Method for suppressing circulating current of multiple intelligent string photovoltaic inverters in parallel connection

Abstract

The application relates to the technical field of photovoltaic power generation control, and discloses a circulating current suppression method for intelligent group string type photovoltaic inverters in multiple-machine parallel connection, wherein the method comprises the following steps: establishing a harmonic fingerprint mapping table of each inverter under multiple power levels; based on real-time circulating current signals and the harmonic fingerprint under the current power, calculating the circulating current contribution weight of each inverter through non-negative matrix decomposition, generating a differentiated feedback compensation distribution scheme; inputting the differentiated compensation task quantity into a repetitive controller to generate a harmonic compensation voltage instruction; calculating a photovoltaic power prediction confidence index; generating a differentiated feedforward compensation distribution scheme based on the harmonic fingerprint under the predicted power; and adaptively weighting and fusing the feedforward and feedback distribution schemes based on the predicted confidence index.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation control technology, and more specifically, to a method for suppressing circulating current in a multi-unit parallel connection of intelligent string photovoltaic inverters. Background Technology

[0002] In a smart string photovoltaic power generation system, multiple string inverters are connected in parallel to a common AC bus. Due to hardware differences such as dead time settings, switching device characteristics, and filter parameter deviations, each inverter generates different harmonic components during operation, forming circulating currents on the parallel bus.

[0003] Existing circulating current suppression methods collect the circulating current signal from the parallel bus, perform spectral decomposition, extract the fundamental frequency, third and fifth harmonics, and carrier sideband components, and uniformly generate compensation commands which are then sent to all inverters for execution. Some systems further incorporate photovoltaic output prediction for feedforward compensation, which is then superimposed on real-time feedback compensation at a fixed ratio to output a comprehensive control command.

[0004] However, under conditions of rapid fluctuations in photovoltaic power (such as sudden changes in irradiance caused by cloud cover), the frequent changes in the inverter's operating point lead to changes in its dead zone effect and core saturation, resulting in a significant drift in the harmonic spectrum characteristics. Existing technologies have the following technical problems: First, the harmonic characteristic reference established at a single rated operating point becomes invalid after power changes, making it impossible to accurately identify the contribution of each inverter to the bus circulating current at the new operating point, leading to a decrease in circulating current source matching accuracy; Second, issuing the same compensation command to all inverters results in unnecessary compensation actions for inverters with small harmonic contributions, increasing switching losses, while inverters with large harmonic contributions receive insufficient compensation, limiting the overall circulating current suppression effect; Third, the feedforward and feedback compensations are superimposed in a fixed ratio. During periods of inaccurate power prediction, the feedforward component becomes an interference source, conflicting with the feedback component and causing control output oscillations. Summary of the Invention

[0005] This invention provides a circulating current suppression method for multiple intelligent string photovoltaic inverters connected in parallel, which solves the circulating current problem caused by the difference in harmonic characteristics when multiple string inverters are connected in parallel in related technologies, and achieves precise differentiated circulating current suppression.

[0006] This invention discloses a circulating current suppression method for multiple parallel intelligent string photovoltaic inverters, comprising the following steps: controlling each string inverter to operate independently at multiple preset power levels, collecting the output current waveforms of each inverter at each power level, performing a fast Fourier transform on each output current waveform, extracting the amplitude and phase of each harmonic component, and establishing a multi-operating-point harmonic fingerprint mapping table for each inverter using the power level as the index key and the amplitude and phase of each harmonic component as the feature value; The system collects the circulating current signal of the parallel bus in real time and extracts the circulating current harmonic spectrum vector. It obtains the current operating power of each inverter, queries the multi-operating-point harmonic fingerprint mapping table based on the current operating power, and obtains the harmonic fingerprint characteristics of each inverter under the current power through interpolation. It performs non-negative matrix decomposition matching on the circulating current harmonic spectrum vector and the current harmonic fingerprint characteristics of each inverter, calculates the circulating current contribution weight of each inverter, and generates the differentiated feedback compensation task of each inverter based on the circulating current contribution weight. The differentiated feedback compensation task of each inverter is input into the corresponding repetitive controller to generate harmonic compensation voltage commands for each inverter. The harmonic compensation voltage commands are then superimposed into the voltage control loop of each inverter to perform directional circulating current suppression.

[0007] Furthermore, the multiple preset power levels are discrete power values ​​obtained by dividing the rated power range of each inverter into equal or non-equal intervals; the phase of each harmonic component is determined modulo 2. Operational constraints are reduced to zero to 2. Within the range; the set of harmonic orders of interest in the harmonic fingerprint mapping table includes the fundamental frequency component, the 3rd harmonic, the 5th harmonic, the 7th harmonic, and the carrier sideband component near the switching frequency.

[0008] Furthermore, the step of obtaining the harmonic fingerprint characteristics of each inverter at the current power through interpolation includes: When the current operating power of the i-th inverter is between two adjacent preset power levels in the multi-operating-point harmonic fingerprint mapping table, the ratio of the difference between the current operating power and the lower preset power level to the difference between the two adjacent preset power levels is used as the interpolation coefficient to perform linear interpolation on the harmonic fingerprint features corresponding to the two adjacent preset power levels, thereby obtaining the harmonic fingerprint features of the inverter at the current power.

[0009] Furthermore, the step of performing nonnegative matrix factorization matching, calculating the circulating current contribution weight of each inverter, and generating differentiated feedback compensation task quantities for each inverter based on the circulating current contribution weight includes: Obtain the harmonic fingerprint amplitude vector of each inverter at the current power, and form a fingerprint basis matrix; with the objective of minimizing the squared 2-norm of the difference between the circulating harmonic spectrum vector and the product of the fingerprint basis matrix and the contribution intensity coefficient vector, solve the contribution intensity coefficient vector under the constraint that each element of the contribution intensity coefficient vector is non-negative; normalize the contribution intensity coefficient vector by dividing each element by the sum of all elements to obtain the feedback contribution weight of each inverter; multiply the feedback contribution weight of each inverter by the circulating harmonic spectrum vector to obtain the feedback compensation task of each inverter.

[0010] Furthermore, when the feedback contribution weight of a certain inverter is lower than the preset contribution threshold, the feedback compensation task of that inverter is set to zero; the contribution threshold ranges from 0.05 to 0.15, and the contribution threshold is set according to the number of inverters connected in parallel.

