Convergence determination method and system in large model driven MMC parameter optimization process

Abstract

The application discloses a convergence judgment method and system in a large model driven MMC parameter optimization process, and relates to the technical field of power electronic device main circuit parameter design, including the following steps: in the standardized parameter space, for each new solution in the solution set dominated by the newly generated main circuit parameters of this round, nearest neighbor search is performed in the main circuit parameter solution set of the historical round, and the nearest neighbor main circuit parameter solution corresponding to each new solution is obtained; the original parameter relative change rate of each new solution relative to the nearest neighbor main circuit parameter solution thereof is obtained; based on the comparison result of the original parameter relative change rate of all new solutions and the preset stable threshold, whether the current iteration round reaches a stable state is judged; the application can adaptively identify the best termination time under the condition that the real frontier is unknown, greatly reduces the calculation power consumption caused by invalid iteration, and effectively improves the convergence accuracy and optimization iteration stability of the optimization result.

Description

Technical Field

[0001] This invention relates to the field of main circuit parameter design technology for power electronic equipment, and in particular to a convergence determination method and system in the process of large model-driven MMC parameter optimization. Background Technology

[0002] Modular multilevel converters (MMCs) have become the mainstream topology for high-voltage, high-capacity flexible DC transmission. In the engineering design of MMCs, the selection of main circuit parameters directly determines the system's size, cost, losses, and dynamic performance. However, MMC parameter design is a typical multi-objective, strongly coupled, and nonlinear optimization problem, with significant constraints between the parameters. Overly conservative main circuit parameter configurations will lead to redundant equipment size and a surge in investment costs; insufficient design margins in the main circuit parameters may induce device overvoltage and overcurrent, or even cause system instability.

[0003] With the development of artificial intelligence technology, using Large Language Models (LLMs) as optimization iterators to explore the parameter space of Multiphysics Common Coding (MMC) in a generative manner has become an emerging and efficient design approach. LLMs can comprehensively consider multiphysics constraints and generate potential optimal parameter combinations.

[0004] Despite LLM's powerful parameter generation capabilities, in actual iterative processes for MMC parameter optimization, traditional iterative convergence determination methods often lead to wasted computational resources, difficulties in normalization, or ineffective identification of convergence states due to issues such as the often unknown true Pareto front, the high computational and costly nature of solution verification, and significant differences in the dimensions of main circuit parameters. This is particularly true in applications designing MMC main circuit parameters, which involve trade-offs between highly nonlinear and tightly coupled objectives such as efficiency, volume, and thermal stress. In this strongly coupled, nonlinear, and front-unknown environment, existing stopping strategies based on fixed iteration counts or single geometric features are often slow to react, easily leading to a large number of invalid "blind searches" or premature convergence, resulting in severe waste of computational resources. This makes it difficult to meet the high-efficiency optimization requirements of LLM as an iterator, severely restricting the convergence accuracy and computational efficiency of MMC parameter optimization design. Summary of the Invention

[0005] Based on the shortcomings of the existing technology, the present invention provides a convergence determination method and system for large model-driven MMC parameter optimization, which solves the existing problems.

[0006] The present invention adopts the following technical solution: In a first aspect, the present invention provides a convergence determination method in the large model-driven MMC parameter optimization process, comprising the following steps: Obtain all solution set data generated in the current iteration round during the large model-driven MMC main circuit parameter optimization process, and separate the newly generated main circuit parameter dominant solution set and the main circuit parameter dominant solution set of the previous round from the original solution set; Based on the statistics of the main circuit parameter dominance solution set of the historical round, the main circuit parameter space is normalized and mapped to obtain the standardized parameter space. In the standardized parameter space, for each new solution in the main circuit parameter dominance solution set newly generated in the current round, a nearest neighbor search is performed in the main circuit parameter dominance solution set of the historical round to obtain the nearest neighbor main circuit parameter solution corresponding to each new solution. Obtain the relative rate of change of the original parameters of each new solution relative to its nearest neighbor main circuit parameter solution; based on the comparison results of the relative rate of change of the original parameters of all new solutions with the preset stability threshold, determine whether the current iteration has reached a stable state; The continuous stability counter is updated based on the determination of the stable state. When the continuous stability counter reaches the preset target stable round threshold, the determination algorithm converges and terminates the iteration, and outputs the final Pareto optimal solution set.

[0007] Preferably, the solution set data includes iterative information, main circuit parameters, and calculated values ​​of constrained electrical quantities.

