Network construction type control method for cascaded SVG (Static Var Generator)

Abstract

The invention relates to the technical field of power electronics and power grid control, and discloses a network construction type control method for a cascaded SVG, and the method comprises the steps: calculating a three-phase reference voltage through a droop control strategy when a virtual synchronous generator meets a phase-locked loop-free condition and a swing equation constraint; virtual impedance is constructed, additional voltage drop is generated based on the virtual impedance, and the three-phase reference voltage is corrected; taking the corrected target three-phase voltage instruction as a reference voltage, constructing a cost function and a constraint condition, converting the cost function and the constraint condition into a finite time domain optimization problem, and solving the finite time domain optimization problem to obtain an optimal equivalent modulation quantity of a three-phase bridge arm; and updating the sub-module modulation quantity in the three-phase bridge arm based on the optimal equivalent modulation quantity, and executing a carrier phase shift pulse width modulation strategy. According to the invention, under the conditions of insufficient power grid supporting capability and obvious voltage fluctuation, accurate adjustment of the grid-connected point voltage and reactive power can be realized, and the voltage stability and the electric energy quality are improved.

Description

Technical Field

[0001] This invention relates to the field of power electronics and power grid control technology, and more specifically, to a grid-type control method for cascaded SVG. Background Technology

[0002] In recent years, with the large-scale integration of renewable energy, traditional rotating power sources in power systems have been gradually replaced by a high proportion of power electronic conversion equipment, resulting in a significant decrease in system inertia support capabilities. Under weak grid or even off-grid operation conditions, voltage is prone to large fluctuations, and the dynamic changes and disturbance sensitivity of grid impedance increase, placing higher demands on the grid-connected stability, dynamic response, and power quality of power electronic equipment. As a key power quality management and voltage support device, the operational capabilities of static var generators (SVG) in high-proportion renewable energy scenarios have attracted widespread attention.

[0003] To address the challenge of accurately locking the synchronous phase in weak power grids, grid-forming control has become a research hotspot in recent years. Among these, virtual synchronous generator (VSG) control provides virtual inertia and damping support to the grid by simulating the rotational characteristics of a synchronous machine, effectively enhancing voltage and frequency stability. However, existing VSG controls are mostly based on fixed virtual parameters, making it difficult to cope with rapid changes in grid impedance, load characteristics, and operating environment. This results in limited ability to suppress three-phase unbalanced currents and may even trigger secondary power oscillations.

[0004] On the other hand, modular multilevel (MMC) and other multi-submodule topologies are increasingly being used in SVG systems, where capacitor voltage balance and current limiting issues directly impact system safety. Traditional capacitor voltage balancing methods mostly rely on simple threshold adjustment or PI control, which struggle to balance voltage stability, dynamic response, and modulation redundancy utilization efficiency. Meanwhile, while virtual impedance can achieve current distribution and imbalance suppression, its parameter selection involves complex trade-offs and it lacks adaptability to system dynamic uncertainties.

[0005] In summary, under weak grid conditions and three-phase imbalance, realizing a grid-based hierarchical collaborative control strategy for cascaded SVG that possesses virtual inertia support, suppresses negative sequence current, and balances capacitance balance and current safety still faces significant challenges. There is an urgent need to propose a grid-based control method for cascaded SVG that can maintain high reliability and high dynamic performance in complex operating scenarios. Summary of the Invention

[0006] In view of the problems in related technologies, this invention proposes a network control method for cascaded SVG to overcome the above-mentioned technical problems existing in the existing related technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows:

[0008] According to a first aspect of the present invention, a network control method for cascaded SVG is provided, the method comprising:

[0009] When the virtual synchronous generator meets the conditions of no phase-locked loop and the constraints of the swing equation, it calculates the three-phase reference voltage through a droop control strategy.

[0010] Based on the pre-acquired grid-connected current amplitude, frequency change rate and bus voltage deviation, a virtual impedance is constructed. An additional voltage drop is generated based on the virtual impedance and the three-phase reference voltage is corrected to obtain the corrected target three-phase voltage command.

[0011] Using the corrected target three-phase voltage command as the reference voltage, a cost function and constraints are constructed, and the cost function and constraints are transformed into a finite-time domain optimization problem based on the equivalent modulation sequence. The optimal equivalent modulation of the three-phase bridge arm is obtained by solving the finite-time domain optimization problem.

[0012] The modulation amount of the three-phase bridge arm neutron module is updated based on the optimal equivalent modulation amount, and the carrier phase-shifting pulse width modulation strategy is executed to achieve capacitor voltage balancing and drive the static var generator output with the three-phase modulation signal.

[0013] Preferably, after updating the modulation amount of the submodule in the three-phase bridge arm based on the equivalent modulation amount and executing the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage equalization and drive the static var generator output with the three-phase modulation signal, the method further includes:

[0014] A deep reinforcement learning adaptive mechanism is constructed based on the current grid connection state variables, and the cost function weights and virtual impedance parameters are adaptively adjusted using the deep reinforcement learning adaptive mechanism.

[0015] Preferably, the step of constructing a virtual impedance based on the pre-acquired grid-connected current amplitude, frequency change rate, and bus voltage deviation, generating an additional voltage drop based on the virtual impedance, and correcting the three-phase reference voltage to obtain the corrected target three-phase voltage command includes:

[0016] Based on the pre-acquired grid-connected current amplitude, frequency change rate and bus voltage deviation, and combined with the fallback mechanism and sequence component acquisition mechanism, a virtual impedance is constructed.

[0017] Using a discrete implementation, the additional voltage drop is calculated in each sampling period based on the grid-connected current amplitude and the current virtual impedance parameters;

[0018] The three-phase reference voltage output by the virtual synchronous generator is corrected by adding a voltage drop, and the corrected target three-phase voltage command is output.

[0019] Preferably, the construction of the virtual impedance based on the pre-acquired grid-connected current amplitude, frequency change rate, and bus voltage deviation, combined with the fallback mechanism and sequence component acquisition mechanism, includes:

[0020] The positive sequence virtual resistance and positive sequence virtual inductance are calculated based on the grid current amplitude, frequency change rate and bus voltage deviation using the positive sequence adaptive law.

[0021] The same adaptive law is extended to the negative-sequence and zero-sequence components by using the amplification factor in negative-sequence suppression and zero-sequence suppression, so as to apply a high-amplitude virtual impedance to unbalanced current and zero-sequence circulating current.

[0022] After the fault or disturbance disappears, a fallback mechanism is executed, using a time constant to smoothly restore the virtual impedance to its initial value, in order to avoid chattering caused by frequent switching of the adaptive impedance near the trigger threshold.