[0011] Furthermore, the repetitive controller is a controller based on the internal model principle. The repetitive controller includes a delay element with a delay length equal to the number of sampling points in the fundamental period. The repetitive controller accumulates and corrects the input differential feedback compensation task quantity cycle by cycle through the delay element, and outputs an inverse cancellation signal corresponding to each harmonic component in the compensation task quantity in the frequency domain as a harmonic compensation voltage command. The harmonic compensation voltage command is superimposed with the original voltage reference signal of the corresponding inverter to obtain a corrected voltage control command. Each inverter drives the power switching device to perform switching actions based on its own corrected voltage control command.

[0012] Furthermore, it also includes the following steps: The system compares the photovoltaic power forecast value of the previous forecast period with the actual power observation value of the corresponding period in real time, calculates the forecast error of each forecast period, incorporates the forecast error of each forecast period into an error sliding window, and calculates the forecast confidence index at the current time based on the normalized mean and normalized standard deviation of the forecast error within the error sliding window. The forecast confidence index is equal to 1 plus the reciprocal of the sum of the mean deviation weighting coefficient, the normalized mean and the standard deviation weighting coefficient, and the normalized standard deviation, where the sum of the mean deviation weighting coefficient and the standard deviation weighting coefficient equals 1. The photovoltaic power prediction value at the current moment is obtained, the expected operating power of each inverter is determined, the multi-operating-point harmonic fingerprint mapping table is queried based on the expected operating power, and the harmonic fingerprint features of each inverter at the expected operating point are obtained by interpolation. Based on the harmonic fingerprint features at the expected operating point, a pre-calculated harmonic synthesis vector is constructed by superimposing complex domain vectors. Non-negative matrix decomposition is performed on the pre-calculated harmonic synthesis vector and the expected fingerprint basis matrix to obtain the expected circulating current contribution weight of each inverter. Based on the expected circulating current contribution weight, the feedforward compensation task of each inverter is generated. The prediction confidence index is used as the feedforward weight coefficient, and 1 minus the prediction confidence index is used as the feedback weight coefficient. The feedforward compensation task amount is scaled using the feedforward weight coefficient, and the feedback compensation task amount is scaled using the feedback weight coefficient. The scaled feedforward compensation task amount and the scaled feedback compensation task amount are superimposed to obtain the adaptive differentiated compensation task amount for each inverter. The adaptive differentiated compensation task amount replaces the differentiated feedback compensation task amount and is input to the corresponding repetitive controller.

[0013] Furthermore, the process of constructing the pre-calculated harmonic synthesis vector includes: for each harmonic in the harmonic order set, representing the amplitude and phase of that harmonic at the expected operating point of each inverter as a complex number, summing the complex contributions of all inverters to that harmonic to obtain a synthesized complex number, and taking the modulus of the synthesized complex number as the component corresponding to that harmonic in the pre-calculated harmonic synthesis vector; the length of the error sliding window ranges from 5 to 20 prediction cycles.

[0014] Furthermore, a first-order low-pass filter is applied to the feedforward weight coefficient and the feedback weight coefficient. The feedforward weight coefficient after the first-order low-pass filter is equal to the sum of the filter coefficient multiplied by the feedforward weight coefficient after filtering at the previous time step and 1 minus the filter coefficient multiplied by the feedforward weight coefficient before filtering at the current time step. The feedback weight coefficient after the first-order low-pass filter is equal to 1 minus the feedforward weight coefficient after the first-order low-pass filter. The value range of the filter coefficient is 0.7 to 0.95.

[0015] This invention provides a circulating current suppression system for multiple intelligent string photovoltaic inverters connected in parallel, comprising: The harmonic fingerprinting module is used to control each string inverter to operate independently at multiple preset power levels, collect the output current waveforms of each inverter at each power level, perform fast Fourier transform on each output current waveform, extract the amplitude and phase of each harmonic component, and establish a multi-operating-point harmonic fingerprint mapping table for each inverter using the power level as the index key and the amplitude and phase of each harmonic component as the feature value. The differentiated compensation allocation module is used to collect the circulating current signal of the parallel bus in real time and extract the circulating current harmonic spectrum vector, obtain the current operating power of each inverter, query the multi-operating point harmonic fingerprint mapping table based on the current operating power and obtain the harmonic fingerprint characteristics of each inverter under the current power through interpolation, perform non-negative matrix decomposition matching on the circulating current harmonic spectrum vector and the current harmonic fingerprint characteristics of each inverter, calculate the circulating current contribution weight of each inverter, and generate the differentiated feedback compensation task amount of each inverter based on the circulating current contribution weight; The circulating current suppression execution module is used to input the differentiated feedback compensation task of each inverter into the corresponding repetitive controller, generate harmonic compensation voltage commands for each inverter, and superimpose each harmonic compensation voltage command into the voltage control loop of each inverter to perform directional circulating current suppression.

[0016] This invention solves the problem of circulating current suppression failure caused by harmonic characteristic drift under rapid photovoltaic power fluctuation conditions by establishing a multi-operating-point harmonic fingerprint mapping table, using a non-negative matrix factorization algorithm for differentiated compensation task allocation, and an adaptive feedforward-feedback fusion mechanism based on prediction confidence index. The invention achieves the following technical effects: First, the acquisition of harmonic characteristic benchmarks after power changes no longer relies on a fixed benchmark at a single rated operating point, maintaining the effectiveness of harmonic contribution identification for each inverter; Second, differentiated allocation concentrates compensation resources on the harmonic source inverters, reducing switching losses in inverters with small harmonic contributions while ensuring sufficient compensation for inverters with large harmonic contributions; Third, it automatically reduces the weight of the feedforward component during periods of unreliable power prediction, avoiding control output oscillations caused by conflicts between the feedforward and feedback components. Attached Figure Description

