Method and system for suppressing and cooperatively stabilizing circulating current of multiple converters of energy storage power station

Abstract

The present application relates to the technical field of energy storage, in particular to a multi-converter parallel circulating current suppression and collaborative stability control method for energy storage power station. The method comprises the following steps: S1, multi-source state acquisition and preprocessing; S2, nonlinear droop initial power distribution; S3, decentralized peer collaborative correction; S4, frequency band impedance remodeling parameter loading; S5, frequency band impedance remodeling voltage correction calculation; S6, voltage and current double loop control and PWM generation; S7, hybrid current limiting mode determination and execution; S8, online impedance identification determination and execution; S9, adaptive virtual impedance parameter adjustment; S10, damping torque analysis and negative damping source positioning; S11, data-driven impedance prediction and feedforward setting; S12, communication state check and topology reconstruction; S13, fault diagnosis and health state update; S14, return to step S1 to execute the next control cycle. Compared with the prior art, the control targets of each frequency band are decoupled by frequency band impedance remodeling, and the parallel circulating current effective value is reduced.

Description

Technical Field

[0001] This invention relates to the field of energy storage technology, specifically to a method and system for suppressing and coordinating stable control of parallel circulating currents in multiple converters of an energy storage power station. Background Technology

[0002] In large-capacity grid-connected energy storage power stations, multiple energy storage converter systems (PCS) are typically connected in parallel via an AC bus to meet the power access requirements of hundreds of megawatts. Due to factors such as manufacturing tolerances, filter inductor deviations, cable impedance differences, inconsistent dead times of switching devices, and varying control sampling delays, the output voltages of each PCS inevitably exhibit differences in amplitude, phase, and harmonic content. These differences drive circulating currents between parallel branches, manifesting as non-functional circulating currents flowing between the converters. This not only increases the conduction losses and thermal stress of power devices but can also lead to uneven power distribution, malfunctioning overcurrent protection, and in severe cases, low-frequency oscillations or even system instability.

[0003] With the widespread application of grid-connected converters (PCSs) in energy storage power stations, the stability issues faced by multiple grid-connected PCSs operating in parallel have become more prominent. Grid-connected PCSs provide inertia and damping support to the grid by simulating the characteristics of synchronous generators. However, when multiple grid-connected PCSs are connected in parallel, the mismatch in virtual inertia, damping coefficient, and impedance characteristics between them can cause different types of stability problems in different frequency bands: power oscillations in the low-frequency band, circulating current resonance in the mid-frequency band, and switching ripple coupling in the high-frequency band.

[0004] Existing multi-PCS parallel control technology has the following defects and shortcomings: First, existing virtual impedance control methods apply the same control effect to all frequency bands using a uniform virtual impedance parameter. However, low-frequency bands require smaller impedances to reduce voltage drop and ensure power distribution accuracy, mid-frequency bands require larger impedances to suppress circulating current resonance, and high-frequency bands require suitable resistive impedance characteristics to attenuate switching ripple coupling. Virtual impedance design for a single frequency band cannot simultaneously meet these three conflicting requirements; there is an inherent design conflict between low-frequency current sharing and high-frequency resonance suppression.

[0005] Second, existing master-slave control schemes rely on a central coordinator to issue synchronization and power allocation commands via a communication network. When the central controller fails or the communication link is interrupted, the entire parallel system loses its coordination capability, and all slave PCs face the risk of going out of control. This single-point-of-failure characteristic of the centralized architecture is fundamentally contradictory to the high reliability requirements of energy storage power stations.

[0006] Third, in existing control schemes, virtual impedance parameters are usually based on offline simulation or factory calibration settings, which cannot track real-time changes in grid impedance, parameter drift caused by equipment aging, and dynamic switching of load conditions during operation. When actual operating conditions deviate from the design conditions, the control effect of fixed parameters decreases significantly.

[0007] Fourth, in existing technologies, circulating current suppression, power distribution, resonant damping, and fault protection are typically designed and debugged as independent control components, lacking a unified multi-level joint control framework. Interference may occur between these control components, causing the overall system performance to fail to reach the expected levels when each component was designed independently.

[0008] Fifth, existing control parameter tuning mainly relies on manual experience or offline optimization, lacking intelligent self-tuning capabilities based on operational data. When the scale of parallel systems expands or operating conditions change frequently, the workload and difficulty of manual parameter tuning increase dramatically.

[0009] Based on the aforementioned defects and shortcomings, this application aims to solve the following technical problems:

[0010] 1. How to achieve full-band suppression of circulating currents across low, medium, and high frequencies when multiple grid-connected energy storage converters are connected in parallel without relying on a central controller, while avoiding mutual interference between control targets of different frequency bands.

[0011] 2. How to establish an online-operable impedance identification and parameter adaptive adjustment closed loop so that the frequency band impedance reshaping parameters can track changes in grid impedance and equipment status evolution.

[0012] 3. How to use a lightweight predictive model driven by runtime data to achieve feedforward pre-adjustment of control parameters and improve the system's response speed to sudden changes in operating conditions.

[0013] 4. How to incorporate power sharing, circulating current suppression, resonant damping, and fault current limiting into a unified hierarchical control framework, and maintain safe system operation when some PCS are offline or communication is interrupted. Summary of the Invention

[0014] To overcome the shortcomings of the prior art, this invention provides a method and system for suppressing and coordinating the stability control of parallel circulating currents of multiple converters in an energy storage power station.

[0015] To achieve the above objectives, a method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station is designed, including the following steps:

[0016] S1, Multi-source state acquisition and preprocessing: based on the acquisition cycle Collect the three-phase output voltage of a certain PCS Three-phase output current DC bus voltage and PCC point voltage The collected data were subjected to Clark transformation to obtain the two-phase stationary coordinate system. and Then perform the Park transformation to obtain the synchronous rotating coordinate system. and Calculate the instantaneous active power of this PCS. and reactive power and the effective value of circulation ;

[0017] S2, Nonlinear droop initial power allocation: Calculation of the frequency reference of the local PCS based on exponential Pf droop mapping. and voltage amplitude reference Simultaneously calculate the voltage offset corresponding to the reactive power;

[0018] S3, Decentralized peer-to-peer collaborative correction: Receive state variables sent by neighboring PCS via the nearest neighbor communication link. Run the consensus algorithm to calculate the active power correction. and reactive power correction amount The correction amount is superimposed on the droop output of step S2 to form a frequency reference and voltage reference after global collaborative correction.

[0019] S4, Loading frequency division impedance reshaping parameters: Reads the currently valid virtual impedance parameters for the frequency division, including the virtual inductance for the low-frequency band. and virtual resistance Mid-frequency virtual resistance High-frequency virtual resistance If feedforward pre-adjustment parameters, adaptive correction parameters, or damping enhancement corrections have been output, these corrections will be superimposed on the virtual impedance parameters of the corresponding frequency band.

[0020] S5: Frequency band impedance reshaping voltage correction calculation: Calculate the output current of this PCS unit. The impedance is directly input into the virtual impedance transfer function of each frequency band, and frequency band isolation is achieved by using the filtering stage inside the transfer function. The corresponding impedance voltage drop is then calculated. , , The total impedance reshaping correction is obtained by superimposing the voltage drops of the three impedance segments. The correction amount is subtracted from the voltage reference output in step S3 to form the final voltage loop reference value;

[0021] S6, Voltage and Current Dual-Loop Control and PWM Generation: The final voltage reference value from step S5 is sent to the voltage outer loop PI controller to generate the current reference; the current inner loop PI controller tracks the current reference and outputs a modulated wave, which is then used to generate the switching drive signals for each bridge arm via space vector pulse width modulation (SVPWM).

[0022] S7, Hybrid Current Limiting Mode Judgment and Execution: After the voltage and current loop outputs, detect whether the instantaneous value of the output current exceeds the current limiting threshold. If the current limit is not exceeded, maintain the current virtual impedance control mode; if it is exceeded, switch to direct current limiting mode to limit the current to within the limit range. Within this range; once the current drops and stabilizes, the virtual impedance mode is restored via a linear transition;

[0023] S8, Online Impedance Identification and Execution: Determines whether the identification cycle has been reached. If it reaches the target value, a low-amplitude wideband detection signal is superimposed on the PCC voltage reference, and the injection duration is [not specified]. Voltage and current response data were collected during the detection period. Frequency domain impedance characteristics were extracted using FFT, and the PCS output impedance in the low-frequency, mid-frequency, and high-frequency bands was calculated. and grid impedance And calculate the real-time short-circuit ratio If identification fails, retain the identification result from the previous cycle;

[0024] S9, Adaptive Virtual Impedance Parameter Adjustment: Based on the output of step S8 and The virtual impedance parameters of each frequency band are recalculated according to the adaptive adjustment rules. The parameter update adopts exponential moving average filtering to avoid sudden changes. The updated parameters take effect in the next control cycle.

[0025] S10, Damping Torque Analysis and Negative Damping Source Location: Based on the identification results of step S8 and the current control parameters, construct the small-signal state matrix of the parallel system, calculate the eigenvalues, and determine the damping ratio of a certain oscillation mode. Below the safety threshold By using participation factor analysis to locate the dominant contribution source of this mode, a damping enhancement correction amount for the corresponding frequency band of this PCS is generated. Send to step S4;

[0026] S11, Data-driven impedance prediction and feedforward tuning: Store the identification results from step S8 into the historical buffer. When the amount of data in the buffer reaches the minimum training length... The lightweight time-series prediction network is used to predict the trend of grid impedance and PCS impedance changes in the next cycle. After the prediction results are projected by physical constraints, the feedforward parameter adjustment is calculated and applied to step S4 in advance before the next identification cycle arrives to achieve feedforward pre-adjustment.

