Photovoltaic cluster control parameter multi-objective optimization method, system, equipment and medium

Abstract

The invention discloses a photovoltaic cluster control parameter multi-objective optimization method, system and device and a medium, and the method comprises the steps: designing a fault scene set covering a plurality of fault types and operation conditions, generating parameter samples, obtaining a transient stability margin index corresponding to each sample through electromagnetic transient simulation, and constructing a transient stability domain agent model; performing global sensitivity analysis on the control parameter set, and screening out a high-sensitivity parameter subset; constructing a multi-objective optimization problem containing four objectives, and introducing constraint conditions; solving by adopting a multi-objective optimization algorithm, and outputting a Pareto optimal solution set; and calculating the comprehensive close degree of each solution by adopting an approximate ideal solution sorting algorithm, applying parameter disturbance to each candidate solution, and calculating a robustness index to obtain final recommended parameter configuration. According to the method, the critical clearing time is remarkably prolonged, the frequency and voltage fluctuation is reduced, and the method has high robustness, efficient computing power and engineering deployability.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic power generation control technology, and in particular to a method, system, equipment and medium for multi-objective optimization of photovoltaic cluster control parameters. Background Technology

[0002] With a high proportion of new energy sources being integrated into the power system, distributed photovoltaic (PV) clusters must not only complete power generation tasks but also possess the ability to actively support the stable operation of the power grid. The control parameters used in PV inverters—such as virtual inertia coefficient, damping coefficient, and droop coefficient—directly determine the transient stability performance of the entire cluster when encountering faults or disturbances. However, current optimization and control technologies for inverter clusters still present numerous challenges.

[0003] Traditional methods often focus on a single objective, either pursuing improvements in transient stability or system economics, making it difficult to achieve a balance between multiple performance indicators in practical applications. Furthermore, handling transient stability constraints is inherently challenging because such stability involves complex nonlinear relationships between multiple variables such as voltage and current. These relationships are difficult to translate into explicit constraints that can be directly used in the optimization model, leading to calculated solutions that often deviate from actual stability requirements. Parameter sensitivity analysis is also insufficiently systematic; existing solutions mostly rely on empirical judgment or local experiments to determine which parameters should be adjusted, forcing the optimization process to involve repeated trial and error, naturally reducing efficiency. In addition, the parameter coupling relationships between multiple inverters within a cluster are often ignored; many technologies simply superimpose the results of individual unit optimizations, which can easily lead to coordination problems such as power oscillations within the cluster. Finally, robustness verification is also clearly insufficient—the optimized parameters are usually only tested in a few scenarios such as rated operating conditions or typical disturbances, without comprehensive verification for complex operating conditions such as frequent fluctuations in grid load and rapid changes in the external environment, which greatly reduces their adaptability in real and changing scenarios. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, device, and medium for multi-objective optimization of photovoltaic cluster control parameters to solve the problem that existing inverter cluster optimization and control methods cannot achieve coordinated optimization and reliable support of multiple performance indicators in scenarios with a high proportion of new energy access under complex and variable operating conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a multi-objective optimization method for photovoltaic cluster control parameters, comprising:

[0008] Acquire system modeling data, establish an electromagnetic transient simulation model of the photovoltaic cluster, and identify the set of control parameters to be optimized and the range of physical constraints;

[0009] Based on the electromagnetic transient simulation model and the control parameter set, a fault scenario set covering multiple fault types and operating conditions is designed, parameter samples are generated, and the transient stability margin index corresponding to each sample is obtained through electromagnetic transient simulation, thus constructing a transient stability domain proxy model.

[0010] Using the transient stability domain proxy model, a global sensitivity analysis is performed on the set of control parameters to filter out a subset of highly sensitive parameters;

[0011] Based on the aforementioned subset of highly sensitive parameters, a multi-objective optimization problem with four objectives is constructed, and constraints are introduced.

[0012] The multi-objective optimization problem is solved using a multi-objective optimization algorithm, and the Pareto optimal solution set is output.

[0013] For the Pareto optimal solution set, the comprehensive closeness of each solution is calculated using the approximation ideal solution sorting algorithm. Then, parameter perturbation is applied to each candidate solution to calculate the robustness index. The solution with high comprehensive closeness and the smallest robustness index is selected as the final recommended parameter configuration.

[0014] As a preferred embodiment of the multi-objective optimization method for photovoltaic cluster control parameters described in this invention, the transient stability domain proxy model is constructed in the following manner:

[0015] Generate parameter samples based on the set of control parameters;

[0016] The parameter samples are subjected to electromagnetic transient simulation in the fault scenario set to obtain the corresponding transient stability margin index, forming a parameter-transient stability margin index training sample pair.

[0017] Using the training sample pairs, a regression model with a radial basis function neural network architecture is trained to construct the transient stable domain proxy model.

[0018] The beneficial effects of this preferred technical solution are that by constructing a transient stability domain proxy model based on radial basis function neural network, the computational overhead of electromagnetic transient simulation in multiple scenarios is significantly reduced, while maintaining a high-precision approximation capability of transient stability margin, providing a feasible basis for subsequent efficient optimization.

