Network configuration type MMC supporting capacity improvement method considering grid voltage drop

Abstract

The application relates to a network-structured MMC supporting capacity improvement method considering power grid voltage drop, and belongs to the field of power electronic converter optimization control. First, an operation condition is initialized and steady-state electrical quantities are calculated; when an overrun occurs, a double-complementary cooperative optimization strategy of double-frequency circulating current and zero sequence signal is introduced, optimal parameters satisfying physical limitations are solved, global optimization is performed through boundary condition traversal, and a mapping table of multi-dimensional operation conditions and optimal parameters is generated; subsequently, a neural network is used to fit the table, a prediction model is generated, and the prediction model is deployed in a controller. When voltage drop occurs in a power grid, the controller can directly output an optimized injection instruction within milliseconds according to a real-time operation condition. The method discards complex online iterative calculation, effectively widens the safe operation boundary of the MMC, and improves the supporting capacity and system stability of the network-structured MMC during power grid faults without increasing hardware cost.

Description

Technical Field

[0001] This invention relates to a method for improving the support capacity of grid-type MMC considering grid voltage dips, and more particularly to a method for multi-dimensional collaborative improvement of the support capacity of grid-type modular multilevel converters under grid voltage dip requirements, belonging to the field of power electronic converter optimization control technology. Background Technology

[0002] With the rapid development of "dual-high" power systems characterized by high proportions of renewable energy and power electronic equipment, problems such as declining system inertia and weakened voltage support capabilities are becoming increasingly prominent. Modular multilevel converters (MMCs), due to their advantages of high modularity, good harmonic characteristics, and low losses, have been widely used in flexible DC transmission and renewable energy grid integration. To improve grid stability and active support capabilities, grid-forming (GFM) control technology is gradually becoming an important development direction for MMCs.

[0003] However, in actual operation, when AC power grid experiences faults such as voltage dips, it is often accompanied by fluctuations in system frequency. At this time, the converter is typically required not only to remain connected to the grid but also to simultaneously provide active and reactive power support to the grid. This is especially true for grid-connected MMCs, whose active role in establishing grid voltage and frequency means that when grid voltage dips, the converter needs to output a large amount of reactive current for voltage support; conversely, when grid frequency decreases, the converter needs to output even more active current to provide inertia and frequency support. This simultaneous and significant increase in both active and reactive power support currents during fault periods leads to drastic changes in the internal electrical quantities of the converter, thus posing a severe challenge to physical safety constraints.

[0004] Currently, the traditional solutions to the problem of internal electrical quantities exceeding limits in MMCs during grid voltage drops fall into two main categories: one is to increase hardware design margins, such as increasing the capacitance of submodule capacitors and improving the withstand voltage rating of switching devices, but this will significantly increase the size and manufacturing cost of the converter; the other is to protect the converter by directly limiting the amplitude or switching to a grid-based control strategy, but this will sacrifice the grid-based MMC's ability to support grid voltage during faults, and cannot meet the high requirements of modern power grids for grid-based equipment.

[0005] Therefore, how to achieve coordinated enhancement of electrical quantities through optimized control strategies, without increasing hardware costs, when grid voltage drops and internal electrical quantities approach or exceed physical limits, so as to ensure safe operation within physical constraints and maximize the combined active and reactive power support capacity of grid-connected MMCs for the grid, is a key technical problem that urgently needs to be solved in the current intersection of power electronics and power systems. Therefore, developing a method for coordinated enhancement of the support capacity of grid-connected MMCs under grid voltage drops, without increasing hardware costs, and simultaneously improving the active and reactive power dual active support capability and equipment operational safety of grid-connected converters during grid faults, is of great significance for ensuring the safe and stable operation of new power systems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a method for enhancing the support capacity of grid-type MMCs that considers grid voltage dips. Without increasing hardware costs, this method employs a collaborative optimization control strategy that combines circulating current and zero-sequence signal complementarity with a neural network prediction model to improve the physical constraint margin of the MMC, thereby enabling the safe operation of the converter within its physical constraints.

[0007] The present invention adopts the following technical solution:

[0008] A method for enhancing the support capacity of grid-type MMC considering grid voltage dips includes the following steps:

[0009] S1, initialize the operating conditions and iteration step size of the grid-type modular multilevel converter (MMC). The operating conditions include grid voltage, active power and reactive power.

[0010] S2, based on the grid voltage, active power and reactive power of the current iteration, maps to the corresponding active current command and reactive current command;

[0011] S3, calculate the steady-state electrical quantities of the MMC without introducing an optimization strategy, and verify the steady-state electrical quantities against multiple physical constraints of the MMC; if none of them exceed the physical constraints, return to S1 and proceed to the next round; if any physical constraint is exceeded, proceed to step S4.

[0012] S4. Introduce a collaborative optimization control strategy and use a global optimization algorithm to find the optimal control strategy parameters that can simultaneously satisfy all physical constraints of the MMC. If the optimal control strategy parameters are successfully solved, record the current operating condition, active current command, reactive current command and the corresponding optimal control strategy parameters. If there is no solution, proceed directly to step S5.