[0008] Preferably, the normalization mapping of the main circuit parameter space based on the statistics of the dominant solution set of the historical main circuit parameters specifically includes the following steps: Calculate the first number in the dominance solution set of the main circuit parameters of the historical cycle. j The statistics of the dimensional parameters include the mean and standard deviation; Based on the aforementioned statistics, any solution of the dominant solution set of the main circuit parameters in the historical rounds and the newly generated dominant solution set of the main circuit parameters in the current round are standardized.

[0009] Preferably, the nearest neighbor search in the historical round main circuit parameter dominance solution set specifically includes the following steps: The Euclidean distance between the new solution and any solution in the solution set dominated by the main circuit parameters of the historical cycle is calculated based on the standardized parameters. The historical solution with the smallest Euclidean distance is selected as the nearest neighbor main circuit parameter solution for the new solution.

[0010] Preferably, the formula for calculating the relative rate of change of the original parameters is: ; In the formula, The relative rate of change of the original parameters. For a new interpretation The j Parameter values, The nearest neighbor main circuit parameter solution for the new solution The j Each parameter value.

[0011] Preferably, determining whether the current iteration has reached a stable state specifically includes the following steps: If no new dominant solution for the main circuit parameters is generated in this round, the current round is directly determined to be in a stable state. If a new dominant solution for the main circuit parameters is generated in this round, it is determined whether the relative rate of change of the original parameters of all new solutions is lower than the preset stability threshold. If so, the current round is determined to be in a stable state; otherwise, it is determined to be in an unstable state.

[0012] Preferably, updating the continuous stable counter based on the stable state determination result specifically includes: If the current round is determined to be in an unstable state, then reset the continuous stability counter; If the current round is determined to be in a stable state, then increment the continuous stability counter by 1; When the continuous stable counter K When the number of stable rounds is greater than or equal to the preset target threshold, the algorithm is considered to have fully converged.

[0013] Secondly, this invention provides a convergence determination system for the large model-driven MMC parameter optimization process, including: The acquisition module is used to acquire all solution set data generated in the current iteration during the optimization of main circuit parameters of MMC driven by a large model, and to separate the newly generated main circuit parameter dominant solution set and the main circuit parameter dominant solution set of the previous iteration from the original solution set; The normalization module is used to normalize the main circuit parameter space based on the statistics of the main circuit parameter dominance solution set of the historical rounds, so as to obtain the standardized parameter space. In the standardized parameter space, for each new solution in the main circuit parameter dominance solution set newly generated in the current round, the nearest neighbor search is performed in the main circuit parameter dominance solution set of the historical rounds to obtain the nearest neighbor main circuit parameter solution corresponding to each new solution. The calculation module is used to obtain the relative rate of change of the original parameters of each new solution relative to its nearest neighbor main circuit parameter solution; based on the comparison results of the relative rate of change of the original parameters of all new solutions with the preset stability threshold, it determines whether the current iteration has reached a stable state. The iteration module is used to update the continuous stability counter based on the determination result of the stable state. When the continuous stability counter reaches the preset target stable round threshold, the determination algorithm converges and terminates the iteration, and outputs the final Pareto optimal solution set.

[0014] Compared with the prior art, the above-mentioned at least one technical solution adopted by the present invention can achieve the following beneficial effects: This invention proposes a method for determining iterator convergence in large-model-driven MMC parameter optimization. This method, even when the true Pareto front of the MMC main circuit parameters is unknown, achieves adaptive termination of the optimization process through a nearest-neighbor search-based front displacement quantization mechanism, significantly reducing invalid iterations and lowering computational resource consumption. Addressing the issue of multidimensional dimensional differences in MMC main circuit parameter design, this method employs Z-Score normalization to ensure the accuracy of convergence calculations. In implementation, this method introduces a continuous stability counting logic to effectively smooth out the inherent random fluctuations of LLM, filtering out interference while maintaining high-precision convergence characteristics and output stability of the final Pareto front. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the convergence determination method in the large model-driven MMC parameter optimization process of the present invention. Figure 2 This is a schematic diagram illustrating the state partitioning of the multi-objective optimization solution set according to the present invention; Figure 3 This is a schematic diagram of the nearest neighbor search in the normalized parameter space of the present invention; Figure 4 This is a graph showing the parameter optimization iteration history and the final Pareto front distribution of the present invention; Figure 5 This is a steady-state waveform diagram of the modulation signal of the upper arm of phase A of the MMC under the rated operating conditions of the present invention; Figure 6 This is a steady-state waveform diagram of the submodule capacitor voltage of the MMC under rated operating conditions according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention incorporates a maximum iteration limit. T maxThe stability determination mechanism based on the Pareto front of the MMC main circuit parameters aims to automatically identify the convergence state in the multi-objective optimization process. Its core logic lies in evaluating convergence by quantifying the displacement characteristics of the Pareto solution set between iterations.