[0023] Through three-phase transformation and low-pass filtering, the three-phase current and three-phase voltage are decomposed into positive-sequence, negative-sequence and zero-sequence quantities, respectively, to obtain time-varying impedance parameters, and virtual impedance is generated based on the time-varying impedance parameters.

[0024] Preferably, the step of using the modified target three-phase voltage command as a reference voltage to construct a cost function and constraints, and then transforming the cost function and constraints into a finite-time domain optimization problem based on an equivalent modulation sequence, and solving the finite-time domain optimization problem to obtain the equivalent modulation of the three-phase bridge arms includes:

[0025] The target three-phase voltage command generated and corrected by the virtual synchronous generator is used as the reference voltage to generate the cost function and constraints.

[0026] The candidate equivalent modulation sequence of the three-phase bridge arm in the preset sampling period is obtained. Based on the current grid-connected state quantity measured in the preset sampling period and the predefined discrete state space model, the candidate equivalent modulation sequence is rolled to predict the predicted value of the three-phase voltage, the predicted amplitude of the current and the voltage of the submodule capacitor at the grid-connected end.

[0027] The predicted values ​​of the three-phase voltage, the predicted amplitude of the current, and the capacitor voltage of the submodule are input into the cost function and constraints to form a finite-time domain optimization problem based on the equivalent modulation sequence.

[0028] An online numerical optimization algorithm is used to solve the finite time domain optimization problem, and the first equivalent modulation amount in the optimal sequence obtained by the solution is taken as the optimal equivalent modulation amount of the three-phase bridge arm.

[0029] Preferably, the step of updating the modulation amount of the submodule in the three-phase bridge arm based on the optimal equivalent modulation amount and executing the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage equalization and drive the static var generator output with the three-phase modulation signal includes:

[0030] The optimal equivalent modulation amount is converted into the equivalent modulation ratio of the three-phase bridge arm, and the optimal equivalent modulation amount is allocated to each sub-module according to the number of sub-modules in the three-phase bridge arm to obtain the reference duty offset of each sub-module.

[0031] The reference duty cycle bias of each submodule is superimposed with the duty cycle bias correction to achieve capacitor voltage balance.

[0032] The fault status of the submodule in the three-phase bridge arm is detected. When a submodule fault is detected, a redundancy bypass and redistribution strategy is executed.

[0033] Preferably, the detection of submodule fault status in the three-phase bridge arm, and the execution of a redundancy bypass and reallocation strategy when a submodule fault is detected, includes:

[0034] When a submodule fault is detected, the main switch drive signal of the faulty submodule is cut off, and the bypass branch switch of the faulty submodule is turned on, so that the capacitor of the faulty submodule is isolated from the power circuit. At the same time, the bridge arm current bypasses the faulty submodule through the bypass branch, so that the output voltage tends to be stable.

[0035] Determine if there is a usable redundant submodule in the current phase bridge arm. If there is a usable redundant submodule in the current phase bridge arm, select a redundant submodule whose capacitor voltage is within the allowable range from the standby set, switch the redundant submodule from standby state to working state, insert it into the original series position of the faulty submodule, and update the original mapping relationship.

[0036] If the redundant submodules in the current phase bridge arm are exhausted, obtain the number of remaining effective working submodules in the current phase bridge arm, and reallocate the optimal equivalent modulation amount of the current sampling period based on the number of effective working submodules.

[0037] Preferably, the current grid-connected status quantities include current grid-connected current, frequency change rate, bus voltage deviation, voltage tracking error, and capacitor voltage deviation of each submodule, etc.

[0038] Preferably, the step of constructing a deep reinforcement learning adaptive mechanism based on the current grid connection state variables, and using the deep reinforcement learning adaptive mechanism to adaptively adjust the cost function weights and virtual impedance parameters includes:

[0039] Within each reinforcement learning update cycle, the current grid-connected state quantity is collected according to the preset sampling period, and a state vector is formed based on the current grid-connected state quantity.

[0040] The action vector is calculated based on the state vector. The action vector is then projected into a predefined feasible region so that the cost function and virtual impedance parameters will operate according to the new parameters in the next few sampling periods.

[0041] During operation, the instant reward is calculated based on voltage tracking error, current limit overrun, and capacitor voltage imbalance. The state vector, action vector, instant reward, and state vector of the next sampling period are used as empirical samples to update the weights of the actor-critic network.

[0042] After one reinforcement learning update cycle, the cost function weights and parameter coefficients in the virtual impedance adaptive law are updated based on the optimal action vector obtained through convergence.

[0043] According to a second aspect of the present invention, a network-type control system for cascaded SVG is provided, the system comprising:

[0044] The upper-level virtual synchronous generator control module is used to calculate the reference frequency and three-phase reference voltage of the virtual synchronous generator through a droop control strategy when the phase-locked loop-free condition and swing equation constraint are met.

[0045] The parallel correction control module is used to construct a virtual impedance based on the pre-acquired grid current amplitude, frequency change rate and bus voltage deviation, generate an additional voltage drop based on the virtual impedance and correct the three-phase reference voltage to obtain the corrected target three-phase voltage command.

[0046] The middle layer voltage capacitor equalization control module is used to construct a cost function and constraints with the modified target three-phase voltage command as the reference voltage, and transform the cost function and constraints into a finite time domain optimization problem based on the equivalent modulation quantity sequence. Solving the finite time domain optimization problem yields the optimal equivalent modulation quantity of the three-phase bridge arm.

[0047] The underlying modulation execution module is used to update the modulation amount of the three-phase bridge arm neutron module based on the optimal equivalent modulation amount, and execute the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage balancing and drive the static var generator output with the three-phase modulation signal.

[0048] According to a third aspect of the present invention, a computer device is provided.

[0049] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0050] According to a fourth aspect of the present invention, a computer-readable storage medium is provided.

[0051] In one embodiment, a computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the above method.

[0052] The beneficial effects of this invention are as follows:

[0053] This invention enables rapid and precise regulation of grid connection point voltage and reactive power under conditions of insufficient grid support capacity and significant voltage fluctuations, thereby improving voltage stability and power quality. Simultaneously, it balances voltage across multiple submodule capacitors and limits current, enhancing system reliability. During grid faults or abnormal disturbances, it ensures the device possesses proactive grid connection and autonomous operation capabilities, achieving stable power supply without external synchronization. Furthermore, it enables smooth grid connection after the system returns to normal, ensuring the safe and efficient operation of the power conversion equipment. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments 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.