[0013] S5, determine whether the iteration is complete, and update the active power, reactive power and grid voltage alternately according to the set boundary conditions. Repeat steps S1 to S5 until the iterative optimization of the global operating condition is completed, and generate a mapping table between the multi-dimensional operating conditions and the optimized control strategy parameters.

[0014] S6. Use a neural network to train and fit the mapping table offline to obtain an optimized control strategy parameter prediction model and deploy it in the controller of the MMC.

[0015] In actual MMC operation, when a voltage drop occurs in the power grid, the controller inputs the real-time detected power grid voltage, active current command, and reactive current command into the optimized control strategy parameter prediction model, and directly outputs the optimized control strategy parameters and injects them into the control system.

[0016] Preferably, in step S1, the initial grid voltage is set. Initial active power and initial reactive power ;

[0017] The iterative process of grid voltage, active power, and reactive power is shown in equations (1) to (3):

[0018] (1)

[0019] (2)

[0020] (3)

[0021] in, The active power iteration step size is... The reactive power iteration step size is... , These are the active power at the nth and (n+1)th cycles, respectively. , These are the reactive power values ​​for the nth and (n+1)th cycles, respectively. The initial grid voltage, , These are the grid voltages for the nth and (n+1)th cycles, respectively.

[0022] Preferably, in step S2, the active current command is mapped. With reactive current command As shown in equations (4) and (5):

[0023] (4)

[0024] (5).

[0025] Preferably, in step S3, the steady-state electrical quantities of the MMC without the introduction of an optimization strategy are calculated according to equations (6) and (7):

[0026] (6)

[0027] (7)

[0028] in, This refers to the capacitor voltage of the submodule. This is the DC component of the capacitor voltage; This refers to the capacitance value of the submodule. This refers to the current flowing through the submodule capacitor. The modulation function for the upper arm of phase A; This is the DC component of the modulated signal; , These are the amplitude and initial phase angle of the k-th harmonic component, respectively; The fundamental angular frequency of the system; It is a time variable; For harmonic order;

[0029] The verification is performed based on three physical constraints: capacitor voltage peak limit, capacitor voltage ripple limit, and upper and lower limits of the bridge arm modulation signal. The verification logic satisfies the following relationship:

[0030] (8)

[0031] Among them This is the rated voltage of the submodule capacitor; This is a binary function, and its logical meaning is that its value is 1 when the corresponding physical constraint inequality is satisfied, and 0 otherwise; This is the status flag for physical limit checks. It is 1 when all physical limit checks meet the requirements, otherwise... The value is 0.

[0032] Preferably, in step S4, a particle swarm optimization algorithm is used for global optimization, and the particle position vector is defined as X=[I c,2ω ,θ c2 The particle velocity V and position X are updated using equations (9) and (10):

[0033] (9)

[0034] (10)

[0035] in, Indicates the first The particle in the first Speed ​​at +1 iteration Indicates the first Speed ​​at the next iteration Indicates the first The particle in the first The position at +1 iteration, Indicates the first The particle in the first The position at the next iteration; This represents the current iteration number of the particle swarm optimization algorithm. Inertial weights; , As a learning factor, , All are random numbers between [0,1]; This represents the historical best position of particle i; The globally optimal position;

[0036] The solution model for the particle swarm optimization algorithm is as follows:

[0037] (11)

[0038] in, F represents the optimized control strategy parameters. T=1 This represents the amplitude I of the second harmonic circulating current at this time. c,2ω Phase angle θ c2 The zero-sequence signal amplitude A3 and phase angle α3 can satisfy T=1, and st means that under the following constraint conditions, the three physical constraints of MMC are satisfied at the same time.

[0039] Preferably, in step S5, the active power P is first compared. n With maximum active power P max and reactive power Q n With maximum reactive power Q max ,like or Then the iteration count n increases by 1, and the program will return to step S1; if and If both conditions are met, then compare U again. s,m The lower bound U of the grid voltage iteration s,min ,like Then the iteration count m increases by 1, n becomes 0, and the program returns to step S1; if The iteration then ends.

[0040] Preferably, in step S7, the optimized control strategy parameter prediction model can reflect the relationship between the circulating current and the zero-sequence signal injection quantity and the changes in grid voltage, active current command, and reactive current command. The function expression is:

[0041] (12)

[0042] in, This represents the neural network fitting function. This represents the required injection optimization control parameters output by the neural network fitting function under operating conditions, including the second harmonic circulating current amplitude I. c,2ω With phase angle θ c2 , as well as the zero-sequence signal amplitude A3 and phase angle α3.