[0027] S12, Communication Status Check and Topology Reconstruction: Check the communication link status with each neighboring PCS. If the communication of a neighboring PCS times out, mark it as offline and remove it from the neighbor set of the consensus algorithm, triggering communication topology reconstruction. If the local PCS loses connection with all neighbors, it falls back to pure local control mode and relies only on frequency band impedance reshaping and droop control to maintain operation.

[0028] S13, Fault Diagnosis and Health Status Update: Based on the operational indicators of each module, calculate the overall health score of this PCS. Update the health status flag and broadcast it to neighboring PCS via the communication link;

[0029] S14, return to step S1 to execute the next control cycle.

[0030] In step S2, the frequency reference of the local PCS is calculated based on the exponential Pf droop mapping. and voltage amplitude reference The specific methods are as follows:

[0031] ,in:

[0032] This is the reference value for the output angular frequency of this PCS, in rad / s;

[0033] This is the system's rated angular frequency; for a 50Hz system, it is... ;

[0034] The active power droop factor has a range of values. ;

[0035] This is the rated active power of this PCS, in watts (W).

[0036] This represents the current active power output of this PCS. Under bidirectional flow conditions, discharge is positive and charging is negative, and the unit is W.

[0037] It is a symbolic function;

[0038] This represents the absolute value of active power.

[0039] This is the exponential droop shape factor, ranging from 1 to 5, and is dimensionless.

[0040] It is the base of the natural logarithm;

[0041] Voltage amplitude droop is:

[0042] ,in:

[0043] This is the reference value for the output voltage amplitude of this PCS, in V;

[0044] The rated voltage amplitude of the system, in volts (V).

[0045] The reactive power droop factor has a range of values. ;

[0046] This represents the current reactive power output of this PCS, measured in Var.

[0047] In step S3, the active power correction is calculated using a consensus algorithm. and reactive power correction amount The specific methods are as follows:

[0048]

[0049] in:

[0050] For the first PCS in Taiwan The active power state estimate for each communication cycle, in W;

[0051] The active power state estimate, after consistency correction, is used for calculation and broadcasting in the next communication cycle, in W.

[0052] This is the consistency step size coefficient, ranging from 0.01 to 0.5, and is dimensionless.

[0053] For the first The nearest neighbor set of PCS contains all nodes that are related to the first node. Each PCS has a PCS number associated with its communication link;

[0054] This is the communication weighting coefficient, ranging from 0 to 1, and is dimensionless; default value. ;

[0055] For the neighboring number PCS in Taiwan The active power state estimate for each communication cycle, in W;

[0056] The active power correction is:

[0057] ,in:

[0058] For the first The active power correction for each PCS, in W;

[0059] The uniform power correction gain has a value ranging from 0.1 to 2.0 and is dimensionless.

[0060] The global average active power obtained when the consensus algorithm converges to equilibrium is, i.e. The unit is W;

[0061] This represents the current actual physical output active power of this PCS, in W.

[0062] The consistency protocol and correction formula for reactive power are the same as those for active power, with active power being replaced by reactive power accordingly.

[0063] In step S5, the low-frequency virtual impedance transfer function is:

[0064] ,in:

[0065] For low-frequency virtual impedance Domain transfer function, in Ω;

[0066] This is a virtual resistance for the low-frequency band, with a value ranging from 0.0010 to 0.050 Ω;

[0067] This is a low-frequency virtual inductance, with a value ranging from 0.1mH to 2mH;

[0068] For the complex frequency variable of the Laplace transform;

[0069] The virtual impedance transfer function in the mid-frequency band is:

[0070] ,in:

[0071] For the mid-frequency virtual impedance Domain transfer function, in Ω;

[0072] This represents the peak value of the virtual resistance in the mid-frequency band, ranging from 0.1Ω to 2Ω.

[0073] This is the low-frequency cutoff angular frequency. The default frequency is 10Hz, and the unit is rad / s;

[0074] It is the high-frequency cutoff angular frequency. The default is 2kHz, and the unit is rad / s;

[0075] The virtual impedance transfer function in the high-frequency band is:

[0076] ,in:

[0077] For high-frequency virtual impedance Domain transfer function, in Ω;

[0078] This is a virtual resistance for the high-frequency band, with a value ranging from 0.05Ω to 0.5Ω;

[0079] It is the high-frequency cutoff angular frequency. The default is 2kHz, and the unit is rad / s;

[0080] The calculation method for the total impedance reshaping correction is as follows:

[0081] ,in:

[0082] For the total voltage correction of impedance reshaping in the frequency band, in In a synchronously rotating coordinate system, the unit is V;

[0083] The output current of this PCS Coordinate system components, including and The unit is A;

[0084] , , These are the virtual impedance transfer functions for low, medium, and high frequency bands, respectively.

[0085] In step S7, the specific method for determining and executing the hybrid rate limiting mode is as follows:

[0086] ,in:

[0087] This is the final output voltage command, in volts (V).

[0088] This is the voltage reference value before impedance reshaping in the frequency band, in V;

[0089] This represents the total correction amount for impedance reshaping across frequency bands, expressed in volts (V).

[0090] The amplitude of the three-phase instantaneous output current, in amperes (A).

[0091] This is the current limiting threshold, which defaults to 1.2 times the peak rated current, in amperes (A).

[0092] This is the equivalent impedance in current-limiting mode, in Ω.

[0093] Mode switching uses a linear transition:

[0094] ,in:

[0095] This is the output voltage command during the transition period, in volts (V).

[0096] This is the output voltage in virtual impedance mode, in V;

[0097] This is the output voltage in direct limiting mode, in volts (V).

[0098] This is a linear transition coefficient that changes linearly from 0 or 1 to 1 or 0 after a mode switch is triggered. The default transition time is [time value missing]. , dimensionless.

[0099] In step S8, the method for calculating the real-time short-circuit ratio is as follows:

[0100] ,in:

[0101] For the first The real-time short-circuit ratio observed by PCS is dimensionless.

[0102] The equivalent impedance of the power grid at the PCC obtained online at the fundamental frequency The amplitude at that point, in Ω;

[0103] The rated voltage of the PCC is expressed in volts (V).

[0104] For the first The rated capacity of a PCS is expressed in VA.

[0105] In step S9, the adaptive adjustment rule is as follows:

[0106] ,in:

[0107] This is the new virtual resistance value for the mid-frequency band after adaptive adjustment, in Ω;

[0108] This is the virtual resistance reference value for the mid-frequency band, provided by the offline design, in Ω;

[0109] The short-circuit ratio is set to the design baseline, with a default value of 10, which is dimensionless.

[0110] The real-time short-circuit ratio obtained through online identification is dimensionless;

[0111] It is an adaptive adjustment index, ranging from 0.3 to 1.0, and is dimensionless;

[0112] The parameter update after exponential moving average filtering is as follows:

[0113] ,in:

[0114] This represents the actual effective virtual resistance value in the mid-frequency band, in Ω.

[0115] This is the exponential moving average filter coefficient, with a value ranging from 0.1 to 0.5, and is dimensionless.

[0116] This is the virtual resistance value in the mid-frequency band of the previous identification cycle, in Ω.

[0117] In step S10, the eigenvalue decomposition of the system state matrix is ​​specifically as follows:

[0118] ,in:

[0119] These are the eigenvalues ​​of the system state matrix;

[0120] It is the identity matrix;

[0121] The small-signal state matrix of the parallel system is composed of the control parameters, filter parameters and grid impedance parameters of each PCS.

[0122] The damping ratio is calculated as follows:

[0123] ,in:

[0124] For the first The damping ratio of each oscillation mode is dimensionless and ranges from 1 to 2. to ;

[0125] For the first The real part of each eigenvalue is represented by 1 / s;

[0126] For the first The imaginary part of each eigenvalue is expressed in rad / s.

[0127] The specific negative damping source involved in the factor localization is as follows:

[0128] ,in:

[0129] For the first The modality pair of the first Participation factors of each state variable, dimensionless, with values ​​ranging from 0 to 1;

[0130] The right eigenvector matrix The Middle Line 1 Column elements;

[0131] Left eigenvector matrix The Middle Line 1 Column elements;

[0132] The specific amount of damping enhancement correction is as follows:

[0133] ,in:

[0134] In order to target the The virtual resistance enhancement correction for each underdamped mode, in Ω;

[0135] The damping enhancement gain ranges from 0.5 to 5.0, and the unit is Ω;

[0136] This is the damping ratio safety threshold, with a default value of 0.05, and is dimensionless.

[0137] For the current number Damping ratio of each mode, dimensionless;

[0138] For the first The maximum participation factor value for each modality is dimensionless.

[0139] In step S11, the data-driven impedance prediction specifically involves:

[0140] ,in:

[0141] The predicted characteristic vector of the grid impedance for the next identification cycle includes impedance magnitudes and phase angles in the low, medium, and high frequency bands, and is dimensionless.

