 and re-execute S2 to S5 to complete the control strategy switching.
[0023] Optionally, in step S1, the grid connection point current sampling and feature extraction specifically includes:
[0024] S1.1 Install a current transformer at the common coupling point to collect the total grid-connected current signal of the multi-inverter parallel system. The sampled data is then sent to the digital signal processor after analog-to-digital conversion.
[0025] S1.2. The acquired three-phase current signals are converted to static state using Clark transformation. coordinate system, to obtain Current components in the coordinate system;
[0026] S1.3, will The current signal in the coordinate system is input to the frequency locking loop module. The frequency locking loop adopts a frequency adaptive structure based on a second-order generalized integrator. It extracts the fundamental angular frequency and the angular frequencies of each harmonic or oscillation component in the grid-connected current through closed-loop adjustment.
[0027] S1.4 Perform spectrum analysis on the current signal using Fast Fourier Transform, automatically identify frequency components exceeding the preset amplitude threshold, form an oscillation frequency component set by combining all frequency components that meet the amplitude condition and whose frequency does not belong to the fundamental wave, and output the fundamental wave angular frequency and the oscillation frequency component set as data streams to S2 and S3 respectively.
[0028] Optionally, in step S2, the oscillation frequency tracking and adaptive locking specifically includes:
[0029] S2.1 Input the stationary αβ coordinate system current signal output from S1 into k parallel phase-locked loop branches respectively. The initial value of the center angular frequency of each branch is set according to the set of oscillation frequency components identified by the FFT spectrum analysis in S1. Each phase-locked loop branch works independently and in parallel.
[0030] S2.2 Each phase-locked loop branch adopts a phase-locked loop structure based on a second-order generalized integrator. It extracts the orthogonal component with the same frequency as the center frequency from the input current signal, and inputs it to the PI regulator after Park transformation. The output of the PI regulator is integrated to form the frequency estimate of the branch.
[0031] S2.3 The frequency signals output by each phase-locked loop branch are synchronously input to the frequency decision module. A data buffer is set at the input of the frequency decision module. The output of each branch is resampled and interpolated based on the system control cycle to ensure that the frequency signals participating in the fusion calculation are time-aligned.
[0032] S2.4 The frequency decision module performs weighted fusion processing on the frequency signals output by each branch. The weight coefficients are allocated according to the amplitude of each frequency component. The dominant oscillation frequency output by the frequency decision module is the weighted sum of the frequencies of each branch. At the same time, the frequency decision module is also equipped with a numerical validity verification submodule, which sets the weight coefficient of the branch determined to be in a lost-lock state to zero and excludes it from the fusion calculation.
[0033] S2.5 The dominant oscillation frequency output by the frequency decision module is output as a data stream to S3. At the same time, the dominant oscillation frequency is fed back to the FFT spectrum analysis module of S1 as the frequency reference for spectrum refinement analysis. The frequency output and amplitude information of each phase-locked loop branch are continuously updated and stored in the system status register for S5 to call.
[0034] Optionally, in S2.4, the weighted fusion process specifically involves: the weighting coefficients λi are adjusted according to the magnitude of each frequency component. Distribute, among which Let be the amplitude of the i-th oscillation frequency component, and let be the dominant oscillation frequency output by the frequency decision module. ;
[0035] in This is the estimated frequency output of the i-th phase-locked loop branch;
[0036] The frequency decision module is also equipped with hysteresis comparison logic. Only when the amplitude of a certain branch is continuously higher than the amplitude of other branches by a set proportion for N consecutive control cycles will the frequency of that branch be determined as the dominant frequency.
[0037] In the numerical validity verification submodule, if the frequency value output by a branch deviates from the center frequency of the branch by more than a preset range or the rate of change exceeds a preset limit, the branch is determined to be in a unlocked state.
[0038] Optionally, in step S3, the frequency-adaptive dynamic impedance reshaping specifically includes:
[0039] S3.1, Based on the dominant oscillation frequency output by S2 Construct a frequency-adaptive virtual impedance:
[0040] Virtual resistance Virtual inductance Virtual capacitor All The function;
[0041] S3.2, Convert the virtual impedance into virtual admittance:
[0042]
[0043] And through the fundamental notch filter After filtering, the transfer function of the fundamental notch filter is:
[0044]
[0045] Where m is the notch filter frequency coefficient, 0 <m<2, The fundamental angular frequency extracted by the frequency-locked loop in S1;
[0046] S3.3, The virtual admittance after notch filtering:
[0047]
[0048] This is superimposed on the output of the current control loop of each inverter, and a feedforward branch is added to the current loop control structure of each inverter. The input of this branch is the common coupling point voltage. The transfer function is ;
[0049] The branch output is a compensation current signal:
[0050]
[0051] The compensation current is superimposed on the given end of the current loop in a negative feedback manner;
[0052] S3.4. The compensation current is added as a feedforward compensation amount to the reference current command value of the current loop to obtain the corrected current command:
[0053]
[0054] The current is fed into the current regulator, which transmits the corrected current commands from each inverter as a data stream to S4.
[0055] Optionally, the multi-inverter coordinated control and damping allocation in S4 specifically includes:
[0056] S4.1. Collect the output power Pi and current operating status of each parallel inverter through the communication network to construct the device state vector:
[0057]
[0058] The device state vector is input as a data stream to the damping distribution coefficient calculation module;
[0059] S4.2 Calculate the damping distribution coefficient of each inverter based on its rated capacity Prated,i and current output power Pi.
[0060]
[0061] The sum of the damping distribution coefficients of all inverters satisfies ;
[0062] For inverters whose status is marked as fault or standby, their damping distribution coefficient is set to zero.
[0063] S4.3, The virtual admittance generated in S3 The damping distribution coefficient ηi is used to allocate power to each inverter.
[0064]
[0065] The allocated virtual admittance commands for each inverter are sent to the local controller of each inverter through the communication network.
[0066] S4.4 Each inverter receives the corresponding virtual admittance command. Each inverter embeds these values into its respective current control loop, and each inverter acquires the PCC point voltage at its own port in real time. Calculate the compensation current of this machine.
[0067]
[0068] The compensation current is subtracted from the reference command of the current loop to obtain the corrected current command, which is then sent to the current regulator.
[0069] Optionally, in step S4, the damping distribution coefficient is updated repeatedly at a fixed period, and the fixed period is adaptively adjusted according to the rate of change of system operating conditions.
[0070] When the system power change rate is detected to exceed 10% / s of the rated power, the update cycle is shortened to 50ms;
[0071] When the system is in a steady state, the update cycle is extended to 500ms;
[0072] The communication network is equipped with a data validity verification step. If the status data has not been updated for more than two communication cycles, the channel of the communication network is determined to be invalid, the inverter is temporarily removed from the coordination control queue and a communication fault alarm is triggered.
[0073] For state data with abrupt changes, first-order low-pass filtering or median filtering is used for smoothing.
[0074] Optionally, in step S5, the adaptive update of the control parameters specifically includes:
[0075] S5.1 Real-time monitoring of total harmonic distortion (THD) and system damping ratio of grid-connected current. Construct a system state evaluation index vector ;
[0076] in The system state evaluation index vector is input as a data stream to the comparison and judgment module as the maximum amplitude value among the oscillation frequency components identified in S1.
[0077] S5.2, The system state evaluation index vector E is compared with the preset stable operation threshold. When comparing, and and If the system is determined to be in a stable operating state, the current control parameters are maintained unchanged.
[0078] If any of the above conditions are not met, a judgment will be made. This triggers adaptive parameter updates;
[0079] S5.3 When the trigger parameter is updated adaptively, it is based on the current dominant oscillation frequency. The virtual impedance reference parameters in S3 are adjusted online using a fuzzy logic controller or particle swarm optimization algorithm, along with the system state evaluation index vector E. , , and the phase-locked loop parameters in S2 The updated parameters are transmitted back as data streams to each phase-locked loop branch of S2 and the virtual impedance building module of S3 to replace the original parameters.
[0080] S5.4 After parameter update, repeat S1 to S4, re-execute current sampling, frequency tracking, virtual impedance construction, and damping allocation under the new control parameters, and then re-enter S5.1 for system state evaluation. If successful, exit the parameter update process; if... If the condition is still met, S5.3 will be executed again for a new round of parameter adjustment until the stability condition is met or the preset maximum number of updates is reached.
[0081] Optionally, the resonance suppression effect evaluation and strategy switching in S6 specifically includes:
[0082] S6.1 After completing S1-S5, continuously monitor the oscillation suppression effect of the system and calculate the resonance suppression rate. The formula for calculating the resonance suppression rate is:
[0083]
[0084] in The maximum amplitude value among the various oscillation frequency components identified before the control strategy is implemented. The real-time amplitude of the same frequency component at the current moment is given. When there are multiple oscillation frequency components in the system, the suppression rate corresponding to each frequency component is calculated separately, and the minimum value is taken as the overall resonance suppression rate of the system.