[0055] Figure 1 This is a flowchart of a network control method for cascaded SVG according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic block diagram of a network control system for cascaded SVG according to an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of the structure of a computer device;

[0058] Figure 4 This is a flowchart of constructing a virtual impedance in a network control method for cascaded SVG according to an embodiment of the present invention;

[0059] Figure 5 This is a flowchart of solving the optimal equivalent modulation amount in a network control method for cascaded SVG according to an embodiment of the present invention.

[0060] In the picture:

[0061] 1. Upper-layer virtual synchronous generator control module; 2. Parallel correction control module; 3. Middle-layer voltage-capacitor balancing control module; 4. Lower-layer modulation execution module. Detailed Implementation

[0062] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0063] According to an embodiment of the present invention, a network control method for cascaded SVG is provided.

[0064] Figure 1 An embodiment of a network control method for cascaded SVG according to the present invention is shown.

[0065] In this optional embodiment, the mesh control method for cascaded SVG includes:

[0066] S1. When the virtual synchronous generator meets the conditions of no phase-locked loop and the constraints of the swing equation, it calculates the three-phase reference voltage through the droop control strategy.

[0067] S2. Based on the pre-acquired grid-connected current amplitude, frequency change rate and bus voltage deviation, construct a virtual impedance, generate an additional voltage drop based on the virtual impedance and correct the three-phase reference voltage to obtain the corrected target three-phase voltage command.

[0068] Among them, such as Figure 4 As shown, the step of constructing a virtual impedance based on the pre-acquired grid-connected current amplitude, frequency change rate, and bus voltage deviation, generating an additional voltage drop based on the virtual impedance, and correcting the three-phase reference voltage to obtain the corrected target three-phase voltage command includes:

[0069] Virtual impedance is constructed based on the pre-acquired grid current amplitude, frequency change rate, and bus voltage deviation, combined with the fallback mechanism and sequence component acquisition mechanism.

[0070] The construction of virtual impedance based on pre-acquired grid-connected current amplitude, frequency change rate, and bus voltage deviation, combined with a fallback mechanism and a sequence component acquisition mechanism, includes:

[0071] The positive sequence virtual resistance and positive sequence virtual inductance are calculated based on the grid current amplitude, frequency change rate and bus voltage deviation using the positive sequence adaptive law.

[0072] The same adaptive law is extended to the negative-sequence and zero-sequence components by using the amplification factor in negative-sequence suppression and zero-sequence suppression, so as to apply a high-amplitude virtual impedance to unbalanced current and zero-sequence circulating current.

[0073] After the fault or disturbance disappears, a fallback mechanism is executed, using a time constant to smoothly restore the virtual impedance to its initial value, in order to avoid chattering caused by frequent switching of the adaptive impedance near the trigger threshold.

[0074] Through three-phase transformation and low-pass filtering, the three-phase current and three-phase voltage are decomposed into positive-sequence, negative-sequence and zero-sequence quantities, respectively, to obtain time-varying impedance parameters, and virtual impedance is generated based on the time-varying impedance parameters.

[0075] Using a discrete implementation, the additional voltage drop is calculated in each sampling period based on the grid-connected current amplitude and the current virtual impedance parameters;

[0076] The three-phase reference voltage output by the virtual synchronous generator is corrected by adding a voltage drop, and the corrected target three-phase voltage command is output.

[0077] S3. Using the corrected target three-phase voltage command as the reference voltage, construct the cost function and constraints, and transform the cost function and constraints into a finite-time domain optimization problem based on the equivalent modulation sequence. Solve the finite-time domain optimization problem to obtain the optimal equivalent modulation of the three-phase bridge arm.

[0078] Among them, such as Figure 5 As shown, the modified target three-phase voltage command is used as the reference voltage to construct a cost function and constraints, which are then transformed into a finite-time domain optimization problem based on an equivalent modulation sequence. Solving the finite-time domain optimization problem yields the equivalent modulation quantities of the three-phase bridge arms, including:

[0079] The target three-phase voltage command generated and corrected by the virtual synchronous generator is used as the reference voltage to generate the cost function and constraints.

[0080] The candidate equivalent modulation sequence of the three-phase bridge arm in the preset sampling period is obtained. Based on the current grid-connected state quantity measured in the preset sampling period and the predefined discrete state space model, the candidate equivalent modulation sequence is rolled to predict the predicted value of the three-phase voltage, the predicted amplitude of the current and the voltage of the submodule capacitor at the grid-connected end.

[0081] The predicted values ​​of the three-phase voltage, the predicted amplitude of the current, and the capacitor voltage of the submodule are input into the cost function and constraints to form a finite-time domain optimization problem based on the equivalent modulation sequence.

[0082] An online numerical optimization algorithm is used to solve the finite time domain optimization problem, and the first equivalent modulation amount in the optimal sequence obtained by the solution is taken as the optimal equivalent modulation amount of the three-phase bridge arm.

[0083] S4. Update the modulation amount of the three-phase bridge arm neutron module based on the optimal equivalent modulation amount, and execute the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage balancing and drive the static var generator output with the three-phase modulation signal.

[0084] The step of updating the modulation amount of the submodule in the three-phase bridge arm based on the optimal equivalent modulation amount and executing the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage equalization and drive the static var generator output with the three-phase modulation signal includes:

[0085] The optimal equivalent modulation amount is converted into the equivalent modulation ratio of the three-phase bridge arm, and the optimal equivalent modulation amount is allocated to each sub-module according to the number of sub-modules in the three-phase bridge arm to obtain the reference duty offset of each sub-module.

[0086] The reference duty cycle bias of each submodule is superimposed with the duty cycle bias correction to achieve capacitor voltage balance.

[0087] The fault status of the submodule in the three-phase bridge arm is detected. When a submodule fault is detected, a redundancy bypass and redistribution strategy is executed.

[0088] The detection of submodule fault status in the three-phase bridge arm, and the execution of redundancy bypass and reallocation strategies when a submodule fault is detected, include:

[0089] When a submodule fault is detected, the main switch drive signal of the faulty submodule is cut off, and the bypass branch switch of the faulty submodule is turned on, so that the capacitor of the faulty submodule is isolated from the power circuit. At the same time, the bridge arm current bypasses the faulty submodule through the bypass branch, so that the output voltage tends to be stable.

[0090] Determine if there is a usable redundant submodule in the current phase bridge arm. If there is a usable redundant submodule in the current phase bridge arm, select a redundant submodule whose capacitor voltage is within the allowable range from the standby set, switch the redundant submodule from standby state to working state, insert it into the original series position of the faulty submodule, and update the original mapping relationship.