[0017] Figure 1 This is a flowchart of the circulating current suppression method for multiple intelligent string photovoltaic inverters connected in parallel, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram comparing the multi-operating-point harmonic amplitudes of INV-1 and INV-3 provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the feedback contribution weight and compensation allocation of each inverter provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the time series analysis of the prediction error sliding window provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the change of the prediction confidence index as a function of error statistics provided in an embodiment of the present invention; Figure 6 This is a schematic diagram showing the comparison of feedforward and feedback contribution weights provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the effective compensation ratio of each inverter after adaptive weighted fusion provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the INV-3 harmonic fingerprint interpolation and operating point adaptability provided in an embodiment of the present invention. Detailed Implementation

[0018] In intelligent string photovoltaic (PV) power generation systems, multiple string inverters are connected in parallel to a common AC bus. Due to hardware differences such as dead-time settings, switching device characteristics, and filter parameter deviations, each inverter generates differentiated harmonic components during operation, forming circulating currents on the parallel bus. Existing circulating current suppression methods collect the circulating current signal from the parallel bus, perform spectral decomposition, extract the fundamental frequency, 3rd and 5th harmonics, and carrier sideband components, and uniformly generate compensation commands that are then sent to all inverters for execution. Some systems further incorporate PV output prediction for feedforward compensation, which is superimposed on real-time feedback compensation at a fixed ratio to output a comprehensive control command.

[0019] However, under conditions of rapid fluctuations in photovoltaic power (such as sudden changes in irradiance caused by cloud cover), the frequent changes in the inverter's operating point lead to alterations in its dead zone effect and core saturation, resulting in a significant drift in the harmonic spectrum characteristics. Existing technologies suffer from the following three technical problems: First, the harmonic characteristic reference established under a single rated operating point becomes invalid after power changes, making it impossible to accurately identify the contribution of each inverter to the bus circulating current under the new operating point, resulting in a decrease in the accuracy of circulating current source matching.

[0020] Second, issuing the same compensation command to all inverters results in inverters with low harmonic contribution performing unnecessary compensation actions, increasing switching losses, while inverters with high harmonic contribution have insufficient compensation, thus limiting the overall circulating current suppression effect.

[0021] Third, the feedforward and feedback compensations are superimposed in a fixed ratio. During periods when power prediction is inaccurate (such as when there is random cloud cover), the feedforward component becomes a source of interference and conflicts with the feedback component, causing the control output to oscillate. The circulating current suppression effect is actually worse than that of pure feedback control.

[0022] According to an embodiment of this invention, a circulating current suppression method for multiple intelligent string photovoltaic inverters connected in parallel is provided to solve the aforementioned technical problems. It should be understood that the method of this embodiment is executed by a central controller communicatively connected to each parallel inverter. The central controller has a data acquisition interface, signal processing capabilities, and a communication channel for issuing control commands to each inverter. Output current sensors and bus voltage and current sensors of each inverter are deployed as signal acquisition hardware in the parallel system.

[0023] At least one embodiment of the present invention discloses a circulating current suppression method for multiple intelligent string photovoltaic inverters connected in parallel, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect the harmonic characteristics of each inverter at multiple power levels and establish a multi-operating-point harmonic fingerprint mapping table; Each string inverter is controlled to operate independently at multiple preset power levels. The output current waveforms of each inverter at each power level are collected. A fast Fourier transform is performed on each output current waveform to extract the amplitude and phase of each harmonic component. Using the power level as the index key and the amplitude and phase of each harmonic component as the feature value, a multi-operating-point harmonic fingerprint mapping table for each inverter is established.

[0024] Specifically, let the parallel system contain a total of Taiwan inverter, preset power level set is ,in For the total number of preset power levels, for the th Taiwan inverter in the first Power Level The collected harmonic fingerprints are represented as follows: in, For the first Taiwanese inverters at power levels Next The amplitude of the second harmonic component, For the corresponding phase, Let be the set of harmonic orders of interest. Therefore, the th The multi-operating-point harmonic fingerprint mapping table of the inverter is denoted as ,in For power level index, .

[0025] Furthermore, the aforementioned phase The range of values ​​is constrained by Within the interval, when the phase value of the Fast Fourier Transform output exceeds this range, it is converted to a modulo-switched state. The operation maps the phase values ​​to a standard range.

[0026] It should be noted that the aforementioned preset power levels refer to discrete power values ​​obtained by dividing the rated power range of each inverter into equal or unequal intervals. For example, for an inverter with a rated power of... The inverter can convert the power range The system is divided into 10 power levels with a step size of 10% of the rated power. In the low-power range where the operating point varies densely, a smaller step size can be used to obtain a finer harmonic fingerprint.

[0027] It should be noted that the above set of harmonic orders , refers to the harmonic order that contributes significantly to the parallel circulating current. In a three-phase inverter system, It typically includes the fundamental frequency component, the 3rd harmonic, the 5th harmonic, the 7th harmonic, and the carrier sideband component near the switching frequency.

[0028] Step 2: Based on the real-time circulating current signal and the harmonic fingerprint under the current power, calculate the circulating current contribution weight of each inverter and generate a differentiated feedback compensation allocation scheme; The system acquires the circulating current signal from the parallel bus in real time, performs a Fast Fourier Transform on the signal to extract the amplitude and phase of each harmonic component, and obtains the circulating harmonic spectrum vector. It also acquires the current operating power of each inverter, queries the multi-operating-point harmonic fingerprint mapping table for each inverter based on the current operating power, and obtains the harmonic fingerprint characteristics of each inverter within the current power range using a linear interpolation algorithm.

[0029] The circulating harmonic spectrum vector is matched with the current harmonic fingerprint characteristics of each inverter using nonnegative matrix factorization (NMF) to calculate the actual contribution weight of each inverter to each harmonic circulating current. The input of the NMF algorithm is the fingerprint basis matrix composed of the circulating harmonic spectrum vector and the harmonic fingerprint amplitude of each inverter, and the output is the contribution intensity coefficient vector of each inverter. Based on the actual contribution weight, a differentiated harmonic compensation feedback allocation scheme is generated. According to the allocation rule of allocating larger compensation amounts to inverters with high contribution weights and smaller or zero compensation amounts to inverters with low contribution weights, the feedback compensation task of each inverter is obtained.