[0085] S6.2, if the resonance suppression rate Greater than the preset inhibition rate threshold If the current control strategy is effective, the current control parameters and strategy will remain unchanged; if the resonance suppression rate is... Less than the preset inhibition rate threshold If the system has a new oscillation mode or its operating conditions have changed significantly, the frequency rescanning mechanism in S2 will be activated.
[0086] S6.3 After activating the frequency rescanning mechanism, the center frequency of each phase-locked loop branch in S2 is reset to the initial state and the frequency search range of each branch is expanded. The FFT spectrum analysis of S1 is re-executed to finely identify the new set of oscillation frequency components. The number of phase-locked loop branches and the initial center frequency in S2 are updated according to the new set of frequency components. All processes from S2 to S5 are re-executed. During the frequency rescanning period, the original control parameters remain unchanged. After the new frequency components are locked and the new virtual impedance parameters are calculated, the new parameter configuration is switched synchronously.
[0087] S6.4 When the system is in a stable operating state for a long time and no new oscillation frequency appears, the control strategy automatically switches to the low power monitoring mode. In the low power monitoring mode, the execution frequency of FFT spectrum analysis in S1, the update rate of the phase-locked loop branch in S2, the calculation frequency of the system state evaluation index in S5, and the update cycle of the damping allocation coefficient in S4 are reduced. Once the resonance suppression rate is detected to drop below the threshold or a new oscillation frequency component is detected, the system immediately exits the low power monitoring mode, returns to the normal operation mode, and re-executes all processes from S1 to S5.
[0088] Optionally, in S6.4, the four conditions for switching the low-power monitoring mode are:
[0089] Resonance suppression rate Persistently greater than The time exceeded the set time And the frequency signals output by each phase-locked loop branch in S2 are The fluctuation range within the time period does not exceed ±2Hz, and the FFT spectrum analysis in S1 is continuous. No new oscillation frequency components exceeding the threshold Ath were detected within the specified time, and the output power of each inverter in S4 was within the specified range. When the rate of change within a time period does not exceed 5% / s of the rated power, and all four conditions above are met simultaneously, the system is determined to have entered a long-term stable state and is switched to low power monitoring mode; the response time to exit low power monitoring mode is required to be within 100ms.
[0090] Compared with the prior art, this application has at least the following beneficial effects:
[0091] This application achieves parallel detection and continuous tracking of multiple oscillation frequency components in the system through the coordinated operation of a frequency-locked loop (PLL) and FFT spectrum analysis. The PLL provides real-time tracking of each frequency component, while the FFT provides global frequency scanning capability. Furthermore, a multi-branch parallel PLL structure is used to independently track each frequency component, and a fusion decision mechanism based on amplitude weighting is employed to output the dominant oscillation frequency. Combined with hysteresis comparison logic and numerical validity verification, this effectively avoids frequent frequency jumps in the critical state and interference from unlocked branches. This mechanism enables the system to adaptively identify and lock onto the most threatening oscillation frequency over a wide frequency band, providing an accurate and reliable frequency reference signal for subsequent impedance dynamic reshaping.
[0092] This application constructs an adaptive function for virtual impedance parameters with the dominant oscillation frequency as the independent variable. The virtual resistance exhibits second-order high-pass filtering characteristics, automatically increasing at higher oscillation frequencies to enhance high-frequency damping and automatically decreasing near the fundamental frequency to avoid affecting fundamental power transmission. The virtual inductance and capacitance automatically decrease with increasing frequency, allowing the RC characteristics of the virtual impedance to dynamically adjust with the oscillation frequency. This mechanism ensures that the virtual impedance maintains optimal damping characteristics across a wide frequency range, overcoming the limitation of traditional fixed-parameter virtual impedances in adapting to frequency shifts. Simultaneously, through filtering by a fundamental notch filter, it is ensured that the introduction of virtual admittance does not affect the fundamental power transmission of the inverter, achieving decoupling between resonance suppression and normal power transmission.
[0093] This application constructs a damping allocation coefficient based on power margin, enabling inverters with larger power margins to undertake more damping tasks and those with smaller power margins to undertake less damping tasks. Virtual admittance commands are synchronously sent to each inverter via a communication network. This mechanism effectively avoids the risk of overmodulation or current saturation caused by a single inverter bearing too much damping load while ensuring resonance suppression. The update cycle of the damping allocation coefficient can be adaptively adjusted according to the rate of change of system operating conditions, and the communication network includes a data validity verification step to ensure the reliability of coordinated control. Each inverter shares the damping task proportionally according to its power margin, resulting in a coordinated and consistent damping characteristic across the multi-inverter parallel system.
[0094] This application constructs a system state evaluation index vector by real-time monitoring of three dimensions: total harmonic distortion (THD), system damping ratio, and maximum oscillation amplitude. When the system state deteriorates, it automatically triggers adaptive parameter updates. Virtual impedance reference parameters and phase-locked loop (PLL) parameters are adjusted online using a fuzzy logic controller or particle swarm optimization algorithm. After the parameter updates, the entire link control process is re-executed to verify the effect, forming a closed-loop adaptive adjustment mechanism of "evaluation-update-verification." This enables the system to proactively identify system state deterioration and automatically optimize control parameters when operating conditions change, such as inverter output power variations, grid short-circuit ratio changes, and PLL parameter offsets, maintaining sufficient damping margin across a wide frequency band.
[0095] This application quantifies the control effect by calculating the resonance suppression rate. When the suppression effect fails to meet the requirements, a frequency rescanning mechanism is automatically activated to re-identify new oscillation frequency components and refresh the control parameters, thus achieving adaptive reconfiguration of the control strategy. Simultaneously, when the system operates stably for an extended period, it automatically switches to a low-power monitoring mode, significantly reducing the computational load on the digital signal processor and minimizing control overhead while ensuring system safety and stability. Attached Figure Description
[0096] Figure 1This is a flowchart of a broadband resonant adaptive control method for multiple inverters in parallel, provided as an embodiment of this application. Detailed Implementation
[0097] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.
[0098] This application provides a wideband resonant adaptive control method for multiple inverters in parallel, such as... Figure 1 As shown, it includes the following steps:
[0099] S1: Grid connection point current sampling and feature extraction, specifically including:
[0100] S1.1 Install a current transformer at the common coupling point to collect the total grid-connected current signal of the multi-inverter parallel system. Sampling frequency The selection of the sampling frequency needs to balance hardware performance and signal analysis requirements, and is generally set to be more than twice the system's highest frequency of interest to satisfy the Nyquist sampling theorem. For typical broadband resonance analysis scenarios, the highest frequency of interest usually covers several kHz, therefore the sampling frequency... The preferred kHz range is 10kHz to 20kHz. The sampled data is converted from analog to digital and then sent to a digital signal processor to form the input data stream for subsequent stages.
[0101] S1.2, The acquired three-phase current signals , , Transformed to static state via Clark transformation Coordinate system, to obtain current components , The matrix form of the Clark transform is:
[0102]
[0103] This transformation maps the time-domain signal in a three-phase stationary coordinate system to a two-phase orthogonal stationary coordinate system, eliminating the coupling relationship between the three-phase variables and facilitating subsequent extraction and analysis of frequency characteristics. The transformed signal... The components retain all the frequency information of the original signal, providing a unified data interface for frequency locking loops and spectrum analysis;
[0104] S1.3, will The current signal in the coordinate system is input as a data stream to the frequency-locked loop module. The frequency-locked loop adopts a frequency adaptive structure based on a second-order generalized integrator, and its transfer function can be expressed as:
[0105]
[0106] in Here, ω is the damping coefficient, and ω is the center angular frequency of the frequency-locked loop. The frequency-locked loop automatically tracks the input signal frequency with its center frequency ω through closed-loop regulation, thereby extracting the fundamental angular frequency from the grid-connected current. and the angular frequencies of each harmonic or oscillating component ;
[0107] Compared to phase-locked loops (PLLs), frequency-locked loops (FLLs) exhibit better anti-interference performance under power grid frequency fluctuations and harmonic distortion conditions, and can accurately lock the actual values of each frequency component. The extracted fundamental angular frequency... On the one hand, it is used for the parameter configuration of the subsequent fundamental notch filter, and on the other hand, it serves as a reference for determining whether the oscillation component belongs to the non-fundamental frequency band;
[0108] S1.4: To further identify the broadband resonance characteristics present in the system, a spectral analysis of the current signal is performed using a Fast Fourier Transform (FFT). The optimal data window length for the FFT operation is 10 fundamental periods (i.e., 0.2s), corresponding to a frequency resolution of 5Hz, which effectively distinguishes adjacent frequency components. The spectral analysis results are output in the form of an amplitude-frequency response curve, and the system uses a preset amplitude threshold... Automatically identify frequency components exceeding this threshold. Threshold The setting needs to comprehensively consider the harmonic background and amplitude characteristics of the resonance state during normal system operation, and is generally taken as 5% to 10% of the rated current amplitude of the system. When a certain frequency component Corresponding amplitude satisfy and If the frequency is not a fundamental frequency, then that frequency component is considered a potential resonant component. All frequency components that meet the conditions are then grouped into a set of oscillating frequency components. The characteristic frequency of the broadband resonance is output to S2;
[0109] It should be noted that the frequency-locked loop extraction in S1.3 and the FFT spectrum analysis in S1.4 are complementary: the frequency-locked loop provides real-time tracking values for each frequency component, offering advantages such as fast response speed and low computational cost, making it suitable for continuous tracking of identified frequencies; the FFT spectrum analysis provides global frequency scanning capability, enabling the discovery of newly emerging oscillation frequency components, suitable for frequency rescanning after changes in system operating conditions. The two work together to ensure the frequency identification capability of S1 under all operating conditions.
