Intelligent self-adaptive suppression method for oscillation of data-driven SVG (static var generator) grid-connected system

Abstract

The invention discloses an intelligent self-adaptive suppression method for oscillation of a data-driven SVG (static var generator) grid-connected system. Obtaining the capacitor voltage of the direct current side of the SVG and the actual voltage and the actual current of the grid connection point of the SVG, and respectively calculating the active power output by the SVG and the voltage amplitude output by the SVG according to the actual voltage and the actual current; a current reference value is generated through a voltage outer loop; the active power output by the SVG, the voltage amplitude output by the SVG and the voltage of the direct current side of the SVG are input into a DeePC controller to be processed, and an additional current reference value is obtained; and then the current error is obtained by adding the current reference value output by the voltage outer loop and subtracting the actual current, and then the current error is input to the voltage inner loop for processing to obtain a modulation wave voltage amplitude signal. According to the method, the oscillation of the converter grid-connected system can be suppressed under the condition that the dynamic characteristics of the converter and the conditions of the power grid are changed, and the problems of insufficient fixed structure and parameter sensitivity of the traditional SVG are solved.

Description

Technical Field

[0001] This invention relates to the field of stable grid-connected operation control technology for converters, and in particular to a data-driven SVG grid-connected system oscillation intelligent adaptive suppression method. Background Technology

[0002] In the field of renewable energy grid connection applications, grid-following (GFL) converters, with their mature control strategies, especially the precise synchronization with grid voltage achieved through phase-locked loops (PLLs), have become the mainstream grid connection interface solution. However, with the increasing penetration rate of renewable energy, the power system exhibits "weak grid" characteristics, with increased equivalent impedance and decreased short-circuit ratio (SCR), making the voltage amplitude and phase at the point of common coupling (PCC) susceptible to the influence of the converter's output current. Under these conditions, a complex interaction occurs between the dynamic characteristics of the PLL and the grid impedance, forming an unfavorable positive feedback path that may trigger small-signal instability in the system. This instability can cause oscillations or distortions in the converter's output current, and may even trigger protection disconnection, reducing power generation. Simultaneously, the oscillation power may propagate to other parts of the grid, threatening the stable operation of adjacent equipment and endangering the safety and reliability of the regional power grid.

[0003] Traditional grid-connected static var generators (SVGs) typically employ fixed-structure control strategies, such as a dual-loop proportional-integral (PI) regulator architecture. While performing well in strong grids or at specific operating points, their rigid design leads to significant limitations. Their control structure is fixed and parameter-sensitive, unable to adjust in real-time according to actual grid operating conditions (such as short-circuit ratio changes and load disturbances). Therefore, in weak grids or when operating conditions deviate from the design point, the controller struggles to maintain optimal damping characteristics, resulting in insufficient voltage and frequency support. Furthermore, traditional control relies on precise mathematical models, but the lack of modeling of converter dynamics (such as switching nonlinearity and measurement delay) and real-time changing grid impedance leads to severe model mismatch, further weakening its adaptability. In weak grids, fixed control may fail to effectively suppress oscillations and may even introduce new problems such as subsynchronous oscillations due to interaction with grid impedance, limiting its role in improving grid stability. Therefore, there is an urgent need to develop new adaptive control strategies to overcome these shortcomings.

[0004] In summary, it is necessary to study a data-driven SVG converter grid-connected system oscillation intelligent adaptive suppression technology. Summary of the Invention

[0005] To address system oscillations caused by phase-locked loops (PLLs) in weak power grids, the parameter sensitivity of traditional SVG fixed control structures, and the shortcomings of conventional methods in dealing with the variability and numerous uncertainties of real power grids, this invention provides a data-driven intelligent adaptive oscillation suppression method for SVG grid-connected systems.

[0006] The technical solution adopted in this invention is:

[0007] The method of the present invention includes the following steps:

[0008] S1. Obtain the capacitor voltage on the DC side of the SVG and the actual voltage and actual current of the SVG grid connection point in the dq coordinate system. Calculate the active power output of the SVG and the voltage amplitude output of the SVG based on the actual voltage and actual current.

[0009] S2. Subtract the preset reference value of the DC side capacitor voltage of the SVG from the DC side capacitor voltage to obtain the DC side voltage deviation. Subtract the preset reference value of the SVG output voltage amplitude from the voltage amplitude of the SVG output to obtain the voltage amplitude deviation. The DC side voltage deviation and the voltage amplitude deviation are used as the voltage outer loop input, and the current reference value is generated through the voltage outer loop.

[0010] S3. The active power output by the SVG, the voltage amplitude output by the SVG, and the DC side voltage of the SVG are used as inputs to the DeePC controller, and the additional current reference value is obtained after processing by the DeePC controller.

[0011] S4. Add the additional current reference value to the current reference value output by the outer voltage loop to obtain the reference value of the inner current loop. Subtract the actual current from the reference value of the inner current loop to obtain the current error, and then input it to the inner voltage loop for processing to obtain the modulated wave voltage amplitude signal.