[0019] As a preferred embodiment of the multi-objective optimization method for photovoltaic cluster control parameters described in this invention, the global sensitivity analysis includes:

[0020] The Sobol algorithm is used to calculate the overall sensitivity index of each control parameter;

[0021] Set a sensitivity threshold;

[0022] Parameters whose total sensitivity index is greater than the threshold are included in the subset of highly sensitive parameters.

[0023] As a preferred embodiment of the multi-objective optimization method for photovoltaic cluster control parameters described in this invention, the step of solving the problem using a multi-objective optimization algorithm includes:

[0024] When initializing the population, the sampling density of the highly sensitive parameter subset is higher than that of the other control parameters;

[0025] During the optimization process, the constraint threshold is dynamically set according to the current iteration number, and ε-constraints are constructed based on the constraint threshold to handle transient stability constraints;

[0026] In the initial stage of several generations of optimization, the fitness of individuals is evaluated using a transient stable domain proxy model. After reaching a pre-set number of switching generations, electromagnetic transient simulation is used to evaluate the fitness of individuals.

[0027] During the optimization process, every pre-set retraining interval algebra, newly generated electromagnetic transient simulation data is added to the training dataset of the surrogate model, and the transient stable domain surrogate model is retrained based on the updated training dataset.

[0028] As a preferred embodiment of the multi-objective optimization method for photovoltaic cluster control parameters described in this invention, the step of calculating the comprehensive closeness of each solution using an algorithm for ranking solutions to approximate the ideal solution includes:

[0029] Decision preference weights were assigned to the four optimization objectives, with transient stability margin having the highest weight and parameter dispersion having the lowest weight.

[0030] Positive ideal solutions are constructed based on the optimal values ​​of each objective in the Pareto optimal solution set, and negative ideal solutions are constructed based on the worst values ​​of each objective.

[0031] Calculate the weighted Euclidean distance from each Pareto solution to the positive and negative ideal solutions;

[0032] The relative proximity of each solution is calculated based on the distance, and the Pareto solutions are sorted according to the relative proximity.

[0033] The beneficial effect of this preferred technical solution is that by introducing decision preference weights and combining them with the weighted distance calculation of positive and negative ideal solutions, a comprehensive quantitative ranking of the multi-objective performance in the Pareto optimal solution set is achieved, which effectively supports the optimal parameter configuration decision-making under the multi-dimensional requirements of taking into account transient stability and cluster coordination.

[0034] As a preferred embodiment of the multi-objective optimization method for photovoltaic cluster control parameters described in this invention, the calculation of robustness indices includes:

[0035] Apply random perturbations within a preset range to the control parameters of each candidate solution, and repeat the simulation multiple times to simulate parameter uncertainty;

[0036] The fluctuation of multi-objective function values ​​under statistical perturbation conditions is evaluated, with the ratio of the standard deviation to the mean of the function values ​​used as a robustness indicator.

[0037] Among the candidate solutions with the highest overall similarity ranking, the solution with the smallest robustness index is selected as the final recommended parameter configuration.

[0038] As a preferred embodiment of the multi-objective optimization method for photovoltaic cluster control parameters described in this invention, the multi-objective optimization problem is constructed in the following manner:

[0039] Based on the aforementioned subset of highly sensitive parameters, four optimization objectives are defined, including: maximizing the minimum transient stability margin index, minimizing the dynamic response overshoot, maximizing the grid support capability, and minimizing the parameter dispersion within the cluster.

[0040] The multi-objective optimization problem introduces constraints, including transient stability hard constraints, safe operation constraints that satisfy system voltage and frequency, and differentiated coordination constraints.

[0041] Secondly, the present invention provides a multi-objective optimization system for photovoltaic cluster control parameters, comprising:

[0042] The system modeling and parameter identification module is used to acquire system modeling data, establish an electromagnetic transient simulation model of the photovoltaic cluster, and identify the set of control parameters to be optimized and the range of physical constraints.

[0043] The transient stability domain proxy modeling module is used to design a set of fault scenarios covering multiple fault types and operating conditions based on the electromagnetic transient simulation model and the set of control parameters, generate parameter samples, obtain the transient stability margin index corresponding to each sample through electromagnetic transient simulation, and construct a transient stability domain proxy model.

[0044] The control parameter sensitivity analysis module is used to perform global sensitivity analysis on the control parameter set using the transient stability domain proxy model, and to filter out a subset of highly sensitive parameters.

[0045] A multi-objective optimization problem construction module is used to construct a multi-objective optimization problem with four objectives based on the subset of highly sensitive parameters, and to introduce constraints.

[0046] The Pareto optimal solution module is used to solve the multi-objective optimization problem using a multi-objective optimization algorithm and output a Pareto optimal solution set.

[0047] The robust decision-making and parameter recommendation module is used to calculate the comprehensive closeness of each solution to the Pareto optimal solution set using an algorithm that approximates the ideal solution ranking algorithm, apply parameter perturbation to each candidate solution, calculate the robustness index, and select the solution with high comprehensive closeness and the smallest robustness index as the final recommended parameter configuration.