[0110] S1 output fundamental angular frequency The center frequency parameter of the fundamental notch filter in S3 will be used as the output oscillation frequency component set Ω, which will be used as the initial tracking target of the multi-branch phase-locked loop in S2, thereby orderly transmitting the data stream of front-end acquisition and feature extraction to the subsequent control and execution links.
[0111] S2: Oscillation frequency tracking and adaptive locking, specifically including:
[0112] S2.1, Set the output of S1 to static. Coordinate system current signal As a data stream, it is input to k parallel phase-locked loop branches. The initial value of the center angular frequency of each branch is based on the set of oscillation frequency components identified by the FFT spectrum analysis in S1. Configure the initial center frequency of the i-th PLL branch. The corresponding frequency component in the set Ω;
[0113] The value of the number of branches k corresponds to the number of oscillation components that exceed the amplitude threshold identified in S1. When a new oscillation frequency component is detected or an existing component decays to below the threshold during system operation, the number of branches k is adjusted in real time. New branches are put into operation with the frequency component as the initial center frequency, while decaying branches are taken out of operation, thereby ensuring that computing resources are always concentrated on effective oscillation frequency tracking.
[0114] Each PLL branch operates independently and in parallel. The frequency output of any branch is only affected by its own input signal and internal feedback loop. There is no coupling between branches, so the dynamic response of a single branch will not interfere with the tracking performance of other branches.
[0115] S2.2 Each PLL branch adopts a phase-locked loop structure based on a second-order generalized integrator. The transfer function of the SOGI link is:
[0116]
[0117] in The damping coefficient is... This is the center angular frequency corresponding to this branch;
[0118] The SOGI stage extracts the frequency relative to the center frequency from the input current signal. The quadrature components of the same frequency are transformed by Park and then input to a PI controller. The output of the PI controller is integrated to form the frequency estimate of this branch. Damping coefficient The selection of frequency selection bandwidth and dynamic response speed of the SOGI stage directly affects the frequency selection bandwidth and dynamic response speed of the SOGI stage. The smaller the value, the narrower the frequency selection bandwidth and the stronger the frequency selectivity, but the slower the dynamic response speed. The larger the value, the faster the dynamic response, but the wider the frequency selection bandwidth, making it easier to introduce interference from adjacent frequency components;
[0119] Preferably, The value should be between 0.5 and 1.5. For scenarios where the oscillation frequencies are closely spaced, a smaller value is used to enhance frequency selectivity; for scenarios with larger frequency intervals, a larger value can be used to speed up the response. (Each branch...) It can be independently set according to the distribution density of the frequency components it tracks in the spectrum, and the branches in the frequency-dense region use smaller values. Branches in sparse frequency ranges use larger [variables / areas]. This achieves an optimal match between frequency selectivity and response speed;
[0120] It is worth noting that the traditional SOGI structure has a certain steady-state error when the frequency offset is large. To overcome this deficiency, a frequency adaptive correction loop can be added to the output of the SOGI stage to feed back the frequency estimation error to the center frequency input of the SOGI, forming a closed-loop correction to ensure error-free tracking under large frequency offset conditions.
[0121] S2.3, Frequency signals output by each PLL branch , … As a data stream, it is synchronously input to the frequency decision module. Each branch's frequency output has a different update timeframe. To ensure time alignment of the input data to the frequency decision module, a data buffer is set at the input end of the frequency decision module to match the system control cycle. Based on this benchmark, the outputs of each branch are resampled and interpolated to ensure that the frequency signals participating in the fusion calculation all correspond to the same time. The data buffer has a depth of at least 2 to store the frequency values of the current and previous times, providing data support for subsequent rate limiting and rate of change detection.
[0122] S2.4 The frequency decision module performs weighted fusion processing on the frequency signals output from each branch. The weighting coefficients λi are allocated according to the amplitude of each frequency component.
[0123]
[0124] in This represents the amplitude of the i-th oscillation frequency component in the S1FFT spectrum analysis, or it can be obtained in real time from the amplitude detection circuit within each PLL branch. Amplitude detection can be achieved by adding an amplitude calculation unit to the PLL branch. This unit uses the quadrature components output by SOGI to calculate the instantaneous amplitude of the frequency component. ,in and These are the two quadrature signals output by SOGI;
[0125] Employing a real-time amplitude update method allows the weighting coefficients to be dynamically adjusted according to the energy changes of each frequency component, improving the timeliness of dominant frequency identification. The dominant oscillation frequency output by the frequency decision module is the weighted sum of the frequencies of each branch:
[0126]
[0127] To prevent frequent frequency jumps between multiple frequency components with similar amplitudes, the frequency decision module incorporates hysteresis comparison logic: the frequency of a branch is determined as the dominant frequency only when its amplitude consistently exceeds that of other branches by a certain percentage (e.g., 20%) for N consecutive control cycles; conversely, the dominant frequency is switched only when the amplitude of the dominant frequency branch continuously decays to a certain percentage below that of the second dominant branch. This hysteresis logic effectively suppresses frequency output oscillations under critical conditions.
[0128] In addition, to prevent individual branches from outputting abnormal frequency values due to phase-lock failure and affecting the fusion results, the frequency decision module is also equipped with a numerical validity verification submodule to verify the rationality of the frequency values output by each branch. If the frequency value output by a branch deviates from the center frequency of the branch by more than a preset range (e.g., ±50Hz) or the rate of change exceeds a preset limit (e.g., 100Hz / s), the branch is determined to be in a state of unlocking, its weight coefficient is temporarily set to zero and it is excluded from the fusion calculation. Its weight is restored after the branch is relocked.
[0129] S2.5, The dominant oscillation frequency output by the frequency decision module. As the impedance reshaping module that outputs data streams to S3, it serves as the basis for adaptive frequency adjustment of virtual impedance parameters. Simultaneously, it controls the dominant oscillation frequency. It is also fed back to the FFT spectrum analysis module of S1 as a frequency benchmark for spectrum refinement analysis, so that FFT can perform higher resolution spectrum analysis near the dominant frequency in subsequent calculations, thereby improving frequency tracking accuracy.
[0130] Frequency output of each PLL branch and amplitude information It is also continuously updated and stored in the system's status register, which is called by the S5 parameter adaptive update module as a basis for evaluating changes in system oscillation modes and adjusting control parameters;
[0131] Through the above-mentioned multi-branch parallel tracking and weighted fusion mechanism, parallel monitoring of multiple oscillation frequency components in the system and adaptive locking of the dominant frequency are realized, providing an accurate and reliable frequency reference signal for subsequent impedance dynamic reshaping.
[0132] At this point, S2 has completed the conversion and locking from the multi-frequency component set output by S1 to a single dominant oscillation frequency, and continuously feeds this dominant frequency to S3 in the form of a data stream, so that the reshaping of the virtual impedance is always aligned with the most threatening oscillation frequency of the system, and realizes the data connection between the frequency tracking link and the control execution link.
[0133] S3: Frequency-adaptive dynamic impedance reshaping, specifically including:
[0134] S3.1, Based on the dominant oscillation frequency output by S2 Construct a frequency-adaptive virtual impedance The virtual impedance is in the form of an RLC series branch, and its expression is:
[0135]
[0136] Virtual resistance Virtual inductance Virtual capacitor All Functions:
[0137]
[0138]
[0139]
[0140] In the formula The cutoff angular frequency, , , Used as a reference parameter;
[0141] The design principle of the above functional relationship is that: when the dominant oscillation frequency... Far below the cutoff frequency Time (i.e.) Virtual resistance Approaching zero, virtual inductance and virtual capacitance Approaching their respective benchmark values and Virtual impedance mainly exhibits LC characteristics; when Much higher Time (i.e.) The virtual resistance approaches the reference value. Meanwhile, virtual inductance and virtual capacitance approach zero, and virtual impedance mainly exhibits pure resistivity;
[0142] Through this frequency-adaptive parameter configuration, the virtual impedance can automatically adjust its resistive-capacitive characteristics according to the oscillation frequency, maintaining good damping performance over a wide frequency range. Cutoff angular frequency The selection of the reference parameter needs to be determined based on the inherent resonant frequency of the system's LCL filter. It is generally set to 0.5 to 1 times the inherent resonant frequency, ensuring that the resistive component of the virtual impedance dominates in the frequency band where the system may oscillate. Reference Parameter , , The tuning needs to balance damping effect and system stability. The larger the virtual resistance, the stronger the system damping, but an excessively large virtual resistance will reduce the system's power transmission efficiency. and Used to compensate for the impedance characteristics of a system in a specific frequency band, its value is generally obtained based on the system impedance measurement results or through online identification by frequency sweeping method;
[0143] It should be noted that, The function form adopts the characteristics of second-order high-pass filtering, which makes the virtual resistance automatically increase when the oscillation frequency is high and automatically decrease near the fundamental frequency. This is consistent with the adaptive adjustment idea of virtual resistance in active dampers.