[0012] S5. Generate a drive signal based on the modulated wave voltage amplitude signal and control the SVG.

[0013] The active power and voltage amplitude of the SVG output in step S1 are obtained by processing according to the following formula:

[0014]

[0015]

[0016] in, This represents the active power output by the SVG. and These represent the actual voltages at... axis, Components of the axis, and These represent the actual currents at... axis, Components of the axis, This indicates the voltage amplitude output by the SVG.

[0017] The DeePC controller in S3 processes the data according to the following steps:

[0018] S3.1. The active power output by the SVG, the voltage amplitude output by the SVG, and the voltage on the DC side of the SVG are used as the system output, and the components of the additional current reference value on the d-axis and q-axis are used as the system input.

[0019] S3.2 Apply two uncorrelated white noise excitations to the SVG system, use the amplitude of the white noise signal as the additional current reference value, collect the active power output of the SVG, the voltage amplitude of the SVG output, the voltage on the DC side of the SVG, and the additional current reference value for a first preset time length, and then construct the system input sequence and the system output sequence in time sequence;

[0020] S3.3 Construct the Hankel matrices of the system input sequence and the system output sequence respectively, and determine whether the Hankel matrix of the system input sequence satisfies the full row rank condition:

[0021] If satisfied, then the Hankel matrices of the system input sequence and the system output sequence are divided into blocks to obtain the block-based Hankel matrices.

[0022] If the condition is not met, proceed to step S3.2;

[0023] S3.4 Collect historical system input data and historical system output data for a second preset time length before the current moment, and then construct the current input sequence and the current output sequence;

[0024] S3.5 Construct an optimization function. Based on the current input sequence, the current output sequence, and the block-based Hankel matrix, construct constraints for the optimization function. Solve the optimization function to obtain the additional current reference sequence. Select the first value in the additional current reference sequence as the additional current reference value at the current moment.

[0025] Specifically, S3.3 involves constructing the system based on the system input sequence and the system output sequence respectively. Given the system input sequence and system output sequence, find the Hankel matrices and determine if the Hankel matrix of the system input sequence satisfies the full row rank condition.

[0026] If satisfied, then perform a block partitioning operation, prepending the Hankel matrices of the system input sequence and the system output sequence. The first row is used as the data matrix for historical input and future input, respectively. The last N rows of the Hankel matrix of the system input sequence and system output sequence are used as the data matrix for historical output and future output, respectively.

[0027] If the condition is not met, proceed to step S3.2.

[0028] The optimization function is set according to the following formula:

[0029]

[0030] in, and Let A and B represent the system's predicted input and predicted output, respectively. and feasible domain, As decision variables, and Let these represent the slack variables of the system's input and output, respectively. As the reference output trajectory of the system, , and They are respectively , and The scaling factor, For regularization terms, Let R represent the weighted L2 norm squared of the predicted input, where R is a predefined positive definite matrix. Let Q represent the weighted L2 norm squared difference between the predicted output and the system's reference output, where Q is a pre-defined positive semi-definite matrix. and Let L2 and L2 represent the squared values ​​of the slack variables of the system's input and output, respectively.

[0031] The constraints of the optimization function are set according to the following formula:

[0032]

[0033] in, , , and These represent data matrices representing historical inputs, historical outputs, future inputs, and future outputs, respectively, with g representing the decision variables. and These represent the current input sequence and the current output sequence, respectively. and Let these represent the slack variables of the system's input and output, respectively. and These represent the system's predicted input and predicted output, respectively.

[0034] The DeePC controller employs a rolling optimization method. The Hankel matrix after the block is constructed for the first time and remains fixed. The additional current reference value at each corresponding time is obtained by updating the current input sequence and the current output sequence at each time step and solving the optimization function.

[0035] This invention embeds a DeePC controller into an SVG (Static Var Generator) and employs behavioral systems theory to construct data-driven predictive control (DeePC). It directly utilizes collected voltage and other data to optimize the control sequence, avoiding the mismatch risk of traditional models. The resulting DeePC-Integrated SVG (DIS) provides effective oscillation damping under varying converter dynamics and grid conditions without requiring an explicit system model. This DIS overcomes the limitations of traditional SVGs with fixed structures and parameter sensitivity. Simultaneously, this model-free approach enhances the system's dynamic performance and anti-interference capabilities, effectively suppressing oscillations and improving the robustness of the SVG under changing grid conditions. It provides an effective solution for improving the small-disturbance stability of renewable energy-dominated grids.

[0036] The beneficial effects of this invention are:

[0037] This invention can suppress grid-connected converter system oscillations under varying converter dynamic characteristics and grid conditions. The DeePC controller uses input and output data to learn system behavior and perform control actions without requiring an explicit model. This data-driven intelligent adaptive oscillation suppression technology for SVG converter grid-connected systems overcomes the shortcomings of traditional SVG fixed structure and parameter sensitivity, and breaks through the limitations of traditional model-based systems. The DeePC controller avoids the mismatch risk of traditional models and can effectively suppress system oscillations caused by phase-locked loops under weak grid conditions. Simultaneously, it can provide a certain voltage and frequency support capability for grid-connected converters. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the method of the present invention.