[0048] Thirdly, the present invention provides an electronic device, comprising:

[0049] Memory, used to store programs;

[0050] A processor is configured to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the multi-objective optimization method for photovoltaic cluster control parameters.

[0051] Fourthly, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the step of implementing the multi-objective optimization method for photovoltaic cluster control parameters. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0053] Figure 1 This is a schematic diagram of the basic process of a multi-objective optimization method for photovoltaic cluster control parameters provided in one embodiment of the present invention. Detailed Implementation

[0054] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0055] Example 1, referring to Figure 1As an embodiment of the present invention, a multi-objective optimization method for photovoltaic cluster control parameters is provided, comprising:

[0056] S100: Acquire system modeling data, establish an electromagnetic transient simulation model of the photovoltaic cluster, and identify the set of control parameters to be optimized and the range of physical constraints;

[0057] S200: Based on the electromagnetic transient simulation model and the control parameter set, design a fault scenario set covering multiple fault types and operating conditions, generate parameter samples, and obtain the transient stability margin index corresponding to each sample through electromagnetic transient simulation, and construct a transient stability domain proxy model.

[0058] S300: Using the transient stability domain proxy model, perform a global sensitivity analysis on the set of control parameters to filter out a subset of highly sensitive parameters;

[0059] S400: Based on the aforementioned subset of highly sensitive parameters, construct a multi-objective optimization problem with four objectives and introduce constraints;

[0060] S500: The multi-objective optimization problem is solved using a multi-objective optimization algorithm, and the Pareto optimal solution set is output.

[0061] S600: For the Pareto optimal solution set, the comprehensive closeness of each solution is calculated using the approximation ideal solution sorting algorithm, and parameter perturbation is applied to each candidate solution to calculate the robustness index. The solution with high comprehensive closeness and the smallest robustness index is selected as the final recommended parameter configuration.

[0062] It should be noted that existing technologies face a series of challenges during photovoltaic cluster operation, including the difficulty of single-objective optimization methods in simultaneously considering transient stability, dynamic response speed, grid support capacity, and cluster coordination, leading to one-sided control performance; transient stability constraints involve complex nonlinear dynamic processes, making it difficult to embed them into the optimization model, often resulting in optimization results that deviate from actual stability requirements; parameter tuning often relies on experience or local trial and error, lacking global sensitivity analysis of high-dimensional parameter spaces, resulting in low optimization efficiency; inverters within the cluster typically use uniform parameter configurations, failing to consider differences in node electrical location and line impedance, which can easily lead to internal power oscillations or suppress local optimal responses; furthermore, optimization schemes are mostly verified only under typical operating conditions, lacking robustness guarantees against real-world disturbances such as cloud cover, communication interruptions, inverter shutdowns, and changes in grid strength, and lacking online adaptive update mechanisms compatible with hardware capabilities and grid connection standards, making it difficult to adapt to dynamic changes in the actual operating environment.

[0063] Therefore, addressing the problem that existing inverter cluster optimization and control methods cannot achieve coordinated optimization and reliable support of multiple performance indicators in scenarios with a high proportion of new energy access under complex and variable operating conditions, the S100-S600 steps, through the integration of transient stability domain proxy modeling, sensitivity-driven multi-objective coordinated optimization and robust decision-making mechanism, achieve the comprehensive optimal configuration of photovoltaic cluster control parameters in terms of stability, dynamic performance, support capability and coordination, and possess efficient computation, strong robustness and engineering deployability.

[0064] Example 2, this is an embodiment of the present invention, which provides a multi-objective optimization method for photovoltaic cluster control parameters based on the previous embodiment, including:

[0065] In this embodiment of the application, the system modeling data in step S100 includes: the power grid topology and line and transformer parameters; the grid connection point location, rated capacity and control architecture type of each inverter; and the irradiance and load level under typical operating conditions.

[0066] In one specific implementation, the set of control parameters to be optimized includes eight key parameters, including the virtual inertia coefficient. Damping coefficient N·m·s / rad, active droop coefficient reactive power droop coefficient Current limiting value PU, LVRT reactive power gain [1.5, 4.0], PLL bandwidth Voltage feedforward coefficient .

[0067] In this embodiment of the application, the transient stable domain proxy model in step S200 is constructed in the following way:

[0068] Generate parameter samples based on the set of control parameters;

[0069] Electromagnetic transient simulations are performed on the parameter samples in the fault scenario set to obtain the corresponding transient stability margin index, forming parameter-transient stability margin index training sample pairs.

[0070] Using training sample pairs, a regression model with a radial basis function neural network architecture is trained to construct a transient stable domain proxy model.

[0071] In this embodiment of the application, the proxy model construction method in step S200 generates no less than 500 sets of control parameter samples through Latin hypercube sampling, obtains transient stability margin index (SSM) by performing electromagnetic transient simulation under 20 fault scenarios, and constructs the mapping relationship between SSM and control parameters by using K-means clustering to determine the center and least squares method to solve the weights.

[0072] In an optional implementation, the proxy model construction method in step S200 can also be as follows: after generating Latin hypercube sampling parameter samples and completing electromagnetic transient simulation to obtain transient stability margin index (SSM), a Gaussian process regression model is trained using the obtained parameter-SSM data. The kernel function hyperparameters (such as the length scale and noise variance of the radial basis kernel) are optimized through maximum likelihood estimation or cross-validation, thereby establishing a nonlinear probability mapping from control parameters to SSM.