[0019] Read and separate the newly generated master circuit parameter dominance solution set in this round. P new The solution set dominated by the main circuit parameters of the historical cycle P prev First, the Z-Score is used to standardize and unify the dimensions of the multidimensional target space. Then, based on the nearest neighbor search algorithm, the newly generated main circuit parameters dominate the solution. P new Searching for the solution set of the main circuit parameters of the historical cycle P prev Find the nearest reference point in the Euclidean distance and calculate the relative error between them. .

[0020] In the judgment process, if no new dominant solution for the main circuit parameters is generated in this round, or if the maximum error of all new solutions relative to the historical dominant reference point is lower than the preset stability threshold. stable The system then determines that the current Pareto front is in a stable state. To further ensure the stability of the Pareto front, this invention introduces a continuous stability counting mechanism, requiring that the stable state must be continuously maintained to reach a target count value. K target This allows the iteration to terminate, effectively filtering out perturbations caused by the randomness of the LLM model and ensuring that the final Pareto front has high-precision convergence characteristics. (Refer to...) Figure 1 This includes the following steps: Step 1: First, read the current global main circuit parameter status data, specifically including: from the first generation to the current generation. t The dominant solution of all main circuit parameters generated by the generation, and the current iteration round number t and current stability count K Stability count K This is to prevent the algorithm from stopping prematurely due to accidental local minima. It also serves as the stopping condition for iteration, i.e., the current stability count K > the target stable iteration threshold. The iteration stops when the time is right.

[0021] Based on the constraints and the relative merits of multiple objectives, these solutions will be divided into three categories: infeasible solutions that violate the constraints, dominated solutions that are covered by better solutions in the objective space, and dominated solutions that are currently at the optimal frontier. Figure 2 As shown, the set of solutions governed by all master circuit parameters is defined as the original solution set. P allEach dominant solution in this set contains iteration information (the number of rounds in which the solution was generated), main circuit parameters, and calculated values ​​of constrained electrical quantities.

[0022] Step 2: Determine the current iteration round t Has the maximum number of iterations been exceeded? T max If equation (1) is not satisfied, the iteration is forcibly terminated and the status is returned as "exceeded the maximum number of iterations". (1); If equation (1) is satisfied, then it is determined whether the current iteration is the first iteration, i.e., whether equation (2) is satisfied. Since there is no historical reference, the displacement cannot be calculated in the first iteration, so it is directly determined that the iteration has not converged and the next iteration continues.

[0023] (2); Step 3: Analyze from round 1 to the current round t The original solution set obtained from round iteration P all Original solution set P all Including up to the t The global dominating solution of the round is then divided into the newly generated master circuit parameter dominating solution set after this round of iteration. P new and the historical cycle main circuit parameter dominance solution set P prev As shown in equations (3) and (4), where equation (4) is the difference operation of sets, that is, from the current original solution set... P all Remove newly generated solutions from this round: (3); (4); In the formula, x This represents a solution vector in the solution space, which is a vector containing multiple parameter values; P t For the first t The dominance set of the wheel, iter ( x ) indicates the current iteration round number of the solution.

[0024] Check whether a valid new solution is generated in this round, i.e. whether it satisfies equation (5). If it is satisfied, proceed to the subsequent quantization calculation process; if it is not satisfied, it is determined that a "stable state" has been reached, and the continuous stable counter counts by 1.

[0025] (5); Step 4: To eliminate the influence of different parameter dimensions on subsequent Euclidean distance calculations, the dominance solution set of the historical wheel main circuit parameters is used. P prev The statistics are standardized using Z-scores, which map parameter spaces of different dimensions to the same space, ensuring the geometric effectiveness of nearest neighbor search.

[0026] Substituting the formula for calculating the statistic into equations (6) and (7), we can obtain... P prev No. j The mean of each parameter j and standard deviation j : (6); (7); In the formula, ( x ) j Representing the solution x The j There are several parameter values, among which j 1,…, D ; D The dimension of the parameter space, i.e. D This indicates that the number of dimensions is equal to the number of parameters.