[0043] This invention first initializes the operating conditions of the grid voltage, active power, and reactive power, and defines the iterative step size and boundary conditions for global optimization. Then, it calculates the steady-state electrical quantities of the MMC based on the current operating conditions and compares them with three physical limits: the peak value of the submodule capacitor voltage, the capacitor voltage ripple, and the upper and lower limits of the bridge arm modulation signal. If no limits are exceeded, the operating conditions are updated according to the set step size; if limits are exceeded, a collaborative optimization control strategy is introduced. In this strategy, a second-harmonic circulating current is injected to effectively suppress the peak value and ripple of the submodule capacitor voltage. However, the injection of circulating current introduces a second harmonic component into the bridge arm modulation signal, which can easily lead to the modulation signal exceeding the limit. Therefore, a zero-sequence signal is injected synchronously, utilizing the characteristics of the third harmonic to compensate for and resolve the overmodulation problem caused by the circulating current injection.

[0044] This dual complementary mechanism significantly enhances the physical constraint margin of the MMC, while simultaneously solving for and recording the optimal control parameters that satisfy the aforementioned physical constraints. Subsequently, active power, reactive power, and grid voltage are alternately updated according to set boundary conditions, and the above state assessment and parameter solving process is cyclically executed until iterative optimization of the global operating condition is completed, thereby generating a mapping table containing multi-dimensional operating conditions and optimal control parameters. Finally, a neural network is used to offline train and fit the mapping table to obtain the optimal control strategy parameter prediction model, which is then deployed in the MMC controller.

[0045] In actual operation, when a voltage drop in the power grid causes the MMC to face extreme support requirements, the controller can output optimized second-harmonic circulating current and zero-sequence signal commands within milliseconds and inject them into the control system based on the real-time detected grid voltage and power without performing complex online iterations. This can effectively suppress the fluctuation of the submodule capacitor voltage and reduce the peak value of the modulation signal, thereby improving the safety margin of the MMC operation, significantly improving the operational stability of the system, and enhancing the ability of the grid-type MMC to support both active and reactive power capacity of the power grid during faults.

[0046] For any details not covered in this invention, please refer to the prior art.

[0047] The beneficial effects of this invention are as follows:

[0048] This invention proposes a method for enhancing the support capacity of grid-connected MMCs considering grid voltage dips. This method, through optimized control strategy injection, improves the physical constraint margin of the MMC in multiple dimensions without increasing the hardware design margin of the converter, broadening the safe operating boundary of the MMC and effectively enhancing the active support capability of grid-connected MMCs during grid faults. Simultaneously, this method eliminates the shortcomings of traditional methods that require complex online iterative calculations during faults. Employing an architecture of "offline global optimization + neural network fitting," when a voltage dip actually occurs in the grid, the controller only needs millisecond-level forward calculations to directly output optimized injection commands, greatly improving the response speed and operational stability of the control system. Attached Figure Description

[0049] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0050] Figure 1 This is a schematic diagram of the overall process of the grid-type MMC support capacity enhancement method considering grid voltage drop of the present invention.

[0051] Figure 2 A diagram illustrating the increased capacity of MMC support under rated grid voltage;

[0052] Figure 3 A three-dimensional spatial surface plot of the second harmonic circulation amplitude obtained by fitting a neural network;

[0053] Figure 4 This is a cross-sectional view of the second harmonic circulating current amplitude at a mains voltage of 0.8 pu;

[0054] Figure 5 The three-dimensional spatial surface plot of the second harmonic circulation phase obtained by neural network fitting;

[0055] Figure 6 This is a cross-sectional view of the second harmonic circulating current phase at a grid voltage of 0.8 pu;

[0056] Figure 7 A three-dimensional spatial surface plot of the zero-sequence signal amplitude obtained by fitting a neural network;

[0057] Figure 8 This is a cross-sectional view of the zero-sequence signal amplitude at a mains voltage of 0.8 pu;

[0058] Figure 9 A three-dimensional spatial surface plot of the phase of the zero-sequence signal obtained by fitting a neural network;

[0059] Figure 10 This is a cross-sectional view of the zero-sequence signal phase at a mains voltage of 0.8 pu;

[0060] Figure 11 This is a control block diagram for a control strategy based on a neural network prediction model.

[0061] Figure 12 The diagram shows the effect of increased MMC support capacity under a 0.8 pu grid voltage drop condition.

[0062] Figure 13 A comparison of the capacitor voltage waveforms of the submodule before and after optimization under the condition of grid voltage drop;

[0063] Figure 14 This is a comparison of the bridge arm modulation signal waveforms before and after optimization under the condition of grid voltage drop. Detailed Implementation

[0064] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. However, this is not the only description; all aspects not described in detail herein are based on conventional techniques in the art.

[0065] Example 1

[0066] A method for increasing the support capacity of grid-type MMC considering grid voltage sag, such as... Figure 1 As shown, it includes the following steps:

[0067] S1, initialize the operating conditions and iteration step size of the grid-type modular multilevel converter (MMC). The operating conditions include grid voltage, active power and reactive power.

[0068] S2, based on the grid voltage, active power and reactive power of the current iteration, maps to the corresponding active current command and reactive current command;

[0069] S3, calculate the steady-state electrical quantities of the MMC without introducing an optimization strategy, and verify the steady-state electrical quantities against multiple physical constraints of the MMC; if none of them exceed the physical constraints, return to S1 and proceed to the next round; if any physical constraint is exceeded, proceed to step S4.