[0142] It is a single-layer gated recurrent unit network with a hidden layer dimension not exceeding 32;

[0143] For the first The characteristic vector of grid impedance for each identification cycle;

[0144] To predict the window length, the value ranges from 5 to 20 and is dimensionless;

[0145] Internal control calculations are specifically as follows:

[0146] ;

[0147] ;

[0148] ;

[0149] ;in:

[0150] To update the gate output vector, the value range is [0,1], and it is dimensionless;

[0151] The reset gate output vector has a value range of [0,1] and is dimensionless;

[0152] is the candidate hidden state vector, with a value range of [-1, 1], and is dimensionless;

[0153] Let be the hidden state vector at the current time step, with a dimension not exceeding 32 and dimensionless;

[0154] This is the hidden state vector of the previous time step;

[0155] The current time step is the input vector, and the impedance characteristics are given.

[0156] , , The weight matrix was obtained through offline training.

[0157] , , It is the bias vector;

[0158] It is the sigmoid activation function;

[0159] It is the hyperbolic tangent activation function;

[0160] This is element-wise multiplication;

[0161] This is a vector concatenation operation;

[0162] The specific method for physical constraint projection is as follows:

[0163] ,in:

[0164] To the stable region The Euclidean projection operator has a stability region defined as the set of impedance parameters that satisfy the following conditions: ,

[0165] This is the predicted value of the power grid impedance after projection with physical constraints.

[0166] To close convex sets Orthogonal projection operator;

[0167] , The upper and lower limits of the impedance amplitude are determined by the system's rated parameters and are expressed in Ω.

[0168] This is the allowable range for the impedance phase angle, with a default value of [-90°, 90°].

[0169] Given impedance parameters Next Damping ratio of each oscillation mode;

[0170] The specific pre-adjustment amount of the feedforward parameter is as follows:

[0171] ,in:

[0172] This is the feedforward pre-adjustment increment of the virtual resistance in the mid-frequency band, in Ω;

[0173] This is the feedforward gain, ranging from 0.1 to 0.8, and is dimensionless.

[0174] The optimal mid-frequency virtual resistance is calculated based on the predicted impedance after projection, in Ω.

[0175] This is the currently active virtual resistance value for the mid-frequency band, in Ω.

[0176] To achieve the above objectives, a parallel circulating current suppression and coordinated stability control system for multiple converters in an energy storage power station is designed, comprising the following modules:

[0177] (1) Module 1, Multi-source status acquisition module: Acquires the three-phase output voltage of this PCS. Three-phase output current DC bus voltage and PCC point voltage The data includes the status variables sent by neighboring PCS via the communication link; after anti-aliasing filtering and analog-to-digital conversion, the collected data is simultaneously sent to the frequency band impedance reshaping module and the online impedance identification module.

[0178] (2) Module 2, frequency band impedance reshaping module: The output impedance of this PCS is divided into three frequency bands: low frequency band, mid frequency band and high frequency band. Each frequency band is configured with an independent virtual impedance transfer function.

[0179] (3) Module 3, Adaptive Virtual Impedance Adjustment Unit: Receives the real-time grid short-circuit ratio RSCR and the equivalent output impedance of the local PCS from the online impedance identification module, and dynamically adjusts the amplitude and phase angle of the virtual impedance of each frequency band in the frequency band impedance reshaping module according to the adaptive adjustment rules.

[0180] (4) Module 4, Decentralized Peer-to-Peer Collaborative Control Module: Each PCS exchanges limited state quantities, including its own active power, only with physically adjacent PCS via CAN bus or fiber optic Ethernet. reactive power RMS value of circulation With the health status indicator H, each PCS runs a discrete-time average consensus algorithm to achieve global power distribution and suppress parallel circulation through multiple rounds of iteration;

[0181] (5) Module 5, Online Impedance Identification Module: According to the configurable identification cycle A low-amplitude broadband detection signal is superimposed on the PCC voltage. By acquiring the voltage and current responses during the detection period, the frequency domain impedance characteristics are extracted using Fast Fourier Transform (FFT), and the equivalent output impedance of this PCS in the low-frequency, mid-frequency, and high-frequency bands is output. and the equivalent impedance of the power grid at PCC The identification results are sent to the adaptive adjustment unit and the data-driven parameter self-tuning module;

[0182] (6) Module 6, Damping Torque Analysis and Negative Damping Source Identification Module: Based on the online impedance identification results and current control parameters, the system state matrix is ​​constructed using a small-signal linearization model, and the damping ratio of each oscillation mode is calculated through eigenvalue decomposition;

[0183] (7) Module 7, Data-Driven Impedance Prediction and Parameter Self-Tuning Module: Acquire the most recent The impedance identification data of each identification cycle is used to predict the trend of grid impedance and PCS output impedance changes in the next identification cycle using a lightweight time-series prediction network; the prediction results are then subjected to physical constraint projection to adjust the frequency band reshaping parameters in advance.

[0184] (8) Module 8, Hybrid Current Limiting Protection Module: Under normal operating conditions, it maintains the voltage source output behavior and limits the steady-state circulating current through frequency-segmented virtual impedance; when the instantaneous value of the output current exceeds the current limiting threshold, it will provide protection. Immediately switch to direct current limiting mode; when the current falls below the threshold and remains stable for a certain period of time... Then, smoothly switch back to virtual impedance mode;

[0185] (9) Module Nine, Power Distribution and Current Sharing Control Module: Based on the nonlinear droop characteristics, the initial distribution of active and reactive power is realized; on the basis of the initial droop distribution, the power correction amount output by the decentralized peer-to-peer collaborative module is superimposed;

[0186] (10) Module 10, Status Monitoring and Fault Diagnosis Module: continuously monitors the operating status of each module of this PCS; when an abnormality is detected, a fault event is generated and recorded, and at the same time, the neighboring PCS is notified through the health status flag H;

[0187] Between the above modules: (1) The data collected by module one is simultaneously sent to (2) module two and (5) module five. (5) The identification results of module five are sent to (3) module three, (6) module six and (7) module seven. (3) module three adjusts the parameters of each frequency band of (2) module two. (6) module six outputs the damping enhancement correction amount to (2) module two. (7) module seven outputs the feedforward pre-adjustment parameters to (2) module two. (4) The collaborative correction amount of module four is sent to (9) module nine and superimposed on the power distribution output. (9) The power command of module nine is sent to (2) module two to form the final voltage reference. (8) module eight implements current limiting protection at the output of module two. (10) module ten receives the status information of all modules.

[0188] Compared with the prior art, the present invention has the following advantages:

[0189] 1. By reshaping the impedance of each frequency band, the control objectives of each frequency band are decoupled, and the effective value of the parallel circulating current can be reduced to less than 2% of the rated current, and the active power distribution deviation between each PCS does not exceed ±1.5%.

[0190] 2. A decentralized peer-to-peer collaborative control protocol is adopted to eliminate the risk of single point of failure of the central controller; when any single PCS goes offline or communication is interrupted, the remaining PCS can complete the power redistribution within 200ms.

[0191] 3. The online impedance identification closed loop ensures that the system remains stable within a range where the grid short-circuit ratio varies from 10 to 3, improving the adaptability to weak grids by more than 50%.

[0192] 4. The data-driven parameter self-tuning module uses historical identification data to predict the optimal parameters, reducing the parameter adjustment response time to 30% of the traditional method when operating conditions change abruptly.

[0193] 5. The hybrid current limiting protection module reduces the peak circulating current surge by more than 60% during a three-phase short-circuit fault, and the system recovery time after fault ride-through does not exceed 500ms. Attached Figure Description

[0194] Figure 1 This is a schematic diagram of the process of the present invention.

[0195] Figure 2 This is a system architecture diagram of the present invention.

[0196] Figure 3 This is a schematic diagram of the decentralized peer-to-peer collaborative communication topology and consistency convergence process in step S3 of the present invention.

[0197] Figure 4 This is a schematic diagram of the Bode plot for frequency band impedance reshaping in step S4 of the present invention.

[0198] Figure 5 This is a triptych of simulation results from Embodiment 1 of the present invention. Detailed Implementation

[0199] The present invention will now be further described with reference to the accompanying drawings.

[0200] The method for suppressing and coordinating the circulatory current of multiple converters in the energy storage power station in this embodiment includes the following steps, which are run in parallel in the local controller of each PCS:

[0201] S1, Multi-source state acquisition and preprocessing: based on the acquisition cycle Collect the three-phase output voltage of a certain PCS Three-phase output current DC bus voltage and PCC point voltage The collected data were subjected to Clark transformation to obtain the two-phase stationary coordinate system. and Then perform the Park transformation to obtain the synchronous rotating coordinate system. and Calculate the instantaneous active power of this PCS. and reactive power and the effective value of circulation ;

[0202] S2, Nonlinear droop initial power allocation: Calculation of the frequency reference of the local PCS based on exponential Pf droop mapping. and voltage amplitude reference It utilizes the exponential droop characteristic to provide a larger frequency margin under light loads and achieve a more accurate distribution effect under heavy loads. Simultaneously, it calculates the voltage offset corresponding to reactive power.

[0203] S3, Decentralized peer-to-peer collaborative correction: Receive state variables sent by neighboring PCS via the nearest neighbor communication link. Run the consensus algorithm to calculate the active power correction. and reactive power correction amount The correction amount is superimposed on the droop output of step S2 to form a frequency reference and voltage reference after global collaborative correction.