[0027] right P new and P prev any solution in x Calculate its first Standardized parameters of dimension : (8); In the formula, To prevent the denominator of a fraction from being zero, it is usually set to 1. e -8 .

[0028] Step 5: After normalization D In the parameter space of the main circuit, two solutions are defined. x a and x b Euclidean distance between d ( x a , x b )for: (9); In the formula, To solve x a No. Standardized parameters of dimension, To solve x b No. Standardized parameters of the dimension.

[0029] Step 6: Perform nearest neighbor search based on the Euclidean distance calculation above, such as... Figure 3 As shown. For each new main circuit parameter solution generated in this round... x new P new In the historical main circuit parameter set P prev Find the old solution that minimizes its standardized Euclidean distance and mark it as the nearest neighbor main circuit parameter solution. x nearest .

[0030] (10); In the formula, argmin represents the independent variable that minimizes the objective function. There are two solutions x new and x The Euclidean distance between them.

[0031] Step 7: Find the nearest neighbor x nearest Then, calculate the relative rate of change of each original parameter of the main circuit. : (11); A new main circuit parameter solution x new There are usually multiple parameters; the solution for the new main circuit parameters is needed. x new If any parameter in the solution changes sufficiently, i.e., satisfies equation (12), the solution is determined to be a "significantly new solution" and is marked as unstable in this round; if and only if P new All main circuit parameters All are less than the rate of change threshold If equation (13) is satisfied, then it is determined that a "stable state" has been reached.

[0032] (12); (13); Step 8: To prevent the algorithm from stopping prematurely due to accidental local minima, this method introduces a continuously stable counter mechanism. The final stop signal is only triggered after the system has demonstrated stability for multiple consecutive rounds.

[0033] Update the continuous stable counter based on the judgment results of the above steps. K If the current flag is marked as unstable, meaning a significant new solution exists in this round, then the counter is reset. K =0; if the current label is stable, including And all new solutions If both values ​​are less than the threshold, then the counter is updated, i.e., a continuously stable counter is generated. K The value is increased by 1.

[0034] According to the counter K Make a convergence decision. If equation (14) is satisfied, the algorithm is determined to have fully converged. If equation (14) is not satisfied, the system will execute a new round of iteration.

[0035] (14); In the formula, K target This represents the target stability round threshold.

[0036] Step 9: Determine if the algorithm has fully converged Or exceeding the maximum number of iterations When the time is reached, the algorithm terminates the iteration and takes the final dominant solution set as the Pareto optimal solution set. P new Output.

[0037] Example To verify the method for determining iterator convergence during large model-driven MMC parameter optimization proposed in this invention, the following verification is conducted in conjunction with specific examples.

[0038] To verify the proposed iteration stopping condition in conjunction with the implementation examples, an MMC main circuit parameter optimization LLM iterator was constructed. By receiving information such as converter fixed parameters, parameters to be optimized, and grid operating conditions, initial parameters are generated. The solution vector containing iteration information, main circuit parameters, and calculated values ​​of constrained electrical quantities is obtained through external calculation tools. The proposed stopping condition for the LLM iterator is used to determine whether the condition is met. If the condition is not met, the iteration continues; if the condition is met, the iteration stops.

[0039] First, for input parameter optimization, the LLM input converter is connected to a voltage level of 500kV, with an output active power of 1200MW and an output reactive power of 300MW. The parameters to be optimized are: C sm and L m Other fixed main circuit parameters are shown in Table 1 in the embodiment.

[0040] Table 1 Other Fixed Main Circuit Parameters

[0041] In this embodiment, the preset stopping condition parameters for the LLM iterator are as follows: maximum number of iterations. T max = 20, relative rate of change threshold Target stable round threshold K target = 2. The number of iterations in the initial state is denoted as . t =1, the continuous stable counter is denoted as K =0.

[0042] After receiving the above information, the parameter optimization LLM, based on its pre-trained knowledge and reasoning ability, outputs 50 sets of main circuit parameters. Each set contains parameters to be optimized and fixed main circuit parameters. Subsequently, a main circuit parameter evaluation tool is used to verify whether these 50 sets of parameters satisfy the constraints, and the 50 sets of data are encapsulated into 50 solution vectors. By comparing each of these 50 solution vectors pairwise, the dominance set for this round, i.e., the current optimal solution set, is selected. P all And input it into the stop condition for judgment.