[0039] Figure 2 This is a schematic diagram of the control structure of the present invention.

[0040] Figure 3 This is a schematic diagram of the simulation model of the parallel operation system of the data-driven SVG and the grid-connected converter in Example 1.

[0041] Figure 4 The waveform diagrams shown are those of the grid-connected converter in Example 1, where (a) represents the active power waveform diagram and (b) represents the reactive power waveform diagram.

[0042] Figure 5 The waveform diagrams shown are those of the grid-connected converter in Example 2, where (a) represents the active power waveform diagram and (b) represents the reactive power waveform diagram. Detailed Implementation

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0045] like Figure 1 As shown, this embodiment includes the following steps:

[0046] S1. Obtain the capacitor voltage on the DC side of the SVG and the actual voltage and actual current of the SVG grid connection point in the dq coordinate system. Calculate the active power output of the SVG and the voltage amplitude output of the SVG based on the actual voltage and actual current.

[0047] Specifically, the voltage and current of the SVG grid connection point and the capacitor voltage on the DC side of the SVG are collected. The voltage and current of the SVG grid connection point are transformed into coordinates through Park transformation to obtain the actual voltage and actual current in the dq coordinate system.

[0048] This involves acquiring the SVG output voltage information, obtaining the controller frequency through a phase-locked loop, and using the controller frequency integral as the phase input for the Park change and the phase signal of the modulated wave. It also involves acquiring the SVG output current and DC-side capacitor voltage information, and then applying a Park change to the SVG output current to obtain the corresponding... The axial component calculates the SVG output power based on the acquired SVG output current and output voltage information.

[0049] S2. Subtract the preset reference value of the DC side capacitor voltage of the SVG from the DC side capacitor voltage to obtain the DC side voltage deviation. Subtract the preset reference value of the SVG output voltage amplitude from the voltage amplitude of the SVG output to obtain the voltage amplitude deviation. The DC side voltage deviation and the voltage amplitude deviation are used as the voltage outer loop input. The current reference value in the dq coordinate system is generated through the voltage outer loop.

[0050] The SVG output current is used as the controlled variable, and a PI controller is designed to control the inner current loop; the SVG output voltage amplitude and the DC-side capacitor voltage are used as the controlled variables, and a PI controller is designed to control the outer voltage loop.

[0051] S3. The active power output by the SVG, the voltage amplitude output by the SVG, and the DC side voltage of the SVG are used as inputs to the DeePC controller, and the additional current reference value is obtained after processing by the DeePC controller.

[0052] This involves designing a DeePC controller, and obtaining the reference value for the current inner loop through the output of the DeePC controller and the output of the outer voltage loop.

[0053] S4. Add the additional current reference value to the current reference value in the dq coordinate system output of the voltage outer loop to obtain the reference value of the current inner loop. Subtract the actual current in the dq coordinate system from the reference value of the current inner loop to obtain the current error, and then input it to the voltage inner loop for processing to obtain the modulated wave voltage amplitude signal.

[0054] That is, set the reference value of SVG output voltage amplitude, the reference value of SVG output active power, the reference value of DC side capacitor voltage, PI controller parameters, and DeePC controller parameters to establish a data-driven SVG, and obtain the modulated wave voltage amplitude signal through current inner loop control.

[0055] Specifically, the reference value for the inner current loop is obtained through the output of the DeePC controller and the output of the outer voltage loop:

[0056]

[0057] In the formula, and This is the reference value for the inner current loop. and The voltage outer loop control output is used. and Output of the DeePC controller.

[0058] Design a PI controller to control the inner current loop:

[0059] The SVG uses an L-type filter, and the filter inductor on the SVG side is... The equivalent resistance of the filter inductor is The SVG output current is The SVG output voltage is The SVG modulated wave signal is According to the circuit law KVL, the filter inductance on the SVG side is obtained. Electrical model:

[0060]

[0061] Performing Park transformations yields Model in coordinate system:

[0062]

[0063] By performing a Laplace transform, we obtain the frequency domain model:

[0064]

[0065] It can be seen that the SVG output current is Coupling exists in the coordinate system, which can be resolved by adding a quantity that is the opposite of the coupling term to the controller. To facilitate the design of a PI controller, coupling can be added to the controller to achieve axis decoupling. shaft current Add coupling to the output of the PI controller ;exist shaft current Add coupling to the output of the PI controller This allows for the decoupling of the dq axis.

[0066] The design of the current loop controller mainly considers system stability and dynamic response speed. The open-loop transfer function and closed-loop transfer function of the current loop are as follows:

[0067]

[0068] In the formula, Let be the open-loop transfer function of the current loop. Let be the closed-loop transfer function of the current loop. For the Laplace operator, and These are the proportional and integral coefficients of the current loop PI controller, respectively. Selected based on the fast response requirements of the current loop. Based on the requirement of minimizing phase lag in PI control, the current loop control design is completed by designing the parameters of the current loop PI controller.