[0070] S4. Introduce a collaborative optimization control strategy and use a global optimization algorithm to find the optimal control strategy parameters that can simultaneously satisfy all physical constraints of the MMC. If the optimal control strategy parameters are successfully solved, record the current operating condition, active current command, reactive current command and the corresponding optimal control strategy parameters. If there is no solution, proceed directly to step S5.

[0071] S5, determine whether the iteration is complete, and update the active power, reactive power and grid voltage alternately according to the set boundary conditions. Repeat steps S1 to S5 until the iterative optimization of the global operating condition is completed, and generate a mapping table between the multi-dimensional operating conditions and the optimized control strategy parameters.

[0072] S6. Use a neural network to train and fit the mapping table offline to obtain an optimized control strategy parameter prediction model and deploy it in the controller of the MMC.

[0073] In actual MMC operation, when a voltage drop occurs in the power grid, the controller inputs the real-time detected power grid voltage, active current command, and reactive current command into the optimized control strategy parameter prediction model, and directly outputs the optimized control strategy parameters and injects them into the control system.

[0074] Example 2

[0075] A method for increasing the support capacity of grid-type MMC considering grid voltage drops, as described in Example 1, differs in that, in step S1, an initial grid voltage is set. Initial active power and initial reactive power Initial grid voltage The normal voltage, i.e., 1.0 pu, is the voltage without any voltage drop; the iteration step size of the grid voltage is 0.1 pu.

[0076] The iterative process of grid voltage, active power, and reactive power is shown in equations (1) to (3):

[0077] (1)

[0078] (2)

[0079] (3)

[0080] in, The active power iteration step size is... The reactive power iteration step size is... , These are the active power at the nth and (n+1)th cycles, respectively. , These are the reactive power at the nth and (n+1)th times, respectively. The initial grid voltage, , These are the grid voltages for the nth and (n+1)th cycles, respectively.

[0081] Example 3

[0082] A method for improving the support capacity of grid-type MMC considering grid voltage dips, as described in Example 2, differs in that, in step S2, before calculating the key steady-state electrical quantities of the MMC, it is necessary to consider the grid voltage of the current iteration. Given active power and reactive power, calculate the corresponding active current i. d With reactive current i q Based on the instantaneous power theory in a synchronous rotating coordinate system, the current is calculated from the scanned power and mapped to an active current command. With reactive current command As shown in equations (4) and (5):

[0083] (4)

[0084] (5).

[0085] The grid voltage can be mapped and the scanned data and The only mapping is the active current command under the current operating condition. With reactive current command This provides accurate current reference commands for the control circuit.

[0086] Example 4

[0087] A method for improving the support capacity of grid-type MMC considering grid voltage dips, as described in Example 3, differs in that, in step S3, the steady-state electrical quantities of the MMC without the introduction of optimization strategies are calculated according to equations (6) and (7):

[0088] (6)

[0089] (7)

[0090] in, This refers to the capacitor voltage of the submodule. This is the DC component of the capacitor voltage; This refers to the capacitance value of the submodule. This refers to the current flowing through the submodule capacitor. The modulation function for the upper arm of phase A; This is the DC component of the modulated signal; , These are the amplitude and initial phase angle of the k-th harmonic component, respectively; The fundamental angular frequency of the system; It is a time variable; For harmonic order;

[0091] The verification is performed based on three physical constraints: capacitor voltage peak limit, capacitor voltage ripple limit, and upper and lower limits of the bridge arm modulation signal. The verification logic satisfies the following relationship:

[0092] (8)

[0093] Among them This is the rated voltage of the submodule capacitor; This is a binary function, and its logical meaning is that its value is 1 when the corresponding physical constraint inequality is satisfied, and 0 otherwise; This is the status flag for physical limit checks. It is 1 when all physical limit checks meet the requirements, otherwise... The value is 0.

[0094] Example 5

[0095] A method for enhancing the supporting capacity of a grid-type MMC considering grid voltage dips is described in Example 4. The difference is that in step S4, a joint injection strategy of second-harmonic circulating current and zero-sequence signal is introduced for the current operating conditions. To obtain the optimal parameter combination that enables the MMC to simultaneously satisfy the triple physical boundary constraints, a particle swarm optimization algorithm is used for global optimization, defining the particle position vector as X=[I c,2ω ,θ c2 The particle velocity V and position X are updated using equations (9) and (10):

[0096] (9)

[0097] (10)

[0098] in, Indicates the first The particle in the first Speed ​​at +1 iteration Indicates the first The particle in the first Speed ​​at the next iteration Indicates the first The particle in the first The position at +1 iteration, Indicates the first The particle in the first The position at the next iteration; This represents the current iteration number of the particle swarm optimization algorithm. Inertial weights; , As a learning factor, , All are random numbers between [0,1]; This represents the historical best position of particle i; The globally optimal position;

[0099] Through continuous iteration, the injection parameters that satisfy the above physical constraints are searched within the solution space. If an optimal control solution satisfying the constraints of equation (11) is successfully obtained... If the data is recorded, proceed to step S5; record the grid voltage under this operating condition. Active power With reactive power Simultaneously, the circulating current amplitude I of the circulating current injection optimization control strategy is recorded. c,2ω ( , , ) and phase angle θ c2 ( , , The zero-sequence signal amplitude A3 of the zero-sequence signal injection optimization control strategy. , , ) and phase angle α3 ( , , If there is no solution under this condition, proceed to step S6.