[0073] In an optional implementation, the proxy model construction method in step S200 can also be as follows: after generating parameter samples through Latin hypercube sampling and obtaining the transient stability margin index (SSM) through electromagnetic transient simulation, a fully connected neural network with 3–5 hidden layers can be trained using the obtained parameter-SSM data. The ReLU activation function and Dropout layer are used to prevent overfitting, and the mean squared error loss is minimized through the Adam optimizer, thereby establishing a high-dimensional nonlinear mapping from control parameters to SSM.

[0074] In this embodiment of the application, the radial basis function neural network (RBFNN) has 120 hidden layer nodes and uses a Gaussian kernel function.

[0075] In this embodiment, the transient stability margin index SSM is defined as the minimum value among voltage safety margin, critical clearing time margin, and power angle stability margin, and its specific expression is as follows:

[0076]

[0077] in, This is the lowest voltage during the fault. This refers to the actual duration of the fault. The value represents the maximum power angle offset, and the others represent the corresponding stability critical values.

[0078] In this embodiment of the application, a regression model based on a radial basis function neural network is trained using training sample pairs, and its output expression is:

[0079]

[0080] in, For the input control parameter vector, For the first The center point of each hidden layer node For radial basis functions, For the corresponding output weights, This represents the total number of hidden layer nodes. The center of each hidden layer node... The output weights are determined from the training samples using K-means clustering. Solve using the least squares method.

[0081] In this embodiment, the transient stability margin index is a normalized index that comprehensively reflects the voltage drop amplitude, critical clearing time redundancy, and power angle offset safety margin. Specifically, it is defined as the minimum value among the three. The transient stability margins under different fault scenarios are weighted according to preset weights, which are allocated according to the probability of occurrence of the fault type. The weights for three-phase short circuit, two-phase short circuit, single-phase grounding, and other fault types are 0.3, 0.25, 0.2, and 0.25, respectively.

[0082] In one specific implementation, the fault scenario set includes 20 operating conditions, covering three-phase short circuit, two-phase short circuit, and single-phase ground fault. The fault location and duration are variable, and the operating illumination intensity range is 600~1000 W / m². Parameter samples are generated using Latin hypercube sampling, with 500 sets of samples, and a total of 10,000 electromagnetic transient simulations are performed. The radial basis function neural network uses a Gaussian kernel function, with 120 hidden layer nodes. The coefficient of determination on the test set after training is... The average relative error is 3.2%.

[0083] In this embodiment of the application, a precision verification mechanism is set during the training of the proxy model, requiring the determination coefficient of the test set to be no less than 0.95 and the average relative error to be no more than 3.5%. If the prediction error in the parameter boundary region exceeds 5%, the region is encrypted and training samples are supplemented.

[0084] In this embodiment of the application, the global sensitivity analysis in step S300 includes:

[0085] The Sobol algorithm is used to calculate the overall sensitivity index of each control parameter;

[0086] Set a sensitivity threshold;

[0087] Parameters whose total sensitivity index is greater than the threshold are included in the subset of highly sensitive parameters.

[0088] In this embodiment of the application, the global sensitivity analysis in step S300 includes using the constructed transient stability domain proxy model to perform large-scale random sampling in the control parameter space, evaluating the influence of each parameter individually and jointly on the transient stability margin through multiple simulations, thereby quantifying the overall sensitivity of each parameter, and selecting a subset of highly sensitive parameters that have a significant impact on system stability based on the set sensitivity threshold.

[0089] In an optional implementation, the global sensitivity analysis in step S300 can also be performed after the transient stability domain proxy model is constructed. This can be done by generating multiple random trajectories in the control parameter space, perturbing individual parameters along each trajectory and calculating their basic effects on the transient stability margin, and then statistically analyzing the mean and standard deviation of the basic effects of each parameter. The mean reflects the overall influence and the standard deviation reflects the nonlinearity or interaction, thereby quickly identifying a subset of highly sensitive parameters.

[0090] In an optional implementation, the global sensitivity analysis in step S300 can also decompose the input signal of the control parameters into harmonic components of different frequencies using Fourier series after constructing the transient stability domain proxy model. By analyzing the energy proportion of each frequency component in the transient stability margin output response, the main effect and total effect sensitivity index of each parameter can be calculated, thereby efficiently identifying a subset of highly sensitive parameters that have a significant impact on system stability.

[0091] In the embodiments of this application, the total sensitivity index The calculation formula is:

[0092]

[0093] in, Indicator of transient stability margin Indicates except the first A vector consisting of all parameters other than the control parameter. Represents variance. Represents the mathematical expectation;

[0094] In this embodiment of the application, the multi-objective optimization problem in step S400 is constructed in the following way:

[0095] Based on the aforementioned subset of highly sensitive parameters, four optimization objectives are defined, including: maximizing the minimum transient stability margin index, minimizing the dynamic response overshoot, maximizing the grid support capability, and minimizing the parameter dispersion within the cluster.