[0204] S4, Loading frequency division impedance reshaping parameters: Reads the currently valid virtual impedance parameters for the frequency division, including the virtual inductance for the low-frequency band. and virtual resistance Mid-frequency virtual resistance High-frequency virtual resistance If module 7 has output feedforward pre-adjustment parameters, or module 3 has output adaptive correction parameters, or module 6 has output damping enhancement correction, then these corrections will be superimposed on the virtual impedance parameters of the corresponding frequency band.

[0205] S5: Frequency band impedance reshaping voltage correction calculation: Calculate the output current of this PCS unit. The impedance is directly input into the virtual impedance transfer function of each frequency band, and frequency band isolation is achieved by using the filtering stage inside the transfer function. The corresponding impedance voltage drop is then calculated. , , The total impedance reshaping correction is obtained by superimposing the voltage drops of the three impedance segments. The correction amount is subtracted from the voltage reference output in step S3 to form the final voltage loop reference value;

[0206] S6, Voltage and Current Dual-Loop Control and PWM Generation: The final voltage reference value from step S5 is sent to the voltage outer loop PI controller to generate the current reference; the current inner loop PI controller tracks the current reference and outputs a modulated wave, which is then used to generate the switching drive signals for each bridge arm via space vector pulse width modulation (SVPWM).

[0207] S7, Hybrid Current Limiting Mode Judgment and Execution: After the voltage and current loop outputs, detect whether the instantaneous value of the output current exceeds the current limiting threshold. If the current limit is not exceeded, maintain the current virtual impedance control mode; if it is exceeded, switch to direct current limiting mode to limit the current to within the limit range. Within this range; once the current drops and stabilizes, the virtual impedance mode is restored via a linear transition;

[0208] S8, Online Impedance Identification and Execution: Determines whether the identification cycle has been reached. If it reaches the target value, a low-amplitude wideband detection signal is superimposed on the PCC voltage reference, and the injection duration is [not specified]. The default time is 200ms. Voltage and current response data are collected during the detection period. Frequency domain impedance features are extracted using FFT, and the PCS output impedance in the low-frequency, mid-frequency, and high-frequency bands is calculated. and grid impedance And calculate the real-time short-circuit ratio If identification fails, for example, due to insufficient signal-to-noise ratio or the converter being in a transient state, the identification result of the previous cycle is retained;

[0209] S9, Adaptive Virtual Impedance Parameter Adjustment: Based on the output of step S8 and The virtual impedance parameters for each frequency band are recalculated according to the adaptive adjustment rules. The parameter updates employ exponential moving average filtering to avoid abrupt changes, and the updated parameters take effect in the next control cycle; the filter coefficients... The default value is 0.3;

[0210] S10, Damping Torque Analysis and Negative Damping Source Location: Based on the identification results of step S8 and the current control parameters, construct the small-signal state matrix of the parallel system, calculate the eigenvalues, and determine the damping ratio of a certain oscillation mode. Below the safety threshold By using participation factor analysis to locate the dominant contribution source of this mode, a damping enhancement correction amount for the corresponding frequency band of this PCS is generated. Send to step S4;

[0211] S11, Data-driven impedance prediction and feedforward tuning: The identification results from step S8 are stored in the historical buffer. A circular buffer is selected as the historical buffer, with a capacity of [missing information]. When the amount of data in the buffer reaches the minimum training length The default cycle is 20 cycles. The lightweight time series prediction network is used to predict the trend of grid impedance and PCS impedance changes in the next cycle. After the prediction results are projected by physical constraints, the feedforward parameter adjustment is calculated and applied to step S4 in advance before the next identification cycle arrives to achieve feedforward pre-adjustment.

[0212] S12, Communication Status Check and Topology Reconfiguration: Check the communication link status with each neighboring PCS. If a communication timeout occurs with a neighboring PCS, exceeding a set threshold, the check will be performed. The default time is 50ms. If a PCS is offline, it is removed from the neighbor set of the consensus algorithm and the communication topology is reconstructed. If the local PCS loses contact with all its neighbors, it falls back to the pure local control mode and relies only on frequency band impedance reshaping and droop control to maintain operation.

[0213] S13, Fault Diagnosis and Health Status Update: Based on the operational indicators of various modules, such as impedance identification success rate, consistency convergence speed, damping ratio margin, and current limiting trigger frequency, calculate the overall health score of this PCS. Update the health status flag and broadcast it to neighboring PCS via the communication link;

[0214] S14, return to step S1 to execute the next control cycle.

[0215] In step S2, the frequency reference of the local PCS is calculated based on the exponential Pf droop mapping. and voltage amplitude reference The specific methods are as follows:

[0216] ,in:

[0217] This is the reference value for the output angular frequency of this PCS, in rad / s;

[0218] This is the system's rated angular frequency; for a 50Hz system, it is... ;

[0219] The active power droop factor has a range of values. This determines the maximum frequency offset;

[0220] This is the rated active power of this PCS, in watts (W).

[0221] This represents the current active power output of this PCS. Under bidirectional flow conditions, discharge is positive and charging is negative, and the unit is W.

[0222] This is the sign function, used to ensure the bidirectional symmetrical droop characteristics of charging and discharging;

[0223] This is the absolute value of the active power to prevent exponential explosions during charging.

[0224] is the exponential drooping shape factor, with a value ranging from 1 to 5, and is dimensionless; When it is close to a linear decline, The larger the value, the more pronounced the exponential characteristic.

[0225] It is the base of the natural logarithm;

[0226] Voltage amplitude droop is:

[0227] ,in:

[0228] This is the reference value for the output voltage amplitude of this PCS, in V;

[0229] The rated voltage amplitude of the system, in volts (V).

[0230] The reactive power droop factor has a range of values. ;

[0231] This represents the current reactive power output of this PCS, measured in Var.

[0232] In step S3, the active power correction is calculated using a consensus algorithm. and reactive power correction amount The specific methods are as follows:

[0233]

[0234] in:

[0235] For the first PCS in Taiwan The active power state estimate for each communication cycle, in W, initially based on the physical power measurement.

[0236] The active power state estimate, after consistency correction, is used for calculation and broadcasting in the next communication cycle, in W.

[0237] The consistency step size coefficient, ranging from 0.01 to 0.5, is dimensionless and determines the trade-off between convergence speed and stability.

[0238] For the first The nearest neighbor set of PCS contains all nodes that are related to the first node. Each PCS has a PCS number associated with its communication link;

[0239] This is a communication weighting coefficient, ranging from 0 to 1, dimensionless, and applicable to undirected communication topologies. ;default ;

[0240] For the neighboring number PCS in Taiwan The active power state estimate for each communication cycle, in W;

[0241] The active power correction is:

[0242] ,in:

[0243] For the first The active power correction for each PCS, in W;

[0244] The uniform power correction gain has a value ranging from 0.1 to 2.0 and is dimensionless.

[0245] The global average active power obtained when the consensus algorithm converges to equilibrium is, i.e. The unit is W;

[0246] This represents the current actual physical output active power of this PCS, in W.

[0247] The consistency protocol and correction formula for reactive power are the same as those for active power, with active power being replaced by reactive power accordingly.

[0248] In step S5, the low-frequency virtual impedance transfer function is:

[0249] ,in:

[0250] For low-frequency virtual impedance Domain transfer function, in Ω;

[0251] This is a virtual resistor for the low-frequency band, with a value ranging from 0.0010 to 0.050Ω, used to provide a small amount of damping while limiting the steady-state voltage drop in the low-frequency band;

[0252] This is a low-frequency virtual inductor, with a value ranging from 0.1mH to 2mH, used to maintain the inductive impedance characteristics in the low-frequency range to ensure power decoupling;

[0253] For the complex frequency variable of the Laplace transform;

[0254] The virtual impedance transfer function in the mid-frequency band is:

[0255] ,in:

[0256] For the mid-frequency virtual impedance Domain transfer function, in Ω;

[0257] This is the peak value of the virtual resistance in the mid-frequency band, ranging from 0.1Ω to 2Ω, used to provide a larger resistive impedance in the mid-frequency band to suppress circulating current resonance;

[0258] This is the low-frequency cutoff angular frequency. The default frequency is 10Hz, and the unit is rad / s;

[0259] It is the high-frequency cutoff angular frequency. The default is 2kHz, and the unit is rad / s;

[0260] The components form a high-pass filter to filter out low-frequency power components;

[0261] The components form a low-pass filter to filter out high-frequency switching components;

[0262] The cascading of two-stage filters makes the virtual impedance in the mid-frequency band only Effective within the scope;

[0263] The virtual impedance transfer function in the high-frequency band is:

[0264] ,in:

[0265] For high-frequency virtual impedance Domain transfer function, in Ω;

[0266] This is a virtual resistor for the high-frequency band, with a value ranging from 0.05Ω to 0.5Ω, used to attenuate switching ripple coupling at high frequencies.

[0267] It is the high-frequency cutoff angular frequency. The default is 2kHz, and the unit is rad / s;

[0268] The terms form a high-pass filter, making the virtual impedance effective only in the high-frequency range;

[0269] The calculation method for the total impedance reshaping correction is as follows:

[0270] ,in:

[0271] For the total voltage correction of impedance reshaping in the frequency band, in In a synchronously rotating coordinate system, the unit is V;

[0272] The output current of this PCS Coordinate system components, including and The unit is A;

[0273] , , These are the virtual impedance transfer functions for the low, mid, and high frequency bands, respectively. Since the frequency response of the three transfer functions decays rapidly outside their respective frequency bands, the cross-interference between the three segments can be ignored.