[0100] The solution model for the particle swarm optimization algorithm is as follows:

[0101] (11)

[0102] in, F represents the optimized control strategy parameters. T=1 This represents the amplitude I of the second harmonic circulating current at this time. c,2ω Phase angle θ c2 The zero-sequence signal amplitude A3 and phase angle α3 can satisfy T=1, and st means that under the following constraint conditions, the three physical constraints of MMC are satisfied at the same time.

[0103] Example 6

[0104] A method for increasing the support capacity of grid-type MMC considering grid voltage dips, as described in Example 5, differs in that, in step S5, the active power P is first compared. n With maximum active power P max and reactive power Q n With maximum reactive power Q max ,like or Then the iteration count n increases by 1, and the program will return to step S1; if and If both conditions are met, then compare U again.s,m The lower bound U of the grid voltage iteration s,min ,like Then the iteration count m increases by 1, n becomes 0, and the program returns to step S1; if The iteration then ends.

[0105] After the above cycle, the system generates a table containing thousands of sets of data on "operating conditions - current commands - optimized control parameters". For each operating condition and current command, a corresponding optimized control parameter can be found. Based on the correspondence of these data, a neural network can be used for fitting and training to obtain a prediction model of the optimized control strategy parameters.

[0106] Example 7

[0107] A method for enhancing the support capacity of grid-type MMC considering grid voltage dips, as described in Example 6, differs in that, in step S7, the optimized control strategy parameter prediction model can quickly and accurately reflect the relationship between circulating current and zero-sequence signal injection with changes in grid voltage, active current command, and reactive current command. The function expression is:

[0108] (12)

[0109] in, This represents the neural network fitting function. This represents the required injection optimization control parameters output by the neural network fitting function under operating conditions, including the second harmonic circulating current amplitude I. c,2ω With phase angle θ c2 , as well as the zero-sequence signal amplitude A3 and phase angle α3.

[0110] To verify the proposed method for multi-dimensional collaborative enhancement of grid-type MMC support capacity under grid voltage drop, the following verification is conducted in conjunction with an embodiment. The main circuit parameters of the MMC in this embodiment are shown in Table 1.

[0111] Table 1. MMC Main Circuit Parameters

[0112]

[0113] Set initial grid voltage For 500kV, this is equivalent to 1.0 pu, the initial active power. The initial reactive power is -2500 MW. The value is -2000 Mvar, the grid voltage iteration step is 10kV, which is 0.02 pu, and the active power iteration step is... The reactive power iteration step size is 50 MW. The maximum active power is 50 MW. The boundary condition for the iteration, i.e., the stopping condition, is the maximum active power.P max It has a capacity of 2500 MW and a maximum reactive power. Q max For 2500 MW, the lowest drop voltage U s,min 350kV is equivalent to 0.7pu.

[0114] First, under the initial operating conditions, i.e., the grid voltage... 500kV, active power The reactive power is -2500 MW. The calculated active current command is -2000 Mvar. -3333A, reactive current command The value is 2666A. Calculation results show that, without optimization, the submodule capacitor voltage peak value is 2738.48V, capacitor voltage ripple is 448.41V, modulation signal valley value is 0.05, and peak value is 0.706. Comparing these values ​​with the system's peak value limit of 2522.9V, ripple value limit of 229.35V, and modulation signal limit of the [0,1] range, it is found that both the capacitor voltage peak value and ripple have significantly exceeded the limits.

[0115] To address the limit violation issue, the control system needs to incorporate a second-harmonic circulating current injection and a zero-sequence signal injection. An attempt is made to calculate the circulating current amplitude I that simultaneously satisfies the three physical limits. c,2ω With phase angle θ c2 And the zero-sequence signal amplitude A3 and phase angle α3. However, the calculation results show that there is no solution under this operating condition, that is, there are no optimal parameters that can restore the system to within the safety limits. Therefore, the system enters the iterative optimization process. First, the active power is compared. With maximum active power P max and active power With maximum reactive power Q max At this point, both active and reactive power are less than the maximum active and maximum reactive power, respectively. Therefore, the system performs a power step size iteration of n+1 and repeats the above steps until the power parameters meet the set boundary conditions. and The capacity improvement diagram under the rated grid voltage is obtained at this time as shown in the figure. Figure 2 As shown. Figure 2In the diagram, the horizontal axis P represents the active power output of the converter, and the vertical axis Q represents the reactive power output of the converter. In step S3, the system calculates the unoptimized steady-state electrical quantities based on the current (P,Q) operating condition. If the calculation results show that none of the three physical limits are exceeded, then the current (P,Q) coordinate point is classified into the "naturally feasible region." The naturally feasible region indicates that within this power range (P,Q) combination, the electrical quantities within the MMC can naturally meet all physical safety limits without the addition of any additional optimization control strategies.