[0096] Multi-objective optimization problems introduce constraints, including hard constraints for transient stability, constraints for safe operation that meet system voltage and frequency, and constraints for differentiated coordination.

[0097] In this embodiment of the application, the transient stability hard constraint requires that the transient stability margin index calculated under all fault scenarios is not lower than the preset minimum allowable value;

[0098] In this embodiment of the application, the system meets the safe operation constraints of system voltage and frequency. During the transient process, the voltage amplitude and frequency deviation of each key bus are always within the safe operation limits specified by the grid connection standard.

[0099] In this embodiment of the application, the differentiated coordination constraint sets the allowable deviation range between the control parameters of each inverter and the average parameters of the cluster based on the electrical importance of the nodes to which each inverter is connected in the photovoltaic cluster. The electrical importance is determined by combining the short-circuit capacity of the corresponding node and the electrical distance from the node to the main grid.

[0100] In one specific implementation, the weighting rules are as follows: If the grid frequency stability requirement is high (e.g., the allowable frequency deviation range is ≤ ±0.2Hz), then the weight of the virtual inertia coefficient J in the support capability target is increased to 0.5; if voltage stability is the core requirement (e.g., the allowable voltage deviation range is ≤ ±5%), then the LVRT reactive power gain... The weight was adjusted to 0.4 to enhance the method's adaptability to different power grid specifications.

[0101] In this embodiment, among the four optimization objectives, the dynamic response overshoot is the weighted sum of the power angle, frequency, and voltage overshoot in the objective function, and the weights are dynamically allocated based on the global sensitivity analysis results.

[0102] In this embodiment, the differentiated coordination constraint further allows each inverter in the cluster to make local fine-tuning based on its baseline parameters, with the fine-tuning range not exceeding 10% of the baseline value, so as to retain local operating condition adaptability and avoid suppressing local optimal response due to forced parameter consistency.

[0103] In this embodiment of the application, step S500 employs a multi-objective optimization algorithm for solving the problem, including:

[0104] When initializing the population, the sampling density of the highly sensitive parameter subset is higher than that of the other control parameters;

[0105] During the optimization process, the constraint threshold is dynamically set according to the current iteration number, and ε-constraints are constructed based on the constraint threshold to handle transient stability constraints;

[0106] In the initial stage of several generations of optimization, the fitness of individuals is evaluated using a transient stable domain proxy model. After reaching a pre-set number of switching generations, electromagnetic transient simulation is used to evaluate the fitness of individuals.

[0107] During the optimization process, every pre-set retraining interval algebra, newly generated electromagnetic transient simulation data is added to the training dataset of the surrogate model, and the transient stable domain surrogate model is retrained based on the updated training dataset.

[0108] In the embodiments of this application, the preset switching generation can be determined according to the trend of the prediction error of the surrogate model on the verification sample. For example, when the relative error of the SSM prediction of the Pareto front solution by the surrogate model for several consecutive generations is less than 5%, the switching to electromagnetic transient simulation is triggered; or, in a preferred embodiment, the switching generation is set to 60% of the total number of iterations (e.g., switching to the 300th generation when the maximum is 500 generations).

[0109] In this embodiment, the pre-set retraining interval can be set to a fixed value (e.g., every 50 generations), or it can be dynamically adjusted according to the prediction deviation of the surrogate model on the latest simulation data: when the relative error between the surrogate prediction value of the newly generated individual and the accurate simulation value exceeds a preset threshold (e.g., 8%), the surrogate model is immediately retrained instead of waiting for a fixed interval.

[0110] It should be noted that, since highly sensitive parameters have a significant impact on system stability, dense sampling of their dimensions can accelerate convergence to a high-quality solution region; dynamically adjusting ε-constraints can gradually tighten the stability margin requirements, avoiding the loss of population diversity due to excessively strict constraints in the early stages; using a surrogate model in the initial stage can significantly reduce computational overhead, while switching to electromagnetic transient simulation in the later stages ensures that the final solution meets the actual physical dynamic characteristics; periodically introducing new simulation data to retrain the surrogate model can effectively correct the prediction bias of the surrogate model in key regions and improve the reliability of the optimization process.

[0111] In this embodiment, the multi-objective optimization in step S500 adopts the improved NSGA-III algorithm. The population is initialized in the search space composed of a subset of highly sensitive parameters. New solutions are generated by combining reference point-guided selection, simulated binary crossover and polynomial mutation. Transient stability hard constraints are handled by dynamic ε-constraints. In the early stage, a surrogate model is used for rapid evaluation, and in the later stage, it is switched to accurate electromagnetic transient simulation for verification. At the same time, the surrogate model is updated online with new simulation data periodically until the convergence criterion is met and the Pareto optimal solution set is output.

[0112] In an optional implementation, the multi-objective optimization algorithm in step S500 can further decompose the four-objective optimization problem into a set of single-objective subproblems in the search space composed of a subset of highly sensitive parameters. Co-evolution is carried out by using a preset weight vector and Tchebycheff or a weighted summation aggregation function. The solution of each subproblem is updated using neighborhood solutions, and constraints such as transient stability are handled in combination with feasibility rules. At the same time, a surrogate model is called in the early stage of optimization to accelerate the evaluation, and the electromagnetic transient simulation is switched to refine the evaluation in the later stage. Finally, a uniformly distributed Pareto optimal solution set is obtained.