[0274] In step S7, the specific method for determining and executing the hybrid rate limiting mode is as follows:

[0275] ,in:

[0276] This is the final output voltage command, in volts (V).

[0277] This is the voltage reference value before impedance reshaping in the frequency band, in V;

[0278] This represents the total correction amount for impedance reshaping across frequency bands, expressed in volts (V).

[0279] The amplitude of the three-phase instantaneous output current, in amperes (A).

[0280] This is the current limiting threshold, which defaults to 1.2 times the peak rated current, in amperes (A).

[0281] This is the equivalent impedance in current-limiting mode, measured in Ω, used to maintain a certain voltage support capability.

[0282] Mode switching uses a linear transition:

[0283] ,in:

[0284] This is the output voltage command during the transition period, in volts (V).

[0285] This is the output voltage in virtual impedance mode, in V;

[0286] This is the output voltage in direct limiting mode, in volts (V).

[0287] This is a linear transition coefficient that changes linearly from 0 or 1 to 1 or 0 after a mode switch is triggered. The default transition time is [time value missing]. , dimensionless.

[0288] In step S8, the method for calculating the real-time short-circuit ratio is as follows:

[0289] ,in:

[0290] For the first The real-time short-circuit ratio observed by PCS is dimensionless.

[0291] The equivalent impedance of the power grid at the PCC obtained online at the fundamental frequency The amplitude at 50Hz, in Ω;

[0292] The rated voltage of the PCC is the effective value of the line voltage, expressed in volts (V).

[0293] For the first The rated capacity of a PCS is expressed in VA.

[0294] In step S9, the adaptive adjustment rule is as follows:

[0295] ,in:

[0296] This is the new virtual resistance value for the mid-frequency band after adaptive adjustment, in Ω;

[0297] This is the virtual resistance reference value for the mid-frequency band, provided by the offline design, in Ω;

[0298] The short-circuit ratio is set to the design baseline, with a default value of 10, which is dimensionless.

[0299] The real-time short-circuit ratio obtained through online identification is dimensionless;

[0300] This is an adaptive adjustment index, ranging from 0.3 to 1.0, and is dimensionless. For linear adjustment, For conservative adjustment, a weak damping network has better adaptability but a larger voltage drop;

[0301] The parameter update after exponential moving average filtering is as follows:

[0302] ,in:

[0303] This represents the actual effective virtual resistance value in the mid-frequency band, in Ω.

[0304] This is the exponential moving average filter coefficient, with a value ranging from 0.1 to 0.5, dimensionless, and a default value of 0.3;

[0305] This is the virtual resistance value in the mid-frequency band of the previous identification cycle, in Ω.

[0306] In step S10, the eigenvalue decomposition of the system state matrix is ​​specifically as follows:

[0307] ,in:

[0308] These are the eigenvalues ​​of the system state matrix, generally in complex form. ;

[0309] It is the identity matrix;

[0310] The small-signal state matrix of the parallel system is composed of the control parameters, filter parameters and grid impedance parameters of each PCS.

[0311] The damping ratio is calculated as follows:

[0312] ,in:

[0313] For the first The damping ratio of each oscillation mode is dimensionless and ranges from 1 to 2. to A positive value indicates stability, and the safety threshold is... =0.05;

[0314] For the first The real part of each eigenvalue is represented by 1 / s, and negative values ​​indicate decay.

[0315] For the first The imaginary part of each eigenvalue, in rad / s, represents the oscillation frequency;

[0316] The specific negative damping source involved in the factor localization is as follows:

[0317] ,in:

[0318] For the first The modality pair of the first Participation factors of each state variable, dimensionless, with values ​​ranging from 0 to 1;

[0319] The right eigenvector matrix The Middle Line 1 Column elements;

[0320] Left eigenvector matrix The Middle Line 1 Column elements;

[0321] The PCS and control loop to which the state variable with the largest participating factor belongs are identified as negative damping sources;

[0322] The specific amount of damping enhancement correction is as follows:

[0323] ,in:

[0324] In order to target the The virtual resistance enhancement correction for each underdamped mode, in Ω;

[0325] The damping enhancement gain ranges from 0.5 to 5.0, with units of Ω, and is determined by the system size and impedance order.

[0326] This is the damping ratio safety threshold, with a default value of 0.05, and is dimensionless.

[0327] For the current number Damping ratio of each mode, dimensionless;

[0328] For the first The maximum participation factor value of each mode, dimensionless, is used to directionally apply the correction to the frequency band where the negative damping source is located.

[0329] In step S11, the data-driven impedance prediction specifically involves:

[0330] ,in:

[0331] The predicted characteristic vector of the grid impedance for the next identification cycle includes impedance amplitude and phase angle in the low, medium and high frequency bands. It is dimensionless and input after normalization.

[0332] It is a single-layer gated recurrent unit network with a hidden layer dimension not exceeding 32;

[0333] For the first The characteristic vector of grid impedance for each identification cycle;

[0334] The value is for the prediction window length, ranging from 5 to 20, dimensionless, with a default value of 10;

[0335] Internal control calculations are specifically as follows:

[0336] ;

[0337] ;

[0338] ;

[0339] ;in:

[0340] To update the gate output vector, the value range is [0,1], and it is dimensionless;

[0341] The reset gate output vector has a value range of [0,1] and is dimensionless;

[0342] is the candidate hidden state vector, with a value range of [-1, 1], and is dimensionless;

[0343] Let be the hidden state vector at the current time step, with a dimension not exceeding 32 and dimensionless;

[0344] This is the hidden state vector of the previous time step;

[0345] The current time step is the input vector, and the impedance characteristics are given.

[0346] , , The weight matrix was obtained through offline training.

[0347] , , It is the bias vector;

[0348] It is the sigmoid activation function;

[0349] It is the hyperbolic tangent activation function;

[0350] This is element-wise multiplication, also known as the Hadamard product;

[0351] This is a vector concatenation operation;

[0352] The specific method for physical constraint projection is as follows:

[0353] ,in:

[0354] To the stable region The Euclidean projection operator has a stability region defined as the set of impedance parameters that satisfy the following conditions: ,

[0355] This is the predicted value of the power grid impedance after projection with physical constraints.

[0356] To close convex sets Orthogonal projection operator;

[0357] , The upper and lower limits of the impedance amplitude are determined by the system's rated parameters and are expressed in Ω.

[0358] This is the allowable range for the impedance phase angle, with a default value of [-90°, 90°].

[0359] Given impedance parameters Next Damping ratio of each oscillation mode;

[0360] The specific pre-adjustment amount of the feedforward parameter is as follows:

[0361] ,in:

[0362] This is the feedforward pre-adjustment increment of the virtual resistance in the mid-frequency band, in Ω;

[0363] The value is the feedforward gain, ranging from 0.1 to 0.8, dimensionless, with smaller values ​​being conservative and larger values ​​being aggressive;

[0364] The optimal mid-frequency virtual resistance is calculated based on the predicted impedance after projection, in Ω.

[0365] This is the currently active virtual resistance value for the mid-frequency band, in Ω.

[0366] The execution order and branching logic of steps S1 to S14 are as follows: Figure 1 As shown, the three independently operating loops are distinguished by different labels: Steps S1 to S7 are fast control loops, operating on a control cycle basis. Steps S8 to S11 are slow identification loops, operating on an identification cycle basis. Steps S3, S12, and S13 are communication coordination loops, operating on a communication cycle basis. The hybrid current limiting mode determination in step S7 is represented by conditional branches. The identification success / failure in step S8 is represented by conditional branches.

[0367] The schematic diagram of the decentralized peer-to-peer collaborative communication topology and consensus convergence process in step S3 is shown below. Figure 3 As shown in the figure, the left side of the figure represents a decentralized ring communication topology with 8 PCSs, where each PCS communicates only with its two nearest neighbors. The right side of the figure shows the power convergence process of the consensus algorithm. The horizontal axis represents the number of communication iterations, from 0 to 15, and the vertical axis represents the active power (kW) of each PCS. The eight curves start from different initial values ​​and converge to the average value after about eight iterations.

[0368] The Bode plot of the frequency band impedance reshaping in step S4 is shown below. Figure 4 As shown, the horizontal axis represents frequency on a logarithmic scale, from 1Hz to 100kHz. The upper half of the vertical axis represents amplitude in dB, and the lower half represents phasing, respectively displaying the virtual impedance in the low-frequency range. Mid-frequency virtual impedance High-frequency virtual impedance The frequency response curve of the total impedance reshaping characteristics after the sum of the three is also shown. The low-frequency range is predominantly inductive, with a phase angle close to +90°. The mid-frequency range is predominantly resistive, with a phase angle close to 0°. The high-frequency range is predominantly resistive. The transition regions between the three ranges are clearly marked.