[0116] After detecting an electrical quantity exceeding the limit, the program proceeds to step S4. In step S4, a particle swarm optimization algorithm is used to globally search for parameters of the second harmonic circulating current and the zero-sequence signal. If the algorithm successfully finds a set of parameter solutions such that the three physical limits of the MMC can be simultaneously satisfied after adding these injected quantities, then the current (P,Q) coordinate point is designated as the "injection feasible region." The injection feasible region indicates that within this power range, without intervention, the internal electrical quantities of the MMC would exceed the limit. However, by injecting the optimized control strategy proposed in this invention, voltage fluctuations and overmodulation can be successfully suppressed, allowing the MMC to return to safe operation within the physical limits.

[0117] During the particle swarm optimization process in step S4, if, after continuous iterative searching, no set of injection parameters can be found that simultaneously satisfies the triple physical boundary constraints of the MMC (i.e., "no solution"), then the current (P,Q) coordinate point is classified into the "infeasible region." The infeasible region indicates that the operating conditions are too extreme within this power range. Even with the optimal injection strategy, it is impossible to suppress the internal electrical quantities of the MMC within safe physical limits. Operating in this region will lead to equipment damage.

[0118] At this time, the grid voltage Even at 500kV, it is still greater than the set minimum drop voltage U. s,min Therefore, the next step is to execute the voltage step size iteration m+1 and reset the power iteration step number n=0. At this time, the operating condition changes to grid voltage. Transformed to 490kV, active power The reactive power is -2500MW. The calculated active current command is -2000Mvar. -3401A, reactive current command The value is 2721A. Recalculation revealed that the electrical quantities, even without optimization, still exceeded the limits, and attempts to inject optimization signals yielded no solution. The system will continue to alternately execute the above inner and outer loop iterative process until the grid voltage, active power, and reactive power all meet the final global convergence condition, i.e. , and After the iteration is complete, the calculation stops, and all recorded data from the entire optimization process is stored.

[0119] After the above cycle, the system generates a mapping table containing thousands of data points: "Operating Condition - Current Command - Optimization Parameters". , , As the input layer, I c,2ω ,θ c2 A3 and α3 are used as output layers, and a neural network is used for offline training and fitting. After training, the neural network prediction model is deployed in the actual controller of the MMC. When a voltage drop actually occurs in the power grid, the controller does not need to perform complex online iterative calculations; it only needs to process the real-time detected voltage. , , Inputting the required circulating current and zero-sequence signal amplitude and phase angle within milliseconds enables the MMC to operate safely without exceeding limits and support its capacity during power grid faults.

[0120] Figures 2 to 10 The second harmonic circulation amplitude I obtained by fitting the model is shown below. c,2ω Second harmonic circulation phase θ c2 The diagram shows the three-dimensional spatial surfaces of the zero-sequence signal amplitude A3 and phase α3, along with a cross-sectional view of the 0.8 pu grid voltage. The blank areas within the surfaces in each diagram represent operating conditions where the system naturally satisfies all physical constraints under normal conditions. Within these regions, the MMC can maintain safe and stable operation without the need for any additional optimization signals. Conversely, the blank areas outside the surfaces represent unsolvable operating conditions, meaning that under these extreme conditions, no optimized control strategy parameters can simultaneously satisfy the three physical constraints of the MMC. Overall, this optimized control strategy parameter prediction model effectively replaces the traditional complex mathematical solution process, ensuring that the real-time control system can quickly and accurately obtain optimized control parameters.

[0121] To further verify the effectiveness of the proposed method, the prediction module obtained by the neural network fitting was embedded into the control loop, and the control strategy is as follows: Figure 11 As shown, a detailed analysis is conducted in conjunction with the MMC main circuit parameter examples shown in Table 2 below.

[0122] To prevent the grid-type MMC from exceeding internal electrical limits due to providing large-capacity active support during AC grid faults such as short circuits, the system employs a dual-track parallel control strategy based on a neural network prediction model. The core of this strategy lies in the deep integration of conventional basic power control with intelligent optimization injection.

[0123] In the control phase, the system rapidly detects the magnitude U of the grid voltage drop. s,m The active and reactive power controllers output the phase θ and amplitude e of the internal electromotive force, and calculate the required current command value through a virtual impedance circuit. and .