[0069] S5. Based on the modulated wave voltage amplitude signal, the three-phase reference voltage is obtained through inverse Parker transformation, and then the drive signal is generated to control the SVG.

[0070] The obtained modulation wave phase signal and modulation wave voltage amplitude signal are combined to construct a modulation wave signal, thereby generating an SVPWM pulse modulation wave signal that acts on the SVG to complete the control of the data-driven SVG.

[0071] Specifically, the following steps are taken: setting the SVG output voltage amplitude reference value, SVG output active power reference value, DC-side capacitor voltage reference value, PI controller parameters, and DeePC controller parameters; establishing a data-driven SVG; and obtaining the modulated wave voltage amplitude signal through current inner loop control.

[0072] Set the reference value for SVG output voltage amplitude. SVG output active power reference value DC side capacitor voltage reference value PI controller parameters: Phase-locked loop PI controller parameters and Current loop PI controller parameters and Voltage loop PI controller parameters , , and DeePC controller parameters: The timing lengths of the input and output trajectories are measured from the system by injecting white noise perturbation signals into the SVG. The time series length of the system's most recent input and output trajectories. Predicted length Control Time Domain Reference for system output trajectory Cost matrix and Regularization weights , and .

[0073] The modulated wave voltage amplitude signal obtained through current inner loop control is:

[0074]

[0075] In the formula, and These are the modulated wave voltage amplitude signals. shaft and Axial components. and These are the proportional and integral coefficients of the current inner-loop PI controller. and This is the reference value for the inner current loop. and For SVG output current shaft and Axial components. This is the filter inductor for the SVG side. and For SVG output voltage shaft and Axial components, This refers to the controller frequency.

[0076] The obtained modulation wave phase signal and modulation wave voltage amplitude signal are combined to construct a modulation wave signal, thereby generating an SVPWM pulse modulation wave signal that acts on the SVG to complete the control of the data-driven SVG. The specific steps include:

[0077] Consider the modulated wave phase signal obtained in 1). The modulated wave voltage amplitude signal obtained from (and 4) , Perform an inverse Park transformation to obtain a three-phase modulated wave signal in the abc coordinate system:

[0078]

[0079] In the formula, , and The modulated wave signal is in the abc coordinate system. and These are the modulated wave signals. shaft and Axial components.

[0080] The three-phase modulated wave signal is applied to the SVPWM generator to generate the pulse signal used to control the SVG, thus completing the control of the data-driven SVG.

[0081] Step S1 is as follows:

[0082] The voltage at the SVG grid connection point is collected using a voltage transformer and used as the SVG output voltage. .right By performing a Park transformation, the output voltage of the SVG is obtained. Axis components:

[0083]

[0084] In the formula, For controller frequency, , and SVG output voltage The voltages of phases A, B, and C, and These are the output voltages of the SVG. Axial components.

[0085] Obtain the SVG controller frequency via a phase-locked loop:

[0086]

[0087] In the formula, For controller frequency, For SVG output voltage Axial components, and These are the proportional and integral coefficients of the phase-locked loop PI controller, respectively.

[0088] Integrating the controller frequency yields the phase input of the Park variation and the phase signal of the modulated wave:

[0089]

[0090] In the formula, This refers to the controller frequency. The phase input for Park variation and the phase signal of the modulated wave.

[0091] Obtain the SVG output current and DC-side capacitor voltage information, and perform Park transformation on the SVG output current to obtain the corresponding... The axial component calculates the SVG output power based on the acquired SVG output current and output voltage information.

[0092] Obtain the DC-side capacitor voltage information of the SVG. The SVG uses an L-type filter for filtering. In a three-phase abc coordinate system, the current of the filter inductor on the SVG side is collected as the output current of the SVG through a current transformer. The SVG output current is then transformed to the dq coordinate system using the following Park transformation. and .

[0093]

[0094] In the formula, For controller frequency, , and The output current of the SVG is respectively The currents of phases A, B, and C, and These are the output currents of the SVG. shaft and Axial components.

[0095] Calculate the SVG output power:

[0096]

[0097] In the formula, and These represent the active power and reactive power output by the SVG, respectively. and These are the output voltages of the SVG. shaft and Axial components. and These are the output currents of the SVG. shaft and Axial components.

[0098] The active power and voltage amplitude of the SVG output in step S1 are obtained by processing them according to the following formula:

[0099]

[0100]

[0101] in, This represents the active power output by the SVG. and These represent the actual voltages at... axis, Components of the axis, and These represent the actual currents at... axis, Components of the axis, This indicates the voltage amplitude output by the SVG.