[0369] The system architecture diagram for implementing the above-mentioned method of parallel circulating current suppression and coordinated stability control of multiple converters in an energy storage power station is shown below. Figure 2As shown, multiple energy storage converters, namely PCS-1 to PCS-n, are connected to a common coupling point PCC via an AC parallel bus and then connected to the power grid through a step-up transformer. Each PCS contains ten functional modules: a multi-source status acquisition module, a frequency band impedance reshaping module, an adaptive virtual impedance adjustment unit, a decentralized peer-to-peer collaborative control module, an online impedance identification module, a damping torque analysis and negative damping source identification module, a data-driven impedance prediction and parameter self-tuning module, a hybrid current limiting protection module, a power distribution and current sharing control module, and a status monitoring and fault diagnosis module. The PCSs are interconnected via proximity communication links, i.e., CAN bus or fiber optic Ethernet, forming a decentralized ring or mesh communication topology. Data flow between modules is indicated by arrowed lines. These ten functional modules are deployed in the local controller of each PCS, and the PCS controllers are connected via proximity communication links to form a decentralized peer-to-peer control network. Specifically:

[0370] (1) Module 1, Multi-source status acquisition module: Acquires the three-phase output voltage of this PCS. Three-phase output current DC bus voltage and PCC point voltage And the status variables sent by neighboring PCS through the communication link, and the collection period. The time is no more than 100μs; the collected data is filtered and converted from analog to digital, and then sent to the frequency band impedance reshaping module and the online impedance identification module.

[0371] (2) Module 2, frequency band impedance reshaping module: The output impedance of this PCS is divided into three frequency bands: low frequency band, mid frequency band and high frequency band. Each frequency band is configured with an independent virtual impedance transfer function. The frequency of the low frequency band is less than 10Hz, the frequency of the mid frequency band is greater than or equal to 10Hz and less than or equal to 2kHz, and the frequency of the high frequency band is greater than 2kHz. The low frequency band is mainly inductive to ensure power distribution accuracy, the mid frequency band is mainly resistive to suppress circulating current resonance, and the high frequency band is mainly resistive to attenuate switching ripple coupling. The virtual impedance outputs of the three frequency bands are superimposed and applied to the voltage loop reference value.

[0372] (3) Module 3, Adaptive Virtual Impedance Adjustment Unit: Receives the real-time grid short-circuit ratio RSCR and the equivalent output impedance of the local PCS from the online impedance identification module, and dynamically adjusts the amplitude and phase angle of the virtual impedance of each frequency band in the frequency band impedance reshaping module according to the adaptive adjustment rules; when RSCR decreases and the grid weakens, it automatically increases the virtual impedance of the mid-frequency band to enhance the circulating current suppression capability; when RSCR increases and the grid strengthens, it appropriately decreases the virtual impedance to reduce voltage drop.

[0373] (4) Module 4, Decentralized Peer-to-Peer Collaborative Control Module: Each PCS exchanges limited state quantities, including its own active power, only with physically adjacent PCS via CAN bus or fiber optic Ethernet. reactive power RMS value of circulation With the health status indicator H, each PCS runs a discrete-time average consensus algorithm to evenly distribute global power and suppress parallel circulating currents through multiple rounds of iteration. This module does not rely on a central controller. When any single or multiple PCSs are offline, the remaining PCS can automatically reconstruct the communication topology and reconverge.

[0374] (5) Module 5, Online Impedance Identification Module: According to the configurable identification cycle The default duration is 30 seconds. A low-amplitude broadband probe signal (i.e., a multi-frequency sinusoidal superposition signal or a pseudo-random binary sequence) is superimposed on the PCC voltage. By acquiring the voltage and current responses during the probe period, the frequency domain impedance characteristics are extracted using Fast Fourier Transform (FFT), and the equivalent output impedance of the local PCS in the low-frequency, mid-frequency, and high-frequency bands is output. and the equivalent impedance of the power grid at PCC The identification results are sent to the adaptive adjustment unit and the data-driven parameter self-tuning module;

[0375] (6) Module Six, Damping Torque Analysis and Negative Damping Source Identification Module: Based on the online impedance identification results and current control parameters, the system state matrix is ​​constructed using a small-signal linearization model, and the damping ratio of each oscillation mode is calculated through eigenvalue decomposition; when the damping ratio of a certain oscillation mode is lower than the safety threshold... When the default value is 0.05, the negative damping source of the mode is located through damping torque analysis. For example, if the virtual inertia loop or power loop gain of a certain PCS is too large, a targeted damping enhancement correction amount is output to the frequency band impedance reshaping module of that PCS.

[0376] (7) Module 7, Data-Driven Impedance Prediction and Parameter Self-Tuning Module: Acquire the most recent Impedance identification data for each identification cycle, by default Based on approximately 50 minutes of operational data, a lightweight time-series prediction network, such as a single-layer gated cyclic unit with a hidden layer dimension not exceeding 32, is used to predict the changing trends of the grid impedance and PCS output impedance in the next identification cycle. The prediction results are then physically constrained and projected to ensure that the predicted impedance is within the stable domain, allowing for advance adjustment of the frequency band reshaping parameters; this achieves parameter feedforward pre-adjustment. The prediction network runs on the local controller, with an inference time not exceeding 2ms.

[0377] (8) Module 8, Hybrid Current Limiting Protection Module: Under normal operating conditions, it maintains the voltage source output behavior and limits the steady-state circulating current through frequency-segmented virtual impedance; when the instantaneous value of the output current exceeds the current limiting threshold, it will provide protection. When the default value is 1.2 times the rated value, it immediately switches to direct current limiting mode to maintain PCC voltage support as much as possible while ensuring current safety; when the current drops below the threshold and remains stable for a certain period of time... The default setting is 100ms before smoothly switching back to virtual impedance mode; the switching process uses a linear transition to avoid transient impacts.

[0378] (9) Module Nine, Power Allocation and Current Sharing Control Module: The initial allocation of active and reactive power is achieved based on the nonlinear droop characteristics; the active power adopts an exponential Pf droop mapping, which provides a larger frequency margin under light load and a more accurate power sharing under heavy load. On the basis of the initial droop allocation, the power correction output of the decentralized peer-to-peer collaborative module is superimposed to achieve global power sharing.

[0379] (10) Module 10, Status Monitoring and Fault Diagnosis Module: Continuously monitors the operating status of each module of this PCS, including indicators such as impedance identification success rate, communication link delay, damping ratio margin, and current limiting mode trigger frequency. When an anomaly is detected, a fault event is generated and recorded, and the neighboring PCS is notified through the health status flag H. Faults are divided into three levels: L1 information level, normal operation; L2 warning level, derating operation; and L3 fault level, active offline and notification for reconfiguration.

[0380] Between the above modules: (1) The data collected by module one is simultaneously sent to (2) module two and (5) module five. (5) The identification results of module five are sent to (3) module three, (6) module six and (7) module seven. (3) module three adjusts the parameters of each frequency band of (2) module two. (6) module six outputs the damping enhancement correction amount to (2) module two. (7) module seven outputs the feedforward pre-adjustment parameters to (2) module two. (4) The collaborative correction amount of module four is sent to (9) module nine and superimposed on the power distribution output. (9) The power command of module nine is sent to (2) module two to form the final voltage reference. (8) module eight implements current limiting protection at the output of module two. (10) module ten receives the status information of all modules.

[0381] In practical use, the parallel circulating current suppression and coordinated stability control method and system of the present invention for energy storage power stations with multiple converters are applicable to the parallel operation scenario of 4 to 20 grid-type PCS in a 100-megawatt grid-connected energy storage power station.

[0382] Example 1:

[0383] This embodiment describes the normal operating conditions of a 100 MW-level energy storage power station. A grid-connected energy storage power station is equipped with eight grid-type PCS units with a rated power of 250kW, connected in parallel via an AC bus and then connected to a 10kV distribution network through a step-up transformer. The grid short-circuit ratio is approximately 8. The power station operates at a rated power of 2MW for frequency regulation ancillary services. The filter inductance deviation range of the eight PCS units is ±5%, with a nominal value of 1.5mH. The cable impedance difference between each PCS unit is ±0.01Ω. The initial system state is uniform power distribution. The grid frequency fluctuates between 49.9Hz and 50.1Hz.

[0384] The control method is as follows: After each PCS is started, the initial impedance identification in step S8 is performed to obtain the PCS output impedance and grid impedance reference values ​​for each frequency band, and the RSCR is calculated to be approximately 8.

[0385] Next, the frequency band impedance reshaping module loads the reference parameters: low frequency band. =0.005Ω t=0.5mH; Mid-frequency band =0.5Ω; High-frequency band =0.1Ω.

[0386] Next, the decentralized peer-to-peer collaboration module runs a consensus algorithm, with each PCS exchanging power and circulating current data every 5ms. Consensus step size. =0.2, and after about 3 rounds of iteration, the active power deviation of the 8 PCS converged to within ±1%.

[0387] Next, an exponential droop mapping is used, with β=3 providing the initial power allocation, and a fine-grained equal distribution is achieved by superimposing a consistency correction.

[0388] Next, online impedance identification is performed every 30 seconds, injecting a 200ms multi-frequency sinusoidal signal. The identification results are stable, and the parameters remain basically unchanged.

[0389] Then, the data-driven GRU network started running after accumulating 20 identification cycles of data, and the deviation between the prediction results and the actual measurements was less than 5%.

[0390] The final effective value of the parallel circulating current is 13A, which is 3.6% of the rated peak current of 360A. This decreases to 5.8A (1.6%) due to the reshaping effect in the frequency division band. The power distribution deviation is initially ±4.2%, which is ±1.1% after collaborative correction. The minimum system damping ratio is 0.12, far exceeding the safety threshold of 0.05.