[0144] It is worth noting that the construction of virtual impedance is not limited to the above-mentioned RLC series form. Parallel RLC branches or more complex impedance network structures can also be adopted according to the specific needs of the system. This embodiment only uses the series RLC branch as a preferred solution for illustration. Those skilled in the art can make equivalent modifications without departing from the concept of the present invention.
[0145] S3.2, virtual impedance Convert to virtual admittance and through the fundamental wave notch filter Filtering is performed. The transfer function of the fundamental notch filter is:
[0146]
[0147] Where m is the notch filter frequency coefficient, 0 <m<2, The fundamental angular frequency extracted by the frequency-locked loop in S1;
[0148] Notch filter at fundamental frequency The frequency exhibits an extremely deep stopband (ideally zero gain), while the gain is close to 1 in other frequency bands away from the fundamental frequency. After processing with this notch filter, the virtual admittance... It presents a low-resistance path only at non-fundamental frequencies (i.e., oscillation frequencies) and a high-resistance state at fundamental frequencies, thereby ensuring that the introduction of virtual admittance does not affect the fundamental power transmission of the inverter.
[0149] The selection of the frequency coefficient m directly affects the stopband width and dynamic response characteristics of the notch filter: the smaller the value of m, the narrower the stopband, and the stronger the attenuation selectivity of signals near the fundamental frequency, but the ability to adapt to power grid frequency fluctuations decreases; the larger the value of m, the wider the stopband, and the stronger the robustness to power grid frequency fluctuations, but it may cause unnecessary attenuation of low-frequency oscillation components of the adjacent fundamental frequency.
[0150] Taking into account both the fundamental frequency fluctuation range (typically ±0.5Hz) and the need to distinguish low-frequency oscillations, the frequency coefficient m is preferably 0.2 to 0.5;
[0151] For weak grid scenarios with large grid frequency fluctuations, an adaptive notch filter structure can be further adopted, enabling the center frequency of the notch filter to track the fundamental angular frequency output by the frequency-locked loop in real time. , that is to say In Replace with time variable This ensures that the notch filter can still be accurately aligned with the fundamental frequency when the grid frequency shifts.
[0152] S3.3, The virtual admittance after notch filter processing As a data stream, it is superimposed on the output terminal of the current control loop of each inverter to realize the dynamic reshaping of the output impedance of the grid-connected inverter;
[0153] The method for superimposing virtual admittance is as follows: a feedforward branch is added to the current loop control structure of each inverter, and the input of this branch is the common coupling point voltage. The transfer function is The branch output is a compensation current signal. The compensation current is superimposed on the given end of the current loop in a negative feedback manner, which is equivalent to connecting a frequency-selective admittance network in parallel at the output of the inverter;
[0154] From an impedance perspective, the equivalent output admittance of the inverter, after introducing virtual admittance, changes from the original... Become That is, an additional admittance branch that adaptively adjusts with the oscillation frequency is connected in parallel on the basis of the original output admittance;
[0155] Because this additional admittance exhibits low impedance characteristics at the oscillation frequency, it can provide an additional flow path for the oscillating current, which is equivalent to adding damping to the system at the oscillation frequency. The amplitude of the virtual admittance needs to be tuned according to the system's damping requirements. If the admittance value is too small, the damping will be insufficient and the suppression effect will be limited; if the admittance value is too large, it may change the low-frequency characteristics of the system and introduce new instabilities.
[0156] Preferably, the amplitude of the virtual admittance is selected according to the admittance level corresponding to 5% to 15% of the rated current of the system, and the specific value can be optimized and determined through small signal modeling and root locus analysis;
[0157] S3.4 In the current loop control of the inverter, the compensation current corresponding to the virtual admittance is... As a feedforward compensation, it is superimposed on the reference current command value of the current loop. Specifically, the voltage at the PCC point is acquired in real time within the digital signal processor. Compare it with the virtual admittance value at the current moment. Multiplying them gives the instantaneous value of the compensation current. Then, this compensation value is taken from the reference command of the current loop. Subtracting the current from the output (equivalent to adding the admittance in parallel) yields the corrected current command. Feed into the current regulator;
[0158] because It only contains non-fundamental frequency components (fundamental components have been filtered out by a notch filter), and this feedforward compensation will not affect the inverter's tracking accuracy for fundamental active and reactive power.
[0159] It is worth noting that the superposition polarity of the compensation current depends on the specific implementation architecture of the control system. If voltage-type control is used (i.e., the current loop output is a modulated voltage signal), the compensation amount corresponding to the virtual admittance should be converted into a voltage compensation amount and superimposed on the modulated wave. If current-type control is used (i.e., the current loop output is directly a current command), it should be superimposed on the current reference command in the above manner.
[0160] The two implementation methods are mathematically equivalent, and those skilled in the art can choose the appropriate implementation scheme according to the specific control system architecture;
[0161] In addition, to suppress the influence of high-frequency noise on virtual admittance calculation, a low-pass filter with a cutoff frequency of 2kHz to 5kHz is added to the PCC voltage sampling channel to ensure that the voltage signal entering the virtual admittance branch does not contain high-frequency noise components.
[0162] At this point, S3 completes the conversion from the dominant oscillation frequency output by S2 to the frequency-adaptive virtual admittance, and embeds this virtual admittance into the current control loop of each inverter in the form of compensation current, realizing the dynamic reshaping of the grid-connected inverter output impedance at the oscillation frequency. Through the above mechanism, the system obtains additional damping that adaptively adjusts with the oscillation frequency over a wide frequency range, providing strong control support for the stable operation of multi-inverter parallel systems. The corrected current commands for each inverter output by S3 are shown below. The data stream will be transmitted to the multi-inverter coordination control and damping allocation module of S4, and S4 will reasonably allocate the damping task according to the power margin of each inverter.
[0163] S4: Multi-inverter coordinated control and damping distribution, specifically including:
[0164] S4.1. Collect the output power PiPi and current operating status of each parallel inverter through the communication network to construct the device state vector. The communication network preferably uses high-speed industrial Ethernet or CAN bus, with a communication baud rate of not less than 1Mbps, to ensure the real-time performance and reliability of status information acquisition;
[0165] For scenarios with a large number of inverters (e.g., n>10), a broadcast communication method based on the CAN bus can be adopted. Each inverter module sends its own status data to the bus at a fixed period (e.g., 100μs~500μs). All modules can receive the complete status vector broadcast on the bus, realizing information sharing and synchronization.
[0166] In the state vector S, The current output active power of the i-th inverter is obtained in real time by the power calculation module inside each inverter. The i-th inverter is identified by its operating status code, which includes at least three states: running, standby, and fault. It may also include auxiliary information such as current loop saturation indication, temperature alarm, and communication timeout flag.
[0167] Considering the potential transmission delays and packet loss in the communication network, a data validity verification step is set up at the status receiver of each inverter: for status data that has not been updated for more than two communication cycles, the channel is determined to be faulty, the inverter is temporarily removed from the coordination control queue, and a communication fault alarm is triggered; for status data with obvious jumps (such as power values changing by more than 50% of the rated capacity in adjacent cycles), a first-order low-pass filter or median filter is used for smoothing to avoid outliers affecting the stability of damping distribution;
[0168] Through the above communication and status acquisition mechanism, the multi-inverter state vector S obtained by S4 will be input as a data stream to the damping distribution coefficient calculation module of S4.2;
[0169] S4.2, Based on the rated capacity of each inverter and current output power Calculate the damping distribution coefficient for each inverter:
[0170]
[0171] The design principle of this allocation factor is based on the principle that the larger the power margin, the more damping work it undertakes. When the current output power of an inverter is low, its power margin... The larger the value, the corresponding damping distribution coefficient. The larger the power margin, the more resonant damping the inverter will be; conversely, when the inverter is running close to full load, its power margin is smaller, and the damping task it undertakes is reduced accordingly, thus avoiding the risk of over-modulation or current saturation caused by adding extra control burden under heavy load conditions.
[0172] The sum of the damping distribution coefficients of all inverters satisfies This ensures that the total virtual admittance is fully distributed among all inverters. To prevent an inverter from having an excessively small power margin, causing its distribution factor to approach zero and completely withdraw from damping participation, a lower limit value for the distribution factor can be set. (e.g., 0.05), when calculated according to the above formula At that time, forced removal The excess coefficient is then redistributed proportionally to other inverters to ensure that each online inverter undertakes a certain damping task, maintaining the redundancy and reliability of the system.