[0102] In step S2, the outer voltage loop is specifically processed according to the following formula:

[0103]

[0104] In the formula, and The voltage outer loop control output is used. and These are the SVG output voltage amplitude and its reference value, respectively. and These are the DC-side capacitor voltage of the SVG and its reference value. , , and These are the proportional and integral coefficients of the voltage outer loop PI controller, respectively. and Selected based on the fast response requirements of the voltage loop. and Based on the requirement of minimizing phase lag in PI control, the voltage loop control design is completed by designing the parameters of the voltage loop PI controller.

[0105] The DeePC controller in S3 is processed according to the following steps:

[0106] S3.1. The active power output by the SVG, the voltage amplitude output by the SVG, and the voltage on the DC side of the SVG are used as the system output, and the components of the additional current reference value on the d-axis and q-axis are used as the system input.

[0107] Specifically, the active power output of the SVG, the amplitude of the SVG output voltage, and the DC-side capacitor voltage of the SVG are taken as the system output and defined as the input of the DeePC controller, i.e.:

[0108]

[0109] In the formula, , and The DeePC controller is respectively in Input at any time This refers to the active power output by the SVG. This refers to the output voltage amplitude of the SVG. Let be the DC-side capacitor voltage of the SVG. Let the number of system outputs be . .

[0110] The additional current reference value is defined as the system input and as the output of the DeePC controller, i.e.

[0111]

[0112] In the formula, and For the DeePC controller in Output at any moment and Add a current reference value to the SVG. Let the number of system inputs be... .

[0113] S3.2 Apply two uncorrelated white noise excitations to the SVG system, and use the amplitudes of the two white noise signals as the components of the additional current reference value on the d-axis and q-axis, respectively. Collect the active power output of the SVG, the voltage amplitude of the SVG output, the voltage on the DC side of the SVG, and the additional current reference value for a first preset time length, and then construct the system input sequence and the system output sequence in time order.

[0114] Specifically, the first preset time length is T, and it is set as follows: ,in The number of system state variables. To predict the step size, The length of the most recent system input and output trajectories. Let be a set of integers. By applying white noise excitation to the SVG and measuring the system response, we obtain a set of integers with lengths of . Input trajectory and output trajectory ,in , , Represent the set of real numbers, defined , Therefore, a data matrix can be constructed based on this:

[0115]

[0116] S3.3 Construct the Hankel matrices of the system input sequence and the system output sequence respectively, and determine whether the Hankel matrix of the system input sequence satisfies the full row rank condition:

[0117] If satisfied, then the Hankel matrices of the system input sequence and the system output sequence are divided into blocks to obtain the block-based Hankel matrices.

[0118] If the condition is not met, proceed to step S3.2;

[0119] Specifically, using system data and Build A row-order Hankel matrix, i.e.:

[0120]

[0121]

[0122] White noise excitation is used to ensure the input signal... yes The order of continuous incentives, i.e. of The order Hankel matrix is ​​of full row rank:

[0123]

[0124] In the formula, A matrix with full row rank guarantees the input signal. yes It provides continuous motivation.

[0125] Then, the Hankel matrix is ​​partitioned into blocks, i.e.:

[0126]

[0127] In the formula, , , , ,and .

[0128] S3.1-S3.3 above are the steps that need to be performed during the first processing. They are not needed in subsequent second processing sessions and beyond.

[0129] S3.4 Collect historical system input data and historical system output data for a second preset time length before the current moment, and then construct the current input sequence and the current output sequence;

[0130] Specifically, in At any given time, the collection length is The system's most recent input and output trajectories, namely:

[0131]

[0132] Construct vectors using the system's most recent input and output trajectories. ,Right now:

[0133]

[0134] S3.5 Construct an optimization function. Based on the current input sequence, the current output sequence, and the block-based Hankel matrix, construct constraints for the optimization function. Solve the optimization function to obtain an additional current reference sequence of length N. Select the first value in the additional current reference sequence as the additional current reference value at the current moment.

[0135] S3.3 specifically involves constructing the system based on the system input sequence and the system output sequence respectively. Given the system input sequence and system output sequence, find the Hankel matrices and determine if the Hankel matrix of the system input sequence satisfies the full row rank condition.

[0136] If satisfied, then perform a block partitioning operation, prepending the Hankel matrices of the system input sequence and the system output sequence. The first row is used as the data matrix for historical input and future input, respectively. The last N rows of the Hankel matrix of the system input sequence and system output sequence are used as the data matrix for historical output and future output, respectively. The block-based Hankel matrix is ​​composed of the data matrices for historical input, future input, historical output, and future output. The preset length of the historical system input data and historical system output data is the second preset time length in S2.4, and N represents the preset prediction time domain length.

[0137] If the condition is not met, proceed to step S3.2.