[0091] If the redundant submodules in the current phase bridge arm are exhausted, obtain the number of remaining effective working submodules in the current phase bridge arm, and reallocate the optimal equivalent modulation amount of the current sampling period based on the number of effective working submodules.

[0092] The process of updating the modulation amount of the submodule in the three-phase bridge arm based on the equivalent modulation amount, and executing the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage balancing and drive the static var generator output with the three-phase modulation signal, further includes:

[0093] A deep reinforcement learning adaptive mechanism is constructed based on the current grid connection state variables, and the cost function weights and three-order virtual impedance parameters are adaptively adjusted using the deep reinforcement learning adaptive mechanism.

[0094] The step of constructing a deep reinforcement learning adaptive mechanism based on the current grid connection state variables, and using the deep reinforcement learning adaptive mechanism to adaptively adjust the cost function weights and the three-order virtual impedance parameters includes:

[0095] Within each reinforcement learning update cycle, the current grid-connected state quantities are collected according to a preset sampling period, and a state vector is formed based on the current grid-connected state quantities; the current grid-connected state quantities include the current grid-connected current, frequency change rate, bus voltage deviation, voltage tracking error, and capacitor voltage deviation of each sub-module, etc.

[0096] The action vector is calculated based on the state vector. The action vector is then projected into a predefined feasible region so that the cost function and virtual impedance parameters will operate according to the new parameters in the next few sampling periods.

[0097] During operation, the instant reward is calculated based on voltage tracking error, current limit overrun, and capacitor voltage imbalance. The state vector, action vector, instant reward, and state vector of the next sampling period are used as empirical samples to update the weights of the actor-critic network.

[0098] After one reinforcement learning update cycle, the cost function weights and parameter coefficients in the virtual impedance adaptive law are updated based on the optimal action vector obtained through convergence.

[0099] Figure 2 An embodiment of a network-type control system for cascaded SVG is shown.

[0100] In this optional embodiment, the network-type control system for cascaded SVG includes:

[0101] The upper-level virtual synchronous generator control module 1 is used to calculate the reference frequency and three-phase reference voltage by means of the droop control strategy when the virtual synchronous generator meets the conditions of no phase-locked loop and the constraints of the swing equation.

[0102] Parallel correction control module 2 is used to construct a three-sequence virtual impedance based on the pre-acquired grid current amplitude, frequency change rate and bus voltage deviation, generate an additional voltage drop based on the three-sequence virtual impedance and correct the three-phase reference voltage to obtain the corrected target three-phase voltage command.

[0103] The middle layer voltage capacitor equalization control module 3 is used to construct a cost function and constraints with the modified target three-phase voltage command as the reference voltage, and transform the cost function and constraints into a finite time domain optimization problem based on the equivalent modulation quantity sequence. Solving the finite time domain optimization problem yields the optimal equivalent modulation quantity of the three-phase bridge arm.

[0104] The underlying modulation execution module 4 is used to update the modulation amount of the three-phase bridge arm neutron module based on the optimal equivalent modulation amount and execute the carrier phase-shifting pulse width modulation strategy to achieve capacitor voltage balancing and drive the static var generator output with the three-phase modulation signal.

[0105] To facilitate understanding of the above technical solutions of the present invention, the following further explains the above technical solutions of the present invention from the perspective of architecture and principle, as follows:

[0106] (1) Upper-level VSG (Virtual Synchronous Generator) control (without PLL): Under the condition of no phase-locked loop, the virtual synchronous generator satisfies the oscillation equation:

[0107]

[0108] In the formula, J is the virtual moment of inertia; D is the damping coefficient; ω is the output angular frequency; θ is the phase angle; ω0 is the rated angular frequency; P e P represents the actual active power. ref For reference only.

[0109] And a reference is obtained through droop control:

[0110] ω ref =ω0-K P (P e -P ref );

[0111] V ref =V0-K Q (Q e -Q ref );

[0112] In the formula, K P ,K Q All are droop coefficients; V0 is the rated voltage amplitude; Q e Q represents the actual reactive power. ref For reactive power reference; ω ref The frequency command output by the upper-level VSG is used to update the virtual rotor angular velocity ω and phase angle θ in the oscillation equation. ref The voltage amplitude command output from the upper-level VSG is used together with the phase angle θ to generate a three-phase reference voltage via the three-phase sine wave generator module.

[0113] (2) Variable virtual impedance current limiting and imbalance suppression:

[0114] Based on the grid-connected current amplitude |i|, the frequency change rate |dω / dt|, and the bus voltage deviation ΔV=V meas -V ref Construct a three-order virtual impedance:

[0115]

[0116] In the formula, For positive / negative / zero sequence virtual impedance; is the time-varying resistive / inductive component; s is the Laplace operator; t is the real-time update time.

[0117] And generate an additional pressure drop Δvvi (t)=Z vi (s,t)i(t), obtained by correcting the VSG three-phase reference:

[0118]

[0119] In the formula, The three-phase reference voltage output by the VSG; v cmd,abc To issue the target three-phase voltage command (and v) to the MPC / PWM ref (equivalent); Δv vi The additional voltage drop is caused by the virtual impedance; the notation k+i|k represents the quantity observed and predicted from time k to k+i (MPC prediction notation).

[0120] In step (2), the discrete implementation of the virtual impedance is as follows:

[0121]

[0122] This voltage drop is used to correct the three-phase voltage reference of the upper VSG output, thereby enhancing damping and unbalanced current suppression capabilities.

[0123] Because the controller of this invention operates on a digital processor such as a DSP / FPGA with a sampling period T s Perform discrete calculations, the virtual impedance given in step (2) Since this is a continuous-domain expression, it cannot be directly implemented in discrete time. Therefore, this continuous virtual resistor-inductor network is equivalent to an additional voltage drop Δv related to the current i[k] and the previous sampled current i[k-1]. vi [k], which corresponds to the first-order differential inductor voltage and series resistor voltage.

[0124] With this discrete implementation, the controller only needs to consider the measured grid-connected current and the current virtual impedance parameter R in each sampling cycle. vi [k],L vi [k] Calculate Δv vi [k], and in accordance with the foregoing By correcting the three-phase reference voltage generated by the upper-level VSG according to the relationship, the equivalent function of continuous-time virtual impedance can be realized under the digital control framework.