[0030] Specifically, let the bus circulating harmonic spectrum vector acquired at the current moment be... ,in Set of harmonic orders The number of harmonic components. Based on the current operating power of each inverter. Query the corresponding multi-operating-point harmonic fingerprint mapping table ,when At two adjacent preset power levels and In between, harmonic fingerprint characteristics at the current power are obtained using a linear interpolation algorithm: Obtain the current harmonic fingerprint amplitude vector of each inverter and form a fingerprint basis matrix. , of which Listed as number The harmonic amplitude vector of the inverter at the current power.

[0031] right and Perform nonnegative matrix decomposition to solve for the contribution intensity coefficient vector. , so that: in, The element Characterizing the first The contribution intensity of the inverter to the harmonic spectrum of the bus circulating current.

[0032] Will Normalization is used to obtain the actual contribution weight of each inverter: in, For the first Feedback contribution weight of the inverter For the inverter index during the summation process, . No. Feedback compensation task of Taiwan inverter Represented as: It should be noted that "smaller or zero compensation amount" in the above allocation rules refers to the feedback contribution weight of a certain inverter. Below the preset contribution threshold When this happens, the feedback compensation task of the inverter is set to zero, thereby avoiding unnecessary compensation actions on the inverter.

[0033] Furthermore, the aforementioned contribution threshold The range of values ​​is Contribution threshold The lower limit of the value range ensures that only inverters with minimal contributions are filtered out; contribution threshold. The upper limit of the value range ensures that inverters with actual contributions are not excessively excluded. In practical applications, the contribution threshold... Based on the number of parallel inverters Configure it when When the value is large, take the smaller value to retain enough compensation participating units. When the value is small, a larger value is used to highlight the compensation responsibility of the main harmonic source.

[0034] Step 3: Input the differential compensation task quantity into the repetitive controller to generate differential harmonic compensation voltage commands and execute circulating current suppression; The differentiated harmonic compensation task of each inverter is input into the corresponding repetitive controller. Each repetitive controller generates a personalized harmonic compensation voltage command based on the input differentiated harmonic compensation task. The harmonic compensation voltage command is superimposed on the voltage control loop of each inverter to drive each inverter to perform directional circulating current suppression.

[0035] Specifically, no. The repetitive controller corresponding to the inverter receives the compensation task. The harmonic compensation voltage command is generated by tracking the frequency cycle period by cycle, which is an integer multiple of the fundamental frequency period. .Will With the original voltage reference signal of the inverter Superimpose to obtain the corrected voltage control command. : Each inverter drives the power switching devices to perform switching actions based on its own modified voltage control command, thereby achieving directional suppression of each harmonic component in the bus circulating current.

[0036] It should be noted that the repetitive controller described above is a controller based on the internal model principle. The core structure of the repetitive controller includes a delay element whose delay length is equal to the number of sampling points in the fundamental period. This delay element enables the repetitive controller to have high gain characteristics for periodic signals, accurately tracking and suppressing harmonic components at integer multiples of the fundamental period. The input to the repetitive controller is the compensation task quantity. The compensation task includes the amplitude and phase information of each harmonic component that needs to be suppressed. The repetitive controller accumulates and corrects the input signal cycle by cycle through a delay element, and outputs a harmonic compensation voltage command. The harmonic compensation voltage command corresponds in the frequency domain to the inverse cancellation signal of each harmonic component in the compensation task.

[0037] In this embodiment of the application, in order to further improve the response speed of circulating current suppression under the condition of rapid power fluctuation, the following steps are also included: introducing an adaptive feedforward compensation path based on prediction confidence, so that the circulating current suppression method can obtain advanced compensation capability during the reliable power prediction period and automatically degenerate into feedback-dominated control mode during the unreliable prediction period.

[0038] Step 4: Calculate the confidence index of photovoltaic power prediction based on the sliding window statistic of prediction error; The system compares the photovoltaic power forecast value of the previous forecast period with the actual power observation value of the corresponding period in real time, calculates the forecast error of each forecast period, incorporates the forecast error of each forecast period into the error sliding window and updates the statistics in the error sliding window, and calculates the forecast confidence index at the current time based on the mean and standard deviation of the forecast error in the error sliding window.

[0039] Specifically, let the first The prediction error for each prediction period is ,in This is the predicted photovoltaic power value for this forecast period. This represents the actual power observations for the corresponding time period. Maintain a data set of length [length missing]. Error sliding window Calculate the mean of the prediction error within the error sliding window. and standard deviation : in, This is the index of the prediction period within the error sliding window. .

[0040] It should be noted that before calculating the prediction confidence index, the mean of the prediction error needs to be calculated. and standard deviation Normalization is performed to eliminate the influence of power dimensions on confidence calculation.

[0041] Based on the normalized mean prediction error and standard deviation Calculate the prediction confidence index at the current moment. : in, The mean deviation weighting coefficient is used. The standard deviation weighting factor is... and This is a preset positive number. When both the mean and standard deviation of the prediction error are small, A value close to 1 indicates high prediction reliability; when the mean or standard deviation of the prediction error is large, Approaching 0 indicates low reliability of the prediction.

[0042] Furthermore, the aforementioned mean deviation weighting coefficient Sum of standard deviation weighting coefficients The range of values ​​for are respectively and And satisfy the constraints. This ensures that the mean deviation and volatility of the prediction error have complementary weights in the confidence assessment. When weather changes in the region where the photovoltaic system is located are mainly characterized by systematic shifts, [the following is taken]. To amplify the impact of mean deviation; when weather changes are characterized primarily by random fluctuations, take To amplify the effect of standard deviation.

[0043] It should be noted that the above-mentioned error sliding window length The selection of the error sliding window length is related to the prediction period and the characteristics of photovoltaic power fluctuations. A shorter error sliding window length makes the prediction confidence index more sensitive to recent changes in prediction error, which is suitable for scenarios with rapid power fluctuations; a longer error sliding window length makes the prediction confidence index more stable, which is suitable for scenarios with slow power changes.

[0044] Furthermore, the aforementioned error sliding window length The range of values ​​is Prediction period, error sliding window length The lower limit of the range of values ​​ensures that the statistic has basic representativeness, and the length of the error sliding window... The upper limit of the value range ensures that the error sliding window maintains sufficient tracking sensitivity to changes in power fluctuation characteristics. When the prediction period is 5 minutes, the error sliding window length of 10 corresponds to 50 minutes of historical observation time, which can cover the duration of typical cloud cover events; when the prediction period is 15 minutes, the error sliding window length of 15 corresponds to about 4 hours of historical observation time, which can reflect the weather change trend on a semi-diurnal scale.