[0173] In addition, for Inverters marked as faulty or in standby mode have their damping allocation coefficients set to zero and do not participate in the damping task allocation. The damping allocation coefficient vector calculated in S4.2... As a virtual admittance allocation module that outputs data stream to S4.3.
[0174] S4.3, The virtual admittance generated in S3 According to the damping distribution coefficient Distributed to each inverter:
[0175]
[0176] That is, the virtual admittance command obtained by each inverter is the product of the global virtual admittance and its allocation coefficient. After allocation, the virtual admittance command of each inverter is... The information is transmitted to the local controller of each inverter via a point-to-point or broadcast method through the communication network.
[0177] To ensure that each inverter executes the damping command on the same time base, the communication network must support a time synchronization mechanism (such as the IEEE 1588 precise time protocol or CAN bus timestamp synchronization), so that each inverter can update its control parameters synchronously after receiving the virtual admittance command, avoiding transient circulating current or reduced damping effect due to inconsistent execution time.
[0178] For scenarios with large communication delays, a command pre-storage and delay compensation mechanism can be set locally on each inverter. This means that the received virtual admittance command is bound to a timestamp, and each inverter calculates the delay based on the deviation between the local clock and the command timestamp. The parameter is then updated at a unified target time, thereby eliminating the impact of inconsistent communication delays on coordinated control.
[0179] S4.4 Each inverter receives the corresponding virtual admittance command. They are then embedded into their respective current control loops. The embedding method is consistent with the implementation method in S3.4: each inverter collects the PCC point voltage at its own port in real time. (In a multi-inverter parallel system, the voltage at each inverter port is theoretically the same. However, in practice, there may be slight deviations due to differences in line impedance. It is preferable to collect the voltage values at each port to improve control accuracy.) Calculate the compensation current of this unit. Then, the compensation current is subtracted from the reference command of the current loop to obtain the corrected current command, which is then sent to the current regulator.
[0180] Through the aforementioned allocation and embedding mechanism, each inverter undertakes the resonant damping task proportionally to its power margin, ensuring that the multi-inverter parallel system exhibits coordinated and consistent damping characteristics. When the power margin of a certain inverter changes with the system operating conditions (such as fluctuations in wind farm output causing some inverters' output power to rise or fall), steps S4.1 to S4.3 are executed repeatedly at a fixed period (e.g., 100ms) to dynamically update the damping allocation coefficient and virtual admittance command, ensuring that the damping allocation always matches the current system operating conditions.
[0181] To simplify calculations and improve response speed, the update cycle of the damping distribution coefficient can be adaptively adjusted according to the rate of change of system operating conditions: when a large change rate of system power is detected (such as exceeding 10% / s of rated power), the update cycle is shortened to 50ms; when the system is in steady state operation, the update cycle is extended to 500ms to reduce communication and computing overhead.
[0182] At this point, S4 has completed the coordinated allocation from the global virtual admittance output by S3 to the personalized virtual admittance commands for each inverter, and has embedded the allocated commands into the current control loop of each inverter.
[0183] Through the above mechanism, the multi-inverter parallel system achieves a reasonable distribution of damping tasks. Inverters with large power margins undertake more damping tasks, while inverters with small power margins undertake less damping tasks. This ensures resonance suppression while avoiding the risk of a single inverter being overloaded due to excessive damping tasks.
[0184] The corrected current commands output by each inverter from S4 will serve as the final control commands, driving the PWM modulation stages of each inverter to generate corresponding switching signals, thereby achieving adaptive suppression of wideband resonance in the multi-inverter parallel system. Simultaneously, the output power of each inverter in S4 is updated in real time. and damping distribution coefficient The parameter adaptive update module will be passed to S5 as a data stream, so that S5 can call it when evaluating the system status and adjusting control parameters;
[0185] S5: Control parameters are updated adaptively, as follows:
[0186] S5.1 Real-time monitoring of total harmonic distortion (THD) of grid-connected current and system damping ratio Construct a system state evaluation index vector ,in This represents the maximum amplitude among the oscillation frequency components identified in S1. The total harmonic distortion (THD) is calculated based on the amplitudes of each harmonic and oscillation component obtained from the FFT spectral analysis in S1, using the following formula: ,in The amplitude of the fundamental current. H represents the amplitude of the h-th harmonic or oscillatory component, where H is the highest harmonic of interest (usually up to the 50th order).
[0187] The update cycle of THD should be consistent with the data window length of FFT, preferably updated once every 0.2 seconds;
[0188] System damping ratio The online estimation employs Prony analysis or eigenvalue-based identification methods: by applying a small pulse disturbance to the grid-connected current or extracting the decay envelope of the current response during the system's natural oscillation, the decay coefficient σ and oscillation angular frequency of the oscillation mode are fitted to obtain the results. Then calculate the damping ratio. ;
[0189] In practical implementation, the damping ratio can be estimated using the frequency signals output from each PLL branch in S2 and their amplitude change rates for simplified calculation. When the amplitude of a certain oscillation frequency component exhibits an exponential decay trend over multiple consecutive cycles, the damping ratio of that mode is calculated based on the time constant of amplitude decay. The damping ratio update period can be set to 0.5s to 1s to balance estimation accuracy and computational burden. Maximum oscillation amplitude. Taken directly from the FFT spectral analysis results in S1 that exceed the threshold The maximum value of the amplitude of each oscillation component, i.e. ;
[0190] The three indicators mentioned above characterize the current stability state of the system from three dimensions: harmonic content, oscillation decay rate, and oscillation intensity, respectively, and constitute the system state evaluation index vector. This serves as the comparison and judgment module for data stream input to S5.2;
[0191] S5.2, System status evaluation indicators With the preset stable operating threshold A comparison is needed. The setting of the stable operation threshold needs to be determined based on the system grid connection standards and actual operating requirements: THD threshold. According to national power quality standards such as IEEE 519 or GB / T 14549, a value of 5% is generally used; damping ratio threshold. The damping ratio reflects the system's ability to dampen oscillations, typically ranging from 0.05 to 0.1. When the system damping ratio is below this threshold, it indicates that the system is in a weakly damped state, posing a risk of oscillation divergence; amplitude threshold. Consistent with the amplitude threshold used to identify oscillation components in S1, it is set to 5% to 10% of the rated current amplitude. The comparison logic is as follows: when... and and When a system is determined to be in a stable operating state, it is considered to be in a stable operating state only if all three conditions are met simultaneously. Maintain the current control parameters unchanged;
[0192] When any one of the above three conditions is not met, that is or or ,determination This triggers adaptive parameter updates;
[0193] To prevent frequent false triggers due to measurement noise or transient disturbances, a hysteresis interval and a delayed confirmation mechanism are set in the comparison logic: the trigger parameter update is only confirmed when the system state evaluation index exceeds the threshold for N consecutive sampling periods (N is preferably 3 to 5), thus avoiding unnecessary parameter adjustments caused by abnormal values in a single period.
[0194] S5.3 When the trigger parameter is updated adaptively, it is based on the current dominant oscillation frequency. and system status assessment indicators The virtual impedance reference parameters in S3 can be adjusted online using a fuzzy logic controller or a particle swarm optimization algorithm. , , and the phase-locked loop parameters in S2 The following describes two preferred implementation schemes:
[0195] Option 1: Fuzzy Logic Controller Scheme. The input variable for the fuzzy logic controller is selected as frequency deviation. (in (The inherent resonant angular frequency of the system's LCL filter) and damping ratio deviation (The difference between the current damping ratio and the threshold), the output variable is the adjustment amount of the virtual impedance reference parameter. and the adjustment amount of the phase-locked loop damping coefficient The fuzzy subsets of each input and output variable are divided into five levels: {negative large (NB), negative small (NS), zero (ZO), positive small (PS), positive large (PB)}, and the membership function adopts a triangular or Gaussian function.
[0196] The fuzzy control rules are formulated based on the following engineering experience: when the frequency deviation... When the value is positive (i.e., the oscillation frequency is much higher than the inherent resonant frequency), the virtual resistance Rv0 should be increased to enhance high-frequency damping, while the virtual inductance should be appropriately reduced. To avoid introducing additional phase hysteresis; when the damping ratio deviation When the current damping ratio is positive (i.e., the current damping ratio is severely insufficient), it should be increased simultaneously. and To provide a stronger damping effect; when jitter occurs in the phase-locked loop tracking, appropriately reduce... To narrow the frequency selection bandwidth of SOGI and enhance frequency selectivity;
[0197] The fuzzy inference employs a Mamdani-type inference engine, and defuzzification uses the centroid method to convert the fuzzy output into precise adjustment values. The inputs and outputs of the fuzzy logic controller are normalized to ensure that all variables perform inference operations within the same domain. The updated parameters are as follows: The parameter update step size needs to be constrained with upper and lower limits to prevent excessive single adjustment from exacerbating system oscillations: the adjustment of each parameter is limited to within ±20% of the reference value. The variable universe fuzzy linear active disturbance rejection control method has been applied to the inner current loop control of energy storage converters. By fuzzy processing the error value and error rate of change and outputting a modulation signal, it solves the technical problems of fixed linear active disturbance rejection control parameters and lack of adaptive capability.