[0113] In an optional implementation, the multi-objective optimization algorithm in step S500 can also regard the control parameter configuration as the action of the agent, and use transient stability margin, dynamic overshoot, support capability and cluster coordination to form a multi-dimensional reward signal. In the simulation environment, it continuously explores the parameter space by interacting with the power grid, and uses Pareto deep Q network (Pareto DQN) or multi-objective reinforcement learning (MORL) mechanism based on preference vector to learn the non-dominated policy set, and finally outputs a set of robust parameter configuration schemes that balance multiple performance indicators.

[0114] In one specific implementation, the algorithm sets the population size to 100, the maximum number of iterations to 500, the crossover probability to 0.9, and the mutation probability to 0.125; the ε-constraint initial value... The attenuation factor α = 2.0, and the final constraint boundary. The first 300 generations of optimization were evaluated using a surrogate model, with precise simulation verification performed every 50 generations. After the 300th generation, the system switched to full-precision simulation and updated the surrogate model online.

[0115] In this embodiment of the application, the optimization process is set with a convergence criterion: when the rate of change of the hypervolume index is less than 0.1% for 20 consecutive generations and the prediction error of the surrogate model is stable within the allowable range, or when the maximum number of iterations is reached, the optimization is terminated.

[0116] In this embodiment of the application, step S600 uses an approximation-to-ideal-solution sorting algorithm to calculate the overall closeness of each solution, including:

[0117] Decision preference weights were assigned to the four optimization objectives, with transient stability margin having the highest weight and parameter dispersion having the lowest weight.

[0118] Positive ideal solutions are constructed based on the optimal values ​​of each objective in the Pareto optimal solution set, and negative ideal solutions are constructed based on the worst values ​​of each objective.

[0119] Calculate the weighted Euclidean distance from each Pareto solution to the positive and negative ideal solutions;

[0120] The relative proximity of each solution is calculated based on the distance, and the Pareto solutions are sorted according to the relative proximity.

[0121] In this embodiment of the application, the calculation of the robustness index in step S600 includes:

[0122] Apply random perturbations within a preset range to the control parameters of each candidate solution, and repeat the simulation multiple times to simulate parameter uncertainty;

[0123] The fluctuation of multi-objective function values ​​under statistical perturbation conditions is evaluated, with the ratio of the standard deviation to the mean of the function values ​​used as a robustness indicator.

[0124] Among the candidate solutions with the highest overall similarity ranking, the solution with the smallest robustness index is selected as the final recommended parameter configuration.

[0125] In this embodiment of the application, the robustness index design in step S600 involves applying random disturbances to the candidate parameter configuration and its power grid environment (such as short-circuit ratio), running simulations multiple times to obtain fluctuation data of multi-objective function values, calculating the ratio of the sum of the standard deviation and the mean, thereby quantifying the stability of performance under parameter uncertainty and external disturbances, and selecting the most robust final recommended scheme from the high comprehensive closeness solutions accordingly.

[0126] In an optional implementation, the robustness index design in step S600 can also systematically search for the worst combination of perturbations that leads to the worst performance of multiple objectives within the preset perturbation range of the parameters of the candidate solution and its operating environment (such as short-circuit ratio and light intensity). The worst performance value is used as the robustness evaluation criterion, and the parameter configuration with the best overall performance under the worst-case scenario is selected as the final recommended solution.

[0127] In an optional implementation, the robustness index design in step S600 can also perform a large number of random sampling simulations within the uncertainty range of candidate parameter configuration and its operating environment (such as short-circuit ratio, illumination, load), statistically analyze the probability that the multi-objective performance index meets the preset safety threshold (such as SSM≥0.15, frequency deviation≤0.5Hz, etc.), and preferentially select the solution with the highest probability of meeting the standard (such as ≥95%) as the final recommended solution.

[0128] In this embodiment, when the grid frequency stability requirement is high (allowable deviation ≤ ±0.2Hz), the weight of the virtual inertia coefficient J in the support capability target is increased to 0.5; when voltage stability is the dominant requirement (allowable deviation ≤ ±5%), The weight is set to 0.4.

[0129] In this embodiment of the application, the robustness index RI is calculated using the following formula:

[0130]

[0131] in, This is the control parameter vector for the candidate solution. For the applied parameter perturbation, The system short-circuit ratio, This represents the short-circuit ratio disturbance. Represents the set of values ​​for a multi-objective function. Indicates standard deviation, This represents the mean;

[0132] In the embodiments of this application, the calculation of robustness index not only considers the random disturbance of control parameters, but also simultaneously introduces external environmental disturbances such as changes in grid short-circuit ratio, step changes in illuminance and load fluctuations, forming a two-dimensional robustness assessment under the joint disturbance of parameters and environment.

[0133] In this application embodiment, the total number of disturbance simulation scenarios is no less than 20, covering typical real-world operational challenges such as three-phase short circuit, rapid cloud cover, communication interruption, partial inverter shutdown, and changes in grid strength.