[0045] Step 5: Based on the harmonic fingerprint under the predicted power, calculate the expected circulating current contribution weight of each inverter and generate a differentiated feedforward compensation allocation scheme. The photovoltaic power prediction curve for the current moment is obtained. Based on the predicted power level, the multi-operating-point harmonic fingerprint mapping table of each inverter is queried. The harmonic fingerprint characteristics of each inverter at the expected operating point are obtained through a linear interpolation algorithm. Based on the expected harmonic fingerprint characteristics, the expected circulating current contribution weight of each inverter to each harmonic circulating current at the predicted operating point is pre-calculated using a non-negative matrix factorization algorithm. Based on the expected circulating current contribution weight, a differentiated harmonic compensation feedforward allocation scheme is generated.

[0046] Specifically, the predicted photovoltaic power value at the current moment is obtained to determine the expected operating power of each inverter. .based on Query the multi-operating-point harmonic fingerprint mapping table for each inverter The harmonic fingerprint characteristics of each inverter at the expected operating point are obtained using the same linear interpolation algorithm as in step 2. To form the expected fingerprint base matrix .

[0047] based on The amplitude and phase of the expected harmonic fingerprints of each inverter are vector-superimposed to form a pre-calculated harmonic synthesis vector. .

[0048] Furthermore, the aforementioned pre-calculated harmonic synthesis vector The construction process is based on the principle of vector superposition in the complex field. For the th The first harmonic component represents the contribution of each inverter to the first harmonic at the expected operating point in complex form. Taiwan inverter for the first The complex contribution of the subharmonic is ,in For amplitude, For phase, The imaginary unit. For all Taiwan inverter Summing the complex numbers of the subharmonic contributions yields the composite complex number. ,in For inverter index, The modulus of the synthesized complex number is taken as the first value of the pre-calculated harmonic synthesis vector. Each component The modulus reflects the value at the expected operating point. The combined amplitude of the second harmonic on the bus. Repeat the above process for all harmonic orders of interest to construct a complete pre-calculated harmonic composition vector. ,in This indicates transpose.

[0049] right and Perform nonnegative matrix factorization to obtain the expected circulating current contribution weights of each inverter at the predicted operating point. Based on feedforward contribution weights Following the same differentiated allocation rules as in step 2, the feedforward compensation task quantity for each inverter is generated. : Step 6: Based on the prediction confidence index, adaptively weightedly fuse the feedforward and feedback allocation schemes to generate adaptive differentiated compensation task volume; Based on steps 2 and 5, the prediction confidence index is mapped to feedforward weight coefficient and feedback weight coefficient. The feedforward compensation allocation scheme is scaled using the feedforward weight coefficient, and the feedback compensation allocation scheme is scaled using the feedback weight coefficient. The two scaled allocation schemes are superimposed and merged to generate the adaptive differentiated harmonic compensation task for each inverter. The adaptive differentiated harmonic compensation task replaces the feedback compensation task in step 2 and is input into the corresponding repetitive controller in step 3.

[0050] Specifically, the prediction confidence index Mapped to feedforward weight coefficients and feedback weight coefficient : The feedforward compensation task and feedback compensation task of each inverter are scaled and summed to obtain the first... The adaptive differential harmonic compensation workload of the inverter : Therefore, when predicting confidence index At higher levels, The feedforward compensation allocation scheme is dominant in the fusion result due to its relatively large size, utilizing the leading information from power prediction to achieve advance compensation; when the prediction confidence index is relatively large... At lower levels, The feedback compensation allocation scheme is dominant in the fusion result, and real-time circulation measurement is relied upon to ensure control stability.

[0051] It should be noted that the sum of the feedforward weight coefficient and the feedback weight coefficient is 1, which means that the fusion process is a convex combination, and the total amount of compensation tasks remains within the range defined by the feedforward compensation allocation scheme and the feedback compensation allocation scheme.

[0052] Furthermore, to avoid frequent weight switching caused by short-term fluctuations in the prediction confidence index, [the following is omitted as it is not relevant to the main point]. and A first-order low-pass filter is applied to limit the rate of change of the weight coefficients. The feedforward weight coefficients after first-order low-pass filtering. Represented as: in, These are the filter coefficients, and their values ​​range from [value range missing]. , The closer the value is to 1, the smoother the weight change. The corresponding feedback weight coefficient is updated as follows: .

[0053] Furthermore, the above filtering coefficients The range of values ​​is Filter coefficients The lower limit of the value range ensures that the first-order low-pass filter has a certain response speed to sudden changes in weights, and the filter coefficients The upper limit of the value range ensures that the weight switching process is smooth enough to avoid abrupt changes in the compensation instruction. When the prediction period is short, the upper limit is set to... Approaching the lower limit to maintain the weight's responsiveness to changes in confidence; when the prediction period is long, take... Approaching the upper limit helps to suppress the excessive influence of single prediction errors on the weights.

[0054] The circulating current suppression method for multiple parallel intelligent string photovoltaic inverters provided in this embodiment solves the problem of circulating current suppression failure caused by harmonic characteristic drift under rapid power change conditions from three levels.

[0055] First, in step 1 of this implementation, a multi-operating-point harmonic fingerprint mapping table is established using power level as the index key, pre-collecting and storing the harmonic spectrum characteristics of each inverter at different power levels. Therefore, in steps 2 and 5, when the operating power of the inverter deviates from the rated operating point due to changes in irradiance, the effective harmonic fingerprint characteristics of each inverter at the current or expected operating point can be obtained by querying the multi-operating-point harmonic fingerprint mapping table and performing interpolation between adjacent power levels. Thus, the acquisition of the harmonic characteristic reference after power change no longer depends on a fixed reference at a single rated operating point, and the circulating current source tracing matching process can continuously track changes in the operating point, maintaining the effectiveness of harmonic contribution identification for each inverter.