[0198] Option 2: Particle Swarm Optimization (PSO) Algorithm. The PSO algorithm uses a comprehensive evaluation function based on the system state evaluation metric EE. As the fitness function, where , , These are the weighting coefficients. Let J be the rated current amplitude of the system, and the optimization objective is to minimize the fitness function J.
[0199] The parameter vector to be optimized is The search range for each parameter is preset based on system parameters and engineering experience. The particle swarm size is set to 20–50, the maximum number of iterations to 30–50, and the inertia weight... The acceleration constant decreases linearly from 0.9 to 0.4. ;
[0200] In each iteration, each particle calculates the corresponding system state evaluation index in the simulation model or actual system based on the current parameter combination, thereby obtaining a fitness value; the particle's velocity and position are iteratively updated according to the standard PSO update formula. To prevent the algorithm from getting trapped in local optima, a mutation operation can be introduced during the iteration process to randomly perturb the particle's position with a certain probability (e.g., 5%).
[0201] Particle swarm optimization (PSO) has proven effective in optimizing virtual impedance parameters. Based on system energy characteristics and eigenvalue indices, PSO optimizes virtual impedance parameters, avoiding the problems of empirical tuning and repeated adjustments required in traditional methods. Virtual resistance particles are determined using the THD of the grid connection point voltage as the fitness function. By iterating repeatedly using velocity and virtual resistance formulas until convergence conditions are met, a fast, accurate, and global search for virtual resistance values can be achieved. Considering the computational complexity of PSO, this approach is preferably implemented during system startup or when significant changes in operating conditions occur (such as changes in grid structure or a significant change in the number of parallel inverters). During steady-state system operation, a fuzzy logic controller scheme is used for small-scale parameter fine-tuning.
[0202] To balance optimization accuracy and computational efficiency, the optimization results can be used as the initial parameters of the fuzzy logic controller after the particle swarm optimization algorithm converges, with subsequent fine-tuning performed online only by the fuzzy logic controller. The virtual impedance optimization control method based on hybrid potential functions and particle swarm optimization algorithm achieves quantitative optimization design of virtual impedance parameters based on system energy characteristics and eigenvalue indices.
[0203] The choice between the two schemes depends on the availability of system computing resources: for systems equipped with high-performance digital signal processors (such as TMS320C6678 or multi-core DSPs), the particle swarm optimization scheme is preferred to obtain globally optimal parameters; for embedded control systems with limited computing resources, the fuzzy logic controller scheme is preferred to ensure real-time performance. Updated parameters , , , As a virtual impedance construction module for the PLL branches in S2 and S3, the data stream is fed back to S2, replacing the original control parameters to form a closed-loop adaptive adjustment. The parameter update adopts an atomic operation method, that is, all parameters are replaced synchronously within the same control cycle, avoiding control chaos caused by asynchronous parameter updates;
[0204] S5.4 After the parameters are updated, repeat S1 to S4, re-execute current sampling, frequency tracking, virtual impedance construction, and damping allocation under the new control parameters, and then re-enter S5.1 to perform system state evaluation and determine... Check if the parameter update is valid. If valid, it indicates the parameter update is effective, the system has stabilized, and the parameter update process exits, maintaining the current parameters and continuing operation; if not... If the condition is still met, then repeat step S5.3 to perform a new round of parameter adjustment until the stability condition is met or the preset maximum number of updates Nmax is reached (preferably 5 to 10 times).
[0205] When the maximum number of updates is reached and the system still fails to meet stability requirements, it is determined that the current system may have a serious fault that cannot be resolved by parameter adjustment (such as hardware damage or communication interruption). At this time, the system protection action is triggered, switching each inverter to standby or fault protection mode and sending an alarm signal to the monitoring system. All data of the parameter update process (including the triggering reason for each update, the parameter value before the update, the parameter value after the update, the system status evaluation index after the update, etc.) are recorded in the system's non-volatile memory for subsequent offline analysis and parameter optimization reference.
[0206] Thus, S5 has completed the evaluation of the execution results from S1 to S4, the adaptive adjustment of parameters, and the closed-loop verification of the effect after adjustment, enabling the entire control strategy to automatically optimize control parameters according to changes in system operating conditions.
[0207] Through the above mechanism, when the system stability deteriorates due to factors such as changes in inverter output power, changes in grid short-circuit ratio, and phase-locked loop parameter offsets, S5 can proactively identify the deterioration of the system state and automatically adjust key control parameters, ensuring that the system maintains sufficient damping margin over a wide frequency range, thereby achieving true adaptive wideband resonance suppression. The final stable parameters output by S5 will be stored in the local controller of each inverter until the next parameter update trigger condition is met, at which point a new round of adaptive adjustment will be performed.
[0208] S6: Evaluation of resonance suppression effect and strategy switching, specifically including:
[0209] S6.1 After completing the above steps, continuously monitor the oscillation suppression effect of the system and calculate the resonance suppression rate. The resonance suppression rate is calculated based on the amplitude of the oscillation frequency components obtained from the FFT spectrum analysis in S1, where... To suppress the maximum amplitude value among the oscillation frequency components identified by S1 before (i.e., when S1 to S5 are not in operation), This refers to the real-time amplitude of the same frequency component at the current moment (i.e., after S1 to S5 are put into operation). To accurately reflect the suppression effect, The value of this parameter needs to be measured and stored in the system register before the control strategy is implemented, serving as a benchmark value for subsequent evaluation; if the system is in a stable operating state without significant oscillations at startup, then... The reference value is set to 1% of the system's rated current amplitude. The formula for calculating the resonance suppression rate is:
[0210]
[0211] When there are multiple oscillation frequency components in the system, calculate the suppression rate corresponding to each frequency component separately. The minimum value among them is taken as the overall resonance suppression rate of the system. This ensures that the evaluation results reflect the worst-case suppression in the system. The update period for the resonance suppression rate is consistent with the update period for the FFT spectrum analysis in S1, preferably once every 0.2s.
[0212] Calculated resonance suppression rate As a data stream input to the comparison and judgment module of S6.2, it is compared with a preset inhibition rate threshold. Comparison. Meanwhile, and The numerical values and their corresponding frequency information are recorded in the system's data log for subsequent offline analysis of the differences in suppression effects of different frequency components and for evaluating the bandwidth coverage of the control strategy. Based on publicly available research, the harmonic content of the grid-connected current can be significantly reduced from 7.03% before and after adopting the impedance reshaping strategy, verifying the feasibility of using resonance suppression rate as an effectiveness evaluation indicator.
[0213] S6.2, If the resonance suppression rate Greater than the preset inhibition rate threshold If the current control strategy is deemed effective, the current control parameters and strategy will remain unchanged. Suppression rate threshold. The settings need to comprehensively consider the system grid connection standards and actual operating requirements: based on the requirements for harmonic distortion rate in national power quality standards such as IEEE 519 or GB / T 14549, and the system's tolerance for oscillation amplitude. The preferred setting is 70%–85%. When the control strategy has suppressed the oscillation amplitude to an acceptable range, the system is in a stable operating state. At this time, the current parameter configuration from S1 to S5 is maintained, and no additional adjustment action is required.
[0214] To avoid frequent fluctuations in the suppression rate around the threshold due to measurement noise or small fluctuations, a hysteresis interval is set in the comparison logic: when From above the threshold to below The strategy switch is triggered only when the time is right. From below the threshold to above Only when the condition is met is the system deemed stable, thus avoiding oscillation switching near the threshold. Under stable operating conditions, S6 continuously monitors the trend of the resonance suppression rate at a low frequency (e.g., once every 1 second). Once a continuous downward trend in the suppression rate is detected (e.g., the suppression rate decreases monotonically within 5 consecutive monitoring cycles), an early warning is issued and the system enters the evaluation process of S6.3, achieving early identification and prevention of oscillation risks.
[0215] S6.3, If the resonance suppression rate Less than the preset inhibition rate threshold If the system is in a new oscillation mode or the operating conditions have changed significantly, the frequency rescanning mechanism in S2 will be activated.
[0216] The triggering conditions for the frequency rescanning mechanism include, but are not limited to, the following:
[0217] (1) The resonance suppression rate remains below the threshold for more than 3 monitoring cycles;
[0218] (2) The FFT spectrum analysis in S1 detected a new oscillation frequency component, which did not appear or had an amplitude below the threshold before suppression;
[0219] (3) The output power of each inverter in S4 undergoes a step change exceeding 20% of its rated capacity;
[0220] (4) System damping ratio in S5 The price has continued to fall below the preset warning threshold.