[0138] The optimization function should be set according to the following formula:

[0139]

[0140] in, and Let A and B represent the system's predicted input and predicted output, respectively. and feasible domain, As decision variables, and Let these represent the slack variables of the system's input and output, respectively. As the reference output trajectory of the system, , and They are respectively , and The scaling factor, For regularization terms, Let R represent the weighted L2 norm squared of the predicted input, where R is a predefined positive definite matrix. Let Q represent the weighted L2 norm squared difference between the predicted output and the system's reference output, where Q is a pre-defined positive semi-definite matrix. and Let L2 and L2 represent the squared values ​​of the slack variables of the system's input and output, respectively.

[0141] For vectors quadratic form , =u or yr, P is the cost matrix, P=R or Q, R and Q are positive definite and positive semi definite matrices respectively; The second norm of vector S The square of S = or , = ; , and They are respectively , and The scaling factor.

[0142] Solving the optimization function will yield the optimal control sequence of the system. Here, Hankel matrix of order 1 Input and output data and Constructed The decision variables are represented as follows: The most recent system input and output trajectories are as follows: and The predicted trajectories are respectively represented as and The predicted time domain length is To enhance system robustness, slack variables are introduced. and and respectively with weighting coefficients and Constraints are applied to handle data compatibility conditions. Additionally, the objective function includes a term determined by the parameters. Scaling regularization terms , usually taken as The system control error and output tracking error are respectively determined by positive definite matrices. and positive semidefinite matrix Weighted penalty, output reference trajectory set The system operation must meet input constraints. and output constraints .

[0143] The constraints of the optimization function are set according to the following formula:

[0144]

[0145] in, , , and These represent data matrices representing historical inputs, historical outputs, future inputs, and future outputs, respectively, with g representing the decision variables. and These represent the initial input sequence and the initial output sequence, respectively. and Let these represent the slack variables of the system's input and output, respectively. y and y represent the system's input and output, respectively.

[0146] The DeePC controller employs a rolling optimization method. The block-based Hankel matrix, which consists of historical inputs, future inputs, historical outputs, and future outputs, remains fixed after its initial construction. The additional current reference value at each corresponding time is obtained by updating the current input sequence and the current output sequence at each time step and solving the optimization function.

[0147] That is, when calculating the additional current reference value for the first time, steps S3.1 to S3.5 are required, but from the next moment onwards, the calculation of the additional current reference value starts from S3.4, and the Hankel matrix after the block is always the one calculated the first time.

[0148] exist At all times Input into the system, where , To control the time domain. Replace with Recollecting length is The system's most recent input and output trajectories are used to reconstruct the system. And re-solve the optimization problem to obtain That is, the optimal control sequence is solved using a rolling optimization method. and .

[0149] like Figure 1 As shown, the first step is to obtain the SVG output voltage information through a voltage transformer and the controller frequency through a phase-locked loop (PLL). The integral of the controller frequency is used as the phase input for the Park change and the phase signal of the modulation wave, providing a foundation for subsequent control. The SVG output current information is then obtained through a current transformer. The DC-side capacitor voltage information of the SVG is also obtained. The SVG output current is then subjected to Park change to obtain the corresponding dq-axis components. Based on the obtained SVG output current and voltage information, the SVG output power is calculated. Using the SVG output current as the controlled variable, a PI controller is designed to control the inner current loop; using the SVG output voltage amplitude and DC-side capacitor voltage as controlled variables, a PI controller is designed to control the outer voltage loop; a DeePC controller is designed, and the reference value for the inner current loop is obtained through the output of the DeePC controller and the output of the outer voltage loop. The reference values ​​for the SVG output voltage amplitude, SVG output active power, DC-side capacitor voltage, PI controller parameters, and DeePC controller parameters are set to establish a data-driven SVG, and the modulation wave voltage amplitude signal is obtained through the inner current loop control. The obtained modulation wave phase signal and modulation wave voltage amplitude signal are combined to construct a modulation wave signal, thereby generating an SVPWM pulse modulation wave signal that acts on the SVG to complete the control of the data-driven SVG. Specific details of each control design can be found in [link to relevant documentation]. Figure 2 .

[0150] Figure 2 This is a control block diagram of the intelligent adaptive oscillation suppression technology for data-driven SVG converter grid-connected systems according to the present invention. Figure 2 middle For SVG side filter inductor, For transmission line inductance, For SVG output voltage, For SVG output current, For modulated wave voltage, This refers to the DC-side capacitor voltage of the SVG. This refers to the output power of the SVG.

[0151] Example 1:

[0152] Specific embodiments of the present invention are as follows:

[0153] Reference Figure 3This is an embodiment of the present invention, which is a system in which a data-driven SVG and a grid-connected converter operate in parallel. The SVG is controlled by both traditional SVG and data-driven SVG, and three different regularization terms are applied to the data-driven SVG. : Quadratic regularization term 1-norm regularization term and projection-based regularization terms The simulation parameters for the parallel operation system of the data-driven SVG and grid-connected converter based on the control strategy are shown in Table 1.

[0154] Table 1. Relevant parameters of the converter grid-connected system in the simulation verification of the embodiment.