[0124] Meanwhile, to overcome the time-consuming nature of traditional online optimization algorithms, the controller will acquire the state [U] in real time. s,m i d i q The data is directly fed into a pre-trained offline neural network prediction model. This model can then directly output the optimal control parameters for the current extreme operating condition—the amplitude of the second-harmonic circulating current—through forward computation within milliseconds. With phase The four sets of commands are then divided into two paths for dual coordinated injection: on the one hand, the circulating current controller tracks the circulating current command to generate a circulating current modulation signal amplitude A2 and phase α2, driving the converter to generate a specific internal circulating current, thereby significantly absorbing and suppressing the peak value and ripple of the submodule capacitor voltage; on the other hand, A3 and α3 are constructed as the third harmonic zero-sequence voltage to compensate for the risk of second harmonic overmodulation caused by the injected circulating current.

[0125] The inner current loop tracks the required current command value. and The fundamental modulation signal amplitude A1 and phase α1 are generated to maintain the basic power transmission. In the synthesis stage, the fundamental modulation signal, the circulating current modulation signal, and the zero-sequence voltage injection signal are mathematically superimposed to generate the final upper arm modulation signal S. ap and lower bridge arm modulation signal S an Through this series of rapid and accurate parameter predictions and collaborative injections, the system utilizes the "peak clipping" characteristics of zero-sequence voltage to strictly limit the modulation wave within a safe physical range. Thus, without affecting the basic power support of the power grid or increasing the hardware design margin, it achieves real-time suppression and safety optimization of the risk of exceeding the limits of electrical quantities within the MMC.

[0126] Table 2 MMC Main Circuit Parameters II

[0127]

[0128] Under the typical operating condition where the voltage drops to 0.8 pu due to faults such as short circuits in the power grid, the capacity improvement effect of the MMC support achieved by the method of this invention is as follows: Figure 12As shown. To further verify the effectiveness and safety of the proposed control strategy, this embodiment compares and analyzes the bridge arm modulation signal of the converter under the proposed optimized strategy and the traditional unoptimized strategy, respectively, for the voltage drop condition. Waveform and submodule capacitor voltage Waveform.

[0129] like Figure 13 As shown in the waveform of the capacitor voltage in the submodule, under the condition that the mains voltage drops to 0.8 pu, the system sets the peak safe operating limit for the capacitor voltage to be 2522.9V and the ripple safe limit to be 229.35V. Without the optimization strategy, the capacitor voltage fluctuates wildly, with the highest peak reaching 2516V, extremely close to the physical limit of 2522.9V, leaving only a safety margin of 6.9V, and the lowest valley value dropping to 2025V; at this time, the capacitor voltage ripple is as high as 245.5V, exceeding the safe operating limit of 229.35V by about 7.04%, posing a great risk of exceeding the limit. After adding the optimization strategy, the highest peak value of the capacitor voltage drops significantly to 2481V, the safety margin increases by 35V, the remaining margin reaches 41.9V, effectively moving away from the peak limit; at the same time, the lowest valley value increases to 2035V, and the capacitor voltage ripple decreases accordingly to 221.5V, successfully falling back within the physical limit. The optimized voltage trajectory not only strictly meets the peak limit, but also significantly suppresses voltage ripple, greatly improving the voltage safety margin of the system.

[0130] Meanwhile, to ensure the normal operation of the converter and avoid overmodulation, the bridge arm modulation signal... It must be strictly limited to the physical range [0,1]. For example... Figure 14 As shown in the waveform of the modulated signal, without the optimization strategy, the peak value of the modulated signal reaches 0.88 and the valley value is 0.082; while after adding the optimization strategy, the peak value of the modulated signal is further suppressed to 0.86 and the valley value is 0.064. This indicates that the optimization strategy effectively suppresses the voltage ripple of the submodule capacitor while also optimizing the modulated signal; it not only successfully reduces the signal peak value and widens the linear modulation upper limit margin of the system, but also ensures that the modulated signal always strictly meets the physical constraints, thereby reducing the risk of overmodulation.

[0131] In summary, under severe fault conditions where the grid voltage drops to 0.8 pu, the neural network-based prediction module can respond quickly and output optimized control parameters. This strategy successfully reduces the submodule capacitor voltage ripple while ensuring that the peak voltage of the submodule capacitor does not exceed the limits of the modulation signal. This ensures that all key electrical quantities of the MMC are strictly kept within safe physical limits, fully verifying the accuracy and effectiveness of the method proposed in this invention.