[0043] The first iteration process: First, check the number of iterations. At this point... t =1, although less than the maximum number of iterations T max =2, but since stability assessment requires historical data as a reference, it does not meet the requirement. t The activation condition is >1, therefore the convergence check is skipped. Let... t +1 means t If the value is 2, proceed to the next round of optimization iteration.

[0044] The second round of iteration process: all data are then input into the parameter optimization LLM. The parameter optimization LLM generates 50 sets of parameters and verifies them using the main circuit parameter determination tool to obtain 50 new solution vectors. The second round of dominance set is obtained by comparing each pair of vectors.

[0045] At this point, the global state data is read to obtain the dominance set from the previous two rounds, i.e., the original solution set. P all and current count K =0. At this time... t =2, meeting the startup condition. The original solution set P all Separate into the newly generated dominant solution set in this round P new There are 11 in total, plus the dominance solution set from the previous round. P prev There are 6 in total. Based on the separated... Pprev The calculated mean values ​​for the parameters are 14414.3 and 45.3, and the standard deviation values ​​are 2897.0 and 10.03. These two statistics are then used to analyze... P new and P prev any solution in x After standardization, the first... j Standardized parameters of dimension This allows for the construction of a normalized two-dimensional parameter space that eliminates dimensional differences. Within this normalized space, all new solutions are compared based on Euclidean distance. x new Perform a nearest neighbor search and calculate the rate of change of the parameters relative to the original parameters. Some new solutions were discovered. L m relative rate of change =0.0952, exceeding the threshold. These solutions are determined to be "significantly new solutions". Therefore, this round is marked as "unstable", and the counter remains unchanged. K =0, less than the target stable round threshold K target = 2, t Accumulate and proceed to the next iteration.

[0046] Ninth iteration (convergence phase): Subsequent iterations repeat the above process. In the ninth iteration, the original solution set is read. P all And the current count, since the eighth round has been marked as stable, the current counter is K =1. At this time... t =9, which satisfies the condition. Then the original solution set P all Separate into P new There are 3 in total. P prev There are 36 in total. Calculate. P prev The set of means for each parameter is 13971.1 and 42.9, and the set of standard deviations is 3117.1 and 8.47. P new and P prev any solution in x After standardization, a normalized two-dimensional parameter space is obtained. The Euclidean distance is then used to evaluate all new solutions. x new Perform a nearest neighbor search and calculate the relative rate of change. The results show that the maximum relative rate of change for any new solution is less than the threshold. This round has once again been marked as "stable". The counter has accumulated to... K =2, satisfying The termination condition is determined when the algorithm has fully converged, at which point the iteration terminates and the current iteration is output. P new As the final Pareto optimal solution set.

[0047] The iteration history and final Pareto front of this parameter optimization process are as follows: Figure 4 As shown, although the initially generated solution set is discrete and contains gray infeasible solutions and blue dominated solutions, after 9 rounds of iterative iteration, it successfully converges and locks into a stable Pareto front (red curve). This front accurately delineates the optimal trade-off boundary between parameters, and it smoothly covers the bridge arm inductance. L m From 27 m H to 56 m H and submodule capacitors C sm From 21400 u F to 10300 u The design space boundary of F. This fully demonstrates that the stopping condition construction method proposed in this invention can effectively eliminate inferior solutions and find the optimal Pareto front, and can converge quickly in fewer rounds, effectively avoiding over-search of the algorithm, thereby maximizing the utilization of computing resources while significantly reducing the experimental burden of subsequent hardware verification.

[0048] To further verify that the device main circuit parameters at the final Pareto front conform to physical constraints, a detailed analysis is provided below in conjunction with the MMC main circuit parameter examples shown in Table 2.

[0049] Table 2 MMC Main Circuit Parameters

[0050] Figure 5 This demonstrates the MMC's output rated active power. P rated =1200MW and reactive power Q rated Under the condition of 300Mvar, the modulation signal of the upper arm of phase A. S ap (t) The steady-state operating waveform. As shown in the figure, in... t= 1.90 s Up to 2.00 s Within the circuit, the MMC maintains a stable sinusoidal modulation state. Waveform data shows that the modulation signal of the upper bridge arm of phase A... S ap (t)The maximum and minimum values ​​are 0.93 and 0.025, respectively. At this point, the modulation waveform is strictly controlled between the upper limit 1 and the lower limit 0 of the modulation signal, without touching the amplitude limiting boundary. Further analysis shows that the modulation signal margin at the lower boundary is... Mar l The upper boundary margin value is 0.025. Mar h for 0.07 This indicates that under the current main circuit parameters, the MMC operates in the linear modulation region, without exceeding the physical limits of the modulation signal, effectively avoiding the risk of waveform distortion caused by overmodulation. The minimum margin value of 0.025 indicates that although the DC voltage utilization is close to its limit, it remains within a safe range, ensuring that the MMC can output a high-quality voltage waveform.