[0155]

[0156] Among them, the SVG output voltage amplitude reference value SVG output active power reference value DC side capacitor voltage reference value Rated frequency value of power grid Phase-locked loop (PLL) PI controller parameters and The parameters of the current loop PI controller are: , The voltage loop PI controller parameters are as follows: , , and DeePC control parameters: The timing lengths of the input and output trajectories are measured from the system by injecting white noise disturbance signals into the SVG. The time series length of the system's most recent input and output trajectories. Predicted length Control Time Domain Reference for system output trajectory Cost matrix and Regularization weights , and .in, Indicates the order is The identity matrix, This represents a column vector with elements 1, 1, and 1.

[0157] To operate the parallel system in grid-connected mode, the converter is subjected to a disturbance at t=0.2s. The simulation is run, and the active and reactive power outputs of the system are recorded as follows: Figure 4 As shown.

[0158] Through observation Figure 4 The results show that the traditional SVG exhibits continuous and severe oscillations in both active and reactive power output from the GFL converter. This indicates that under weak grid conditions, the interaction between the PLL within the GFL converter and the grid impedance leads to small-signal instability in the system. Traditional fixed-parameter control strategies (i.e., the traditional SVG mode) cannot detect and suppress this oscillation. The data-driven SVG, after activating the DeePC controller at t=1.7s, rapidly and effectively suppresses the oscillation. The data-driven SVG, employing three different regularization terms, successfully smooths the oscillations, enabling the active and reactive power outputs of the GFL converter to quickly return to a stable state. Furthermore, the projection-based and quadratic regularization terms outperform the L1 regularization term. This highlights the core advantage of the data-driven approach: it does not rely on an accurate analytical system model but optimizes the control output online through real-time data, thus adaptively addressing complex instabilities caused by the interaction between grid impedance and the PLL. Simultaneously, under weak grid conditions, the data-driven SVG can provide certain voltage and frequency support for grid-connected converters.

[0159] Experimental simulations demonstrate that the intelligent adaptive oscillation suppression technology for data-driven SVG converter grid-connected systems proposed in this invention can provide effective oscillation damping under varying SVG dynamic characteristics and grid conditions, overcoming the limitations of traditional SVG's fixed structure and parameter sensitivity. The employed DeePC controller avoids the mismatch risk of traditional models and can effectively suppress system oscillations caused by phase-locked loops under weak grid conditions. Simultaneously, it can provide a certain voltage and frequency support capability for grid-connected converters.

[0160] Example 2:

[0161] To further illustrate, the intelligent adaptive oscillation suppression method for the data-driven SVG grid-connected system described in this invention still possesses oscillation suppression functionality even when the phase-locked loop (PLL) is affected by amplitude limiting. This is achieved by changing the PLL PI controller parameters in Example 1 to... and Other control parameters remain consistent with those in Example 1. Specifically, the SVG is controlled using both traditional SVG and data-driven SVG, with different regularization terms applied to the data-driven SVG for each. : Quadratic regularization term and projection-based regularization terms .

[0162] Run the parallel system in grid-connected mode, perform a simulation, and record the system's output active and reactive power, such as... Figure 5 As shown.

[0163] Through observation Figure 4The results show that under the traditional SVG operating mode, the active and reactive power output of the GFL converter exhibits continuous and significant fluctuations. This phenomenon reflects the interaction between the phase-locked loop (PLL) inside the GFL converter and the grid impedance in a weak grid environment, causing small-signal instability in the system. Fixed-parameter control strategies (i.e., traditional SVG operation) are insufficient to detect and suppress such oscillations. However, after the DeePC controller is activated in the data-driven SVG, the oscillations are quickly and effectively suppressed. Both data-driven SVGs using two different regularization terms successfully suppress the oscillations, allowing the active and reactive power output of the GFL converter to quickly return to a stable state. Furthermore, the projection-based regularization term outperforms the quadratic regularization term. This fully demonstrates the significant value of the data-driven strategy: this method does not rely on an accurate system analytical model but instead uses real-time data for online optimization, dynamically adjusting control commands to flexibly adapt to and effectively suppress dynamic instability induced by the interaction between grid impedance and the PLL. In addition, in weak grid operating environments, data-driven SVG can also provide the necessary voltage and frequency support for grid-connected converters (GFLs).

[0164] The above detailed embodiments illustrate the technical solution and beneficial effects of the present invention. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data-driven SVG grid-connected system oscillation intelligent adaptive suppression method, characterized in that, The method comprises the following steps: S1, obtaining the capacitor voltage of the SVG DC side and the actual voltage and current of the SVG grid-connected point in the dq coordinate system, and calculating the active power output by the SVG and the voltage amplitude output by the SVG according to the actual voltage and current respectively; S2, subtracting the reference value of the preset SVG DC side capacitor voltage from the capacitor voltage of the SVG DC side to obtain the DC side voltage deviation, subtracting the reference value of the preset SVG output voltage amplitude from the voltage amplitude output by the SVG to obtain the voltage amplitude deviation, and taking the DC side voltage deviation and the voltage amplitude deviation as the voltage outer loop input to generate the current reference value through the voltage outer loop; S3, taking the active power output by the SVG, the voltage amplitude output by the SVG and the voltage of the SVG DC side as the DeePC controller input, and obtaining the additional current reference value through DeePC controller processing; S4, adding the additional current reference value and the current reference value output by the voltage outer loop to obtain the reference value of the current inner loop, subtracting the actual current from the reference value of the current inner loop to obtain the current error, and then inputting to the voltage inner loop to obtain the modulation wave voltage amplitude signal; S5, generating a driving signal according to the modulation wave voltage amplitude signal and controlling the SVG.