[0056] Secondly, in both steps 2 and 5 of this implementation, a non-negative matrix factorization algorithm is used to calculate the contribution weight of each inverter to the bus circulating harmonics, and differentiated compensation tasks are allocated based on the contribution weight. Therefore, inverters with larger harmonic contributions are assigned a larger compensation task and bear the main responsibility for circulating current suppression, while inverters with smaller harmonic contributions are assigned a smaller or zero compensation task, avoiding unnecessary compensation actions. Compared to uniformly issuing the same compensation command, differentiated allocation concentrates compensation resources on the harmonic source inverters, reducing switching losses caused by unnecessary compensation performed by inverters with small harmonic contributions, while ensuring that inverters with large harmonic contributions receive sufficient compensation.

[0057] Furthermore, in this embodiment, the prediction confidence index is calculated based on the error sliding window statistic of the prediction error in step 4, and in step 6, the prediction confidence index is mapped to the weight coefficients of the feedforward and feedback components for adaptive fusion. Therefore, during periods when photovoltaic power prediction is reliable, the feedforward weight coefficient is larger, and the advance compensation information of the feedforward path dominates the fusion result, enabling the system to acquire the ability to respond to power changes in advance. During periods when photovoltaic power prediction is unreliable (such as when random cloud cover causes an increase in prediction error), the prediction confidence index decreases, the feedback weight coefficient automatically increases, and the system degenerates into a feedback-dominated control mode, relying on real-time circulating current measurements for compensation. Thus, the weight of the feedforward component is adaptively reduced during periods of prediction inaccuracy, avoiding the situation where the feedforward component acts as an interference source and conflicts with the feedback component, leading to oscillations in the control output.

[0058] A photovoltaic power station (code: PV-STATION-A) is equipped with four string inverters (codes: INV-1 to INV-4) with a rated power of 50kW, connected in parallel to the same AC bus (380V, 50Hz). The power station is located in a hilly area with frequently changing cloud cover, and sudden changes in irradiance occur approximately 6 to 8 times per day. Due to different manufacturing batches, the dead time of each inverter is set between 2.1μs and 2.8μs, and the filter inductor parameter deviation is approximately 5% to 12%, generating differentiated harmonic components and forming bus circulating current during parallel operation. The central controller continuously collects the operating data of each inverter and the bus circulating current signal with a prediction cycle of 5 minutes. The harmonic order set of the system of interest includes the fundamental frequency (1st), 3rd, 5th, and 7th harmonics and carrier sideband components (a total of Q=5 components), and the number of inverters connected in parallel is N=4.

[0059] The central controller sequentially controls INV-1 to INV-4 to operate in individual grid-connected mode. It acquires output current waveforms at six preset power levels (5kW, 10kW, 20kW, 30kW, 40kW, and 50kW) in 10% increments of the rated power of 50kW, and performs Fast Fourier Transform to extract the amplitude and phase of each harmonic component. The harmonic fingerprints of INV-1 and INV-3 at power levels of 30kW and 40kW are shown below as examples.

[0060] Table 1. Multi-operating-point harmonic fingerprint data of INV-1 and INV-3 (partial): Therefore, the multi-operating-point harmonic fingerprint mapping table of INV-1 stores the harmonic characteristics at all six power levels using the power level as the index key. The mapping table for INV-3 is similar, as are those for the other inverters. It can be observed that the amplitudes of the 3rd and 5th harmonics of INV-3 at all power levels are significantly higher than those of INV-1, reflecting that INV-3 has stronger low-order harmonic emission characteristics due to its larger dead time (2.8μs). This difference will be quantified and identified in subsequent steps.

[0061] At 10:23 AM on a certain day in 20XX, partial cloud cover caused a sudden drop in irradiance. The current operating power of the four inverters was: INV-1 35kW, INV-2 33kW, INV-3 36kW, and INV-4 34kW. The central controller acquired the bus circulating current signal in real time and performed a fast Fourier transform to obtain the circulating current harmonic spectrum vector v.

[0062] Taking INV-3 as an example, its current power of 36kW is between the adjacent preset power levels of 30kW and 40kW. Linear interpolation is performed on the amplitude of the third harmonic: After interpolating all five harmonic components of the four inverters, a fingerprint basis matrix W (5 rows × 4 columns) is formed. Non-negative matrix decomposition is performed on v and W to solve for the contribution intensity coefficient vector h, and then normalization is performed to obtain the feedback contribution weight of each inverter.

[0063] Table 2. Feedback contribution weights and compensation task allocation for each inverter: The feedback contribution weight of INV-2, 0.029, is less than the threshold of 0.08. Therefore, the feedback compensation task of INV-2 is set to zero to avoid it performing unnecessary compensation actions. INV-3 has the highest contribution weight (0.541) and will undertake the largest feedback compensation task, i.e., 0.541×v.

[0064] INV-1, INV-3, and INV-4 each receive their respective feedback compensation task quantities and input them into the corresponding repetitive controller. Taking INV-3 as an example, its repetitive controller receives the feedback compensation task quantities (including the amplitude and phase information of the 3rd, 5th, and 7th harmonics and carrier sideband components), and accumulates and corrects them cycle by cycle through a delay element with a delay length equal to the number of sampling points in the fundamental frequency period (200 sampling points at a 50Hz fundamental frequency and a 10kHz sampling rate), and outputs the harmonic compensation voltage command.

[0065] The revised voltage control command is as follows:

[0066] Since the compensation task of INV-2 is zero, its harmonic compensation voltage command is zero, and the voltage control command remains unchanged at the original reference value, without generating any additional switching action.

[0067] The system uses a 5-minute prediction period and an error sliding window length of L=10 (corresponding to 50 minutes of historical observations). At 10:23 (corresponding to prediction period t=126), the prediction error sequence in the current window comes from records of 10 consecutive prediction periods, and this period happens to have experienced a cloud cover process.

[0068] Table 3. Prediction error records within the error sliding window (prediction periods 117 to 126): Calculate the mean error within the window: Using a total power of 200kW as the normalization benchmark, the mean normalization error is 7.29 / 200 = 0.0365, and the standard deviation of the normalization error is 4.51 / 200 = 0.0226. The weighting coefficient α is set to 0.5. =0.5 (balancing systematic shifts and random fluctuations in the region), calculate the prediction confidence index: The above calculation results correspond to the period with relatively light cloud cover. Under the larger cloud cover condition at 10:23, both the mean and standard deviation of the normalized error increase significantly. Substituting these values ​​into the same formula, the actual confidence score is calculated as follows: (126) = 0.38, which will be used directly in step 6.