[0125] (3) Mid-level MPC (Model Predictive Control) tracking control:

[0126] The target three-phase voltage command v is generated by the upper-level VSG and corrected by virtual impedance. cmd,abc (k+i|k) is used as the reference voltage:

[0127] v ref (k+i|k)=v cmd,abc (k+i|k);

[0128] Using the predicted grid-connected voltage as the tracking object, minimize the cost function:

[0129]

[0130] In the formula, v ac (k+i|k) represents the predicted three-phase voltage at the grid connection point; v ref (k+i|k) represents the reference three-phase voltage; i(k+i|k) represents the predicted current amplitude; v cj (k+i|k) represents the capacitor voltage of the j-th submodule; The average voltage in phase; w v ,w i ,w b For tracking / current limiting / capacitor balancing weights; N p For predicting the step size; N SM This represents the number of single-phase submodules.

[0131] And satisfy the following constraints:

[0132] |i(k+i∣k)|≤I max

[0133] d min ≤d j (k+i|k)≤d max ;

[0134] v c,min ≤v cj (k+i|k)≤v c,max

[0135] In the formula, d j (k) represents the duty cycle offset of the j-th submodule; v c,min ,v c,max For capacitor voltage safety limits; d j (k+i|k) represents the duty cycle offset for each submodule; d min and d max These are the minimum and maximum allowable values ​​for the duty cycle offset, respectively; I max This represents the maximum allowable amplitude of the bridge arm current.

[0136] In this embodiment, the optimization variable for the middle-layer MPC is selected as the future N. p The equivalent modulation sequence of each phase arm within a prediction step. Let the equivalent modulation of phases a, b, and c at time k+i be:

[0137] m(k+i|k)=[m a(k+i∣k),m b (k+i∣k),m c (k+i∣k)] T i = 0, ..., N p -1;

[0138] In the formula, m a (k+i|k) is the equivalent modulation amount of phase a at time k+i, m b (k+i|k) is the equivalent modulation amount of phase b at time k+i, m c (k+i|k) is the equivalent modulation amount of phase c at time k+i, and m(k+i|k) is related to the duty cycle offset d of each submodule. j (k+i|k) has a linear relationship, for example, for each phase arm (taking phase a as an example):

[0139]

[0140] Among them, phases b and c are similar, so that the equivalent modulation amount m(k+i|k) can be regarded as the "average control amount" of the duty cycle of all sub-modules in each phase.

[0141] At sampling time k, the currently measured state variables x(k) (including grid current, capacitor voltages of each submodule, etc.) are first used, combined with the discrete state-space model described in step (5), to process a given set of candidate equivalent modulation quantity sequences. Perform rolling prediction to obtain the corresponding v ac (k+i|k), i(k+i|k), v cj Substituting the predicted values ​​into the cost function and constraints given in this step, we arrive at a finite-time optimization problem concerning the equivalent modulation sequence:

[0142]

[0143] The optimization problem can be solved using online numerical optimization algorithms. For example, when the model is linear, the cost function is quadratic, and the constraints are linear, quadratic programming (QP) can be used; in the nonlinear case, sequential quadratic programming (SQP) or other mature MPC solvers can be used. The optimal sequence obtained by the solution... The first control variable m in * (k|k) represents the optimal equivalent modulation amount for the current sampling period. In actual execution, m... * (k|k) is converted into the equivalent modulation ratio of the three-phase bridge arm (i.e., the modulation coefficient normalized to the carrier amplitude), and then calculated according to the number of submodules N in each phase bridge arm. SM The allocation is performed to obtain the baseline duty cycle offset d for each submodule. j,0 (k), and then in step (4) the duty cycle offset correction Δd is superimposed.j (k) to achieve capacitor voltage equalization.

[0144] To achieve the mid-level MPC tracking control described in step (3) above, this embodiment adopts the following prediction model based on discrete state space:

[0145] MPC is based on a discrete state-space model:

[0146] x(k+1)=Ax(k)+Bu(k), y(k)=Cx(k);

[0147] In the formula, x(k) is the sum of grid-connected current and voltage of each SM capacitor; u(k) is the sum of duty cycle bias commands of each submodule; y(k) is the predicted voltage state value of the grid-connected side; x(k+1) is the state vector at the next sampling time; A, B, and C are the discretized state matrix, input matrix, and output matrix, respectively.

[0148] The discrete state-space model shown is used to characterize the cascaded SVG during the control period T. s The dynamic behavior within the middle-layer MPC is the basis for rolling prediction.

[0149] Specifically, in step (3), MPC needs to provide a set of future N values ​​at the current sampling time k. p The control variables (i.e., equivalent modulation amount or duty cycle bias command sequence) for each sampling period are used to predict the grid-connected three-phase voltage v for each future step based on the current state x(k). ac (k+i|k), current amplitude i(k+i|k), and capacitor voltage v of each submodule cj (k+i|k), the above predicted quantities are all calculated recursively by the discrete state-space model, and then substituted into the cost function and constraints given in step (3) to form a finite-time optimization problem.

[0150] Therefore, the purpose of step (5) is to provide a unified predictive model for the MPC tracking control described in step (3), enabling the controller to predict the future N... p Simultaneously considering voltage tracking, current limiting constraints, and capacitor voltage safety range within each sampling period, this enables proactive optimization control of the dynamic process of cascaded SVG.

[0151] (4) Execution Layer - Execution and Balance Control:

[0152] The equivalent modulation amount obtained from MPC is allocated to each submodule and executed via carrier phase-shifted PWM. Duty cycle bias correction is used to achieve capacitor voltage equalization.

[0153]

[0154] In the formula, Δdj (k) is the duty cycle offset correction term, k b The proportional coefficient for capacitor voltage equalization control, v cj (k) represents the capacitor voltage of the j-th submodule at time k. Δd is the average value of the capacitor voltages of all submodules. max This represents the maximum permissible amplitude of the duty cycle offset correction during a single adjustment.

[0155] When a submodule failure is detected, a redundancy bypass and redistribution strategy is executed.

[0156] 1) Redundant Bypass: Immediately shut off the main switch drive signal of the faulty submodule j and turn on its bypass branch switch, isolating the capacitor of the submodule from the power circuit. At the same time, the bridge arm current bypasses the submodule through the bypass branch, thus no longer affecting the system output voltage. At this time, the submodule no longer participates in subsequent duty cycle allocation and capacitor equalization control.

[0157] 2) Redundant Submodule Access and Mapping Update: If there are still available redundant submodules in the current phase bridge arm, select a redundant submodule j whose capacitor voltage is within the allowable range from the spare set. red Switch it from standby to working state, insert it into the serial position where the faulty submodule was originally located, and update the original mapping relationship as follows:

[0158]

[0159]

[0160] In the formula, j red d represents the redundant submodule number. j,0 (k) The reference duty cycle offset for submodule j; Δd j (k) This is the duty cycle offset correction term for submodule j; the symbol "←" indicates mapping or assigning the value of the right-hand quantity to the left-hand quantity.