[0051] Figure 6 This demonstrates the MMC's output rated active power. P rated = 1200 MW and reactive power Q rated =300 Mvar operating condition, submodule capacitor voltage U cap ( t The steady-state operating waveform of the capacitor is shown in the figure. As can be seen from the figure, due to the periodic charging and discharging of the bridge arm current, the capacitor voltage exhibits a fundamental frequency of... 50 The voltage fluctuates periodically in Hz. The data in the figure shows that, under the current main circuit parameter configuration, the submodule capacitor voltage... U cap ( t The maximum peak value reached 2488V, while the minimum valley value after discharge was 2077V. The peak value constraint for the submodule capacitor was set at 2523V, and the capacitor voltage ripple constraint at 229V. Based on this analysis, the current peak margin is only 35V, and the peak margin percentage is 1.38%, indicating that the capacitor voltage peak value is very close to the safety constraint boundary. The current voltage ripple peak value is 411V, that is, the single-sided ripple amplitude is 205.5V, the ripple margin is 23.5V, and the remaining percentage of the ripple margin is 10.2%. A comparison shows that the peak margin is much smaller than the ripple margin. This result reveals that under the current operating conditions, the submodule capacitor peak constraint is the bottleneck limiting further optimization of the main circuit parameters. If the parameters to be optimized are further reduced, the constraints will be exceeded. This result fully proves that the found Pareto front set is optimal, thus further verifying the accuracy and effectiveness of the iterator convergence determination method for large model-driven MMC parameter optimization constructed in this invention.

[0052] Based on the same concept, this invention also provides a convergence determination system in the large model-driven MMC parameter optimization process, including an acquisition module, a normalization module, a calculation module, and an iteration module.

[0053] The acquisition module is used to acquire all solution set data generated in the current iteration during the optimization of main circuit parameters of MMC driven by a large model, and to separate the newly generated main circuit parameter dominant solution set and the main circuit parameter dominant solution set of the previous iteration from the original solution set.

[0054] The normalization module is used to normalize the main circuit parameter space based on the statistics of the main circuit parameter dominance solution set of the historical rounds, so as to obtain the standardized parameter space. In the standardized parameter space, for each new solution in the main circuit parameter dominance solution set newly generated in the current round, the nearest neighbor search is performed in the main circuit parameter dominance solution set of the historical rounds to obtain the nearest neighbor main circuit parameter solution corresponding to each new solution.

[0055] The calculation module is used to obtain the relative rate of change of the original parameters of each new solution relative to its nearest neighbor main circuit parameter solution; based on the comparison results of the relative rate of change of the original parameters of all new solutions with the preset stability threshold, it determines whether the current iteration has reached a stable state.

[0056] The iteration module is used to update the continuous stability counter based on the determination result of the stable state. When the continuous stability counter reaches the preset target stable round threshold, the determination algorithm converges and terminates the iteration, and outputs the final Pareto optimal solution set.

[0057] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0058] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A convergence determination method in the large model-driven MMC parameter optimization process, characterized in that, Includes the following steps: Obtain all solution set data generated in the current iteration round during the large model-driven MMC main circuit parameter optimization process, and separate the newly generated main circuit parameter dominant solution set and the main circuit parameter dominant solution set of the previous round from the original solution set; Based on the statistics of the main circuit parameter dominance solution set of the historical round, the main circuit parameter space is normalized and mapped to obtain the standardized parameter space. In the standardized parameter space, for each new solution in the main circuit parameter dominance solution set newly generated in the current round, a nearest neighbor search is performed in the main circuit parameter dominance solution set of the historical round to obtain the nearest neighbor main circuit parameter solution corresponding to each new solution. Obtain the relative rate of change of the original parameters of each new solution relative to its nearest neighbor main circuit parameter solution; based on the comparison results of the relative rate of change of the original parameters of all new solutions with the preset stability threshold, determine whether the current iteration has reached a stable state; The continuous stability counter is updated based on the determination result of the stable state. When the continuous stability counter reaches the preset target stable round threshold, the determination algorithm converges and terminates the iteration, and outputs the final Pareto optimal solution set. The statistical measures based on the dominant solution set of the historical main circuit parameters are used to normalize and map the main circuit parameter space, specifically including the following steps: Calculate the first number in the dominance solution set of the main circuit parameters of the historical cycle. j The statistics of the dimensional parameters include the mean and standard deviation; Based on the aforementioned statistics, any solution of the dominant solution set of the main circuit parameters in the historical rounds and the newly generated dominant solution set of the main circuit parameters in the current round are standardized.