[0069] The central controller obtains the photovoltaic power forecast for the next forecast period (10:28). The expected operating power of each inverter is: INV-1 42kW, INV-2 40kW, INV-3 43kW, and INV-4 41kW. Based on the multi-operating-point harmonic fingerprint mapping table of each inverter, the same linear interpolation algorithm as in step 2 is used to obtain the harmonic fingerprint features of each inverter at the expected operating point, forming the expected fingerprint base matrix.

[0070] Taking the third harmonic of INV-3 at a expected power of 43kW as an example, 43kW falls between 40kW and 50kW. Referring to Table 1, the amplitude of the third harmonic of INV-3 at 50kW is 4.21A. The interpolation result is: Complex domain vector superposition is performed on all harmonic components of all inverters to construct a pre-calculated harmonic synthesis vector, and then non-negative matrix decomposition is performed to obtain the feedforward contribution weight.

[0071] Table 4. Feedforward contribution weights and feedforward compensation task allocation for each inverter: INV-3 remains the primary harmonic contributor at the predicted operating point. The feedforward weight (0.558) is highly consistent with the feedback weight (0.541) in step 2, indicating that the harmonic source identification results maintain consistency between the predicted operating point and the current operating point. INV-2 is also set to zero compensation in the feedforward path.

[0072] Current time prediction confidence index (126) = 0.38 (cloud cover causes large prediction error), filter coefficient =0.85 (prediction period is 5 minutes, taking the median value). Assuming the feedforward weight after filtering at the previous time step is 0.62, perform a first-order low-pass filter: Taking INV-3 as an example, calculate the workload of adaptive differential harmonic compensation: Table 5. Results of adaptive differential compensation workload fusion for each inverter: because The value of (126) = 0.38 is relatively low, and the feedback weight (0.416) has been significantly increased, indicating that the system is biased towards a feedback-dominated mode. If the prediction confidence further decreases to below 0.10, the feedback weight will converge to close to 1, and the system will completely degenerate into feedback-dominated control. The feedforward component will no longer have a substantial impact on the compensation command, thereby avoiding inaccurate prediction information from interfering with the circulation suppression effect.

[0073] The integrated adaptive differential compensation task replaces the pure feedback compensation task in step 2, and is input to the repetitive controller corresponding to each inverter in step 3 to generate the final harmonic compensation voltage command and superimpose it onto the voltage control loop of each inverter to perform directional circulating current suppression.

[0074] The data flow logic of the entire implementation process is as follows: Step 1: In the offline stage, harmonic fingerprints of 4 inverters at 6 power levels are collected and a mapping table is established (including 50kW operating point data, providing complete data support for the interpolation calculation in Step 5), providing a harmonic characteristic benchmark for operating point adaptation for all subsequent online steps; Step 2: The mapping table is queried online and the current operating point fingerprint is obtained through linear interpolation. INV-3 is identified as the main harmonic contributor (weight 0.541) through non-negative matrix decomposition, while INV-2 is set to zero due to its low weight (0.029), generating a differentiated feedback compensation allocation; Step 3: The differentiated task quantity is input into the repeating controller of each inverter to realize... Directional circulation suppression; Step 4 calculates the confidence index (0.38) based on the error statistics of 10 historical prediction periods, using a weighted combination of the normalized mean and standard deviation of the error, to quantify the unreliability of the prediction under the current cloud cover condition; Step 5 generates feedforward compensation allocation based on the prediction power lookup mapping table. The INV-3 feedforward weight (0.558) and feedback weight (0.541) are highly consistent, verifying the consistency of harmonic source identification; Step 6 maps the confidence index to a weight pair (0.584, 0.416) through a first-order low-pass filter, performs weighted fusion of the feedforward and feedback schemes, and generates the final adaptive differentiated compensation task amount, which is then input back to Step 3 for execution. The entire process, from offline fingerprint acquisition, real-time circulation measurement, power prediction error statistics to compensation weight fusion output, forms a complete and logically self-consistent data flow link.

[0075] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for suppressing circulating current in a multi-unit parallel intelligent string photovoltaic inverter, characterized in that, Includes the following steps: Control each string inverter to operate independently at multiple preset power levels, collect the output current waveforms of each inverter at each power level, perform fast Fourier transform on each output current waveform, extract the amplitude and phase of each harmonic component, and establish a multi-operating-point harmonic fingerprint mapping table for each inverter using the power level as the index key and the amplitude and phase of each harmonic component as the feature value. The system collects the circulating current signal of the parallel bus in real time and extracts the circulating current harmonic spectrum vector. It obtains the current operating power of each inverter, queries the multi-operating-point harmonic fingerprint mapping table based on the current operating power, and obtains the harmonic fingerprint characteristics of each inverter under the current power through interpolation. It performs non-negative matrix decomposition matching on the circulating current harmonic spectrum vector and the current harmonic fingerprint characteristics of each inverter, calculates the circulating current contribution weight of each inverter, and generates the differentiated feedback compensation task of each inverter based on the circulating current contribution weight. The differentiated feedback compensation task of each inverter is input into the corresponding repetitive controller to generate harmonic compensation voltage commands for each inverter. The harmonic compensation voltage commands are then superimposed into the voltage control loop of each inverter to perform directional circulating current suppression.

2. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 1, characterized in that, The multiple preset power levels are discrete power values ​​obtained by dividing the rated power range of each inverter into equal or non-equal intervals; the phase of each harmonic component is obtained by modulo 2... Operational constraints from zero to 2 Within the range; the set of harmonic orders of interest in the harmonic fingerprint mapping table includes the fundamental frequency component, the 3rd harmonic, the 5th harmonic, the 7th harmonic, and the carrier sideband component near the switching frequency.

3. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 1, characterized in that, The process of obtaining the harmonic fingerprint characteristics of each inverter at the current power through interpolation includes: When the current operating power of the i-th inverter is between two adjacent preset power levels in the multi-operating-point harmonic fingerprint mapping table, the ratio of the difference between the current operating power and the lower preset power level to the difference between the two adjacent preset power levels is used as the interpolation coefficient to perform linear interpolation on the harmonic fingerprint features corresponding to the two adjacent preset power levels, thereby obtaining the harmonic fingerprint features of the inverter at the current power.

4. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 1, characterized in that, The process of performing nonnegative matrix factorization matching, calculating the circulating current contribution weight of each inverter, and generating differentiated feedback compensation task quantities for each inverter based on the circulating current contribution weight includes: Obtain the harmonic fingerprint amplitude vector of each inverter at the current power, and form a fingerprint basis matrix; with the objective of minimizing the squared 2-norm of the difference between the circulating harmonic spectrum vector and the product of the fingerprint basis matrix and the contribution intensity coefficient vector, solve the contribution intensity coefficient vector under the constraint that each element of the contribution intensity coefficient vector is non-negative; normalize the contribution intensity coefficient vector by dividing each element by the sum of all elements to obtain the feedback contribution weight of each inverter; multiply the feedback contribution weight of each inverter by the circulating harmonic spectrum vector to obtain the feedback compensation task of each inverter.

5. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 4, characterized in that, When the feedback contribution weight of a certain inverter is lower than the preset contribution threshold, the feedback compensation task of that inverter is set to zero; the contribution threshold ranges from 0.05 to 0.15, and the contribution threshold is set according to the number of inverters connected in parallel.

6. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 1, characterized in that, The repetitive controller is a controller based on the internal model principle. The repetitive controller includes a delay element with a delay length equal to the number of sampling points in the fundamental period. The repetitive controller accumulates and corrects the input differential feedback compensation task quantity cycle by cycle through the delay element, and outputs an inverse cancellation signal corresponding to each harmonic component in the compensation task quantity in the frequency domain as a harmonic compensation voltage command. The harmonic compensation voltage command is superimposed with the original voltage reference signal of the corresponding inverter to obtain a corrected voltage control command. Each inverter drives the power switching device to perform switching actions based on its own corrected voltage control command.

7. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel operation according to any one of claims 1 to 6, characterized in that, It also includes the following steps: The system compares the photovoltaic power forecast value of the previous forecast period with the actual power observation value of the corresponding period in real time, calculates the forecast error of each forecast period, incorporates the forecast error of each forecast period into an error sliding window, and calculates the forecast confidence index at the current time based on the normalized mean and normalized standard deviation of the forecast error within the error sliding window. The forecast confidence index is equal to 1 plus the reciprocal of the sum of the mean deviation weighting coefficient, the normalized mean and the standard deviation weighting coefficient, and the normalized standard deviation, where the sum of the mean deviation weighting coefficient and the standard deviation weighting coefficient equals 1. The photovoltaic power prediction value at the current moment is obtained, the expected operating power of each inverter is determined, the multi-operating-point harmonic fingerprint mapping table is queried based on the expected operating power, and the harmonic fingerprint features of each inverter at the expected operating point are obtained by interpolation. Based on the harmonic fingerprint features at the expected operating point, a pre-calculated harmonic synthesis vector is constructed by superimposing complex domain vectors. Non-negative matrix decomposition is performed on the pre-calculated harmonic synthesis vector and the expected fingerprint basis matrix to obtain the expected circulating current contribution weight of each inverter. Based on the expected circulating current contribution weight, the feedforward compensation task of each inverter is generated. The prediction confidence index is used as the feedforward weight coefficient, and 1 minus the prediction confidence index is used as the feedback weight coefficient. The feedforward compensation task amount is scaled using the feedforward weight coefficient, and the feedback compensation task amount is scaled using the feedback weight coefficient. The scaled feedforward compensation task amount and the scaled feedback compensation task amount are superimposed to obtain the adaptive differentiated compensation task amount for each inverter. The adaptive differentiated compensation task amount replaces the differentiated feedback compensation task amount and is input to the corresponding repetitive controller.

8. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 7, characterized in that, The process of constructing the pre-calculated harmonic synthesis vector includes: for each harmonic in the harmonic order set, representing the amplitude and phase of that harmonic at the expected operating point of each inverter as a complex number, summing the complex contributions of all inverters to that harmonic to obtain a synthesized complex number, and taking the magnitude of the synthesized complex number as the component corresponding to that harmonic in the pre-calculated harmonic synthesis vector; the length of the error sliding window ranges from 5 to 20 prediction cycles.

9. The circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in claim 7, characterized in that, The feedforward weight coefficient and the feedback weight coefficient are subjected to a first-order low-pass filter. The feedforward weight coefficient after the first-order low-pass filter is equal to the sum of the filter coefficient multiplied by the feedforward weight coefficient after filtering at the previous time step and 1 minus the filter coefficient multiplied by the feedforward weight coefficient before filtering at the current time step. The feedback weight coefficient after the first-order low-pass filter is equal to 1 minus the feedforward weight coefficient after the first-order low-pass filter. The value range of the filter coefficient is 0.7 to 0.

95.

10. A circulating current suppression system for multiple intelligent string photovoltaic inverters in parallel, used to execute the circulating current suppression method for multiple intelligent string photovoltaic inverters in parallel as described in any one of claims 1 to 9, characterized in that, include: The harmonic fingerprinting module is used to control each string inverter to operate independently at multiple preset power levels, collect the output current waveforms of each inverter at each power level, perform fast Fourier transform on each output current waveform, extract the amplitude and phase of each harmonic component, and establish a multi-operating-point harmonic fingerprint mapping table for each inverter using the power level as the index key and the amplitude and phase of each harmonic component as the feature value. The differentiated compensation allocation module is used to collect the circulating current signal of the parallel bus in real time and extract the circulating current harmonic spectrum vector, obtain the current operating power of each inverter, query the multi-operating point harmonic fingerprint mapping table based on the current operating power and obtain the harmonic fingerprint characteristics of each inverter under the current power through interpolation, perform non-negative matrix decomposition matching on the circulating current harmonic spectrum vector and the current harmonic fingerprint characteristics of each inverter, calculate the circulating current contribution weight of each inverter, and generate the differentiated feedback compensation task amount of each inverter based on the circulating current contribution weight; The circulating current suppression execution module is used to input the differentiated feedback compensation task of each inverter into the corresponding repetitive controller, generate harmonic compensation voltage commands for each inverter, and superimpose each harmonic compensation voltage command into the voltage control loop of each inverter to perform directional circulating current suppression.

Images