[0134] In this embodiment, the final recommended parameter configuration is verified under various disturbance scenarios. The verified final recommended parameter configuration is then subjected to hardware compatibility verification and grid connection standard compliance checks, and finally distributed to the photovoltaic cluster.

[0135] In this embodiment of the application, before the parameters are issued, a hardware feasibility check is performed on the final recommended parameter configuration, including that the inverter sampling frequency is not less than 1kHz, the floating-point operation capability is not less than 100MIPS, and the communication bandwidth is not less than 1Mbps; for parameters that exceed the range supported by the device, they are adjusted to the feasible boundary.

[0136] In this embodiment of the application, the parameter distribution adopts the IEC 61850 communication protocol that conforms to the IEEE 1547 and GB / T 36547 standards to ensure compatibility with the interface of the power grid dispatching system and the inverter controller;

[0137] In this embodiment, the system further deploys an online monitoring module to build a simplified stability margin evaluation model based on real-time voltage and frequency data, reducing computational overhead by 70%. When the approximate stability margin is less than 0.12 for five consecutive sampling periods, or the frequency deviation continues to exceed 0.3Hz or the voltage deviation continues to exceed 8%, it is determined to be a performance degradation.

[0138] In this embodiment of the application, an adaptive parameter update process is triggered when any of the following conditions are met: photovoltaic output change exceeds 30%, the number of inverters shut down exceeds 10, performance deterioration occurs, or grid topology is adjusted.

[0139] In this embodiment of the application, the parameter update adopts a partitioned synchronization strategy, which divides the photovoltaic cluster into 3 to 5 regions and updates the parameters of each region sequentially. The update time interval between adjacent regions is not less than 1 second, so as to suppress the power oscillation that may be caused by asynchronous updates of the entire cluster.

[0140] Example 3 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a multi-objective optimization system for photovoltaic cluster control parameters.

[0141] It should be noted that the technical solution of the photovoltaic cluster control parameter multi-objective optimization system is based on the same concept as the above-mentioned photovoltaic cluster control parameter multi-objective optimization method. For details not described in detail in this embodiment, please refer to the description of the above-mentioned photovoltaic cluster control parameter multi-objective optimization method.

[0142] This embodiment provides a multi-objective optimization system for photovoltaic cluster control parameters, comprising:

[0143] The system modeling and parameter identification module is used to acquire system modeling data, establish an electromagnetic transient simulation model of the photovoltaic cluster, and identify the set of control parameters to be optimized and the range of physical constraints.

[0144] The transient stability domain proxy modeling module is used to design a set of fault scenarios covering multiple fault types and operating conditions based on the electromagnetic transient simulation model and the set of control parameters, generate parameter samples, obtain the transient stability margin index corresponding to each sample through electromagnetic transient simulation, and construct a transient stability domain proxy model.

[0145] The control parameter sensitivity analysis module is used to perform global sensitivity analysis on the control parameter set using the transient stability domain proxy model, and to filter out a subset of highly sensitive parameters.

[0146] A multi-objective optimization problem construction module is used to construct a multi-objective optimization problem with four objectives based on the subset of highly sensitive parameters, and to introduce constraints.

[0147] The Pareto optimal solution module is used to solve the multi-objective optimization problem using a multi-objective optimization algorithm and output a Pareto optimal solution set.

[0148] The robust decision-making and parameter recommendation module is used to calculate the comprehensive closeness of each solution to the Pareto optimal solution set using an algorithm that approximates the ideal solution ranking algorithm, apply parameter perturbation to each candidate solution, calculate the robustness index, and select the solution with high comprehensive closeness and the smallest robustness index as the final recommended parameter configuration.

[0149] This embodiment also provides an electronic device applicable to a multi-objective optimization method for photovoltaic cluster control parameters, including:

[0150] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a multi-objective optimization method for photovoltaic cluster control parameters as proposed in the above embodiments.

[0151] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a multi-objective optimization method for photovoltaic cluster control parameters as proposed in the above embodiments.

[0152] The storage medium proposed in this embodiment and the method for multi-objective optimization of photovoltaic cluster control parameters proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0153] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-objective optimization method for photovoltaic cluster control parameters, characterized in that, include: Acquire system modeling data, establish an electromagnetic transient simulation model of the photovoltaic cluster, and identify the set of control parameters to be optimized and the range of physical constraints; Based on the electromagnetic transient simulation model and the control parameter set, a fault scenario set covering multiple fault types and operating conditions is designed, parameter samples are generated, and the transient stability margin index corresponding to each sample is obtained through electromagnetic transient simulation, and a transient stability domain proxy model is constructed. Using the transient stability domain proxy model, a global sensitivity analysis is performed on the set of control parameters to filter out a subset of highly sensitive parameters; Based on the aforementioned subset of highly sensitive parameters, a multi-objective optimization problem with four objectives is constructed, and constraints are introduced. The multi-objective optimization problem is solved using a multi-objective optimization algorithm, and the Pareto optimal solution set is output. For the Pareto optimal solution set, the comprehensive closeness of each solution is calculated using the approximation ideal solution sorting algorithm. Then, parameter perturbation is applied to each candidate solution to calculate the robustness index. The solution with high comprehensive closeness and the smallest robustness index is selected as the final recommended parameter configuration.