[0161] In carrier phase-shift PWM, the carrier phase and modulation index originally corresponding to the faulty submodule are also transferred to the redundant submodule to ensure that the equivalent modulation amount of the bridge arm does not change abruptly. The faulty submodule remains in bypass state permanently until maintenance.

[0162] 3) Modulation redistribution without redundancy: When the redundant submodules in a certain phase arm are exhausted, the number of remaining effective working submodules in that phase arm is denoted as... The controller reallocates the equivalent modulation amount according to the remaining sub-modules. Taking phase a as an example, if the equivalent modulation amount of this phase in the current sampling period is... The baseline duty cycle offset for each working submodule is then recalculated as follows:

[0163]

[0164] In the formula, d j,0 (k) represents the baseline duty cycle offset for each working submodule. N min These represent the rated number of active working submodules, the current number of active working submodules, and the preset minimum number of active working submodules for a certain phase arm. The optimal equivalent modulation amount of phase a is obtained by MPC at sampling time k.

[0165] At the same time, maintain the duty cycle offset correction term Δd j The structure of (k) remains unchanged so that capacitor equalization control continues to be effective on the new set of submodules.

[0166] If necessary, the DC-side target voltage or the maximum allowable modulation ratio of that phase arm can be adjusted accordingly to avoid exceeding the voltage limit of a single submodule. When the number of effective submodules in a certain phase arm drops to the preset lower limit N... min When the following occurs, the controller issues a derating operation or shutdown signal and enters safe mode.

[0167] Virtual impedance adaptation and order component injection include:

[0168] (1) Positive sequence adaptive law (current limiting and anti-dynamic):

[0169]

[0170] In the formula, This is a positive-order adaptive law; The initial value of the positive-sequence virtual impedance; α1, α2, α3, β1 are all adjustment coefficients; |i| is the amplitude of the grid-connected current; I th This is the threshold for triggering rate limiting; ΔV = V meas -V ref This refers to the bus voltage deviation. This represents the rate of change of frequency.

[0171] (2) Negative / zero order suppression:

[0172]

[0173] In the formula, γ R ,γ L ,η R ,η L All are unbalanced and zero-sequence circulation suppression coefficients, and satisfy γ R ,γ L ,η R ,η L ≥1; These are the positive-sequence virtual impedance and positive-sequence virtual inductance, respectively, given by the upper-level controller; These are the negative-sequence virtual impedance and the negative-sequence virtual inductance, respectively. These are the zero-sequence virtual impedance and the zero-sequence virtual inductance, respectively. The update rate (i.e., the differential change at the sampling time) of the positive-sequence virtual impedance and positive-sequence virtual inductance in the fallback suppression circuit is used to achieve fallback control; among them, the three-sequence virtual impedances are all control quantities for suppressing imbalance, voltage distortion and low-order circulating current.

[0174] (3) Fallback mechanism to avoid chattering: when |i| th When -δ and |ΔV|<ε,

[0175] In the formula, δ,ε,τ>0, i[k] is the grid-connected current at discrete sampling time k, and T s R is the sampling period. vi [k],L vi [k] represents the resistive / inductive parameter at discrete time k, τ is the time constant, and Δv is calculated. vi [k] is used to correct the reference voltage.

[0176] (4) Sequence component acquisition: Positive / negative / zero sequence quantities are obtained by three-phase-αβ0 transformation and low-pass filtering, or by using a sequence component separator in a synchronous rotating coordinate system.

[0177] In this optional embodiment, the "positive sequence adaptive law," "negative / zero sequence suppression," "fallback mechanism to avoid chattering," and "sequence component acquisition" together constitute the virtual impedance. The specific implementation method is as follows: First, the positive-sequence adaptive law calculates the positive-sequence virtual resistance and inductance online based on the grid current amplitude |i|, the frequency change rate |dω / dt|, and the bus voltage deviation ΔV. This is used to automatically increase the equivalent damping when the current is too large or the disturbance is strong, thereby achieving current limiting and transient performance enhancement. On this basis, the amplification factor γ in the negative / zero sequence suppression is used. R ,γ L ,η R ,η L The ≥1 parameter extends the same adaptive law to the negative and zero-sequence components, thereby applying a stronger virtual impedance to unbalanced currents and zero-sequence circulating currents. To avoid chattering caused by frequent switching of the adaptive impedance near the trigger threshold, a fallback mechanism uses a time constant τ to smoothly restore the virtual impedance to its initial value after the fault or large disturbance disappears. Sequence component acquisition is achieved through three-phase αβ0 transformation and low-pass filtering, or a sequence component separator in a synchronous rotating coordinate system, which splits the three-phase current / voltage into positive, negative, and zero-sequence quantities, ensuring that each adaptive law and suppression coefficient accurately acts on the corresponding sequence component. The resulting time-varying impedance parameters are... Substituting the discrete additional voltage drop Δv vi ​The calculation of [k] is further performed through... Modify the three-phase voltage reference of the upper-level VSG.

[0178] The deep reinforcement learning adaptive mechanism involves setting up a DRL adaptive layer between the VSG and MPC / virtual impedance to adjust the objective function weights and virtual impedance parameters online, including:

[0179] (1) State vector (in T) s For sampling period, T rl |>>T s (For update cycle):

[0180]

[0181] In the formula, fault_flag is a fault / imbalance indicator (0 / 1 or multiple types); the other symbols are the same as in the previous formula.

[0182] (2) Motion vector (adjustable parameter):

[0183] a k =[w v ,w i ,w b ,I th ,α1,α2,α3,β1,γ R ,γ L ,η R ,η L ] T ;

[0184] In the formula, a k =[w v ,w i ,w b ,I th ,α1,α2,α3,β1,γ R ,γ L ,η R ,η L ] T This represents a vector of adjustable parameters. This indicates that the action is projected onto the feasible region Ω.

[0185] The example feasible domain is:

[0186]

[0187] And through projection Π Ω Maintain feasible domain:

[0188]

[0189] (3) The instant reward is the same as J:

[0190]

[0191] λ v ,λ i ,λ b ,λ a >0;

[0192] (4) Learning and online writing: The actor-critic algorithm with entropy regularization is used for pre-training and online updating; every T rl The cycle will be optimal Write:

[0193] (i) use Replace w in the minimum cost function F v ,w i ,w b Weight;

[0194] (ii) using Replace the positive order adaptive law coefficient I th ,α1,α2,α3,β1;

[0195] (iii) Use Update negative / zero sequence amplification factor γ R ,γ L ,η R ,η L .