[0221] After triggering a frequency rescan, the control strategy is re-executed according to the following procedure: First, the center frequencies of each PLL branch in S2 are reset to their initial states, and the frequency search range of each branch is expanded (from ±10Hz to ±50Hz) to cover any new oscillation frequencies that may appear. Second, the FFT spectrum analysis of S1 is re-executed to perform a fine spectrum analysis of the current signal with a higher frequency resolution (extending the data window length from 0.2s to 0.5s and increasing the frequency resolution from 5Hz to 2Hz) to identify new sets of oscillation frequency components. Then, based on the new set of frequency components, the number of PLL branches k and the initial center frequency in S2 are updated, and the entire process from S2 to S5 is re-executed. During the frequency rescan, the original control parameters remain unchanged. Once the new frequency components are locked and the new virtual impedance parameters are calculated, the new parameter configuration is switched synchronously in the next control cycle to achieve a smooth transition of the control strategy.
[0222] According to relevant research, with the strategy of active damper combined with improved feedforward, the resonance of multi-inverter grid-connected system can be effectively suppressed to the allowable range. This further illustrates that when the resonance suppression rate does not meet the requirements, it is necessary to re-achieve effective control of system resonance through strategy switching and parameter readjustment.
[0223] Set an upper limit on the number of rescans. (Preferably 3 to 5 times), if the resonance suppression rate still cannot be satisfied after reaching the upper limit of the number of times. If the system is found to have a hardware failure or a serious stability problem that cannot be solved by control strategies, the system protection action is triggered, each inverter is switched to standby mode and a fault alarm is sent to the monitoring system.
[0224] S6.4 When the system operates stably for an extended period without any new oscillation frequency appearing, the control strategy automatically switches to low-power monitoring mode. Entering low-power monitoring mode requires the following conditions to be met simultaneously:
[0225] (1) Resonance suppression rate Persistently greater than The time exceeded the set time , The preferred setting is 30s to 60s;
[0226] (2) The frequency signals output by each PLL branch in S2 are in The fluctuation range within the time period shall not exceed ±2Hz;
[0227] (3) FFT spectrum analysis in S1 in continuous No new values exceeding the threshold were detected within the specified time. The oscillation frequency component;
[0228] (4) The output power of each inverter in S4 is The rate of change within a time period shall not exceed 5% / s of the rated power;
[0229] When all four conditions above are met simultaneously, the system is determined to have entered a long-term stable state, and the control strategy switches to low-power monitoring mode. Specific measures under low-power monitoring mode include:
[0230] (1) Reduce the execution frequency of FFT spectrum analysis in S1 from once every 0.2s to once every 2s;
[0231] (2) Reduce the update rate of the PLL branch in S2 from once per control cycle to once every 10 control cycles;
[0232] (3) Reduce the calculation frequency of the system state evaluation index in S5 from once every 0.2s to once every 1s;
[0233] (4) Extend the update cycle of the damping distribution coefficient in S4 from 100ms to 500ms.
[0234] Through the above measures, the computational load of the digital signal processor can be reduced by about 60% to 80%, effectively reducing the system's control overhead and power consumption.
[0235] In low-power monitoring mode, S6 continues to monitor the resonance rejection ratio at a reduced frequency. Changes, once detected If the power consumption drops below (ηth−5%) or a new oscillation frequency component is detected, immediately exit the low-power monitoring mode, return to normal operation mode, and re-execute all processes from S1 to S5. The response time for exiting the low-power monitoring mode must be within 100ms to ensure that the system can respond quickly to sudden oscillation events.
[0236] The operating status of the low-power monitoring mode is reported to the monitoring system via the communication network, so that operation and maintenance personnel can understand the current operating mode and control strategy status of the system.
[0237] Existing research has shown that inverter systems can switch from current control mode to low power mode through mode selection, which provides a technical reference for the engineering implementation of low power monitoring mode in this step.
[0238] Thus, S6 has completed the continuous evaluation of the overall control strategy's execution effect, strategy switching decisions, and optimized management of operating modes. Through real-time monitoring and threshold comparison of the resonance suppression rate, S6 ensures that the effectiveness of the control strategy can be verified in a timely manner; through the activation of the frequency rescanning mechanism, S6 gives the system the ability to reconfigure control parameters when operating conditions change; and through intelligent switching of the low-power monitoring mode, S6 minimizes control overhead while ensuring system safety and stability.
[0239] The system operating status output by S6 (including normal operation mode, frequency rescan mode, low power monitoring mode, fault protection mode, etc.) serves as the final policy status information and is uploaded to the upper-level monitoring system via the communication network, allowing maintenance personnel to monitor the system's operational status in real time. Simultaneously, historical data on resonance suppression rate, frequency rescan trigger records, and mode switching logs recorded by S6 are stored in the system's non-volatile memory, providing data support for subsequent system performance evaluation, parameter optimization, and fault diagnosis.
[0240] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A broadband resonant adaptive control method for multiple inverters in parallel, characterized in that, The method includes: S1: Grid-connected point current sampling and feature extraction: Collect the total grid-connected current signal of the multi-inverter parallel system, extract the fundamental angular frequency through coordinate transformation and frequency-locked loop processing, and identify the oscillation frequency components exceeding the amplitude threshold through spectrum analysis to form an oscillation frequency component set. Output the fundamental angular frequency and the oscillation frequency component set as a data stream. S2: Oscillation frequency tracking and adaptive locking. The current signal is input to multiple parallel phase-locked loop branches that correspond one-to-one with each component in the oscillation frequency component set. Each branch tracks the frequency with the corresponding oscillation frequency component as the initial center frequency. The frequency signals output by each branch are weighted and fused to obtain the dominant oscillation frequency. The dominant oscillation frequency is output as a data stream. S3: Based on frequency adaptive impedance dynamic reshaping, a frequency adaptive virtual impedance is constructed according to the dominant oscillation frequency, the virtual impedance is converted into virtual admittance, and filtered by a fundamental notch filter. The filtered virtual admittance is superimposed on the current control loop output of each inverter to realize dynamic reshaping of the grid-connected inverter output impedance, and the corrected current command is output as a data stream. S4: Multi-inverter coordinated control and damping allocation: Collect the output power and operating status of each parallel inverter, calculate the damping allocation coefficient of each inverter based on the rated capacity and current output power of each inverter, allocate the virtual admittance to each inverter according to the damping allocation coefficient, each inverter embeds the allocated virtual admittance into its own current control loop, and outputs the updated output power and damping allocation coefficient of each inverter as a data stream; S5: Control parameters are adaptively updated. The total harmonic distortion rate of the grid-connected current, the system damping ratio, and the maximum oscillation amplitude are monitored in real time. A system state evaluation index vector is constructed. When the system state evaluation index vector exceeds the preset stable operation threshold, the reference parameters of the phase-locked loop in S2 and the virtual impedance in S3 are adjusted online according to the current dominant oscillation frequency and the system state evaluation index vector through a fuzzy logic controller or particle swarm optimization algorithm. The updated parameters are sent back to S2 and S3 as a data stream to replace the original parameters. S6: Resonance suppression effect evaluation and strategy switching. Calculate the resonance suppression rate. If the resonance suppression rate is greater than the preset suppression rate threshold, maintain the current control parameters and strategy unchanged. If the resonance suppression rate is less than the preset suppression rate threshold, activate the frequency rescanning mechanism in S2 and re-execute S2 to S5 to complete the control strategy switching.
2. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 1, characterized in that, In step S1, the grid connection point current sampling and feature extraction specifically include: S1.1 Install a current transformer at the common coupling point to collect the total grid-connected current signal of the multi-inverter parallel system. The sampled data is then sent to the digital signal processor after analog-to-digital conversion. S1.
2. The acquired three-phase current signals are converted to static state using Clark transformation. coordinate system, to obtain Current components in the coordinate system; S1.3, will The current signal in the coordinate system is input to the frequency locking loop module. The frequency locking loop adopts a frequency adaptive structure based on a second-order generalized integrator. It extracts the fundamental angular frequency and the angular frequencies of each harmonic or oscillation component in the grid-connected current through closed-loop adjustment. S1.4 Perform spectrum analysis on the current signal using Fast Fourier Transform, automatically identify frequency components exceeding the preset amplitude threshold, form an oscillation frequency component set by combining all frequency components that meet the amplitude condition and whose frequency does not belong to the fundamental wave, and output the fundamental wave angular frequency and the oscillation frequency component set as data streams to S2 and S3 respectively.
3. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 1, characterized in that, In step S2, the oscillation frequency tracking and adaptive locking specifically include: S2.1 Input the stationary αβ coordinate system current signal output from S1 into k parallel phase-locked loop branches respectively. The initial value of the center angular frequency of each branch is set according to the set of oscillation frequency components identified by the FFT spectrum analysis in S1. Each phase-locked loop branch works independently and in parallel. S2.2 Each phase-locked loop branch adopts a phase-locked loop structure based on a second-order generalized integrator. It extracts the orthogonal component with the same frequency as the center frequency from the input current signal, and inputs it to the PI regulator after Park transformation. The output of the PI regulator is integrated to form the frequency estimate of the branch. S2.3 The frequency signals output by each phase-locked loop branch are synchronously input to the frequency decision module. A data buffer is set at the input of the frequency decision module. The output of each branch is resampled and interpolated based on the system control cycle to ensure that the frequency signals participating in the fusion calculation are time-aligned. S2.4 The frequency decision module performs weighted fusion processing on the frequency signals output by each branch. The weight coefficients are allocated according to the amplitude of each frequency component. The dominant oscillation frequency output by the frequency decision module is the weighted sum of the frequencies of each branch. At the same time, the frequency decision module is also equipped with a numerical validity verification submodule, which sets the weight coefficient of the branch determined to be in a lost-lock state to zero and excludes it from the fusion calculation. S2.5 The dominant oscillation frequency output by the frequency decision module is output as a data stream to S3. At the same time, the dominant oscillation frequency is fed back to the FFT spectrum analysis module of S1 as the frequency reference for spectrum refinement analysis. The frequency output and amplitude information of each phase-locked loop branch are continuously updated and stored in the system status register for S5 to call.
4. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 3, characterized in that, In S2.4, the weighted fusion process specifically involves: the weighting coefficients λi are adjusted according to the magnitude of each frequency component. Distribute, among which Let be the amplitude of the i-th oscillation frequency component, and let be the dominant oscillation frequency output by the frequency decision module. ; in This is the estimated frequency output of the i-th phase-locked loop branch; The frequency decision module is also equipped with hysteresis comparison logic. Only when the amplitude of a certain branch is continuously higher than the amplitude of other branches by a set proportion for N consecutive control cycles will the frequency of that branch be determined as the dominant frequency. In the numerical validity verification submodule, if the frequency value output by a branch deviates from the center frequency of the branch by more than a preset range or the rate of change exceeds a preset limit, the branch is determined to be in a unlocked state.
5. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 1, characterized in that, In S3, the frequency-adaptive dynamic impedance reshaping specifically includes: S3.1, Based on the dominant oscillation frequency output by S2 Construct a frequency-adaptive virtual impedance: Virtual resistance Virtual inductance Virtual capacitor All The function; S3.2, Convert the virtual impedance into virtual admittance: And through the fundamental notch filter After filtering, the transfer function of the fundamental notch filter is: Where m is the notch filter frequency coefficient, 0 <m<2, The fundamental angular frequency extracted by the frequency-locked loop in S1; S3.3, The virtual admittance after notch filtering: This is superimposed on the output of the current control loop of each inverter, and a feedforward branch is added to the current loop control structure of each inverter. The input of this branch is the common coupling point voltage. The transfer function is ; The branch output is a compensation current signal: The compensation current is superimposed on the given end of the current loop in a negative feedback manner; S3.
4. The compensation current is added as a feedforward compensation amount to the reference current command value of the current loop to obtain the corrected current command: The current is fed into the current regulator, which transmits the corrected current commands from each inverter as a data stream to S4.
6. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 1, characterized in that, The multi-inverter coordinated control and damping allocation in S4 specifically includes: S4.
1. Collect the output power Pi and current operating status of each parallel inverter through the communication network to construct the device state vector: The device state vector is input as a data stream to the damping distribution coefficient calculation module; S4.2 Calculate the damping distribution coefficient of each inverter based on its rated capacity Prated,i and current output power Pi. The sum of the damping distribution coefficients of all inverters satisfies ; For inverters whose status is marked as fault or standby, their damping distribution coefficient is set to zero. S4.3, The virtual admittance generated in S3 The damping distribution coefficient ηi is used to allocate power to each inverter. The allocated virtual admittance commands for each inverter are sent to the local controller of each inverter through the communication network. S4.4 Each inverter receives the corresponding virtual admittance command. Each inverter embeds these values into its respective current control loop, and each inverter acquires the PCC point voltage at its own port in real time. Calculate the compensation current of this machine. The compensation current is subtracted from the reference command of the current loop to obtain the corrected current command, which is then sent to the current regulator.
7. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 6, characterized in that, In step S4, the damping distribution coefficient is updated repeatedly at a fixed period, and the fixed period is adaptively adjusted according to the rate of change of system operating conditions. When the system power change rate is detected to exceed 10% / s of the rated power, the update cycle is shortened to 50ms; When the system is in a steady state, the update cycle is extended to 500ms; The communication network is equipped with a data validity verification step. If the status data has not been updated for more than two communication cycles, the channel of the communication network is determined to be invalid, the inverter is temporarily removed from the coordination control queue and a communication fault alarm is triggered. For state data with abrupt changes, first-order low-pass filtering or median filtering is used for smoothing.
8. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 1, characterized in that, In step S5, the adaptive update of the control parameters specifically includes: S5.1 Real-time monitoring of total harmonic distortion (THD) and system damping ratio of grid-connected current. Construct a system state evaluation index vector ; in The system state evaluation index vector is input as a data stream to the comparison and judgment module as the maximum amplitude value among the oscillation frequency components identified in S1. S5.2, The system state evaluation index vector E is compared with the preset stable operation threshold. When comparing, and and If the system is determined to be in a stable operating state, the current control parameters are maintained unchanged. If any of the above conditions are not met, a judgment will be made. This triggers adaptive parameter updates; S5.3 When the trigger parameter is updated adaptively, it is based on the current dominant oscillation frequency. The virtual impedance reference parameters in S3 are adjusted online using a fuzzy logic controller or particle swarm optimization algorithm, along with the system state evaluation index vector E. , , and the phase-locked loop parameters in S2 The updated parameters are transmitted back as data streams to each phase-locked loop branch of S2 and the virtual impedance building module of S3 to replace the original parameters. S5.4 After parameter update, repeat S1 to S4, re-execute current sampling, frequency tracking, virtual impedance construction, and damping allocation under the new control parameters, and then re-enter S5.1 for system state evaluation. If successful, exit the parameter update process; if... If the condition is still met, S5.3 will be executed again for a new round of parameter adjustment until the stability condition is met or the preset maximum number of updates is reached.
9. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 1, characterized in that, The resonance suppression effect evaluation and strategy switching in S6 specifically include: S6.1 After completing S1-S5, continuously monitor the oscillation suppression effect of the system and calculate the resonance suppression rate. The formula for calculating the resonance suppression rate is: in The maximum amplitude value among the various oscillation frequency components identified before the control strategy is implemented. The real-time amplitude of the same frequency component at the current moment is given. When there are multiple oscillation frequency components in the system, the suppression rate corresponding to each frequency component is calculated separately, and the minimum value is taken as the overall resonance suppression rate of the system. S6.2, if the resonance suppression rate Greater than the preset inhibition rate threshold If the current control strategy is effective, the current control parameters and strategy will remain unchanged; if the resonance suppression rate is... Less than the preset inhibition rate threshold If the system has a new oscillation mode or its operating conditions have changed significantly, the frequency rescanning mechanism in S2 will be activated. S6.3 After activating the frequency rescanning mechanism, the center frequency of each phase-locked loop branch in S2 is reset to the initial state and the frequency search range of each branch is expanded. The FFT spectrum analysis of S1 is re-executed to finely identify the new set of oscillation frequency components. The number of phase-locked loop branches and the initial center frequency in S2 are updated according to the new set of frequency components. All processes from S2 to S5 are re-executed. During the frequency rescanning period, the original control parameters remain unchanged. After the new frequency components are locked and the new virtual impedance parameters are calculated, the new parameter configuration is switched synchronously. S6.4 When the system is in a stable operating state for a long time and no new oscillation frequency appears, the control strategy automatically switches to the low power monitoring mode. In the low power monitoring mode, the execution frequency of FFT spectrum analysis in S1, the update rate of the phase-locked loop branch in S2, the calculation frequency of the system state evaluation index in S5, and the update cycle of the damping allocation coefficient in S4 are reduced. Once the resonance suppression rate is detected to drop below the threshold or a new oscillation frequency component is detected, the system immediately exits the low power monitoring mode, returns to the normal operation mode, and re-executes all processes from S1 to S5.
10. The broadband resonant adaptive control method for multiple inverters in parallel as described in claim 9, characterized in that, In S6.4, the four conditions for switching the low-power monitoring mode are: Resonance suppression rate Persistently greater than The time exceeded the set time And the frequency signals output by each phase-locked loop branch in S2 are The fluctuation range within the time period does not exceed ±2Hz, and the FFT spectrum analysis in S1 is continuous. No new oscillation frequency components exceeding the threshold Ath were detected within the specified time, and the output power of each inverter in S4 was within the specified range. When the rate of change within a time period does not exceed 5% / s of the rated power, and all four conditions above are met simultaneously, the system is determined to have entered a long-term stable state and is switched to low power monitoring mode; the response time to exit low power monitoring mode is required to be within 100ms.