2. The multi-objective optimization method for photovoltaic cluster control parameters as described in claim 1, characterized in that: The transient stable domain proxy model is constructed in the following way: Generate parameter samples based on the set of control parameters; The parameter samples are subjected to electromagnetic transient simulation in the fault scenario set to obtain the corresponding transient stability margin index, forming a parameter-transient stability margin index training sample pair. Using the training sample pairs, a regression model with a radial basis function neural network architecture is trained to construct the transient stable domain proxy model.

3. The multi-objective optimization method for photovoltaic cluster control parameters as described in claim 1 or 2, characterized in that: The global sensitivity analysis includes: The Sobol algorithm is used to calculate the overall sensitivity index of each control parameter; Set a sensitivity threshold; Parameters whose total sensitivity index is greater than the threshold are included in the subset of highly sensitive parameters.

4. The multi-objective optimization method for photovoltaic cluster control parameters as described in claim 3, characterized in that: The method of solving the problem using a multi-objective optimization algorithm includes: When initializing the population, the sampling density of the highly sensitive parameter subset is higher than that of the other control parameters; During the optimization process, the constraint threshold is dynamically set according to the current iteration number, and ε-constraints are constructed based on the constraint threshold to handle transient stability constraints; In the initial stage of several generations of optimization, the fitness of individuals is evaluated using a transient stable domain proxy model. After reaching a pre-set number of switching generations, electromagnetic transient simulation is used to evaluate the fitness of individuals. During the optimization process, every pre-set retraining interval algebra, newly generated electromagnetic transient simulation data is added to the training dataset of the surrogate model, and the transient stable domain surrogate model is retrained based on the updated training dataset.

5. The multi-objective optimization method for photovoltaic cluster control parameters as described in claim 4, characterized in that: The algorithm for ranking solutions to approximate the ideal solution is used to calculate the overall similarity of each solution, including: Decision preference weights were assigned to the four optimization objectives, with transient stability margin having the highest weight and parameter dispersion having the lowest weight. Positive ideal solutions are constructed based on the optimal values ​​of each objective in the Pareto optimal solution set, and negative ideal solutions are constructed based on the worst values ​​of each objective. Calculate the weighted Euclidean distance from each Pareto solution to the positive and negative ideal solutions; The relative proximity of each solution is calculated based on the distance, and the Pareto solutions are sorted according to the relative proximity.

6. The multi-objective optimization method for photovoltaic cluster control parameters as described in claim 5, characterized in that: The calculation of robustness indices includes: Apply random perturbations within a preset range to the control parameters of each candidate solution, and repeat the simulation multiple times to simulate parameter uncertainty; The fluctuation of multi-objective function values ​​under statistical perturbation conditions is evaluated, with the ratio of the standard deviation to the mean of the function values ​​used as a robustness indicator. Among the candidate solutions with the highest overall similarity ranking, the solution with the smallest robustness index is selected as the final recommended parameter configuration.

7. The multi-objective optimization method for photovoltaic cluster control parameters as described in claim 6, characterized in that: The multi-objective optimization problem is constructed in the following way: Based on the aforementioned subset of highly sensitive parameters, four optimization objectives are defined, including: maximizing the minimum transient stability margin index, minimizing the dynamic response overshoot, maximizing the grid support capability, and minimizing the parameter dispersion within the cluster. The multi-objective optimization problem introduces constraints, including transient stability hard constraints, safe operation constraints that satisfy system voltage and frequency, and differentiated coordination constraints.

8. A multi-objective optimization system for photovoltaic cluster control parameters, using the method described in any one of claims 1-7, characterized in that, include: The system modeling and parameter identification module is used to acquire system modeling data, establish an electromagnetic transient simulation model of the photovoltaic cluster, and identify the set of control parameters to be optimized and the range of physical constraints. The transient stability domain proxy modeling module is used to design a set of fault scenarios covering multiple fault types and operating conditions based on the electromagnetic transient simulation model and the set of control parameters, generate parameter samples, obtain the transient stability margin index corresponding to each sample through electromagnetic transient simulation, and construct a transient stability domain proxy model. The control parameter sensitivity analysis module is used to perform global sensitivity analysis on the control parameter set using the transient stability domain proxy model, and to filter out a subset of highly sensitive parameters. A multi-objective optimization problem construction module is used to construct a multi-objective optimization problem with four objectives based on the subset of highly sensitive parameters, and to introduce constraints. The Pareto optimal solution module is used to solve the multi-objective optimization problem using a multi-objective optimization algorithm and output a Pareto optimal solution set. The robust decision-making and parameter recommendation module is used to calculate the comprehensive closeness of each solution to the Pareto optimal solution set using an algorithm that approximates the ideal solution ranking algorithm, apply parameter perturbation to each candidate solution, calculate the robustness index, and select the solution with high comprehensive closeness and the smallest robustness index as the final recommended parameter configuration.

9. An electronic device, characterized in that, include: Memory, used to store programs; A processor for loading the program to perform the steps of the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images