[0196] Keep Δv vi (t)=Z vi The structure of (s,t)i(t) remains unchanged.

[0197] In this optional embodiment, calculating the state vector, calculating the action vector, and learning and online writing constitute a complete deep reinforcement learning closed-loop process: First, the controller performs each reinforcement learning update cycle T rl Within, according to the sampling period T s The system collects current grid-connected current, frequency change rate, bus voltage deviation, voltage tracking error, and capacitor voltage deviation of each submodule, and uses these data to form a state vector s. k This serves as the input to the DRL adaptive layer; then, the policy network adjusts the input based on the input. k The action vector a is calculated. k Each component w v ,w i ,w b ,I th ,α1,α2,α3,β1,γ R ,γ L ,η R ,η LThe adjustable parameters corresponding to the cost function weights and the virtual impedance adaptive law, after projection It then falls into the predefined feasible region Ω and is temporarily written into the controller, so that the MPC cost function J and the virtual impedance parameter in the subsequent sampling periods are related. Operate according to the new parameters; during operation, the controller calculates the real-time reward r based on voltage tracking error, current limit overrun, and capacitor voltage imbalance. k , will (s k ,a k ,r k ,s k+1 This serves as an empirical sample used to update the weights of the actor-critic network, enabling learning and online updates; after one update cycle T... rl Afterwards, the optimal action obtained by DRL convergence. It was officially written back, that is, it was used Replace the cost function weights, using The parameter coefficients in the virtual impedance adaptive law are replaced to drive the next stage of VSG-MPC-virtual impedance joint control, thereby realizing adaptive optimization of automatically adjusting control weights and impedance parameters according to the system operating status.

[0198] The "actor-critic network" in this invention adopts a typical Actor-Critic architecture, including:

[0199] Actor Network: Based on the current network state variable s k As input, the action vector a is output through a multi-layer fully connected neural network. k Its components correspond to the cost function weights and adjustable parameters in the virtual impedance adaptive law.

[0200] Critic Network: with (s k ,a k Given the input , the output is a state-action value estimate Q(s). k ,a k The network parameters are updated by minimizing the time-series difference error.

[0201] Update mechanism: In each reinforcement learning update cycle T rl >>T s Within, from the empirical sample (s) k ,a k ,r k ,s k+1 In batch learning, the critic network is updated first, and then the actor network parameters are updated according to the policy gradient. The optimal action obtained is written to replace the cost function weights and virtual impedance parameters, realizing online adaptive optimization of VSG-MPC-virtual impedance joint control.

[0202] Furthermore, the numerical range and implementation conditions in this invention also include:

[0203] T s ∈[50,200]μs;

[0204] N p ∈[8,20];

[0205] I max =1.1~1.5I rated ;

[0206] Select a value that ensures the increase in the equivalent impedance at the grid connection end does not exceed 20% of the system short-circuit reactance.

[0207] In summary, this invention achieves hierarchical collaborative control of the static var generator (SGR) through upper-level virtual synchronous generator control, mid-level voltage tracking and capacitor balancing control, and lower-level modulation execution. It utilizes a virtual synchronous machine to actively support grid voltage and frequency, introduces variable virtual impedance to improve three-phase unbalanced current and dynamic damping performance in weak grids, optimizes multiple objectives based on model predictive control to achieve precise voltage tracking, current limiting, and submodule capacitor voltage balancing, and employs duty cycle bias allocation and carrier phase-shifted PWM to achieve reliable modulation and redundancy for faulty submodules. This system can autonomously establish a grid during grid faults and smoothly reconnect to the grid after restoration, improving system stability, power quality, and equipment reliability. It is suitable for grid support and power quality management in scenarios with a high proportion of renewable energy integration.

[0208] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0209] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0210] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0211] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0212] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0213] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A network configuration type control method for a cascaded SVG, characterized by, The method comprises: The virtual synchronous generator meets the condition of no phase-locked loop and swing equation constraint, and calculates the three-phase reference voltage through the droop control strategy; According to the pre-acquired grid-connected current amplitude, frequency rate of change and bus voltage deviation, a virtual impedance is constructed, an additional voltage drop is generated based on the virtual impedance, and the three-phase reference voltage is corrected to obtain a corrected target three-phase voltage instruction; The corrected target three-phase voltage instruction is taken as a reference voltage, a cost function and a constraint condition are constructed, the cost function and the constraint condition are converted into a finite time domain optimization problem based on an equivalent modulation quantity sequence, and the optimal equivalent modulation quantity of the three-phase bridge arm is obtained by solving the finite time domain optimization problem; Based on the optimal equivalent modulation quantity, the modulation quantity of the sub-module in the three-phase bridge arm is updated, and the carrier phase-shifted pulse width modulation strategy is executed to realize capacitor voltage balancing and three-phase modulation signal driving static reactive power generator output.

2. The network configuration type control method for a cascaded SVG according to claim 1, characterized in that, After the three-phase bridge arm modulation quantity is updated based on the equivalent modulation quantity, the carrier phase-shifted pulse width modulation strategy is executed to realize capacitor voltage balancing and three-phase modulation signal driving static reactive power generator output, it further comprises: A deep reinforcement learning adaptive mechanism is constructed according to the current grid-connected state quantity, and the deep reinforcement learning adaptive mechanism is used to adaptively adjust the cost function weight and the virtual impedance parameter.

3. The network configuration type control method for a cascaded SVG according to claim 2, characterized in that, The current grid-connected state quantity includes the current grid-connected current, the frequency rate of change, the bus voltage deviation, the voltage tracking error and the capacitor voltage deviation of each sub-module.

4. The network configuration type control method for a cascaded SVG according to claim 1, characterized in that, The corrected target three-phase voltage instruction includes: Based on the pre-acquired grid-connected current amplitude, frequency rate of change and bus voltage deviation, and combined with a back-falling mechanism and a sequence component acquisition mechanism, a virtual impedance is constructed; In each sampling period, the additional voltage drop is calculated according to the grid-connected current amplitude and the current virtual impedance parameter by using a discrete implementation method; The three-phase reference voltage output by the virtual synchronous generator is corrected by the additional voltage drop, and the corrected target three-phase voltage instruction is output.