[0132] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for enhancing the support capacity of grid-type MMC considering grid voltage dips, characterized in that, Includes the following steps: S1, Initialize the operating conditions and iteration step size of the grid-type modular multilevel converter. The operating conditions include grid voltage, active power and reactive power. S2, based on the grid voltage, active power and reactive power of the current iteration, maps to the corresponding active current command and reactive current command; S3, calculate the steady-state electrical quantities of the MMC without introducing an optimization strategy, and verify the steady-state electrical quantities against multiple physical constraints of the MMC; if none of them exceed the physical constraints, return to S1 and proceed to the next round; if any physical constraint is exceeded, proceed to step S4. S4. Introduce a collaborative optimization control strategy and use a global optimization algorithm to find the optimal control strategy parameters that can simultaneously satisfy all physical constraints of the MMC. If the optimal control strategy parameters are successfully solved, record the current operating condition, active current command, reactive current command and the corresponding optimal control strategy parameters. If there is no solution, proceed directly to step S5. S5, determine whether the iteration is complete, and update the active power, reactive power and grid voltage alternately according to the set boundary conditions. Repeat steps S1 to S5 until the iterative optimization of the global operating condition is completed, and generate a mapping table between the multi-dimensional operating conditions and the optimized control strategy parameters. S6. Use a neural network to train and fit the mapping table offline to obtain an optimized control strategy parameter prediction model and deploy it in the controller of the MMC. In actual MMC operation, when a voltage drop occurs in the power grid, the controller inputs the real-time detected power grid voltage, active current command, and reactive current command into the optimized control strategy parameter prediction model, and directly outputs the optimized control strategy parameters and injects them into the control system.

2. The method for enhancing the support capacity of grid-type MMC considering grid voltage dips according to claim 1, characterized in that, In step S1, the initial grid voltage is set. Initial active power and initial reactive power ; The iterative process of grid voltage, active power, and reactive power is shown in equations (1) to (3): （1） （2） （3） in, The active power iteration step size is... The reactive power iteration step size is... , These are the active power at the nth and (n+1)th cycles, respectively. , These are the reactive power values ​​for the nth and (n+1)th cycles, respectively. The initial grid voltage, , These are the grid voltages for the nth and (n+1)th cycles, respectively.

3. The method for enhancing the support capacity of grid-type MMC considering grid voltage dips according to claim 2, characterized in that, In step S2, the active current command is mapped. With reactive current command As shown in equations (4) and (5): （4） （5）。 4. The method for increasing the support capacity of grid-type MMC considering grid voltage dips according to claim 3, characterized in that, In step S3, the steady-state electrical quantities of the MMC without the introduction of an optimization strategy are calculated according to equations (6) and (7): （6） （7） in, This refers to the capacitor voltage of the submodule. This is the DC component of the capacitor voltage; This refers to the capacitance value of the submodule. This refers to the current flowing through the submodule capacitor. The modulation function for the upper arm of phase A; This is the DC component of the modulated signal; , These are the amplitude and initial phase angle of the k-th harmonic component, respectively; The fundamental angular frequency of the system; It is a time variable; For harmonic order; The verification is performed based on three physical constraints: capacitor voltage peak limit, capacitor voltage ripple limit, and upper and lower limits of the bridge arm modulation signal. The verification logic satisfies the following relationship: （8） Among them This is the rated voltage of the submodule capacitor; This is a binary function, and its logical meaning is that its value is 1 when the corresponding physical constraint inequality is satisfied, and 0 otherwise; This is the status flag for physical limit checks. It is 1 when all physical limit checks meet the requirements, otherwise... The value is 0.

5. The method for enhancing the support capacity of grid-type MMC considering grid voltage dips according to claim 4, characterized in that, In step S4, a particle swarm optimization algorithm is used for global optimization, and the particle position vector is defined as X=[I c,2ω ,θ c2 The particle velocity V and position X are updated using equations (9) and (10): （9） （10） in, Indicates the first The particle in the first Speed ​​at +1 iteration Indicates the first The particle in the first Speed ​​at the next iteration Indicates the first The particle in the first The position at +1 iteration, Indicates the first The particle in the first The position at the next iteration; This represents the current iteration number of the particle swarm optimization algorithm. Inertial weights; , As a learning factor, , All are random numbers between [0, 1]; This represents the historical best position of particle i; The globally optimal position; The solution model for the particle swarm optimization algorithm is as follows: （11） in, F represents the optimized control strategy parameters. T=1 This represents the amplitude I of the second harmonic circulating current at this time. c,2ω Phase angle θ c2 The zero-sequence signal amplitude A3 and phase angle α3 can satisfy T=1, and st means that under the following constraint conditions, the three physical constraints of MMC are satisfied at the same time.

6. The method for enhancing the support capacity of grid-type MMC considering grid voltage dips according to claim 5, characterized in that, In step S5, the active power P is first compared. n With maximum active power P max and reactive power Q n With maximum reactive power Q max， like or Then the iteration count n increases by 1, and the program will return to step S1; if and If both conditions are met, then compare U again. s,m The lower bound U of the grid voltage iteration s,min ,like Then the iteration count m increases by 1, n becomes 0, and the program returns to step S1; if Then the iteration ends.

7. The method for enhancing the support capacity of grid-type MMC considering grid voltage dips according to claim 6, characterized in that, In step S7, the optimized control strategy parameter prediction model can reflect the relationship between the circulating current and zero-sequence signal injection quantity and the changes in grid voltage, active current command, and reactive current command. The function expression is: （12） in, This represents the neural network fitting function. This represents the required injection optimization control parameters output by the neural network fitting function under operating conditions, including the second harmonic circulating current amplitude I. c,2ω With phase angle θ c2 , as well as the zero-sequence signal amplitude A3 and phase angle α3.

Images