2. The data-driven SVG grid-connected system oscillation intelligent adaptive damping method according to claim 1, characterized in that: The active power output by the SVG and the voltage amplitude output by the SVG in step S1 are processed according to the following formula: wherein represents the active power output by the SVG, and represent the actual voltage components in the axis, axis, and represent the actual current components in the axis, axis, represents the voltage magnitude output by the SVG.

3. The data-driven SVG grid-connected system oscillation intelligent adaptive damping method according to claim 1, characterized in that: The DeePC controller in S3 is processed according to the following steps: S3.1, taking the active power output by the SVG, the voltage amplitude output by the SVG and the voltage of the SVG DC side as the system output, and taking the components of the additional current reference value on the d-axis and the q-axis as the system input; S3.2, applying two unrelated white noise excitations to the SVG system, taking the white noise signal amplitude as the additional current reference value, collecting the active power output by the SVG, the voltage amplitude output by the SVG, the voltage of the SVG DC side and the additional current reference value for a first preset length of time, and then constructing the system input sequence and the system output sequence in time sequence; S3.3, constructing the Hankel matrix of the system input sequence and the system output sequence according to the system input sequence and the system output sequence respectively, and judging whether the Hankel matrix of the system input sequence satisfies the row full rank condition: If it is satisfied, the Hankel matrix after blocking is obtained by performing blocking operation on the Hankel matrix of the system input sequence and the system output sequence respectively; If it is not satisfied, jump to step S3.2; S3.4, collecting the historical system input data and the historical system output data for a second preset length of time before the current time, and then constructing the current input sequence and the current output sequence; S3.5, constructing an optimization function, constructing the constraint of the optimization function according to the current input sequence, the current output sequence and the Hankel matrix after blocking, solving the optimization function to obtain an additional current reference sequence, and selecting the first value in the additional current reference sequence as the additional current reference value at the current time.

4. The data-driven SVG grid-connected system oscillation intelligent adaptive damping method according to claim 3, characterized in that: The S3.3 is specifically: constructing Hankel matrixes of the system input sequence and the system output sequence according to the system input sequence and the system output sequence respectively, judging whether the Hankel matrix of the system input sequence satisfies the row full rank condition: the row full rank condition of the Hankel matrix of the system input sequence: If yes, then proceed to the block operation, taking the first N rows of the Hankel matrix of the system input sequence and the system output sequence as the data matrix of the historical input and the future input respectively, and taking the last N rows of the Hankel matrix of the system input sequence and the system output sequence as the data matrix of the historical output and the future output respectively. If yes, then proceed to the block operation, taking the first N rows of the Hankel matrix of the system input sequence and the system output sequence as the data matrix of the historical input and the future input respectively, and taking the last N rows of the Hankel matrix of the system input sequence and the system output sequence as the data matrix of the historical output and If it is not satisfied, jump to step S3.

2.

5. The data-driven SVG grid-connected system oscillation intelligent adaptive damping method according to claim 3, characterized in that: The optimization function is set according to the following formula: wherein, and respectively represent the predicted input and the predicted output of the system, A and B respectively represent and the feasible region of is the decision variable, and respectively represent the slack variables of the input and the output of the system, is the reference output trajectory of the system, , and are the scaling coefficients of , and respectively, is the regularization term, represents the weighted two-norm square of the predicted input, R is a pre-set positive definite matrix, represents the weighted two-norm square of the difference between the predicted output and the reference output of the system, Q is a pre-set semi-positive definite matrix, and respectively represent the two-norm square of the slack variables of the input and the output of the system.

6. The data-driven SVG grid-connected system oscillation intelligent adaptive damping method of claim 1, wherein: The constraint of the optimization function is set according to the following formula: wherein, , , and denote data matrices of historical inputs, historical outputs, future inputs and future outputs, respectively, g denotes a decision variable, and denote current input and output sequences, respectively, and denote slack variables for the input and output of the system, respectively, and denote predicted input and output of the system, respectively.

7. The data-driven SVG grid-connected system oscillation intelligent adaptive damping method according to claim 3, characterized in that: The DeePC controller adopts a method of rolling optimization, the Hankel matrix after being divided into blocks is kept fixed after being first constructed, the current input sequence and the current output sequence are updated at each time, and the additional current reference value at each corresponding time is obtained by solving an optimization function.

Images