2. The convergence determination method in the large model-driven MMC parameter optimization process as described in claim 1, characterized in that, The solution set data includes iterative information, main circuit parameters, and calculated values ​​of constrained electrical quantities.

3. The convergence determination method in the large model-driven MMC parameter optimization process as described in claim 1, characterized in that, The nearest neighbor search in the solution set dominated by the main circuit parameters of the historical round specifically includes the following steps: The Euclidean distance between the new solution and any solution in the solution set dominated by the main circuit parameters of the historical cycle is calculated based on the standardized parameters. The historical solution with the smallest Euclidean distance is selected as the nearest neighbor main circuit parameter solution for the new solution.

4. The convergence determination method in the large model-driven MMC parameter optimization process as described in claim 1, characterized in that, The formula for calculating the relative rate of change of the original parameters is: ； In the formula, The relative rate of change of the original parameters. For a new interpretation The j Parameter values, The nearest neighbor main circuit parameter solution for the new solution The j Each parameter value.

5. The convergence determination method in the large model-driven MMC parameter optimization process as described in claim 1, characterized in that, The determination of whether the current iteration has reached a stable state specifically includes the following steps: If no new dominant solution for the main circuit parameters is generated in this round, the current round is directly determined to be in a stable state. If a new dominant solution for the main circuit parameters is generated in this round, it is determined whether the relative rate of change of the original parameters of all new solutions is lower than the preset stability threshold. If so, the current round is determined to be in a stable state; otherwise, it is determined to be in an unstable state.

6. The convergence determination method in the large model-driven MMC parameter optimization process as described in claim 5, characterized in that, The step of updating the continuous stability counter based on the determination result of the stable state specifically includes: If the current round is determined to be in an unstable state, then reset the continuous stability counter; If the current round is determined to be in a stable state, then increment the continuous stability counter by 1; When the continuous stable counter K When the number of stable rounds is greater than or equal to the preset target threshold, the algorithm is considered to have fully converged.

7. A convergence determination system for large model-driven MMC parameter optimization, characterized in that, include: The acquisition module is used to acquire all solution set data generated in the current iteration during the optimization of main circuit parameters of MMC driven by a large model, and to separate the newly generated main circuit parameter dominant solution set and the main circuit parameter dominant solution set of the previous iteration from the original solution set; The normalization module is used to normalize the main circuit parameter space based on the statistics of the main circuit parameter dominance solution set of the historical rounds, so as to obtain the standardized parameter space. In the standardized parameter space, for each new solution in the main circuit parameter dominance solution set newly generated in the current round, the nearest neighbor search is performed in the main circuit parameter dominance solution set of the historical rounds to obtain the nearest neighbor main circuit parameter solution corresponding to each new solution. The calculation module is used to obtain the relative rate of change of the original parameters of each new solution relative to its nearest neighbor main circuit parameter solution; based on the comparison results of the relative rate of change of the original parameters of all new solutions with the preset stability threshold, it determines whether the current iteration has reached a stable state. The iteration module is used to update the continuous stability counter based on the determination result of the stable state. When the continuous stability counter reaches the preset target stable round threshold, the determination algorithm converges and terminates the iteration, and outputs the final Pareto optimal solution set. The statistical measures based on the dominant solution set of the historical main circuit parameters are used to normalize and map the main circuit parameter space, specifically including the following steps: Calculate the first number in the dominance solution set of the main circuit parameters of the historical cycle. j The statistics of the dimensional parameters include the mean and standard deviation; Based on the aforementioned statistics, any solution of the dominant solution set of the main circuit parameters in the historical rounds and the newly generated dominant solution set of the main circuit parameters in the current round are standardized.

Images