[0391] Under normal operating conditions, the frequency band impedance reshaping reduces the effective value of the circulating current by more than 55%, and the peer-to-peer collaborative correction reduces the power deviation by 74%, providing sufficient system stability margin.

[0392] The simulation results of this embodiment are shown in the triptych. Figure 5 As shown in the diagram. The top panel displays the active power time curves of the eight PCS units, illustrating the process from startup to steady-state equalization. The middle panel shows the RMS value time curve of the parallel circulating current, demonstrating the comparison before and after the frequency band impedance reshaping is enabled. The bottom panel shows a partial magnification of the output current waveforms of each PCS, showcasing the circulating current suppression effect.

Claims

1. A method for suppressing parallel circulating current and coordinating stable control of multiple converters in an energy storage power station, characterized in that: Includes the following steps: S1, Multi-source state acquisition and preprocessing: based on the acquisition cycle Collect the three-phase output voltage of a certain PCS Three-phase output current DC bus voltage and PCC point voltage The collected data were subjected to Clark transformation to obtain the two-phase stationary coordinate system. and Then perform the Park transformation to obtain the synchronous rotating coordinate system. and Calculate the instantaneous active power of this PCS. and reactive power and the effective value of circulation ; S2, Nonlinear droop initial power allocation: Calculation of the frequency reference of the local PCS based on exponential Pf droop mapping. and voltage amplitude reference Simultaneously calculate the voltage offset corresponding to the reactive power; S3, Decentralized peer-to-peer collaborative correction: Receive state variables sent by neighboring PCS via the nearest neighbor communication link. Run the consensus algorithm to calculate the active power correction. and reactive power correction amount The correction amount is superimposed on the droop output of step S2 to form a frequency reference and voltage reference after global collaborative correction. S4, Loading frequency division impedance reshaping parameters: Reads the currently valid virtual impedance parameters for the frequency division, including the virtual inductance for the low-frequency band. and virtual resistance Mid-frequency virtual resistance High-frequency virtual resistance If feedforward pre-adjustment parameters, adaptive correction parameters, or damping enhancement corrections have been output, these corrections will be superimposed on the virtual impedance parameters of the corresponding frequency band. S5: Frequency band impedance reshaping voltage correction calculation: Calculate the output current of this PCS unit. The impedance is directly input into the virtual impedance transfer function of each frequency band, and frequency band isolation is achieved by using the filtering stage inside the transfer function. The corresponding impedance voltage drop is then calculated. , , The total impedance reshaping correction is obtained by superimposing the voltage drops of the three impedance segments. The correction amount is subtracted from the voltage reference output in step S3 to form the final voltage loop reference value; S6, Voltage and Current Dual-Loop Control and PWM Generation: The final voltage reference value from step S5 is sent to the outer voltage loop PI controller to generate the current reference. The inner current loop PI controller tracks the current reference and outputs a modulated wave, which is then used to generate the switch drive signals for each bridge arm via space vector pulse width modulation (SVPWM). S7, Hybrid Current Limiting Mode Judgment and Execution: After the voltage and current loop outputs, detect whether the instantaneous value of the output current exceeds the current limiting threshold. If the limit is not exceeded, maintain the current virtual impedance control mode; If the current is exceeded, switch to direct current limiting mode to limit the current to a certain level. Within this range; once the current drops and stabilizes, the virtual impedance mode is restored via a linear transition; S8, Online Impedance Identification and Execution: Determines whether the identification cycle has been reached. If it reaches the target value, a low-amplitude wideband detection signal is superimposed on the PCC voltage reference, and the injection duration is [not specified]. Voltage and current response data were collected during the detection period. Frequency domain impedance characteristics were extracted using FFT, and the PCS output impedance in the low-frequency, mid-frequency, and high-frequency bands was calculated. and grid impedance And calculate the real-time short-circuit ratio If identification fails, retain the identification result from the previous cycle; S9, Adaptive Virtual Impedance Parameter Adjustment: Based on the output of step S8 and The virtual impedance parameters of each frequency band are recalculated according to the adaptive adjustment rules. The parameter update adopts exponential moving average filtering to avoid sudden changes. The updated parameters take effect in the next control cycle. S10, Damping Torque Analysis and Negative Damping Source Location: Based on the identification results of step S8 and the current control parameters, construct the small-signal state matrix of the parallel system, calculate the eigenvalues, and determine the damping ratio of a certain oscillation mode. Below the safety threshold By using participation factor analysis to locate the dominant contribution source of this mode, a damping enhancement correction amount for the corresponding frequency band of this PCS is generated. Send to step S4; S11, Data-driven impedance prediction and feedforward tuning: Store the identification results from step S8 into the historical buffer. When the amount of data in the buffer reaches the minimum training length... The lightweight time-series prediction network is used to predict the trend of grid impedance and PCS impedance changes in the next cycle. After the prediction results are projected by physical constraints, the feedforward parameter adjustment is calculated and applied to step S4 in advance before the next identification cycle arrives to achieve feedforward pre-adjustment. S12, Communication Status Check and Topology Reconstruction: Check the communication link status with each neighboring PCS. If the communication of a neighboring PCS times out, mark it as offline and remove it from the neighbor set of the consensus algorithm, triggering communication topology reconstruction. If the local PCS loses connection with all neighbors, it falls back to pure local control mode and relies only on frequency band impedance reshaping and droop control to maintain operation. S13, Fault Diagnosis and Health Status Update: Based on the operational indicators of each module, calculate the overall health score of this PCS. Update the health status flag and broadcast it to neighboring PCS via the communication link; S14, return to step S1 to execute the next control cycle.

2. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S2, the frequency reference of the local PCS is calculated based on the exponential Pf droop mapping. and voltage amplitude reference The specific methods are as follows: ,in: This is the reference value for the output angular frequency of this PCS, in rad / s; This is the system's rated angular frequency; for a 50Hz system, it is... ; The active power droop factor has a range of values. ; This is the rated active power of this PCS, in watts (W). This represents the current active power output of this PCS. Under bidirectional flow conditions, discharge is positive and charging is negative, and the unit is W. It is a symbolic function; This is the absolute value of active power; is the exponential drooping shape factor, with a value ranging from 1 to 5, and is dimensionless; It is the base of the natural logarithm; Voltage amplitude droop is: ,in: This is the reference value for the output voltage amplitude of this PCS, in V; The rated voltage amplitude of the system, in volts (V). The reactive power droop factor has a range of values. ; This represents the current reactive power output of this PCS, measured in Var.

3. The method for suppressing parallel circulating current and coordinating stable control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S3, the active power correction is calculated using a consensus algorithm. and reactive power correction amount The specific methods are as follows: ， in: For the first PCS in Taiwan The active power state estimate for each communication cycle, in W; The active power state estimate, after consistency correction, is used for calculation and broadcasting in the next communication cycle, in W. This is the consistency step size coefficient, ranging from 0.01 to 0.5, and is dimensionless. For the first The nearest neighbor set of PCS contains all nodes that are related to the first node. Each PCS has a PCS number associated with its communication link; This is the communication weighting coefficient, ranging from 0 to 1, and is dimensionless; default value. ; For the neighboring number PCS in Taiwan The active power state estimate for each communication cycle, in W; The active power correction is: ,in: For the first The active power correction for each PCS, in W; The uniform power correction gain has a value ranging from 0.1 to 2.0 and is dimensionless. The global average active power obtained when the consensus algorithm converges to equilibrium is, i.e. The unit is W; This represents the current actual physical output active power of this PCS, in W. The consistency protocol and correction formula for reactive power are the same as those for active power, with active power being replaced by reactive power accordingly.

4. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S5, the low-frequency virtual impedance transfer function is: ,in: For low-frequency virtual impedance Domain transfer function, in Ω; This is a virtual resistance for the low-frequency band, with a value ranging from 0.0010 to 0.050 Ω; This is a low-frequency virtual inductance, with a value ranging from 0.1mH to 2mH; For the complex frequency variable of the Laplace transform; The virtual impedance transfer function in the mid-frequency band is: ,in: For the mid-frequency virtual impedance Domain transfer function, in Ω; This represents the peak value of the virtual resistance in the mid-frequency band, ranging from 0.1Ω to 2Ω. This is the low-frequency cutoff angular frequency. The default frequency is 10Hz, and the unit is rad / s; It is the high-frequency cutoff angular frequency. The default is 2kHz, and the unit is rad / s; The virtual impedance transfer function in the high-frequency band is: ,in: For high-frequency virtual impedance Domain transfer function, in Ω; This is a virtual resistance for the high-frequency band, with a value ranging from 0.05Ω to 0.5Ω; It is the high-frequency cutoff angular frequency. The default is 2kHz, and the unit is rad / s; The calculation method for the total impedance reshaping correction is as follows: ,in: For the total voltage correction of impedance reshaping in the frequency band, in In a synchronously rotating coordinate system, the unit is V; The output current of this PCS Coordinate system components, including and The unit is A; , , These are the virtual impedance transfer functions for low, medium, and high frequency bands, respectively.

5. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S7, the specific method for determining and executing the hybrid rate limiting mode is as follows: ,in: This is the final output voltage command, in volts (V). This is the voltage reference value before impedance reshaping in the frequency band, in V; This represents the total correction amount for impedance reshaping across frequency bands, expressed in volts (V). The amplitude of the three-phase instantaneous output current, in amperes (A). This is the current limiting threshold, which defaults to 1.2 times the peak rated current, in amperes (A). This is the equivalent impedance in current-limiting mode, in Ω. Mode switching uses a linear transition: ,in: This is the output voltage command during the transition period, in volts (V). This is the output voltage in virtual impedance mode, in V; This is the output voltage in direct limiting mode, in volts (V). This is a linear transition coefficient that changes linearly from 0 or 1 to 1 or 0 after a mode switch is triggered. The default transition time is [time value missing]. , dimensionless.

6. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S8, the method for calculating the real-time short-circuit ratio is as follows: ,in: For the first The real-time short-circuit ratio observed by PCS is dimensionless. The equivalent impedance of the power grid at the PCC obtained online at the fundamental frequency The amplitude at that point, in Ω; The rated voltage of the PCC is expressed in volts (V). For the first The rated capacity of a PCS is expressed in VA.

7. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S9, the adaptive adjustment rule is as follows: ,in: This is the new virtual resistance value for the mid-frequency band after adaptive adjustment, in Ω; This is the virtual resistance reference value for the mid-frequency band, provided by the offline design, in Ω; The short-circuit ratio is set to the design baseline, with a default value of 10, which is dimensionless. The real-time short-circuit ratio obtained through online identification is dimensionless; It is an adaptive adjustment index, ranging from 0.3 to 1.0, and is dimensionless; The parameter update after exponential moving average filtering is as follows: ,in: This represents the actual effective virtual resistance value in the mid-frequency band, in Ω. This is the exponential moving average filter coefficient, with a value ranging from 0.1 to 0.5, and is dimensionless. This is the virtual resistance value in the mid-frequency band of the previous identification cycle, in Ω.

8. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S10, the eigenvalue decomposition of the system state matrix is ​​specifically as follows: ,in: These are the eigenvalues ​​of the system state matrix; It is the identity matrix; The small-signal state matrix of the parallel system is composed of the control parameters, filter parameters and grid impedance parameters of each PCS. The damping ratio is calculated as follows: ,in: For the first The damping ratio of each oscillation mode is dimensionless and ranges from 1 to 2. to ; For the first The real part of each eigenvalue is represented by 1 / s; For the first The imaginary part of each eigenvalue is expressed in rad / s. The specific negative damping source involved in the factor localization is as follows: ,in: For the first The modality pair of the first Participation factors of each state variable, dimensionless, with values ​​ranging from 0 to 1; The right eigenvector matrix The Middle Line 1 Column elements; Left eigenvector matrix The Middle Line 1 Column elements; The specific amount of damping enhancement correction is as follows: ,in: In order to target the The virtual resistance enhancement correction for each underdamped mode, in Ω; The damping enhancement gain ranges from 0.5 to 5.0, and the unit is Ω; This is the damping ratio safety threshold, with a default value of 0.05, and is dimensionless. For the current number Damping ratio of each mode, dimensionless; For the first The maximum participation factor value for each modality is dimensionless.

9. The method for suppressing parallel circulating current and coordinating stability control of multiple converters in an energy storage power station according to claim 1, characterized in that: In step S11, the data-driven impedance prediction specifically involves: ,in: The predicted characteristic vector of the grid impedance for the next identification cycle includes impedance magnitudes and phase angles in the low, medium, and high frequency bands, and is dimensionless. It is a single-layer gated recurrent unit network with a hidden layer dimension not exceeding 32; For the first The characteristic vector of grid impedance for each identification cycle; To predict the window length, the value ranges from 5 to 20 and is dimensionless; Internal control calculations are specifically as follows: ； ； ； ;in: To update the gate output vector, the value range is [0,1], and it is dimensionless; The reset gate output vector has a value range of [0,1] and is dimensionless; is the candidate hidden state vector, with a value range of [-1, 1], and is dimensionless; Let be the hidden state vector at the current time step, with a dimension not exceeding 32 and dimensionless; This is the hidden state vector of the previous time step; The current time step is the input vector, and the impedance characteristics are given. , , The weight matrix was obtained through offline training. , , It is the bias vector; It is the sigmoid activation function; It is the hyperbolic tangent activation function; This is element-wise multiplication; This is a vector concatenation operation; The specific method for physical constraint projection is as follows: ,in: To the stable region The Euclidean projection operator has a stability region defined as the set of impedance parameters that satisfy the following conditions: ， This is the predicted value of the power grid impedance after projection with physical constraints. To close convex sets Orthogonal projection operator; , The upper and lower limits of the impedance amplitude are determined by the system's rated parameters and are expressed in Ω. This is the allowable range for the impedance phase angle, with a default value of [-90°, 90°]. Given impedance parameters Next Damping ratio of each oscillation mode; The specific pre-adjustment amount of the feedforward parameter is as follows: ,in: This is the feedforward pre-adjustment increment of the virtual resistance in the mid-frequency band, in Ω; This is the feedforward gain, ranging from 0.1 to 0.8, and is dimensionless. The optimal mid-frequency virtual resistance is calculated based on the predicted impedance after projection, in Ω. This is the currently active virtual resistance value for the mid-frequency band, in Ω.

10. A parallel circulating current suppression and coordinated stability control system for multiple converters in an energy storage power station, characterized in that: Includes the following modules: (1) Module 1, Multi-source status acquisition module: Acquires the three-phase output voltage of this PCS. Three-phase output current DC bus voltage and PCC point voltage The data includes the status variables sent by neighboring PCS via the communication link; after anti-aliasing filtering and analog-to-digital conversion, the collected data is simultaneously sent to the frequency band impedance reshaping module and the online impedance identification module. (2) Module 2, frequency band impedance reshaping module: The output impedance of this PCS is divided into three frequency bands: low frequency band, mid frequency band and high frequency band. Each frequency band is configured with an independent virtual impedance transfer function. (3) Module 3, Adaptive Virtual Impedance Adjustment Unit: Receives the real-time grid short-circuit ratio RSCR and the equivalent output impedance of the local PCS from the online impedance identification module, and dynamically adjusts the amplitude and phase angle of the virtual impedance of each frequency band in the frequency band impedance reshaping module according to the adaptive adjustment rules. (4) Module 4, Decentralized Peer-to-Peer Collaborative Control Module: Each PCS exchanges limited state quantities, including its own active power, only with physically adjacent PCS via CAN bus or fiber optic Ethernet. reactive power RMS value of circulation With the health status indicator H, each PCS runs a discrete-time average consensus algorithm to achieve global power distribution and suppress parallel circulation through multiple rounds of iteration; (5) Module 5, Online Impedance Identification Module: According to the configurable identification cycle A low-amplitude broadband probe signal is superimposed on the PCC voltage. By acquiring the voltage and current responses during the probe period, the frequency domain impedance characteristics are extracted using Fast Fourier Transform (FFT), and the equivalent output impedance of the local PCS in the low-frequency, mid-frequency, and high-frequency bands is output. and the equivalent impedance of the power grid at PCC The identification results are sent to the adaptive adjustment unit and the data-driven parameter self-tuning module; (6) Module 6, Damping Torque Analysis and Negative Damping Source Identification Module: Based on the online impedance identification results and current control parameters, the system state matrix is ​​constructed using a small-signal linearization model, and the damping ratio of each oscillation mode is calculated through eigenvalue decomposition; (7) Module 7, Data-Driven Impedance Prediction and Parameter Self-Tuning Module: Acquire the most recent The impedance identification data of each identification cycle is used to predict the trend of grid impedance and PCS output impedance changes in the next identification cycle using a lightweight time-series prediction network; the prediction results are then subjected to physical constraint projection to adjust the frequency band reshaping parameters in advance. (8) Module 8, Hybrid Current Limiting Protection Module: Under normal operating conditions, it maintains the voltage source output behavior and limits the steady-state circulating current through frequency-segmented virtual impedance; when the instantaneous value of the output current exceeds the current limiting threshold, it will provide protection. Immediately switch to direct current limiting mode; when the current falls below the threshold and remains stable for a certain period of time... Then, smoothly switch back to virtual impedance mode; (9) Module Nine, Power Distribution and Current Sharing Control Module: Based on the nonlinear droop characteristics, the initial distribution of active and reactive power is realized; on the basis of the initial droop distribution, the power correction amount output by the decentralized peer-to-peer collaborative module is superimposed; (10) Module 10, Status Monitoring and Fault Diagnosis Module: continuously monitors the operating status of each module of this PCS; when an abnormality is detected, a fault event is generated and recorded, and at the same time, the neighboring PCS is notified through the health status flag H; Between the above modules: (1) The data collected by module one is simultaneously sent to (2) module two and (5) module five. (5) The identification results of module five are sent to (3) module three, (6) module six and (7) module seven. (3) module three adjusts the parameters of each frequency band of (2) module two. (6) module six outputs the damping enhancement correction amount to (2) module two. (7) module seven outputs the feedforward pre-adjustment parameters to (2) module two. (4) The collaborative correction amount of module four is sent to (9) module nine and superimposed on the power distribution output. (9) The power command of module nine is sent to (2) module two to form the final voltage reference. (8) module eight implements current limiting protection at the output of module two. (10) module ten receives the status information of all modules.

Images