5. The network configuration type control method for a cascaded SVG according to claim 4, characterized in that, The virtual impedance is constructed based on the pre-acquired grid-connected current amplitude, frequency rate of change and bus voltage deviation, and combined with a back-falling mechanism and a sequence component acquisition mechanism, which includes: The positive sequence adaptive law is used to calculate the positive sequence virtual resistance and the positive sequence virtual inductance according to the grid-connected current amplitude, the frequency rate of change and the bus voltage deviation; The same adaptive law is extended to the negative sequence component and the zero sequence component through the amplification coefficients in the negative sequence suppression and the zero sequence suppression, so as to apply high-amplitude virtual impedance to the unbalanced current and the zero sequence current; After the fault or disturbance disappears, the back-falling mechanism is executed, and the time constant is used to make the virtual impedance smoothly recover to the initial value, so as to avoid the chattering caused by the frequent switching of the adaptive impedance near the trigger threshold; Through three-phase transformation and low-pass filtering mechanism, the three-phase current and the three-phase voltage are respectively split into positive sequence, negative sequence and zero sequence, time-varying impedance parameters are obtained, and the virtual impedance is generated based on the time-varying impedance parameters.

6. The network configuration type control method for a cascaded SVG according to claim 1, characterized in that, The modified target three-phase voltage instruction is taken as a reference voltage, a cost function and a constraint condition are constructed, and the cost function and the constraint condition are converted into a finite time domain optimization problem based on an equivalent modulation quantity sequence, and an equivalent modulation quantity of the three-phase bridge arm is obtained by solving the finite time domain optimization problem, including: The virtual synchronous generator generates and modifies the target three-phase voltage instruction as the reference voltage, and generates the cost function and the constraint condition; Obtain the candidate equivalent modulation quantity sequence of the three-phase bridge arm in the preset sampling period, and perform rolling prediction on the candidate equivalent modulation quantity sequence based on the current grid-connected state quantity measured in the preset sampling period and the pre-defined discrete state space model, to obtain the predicted value of the three-phase voltage at the grid-connected end, the predicted amplitude of the current, and the capacitor voltage of the sub-module; The predicted value of the three-phase voltage at the grid-connected end, the predicted amplitude of the current, and the capacitor voltage of the sub-module are input into the cost function and the constraint condition to form a finite time domain optimization problem based on the equivalent modulation quantity sequence; The optimal sequence obtained by solving the finite time domain optimization problem is used as the optimal equivalent modulation quantity of the three-phase bridge arm.

7. The network configuration type control method for a cascaded SVG according to claim 1, characterized by, The optimal equivalent modulation quantity is used to update the modulation quantity of the sub-module in the three-phase bridge arm, and a carrier phase-shifted pulse width modulation strategy is executed to realize capacitor voltage balancing and three-phase modulation signal driving static reactive power generator output, including: The optimal equivalent modulation quantity is converted into an equivalent modulation ratio of the three-phase bridge arm, and the optimal equivalent modulation quantity is distributed to each sub-module according to the number of sub-modules in the three-phase bridge arm to obtain a reference duty cycle bias of each sub-module; The reference duty cycle bias of each sub-module is superimposed with a duty cycle bias correction to realize capacitor voltage balancing; Detect the fault state of the sub-module in the three-phase bridge arm, and execute a redundancy bypass and redistribution strategy when a sub-module fault is detected.

8. The network configuration type control method for a cascaded SVG according to claim 7, characterized by, The detection of the fault state of the sub-module in the three-phase bridge arm, and the execution of the redundancy bypass and redistribution strategy when a sub-module fault is detected, include: When a sub-module fault is detected, the main switch driving signal of the fault sub-module is cut off, the bypass branch switch of the fault sub-module is turned on, the capacitor of the fault sub-module is isolated from the power circuit, and the bridge arm current bypasses the fault sub-module through the bypass branch to make the output voltage tend to a stable state; Determine whether there is a usable redundant sub-module in the current phase bridge arm, if there is a usable redundant sub-module in the current phase bridge arm, select a redundant sub-module with a capacitor voltage within the allowed range from the standby set, switch the redundant sub-module from the standby state to the working state, insert it into the original series position of the fault sub-module, and update the original mapping relationship; If the redundant sub-modules in the current phase bridge arm are exhausted, obtain the number of valid working sub-modules remaining in the current phase bridge arm, and redistribute the optimal equivalent modulation quantity of the current sampling period according to the number of valid working sub-modules.

9. The network configuration type control method for a cascaded SVG according to claim 3, characterized by, The deep reinforcement learning adaptive mechanism is constructed according to the current grid-connected state quantity, and the deep reinforcement learning adaptive mechanism is used to adaptively adjust the cost function weight and the virtual impedance parameter, including: In each reinforcement learning update period, the current grid-connected state quantity is collected according to the preset sampling period, and a state vector is formed according to the current grid-connected state quantity; The action vector is calculated according to the state vector, and the action vector falls into a predefined feasible region after projection, so that the cost function in the future sampling period and the virtual impedance parameter run according to the new parameters; During operation, the instantaneous reward is calculated according to the voltage tracking error, the current limiting overrun amount and the capacitor voltage imbalance degree, the state vector, the action vector, the instantaneous reward and the state vector of the next sampling period are used as experience samples to update the weights of the actor-critic network; After a reinforcement learning update cycle, the optimal action vector obtained through convergence is used to update the weight of the cost function and the parameter coefficient in the virtual impedance adaptive law.

10. A network-forming control system for a cascaded SVG, for implementing the network-forming control method for a cascaded SVG according to any one of claims 1 to 9, characterized by The system comprises: The upper virtual synchronous generator control module is used to calculate the reference frequency and the three-phase reference voltage through the droop control strategy when the virtual synchronous generator meets the condition of no phase-locked loop and the swing equation constraint; The parallel correction control module is used to construct a virtual impedance according to the pre-acquired grid-connected current amplitude, frequency change rate and bus voltage deviation, generate an additional voltage drop based on the virtual impedance and correct the three-phase reference voltage to obtain a corrected target three-phase voltage instruction; The middle-layer voltage capacitor equalization control module is used to construct a cost function and a constraint condition with the corrected target three-phase voltage instruction as the reference voltage, convert the cost function and the constraint condition into a finite time domain optimization problem based on an equivalent modulation quantity sequence, and solve the finite time domain optimization problem to obtain the optimal equivalent modulation quantity of the three-phase bridge arm; The bottom-layer modulation execution module is used to update the sub-module modulation quantity of the three-phase bridge arm based on the optimal equivalent modulation quantity, execute the carrier phase-shifted pulse width modulation strategy, and realize capacitor voltage equalization and three-phase modulation signal driving static reactive power generator output.

Images