Wind power plant voltage and frequency collaborative tuning control method

Abstract

The invention discloses a wind power plant voltage and frequency cooperative tuning control method, and relates to the technical field of tuning control, and the method comprises the steps: obtaining the real-time operation data of a wind power plant, carrying out the preprocessing of the real-time operation data of the wind power plant, carrying out the voltage fluctuation and frequency disturbance prediction through a cellular automaton model in combination with the pre-obtained power grid data, and carrying out the voltage fluctuation and frequency disturbance prediction. Obtaining a prediction state matrix; carrying out multi-objective optimization solution in combination with a pre-constructed multi-objective optimization problem and constraint conditions, and obtaining an optimal droop coefficient vector in combination with a preset rule; and the optimal droop coefficient vector is used as a control reference, and voltage and frequency cooperative tuning is carried out in combination with grid-connected point voltage and frequency obtained in real time and communication network topology data. According to the method, prediction is carried out through the cellular automaton model, the rapid and cooperative regulation capability of the voltage and the frequency of the wind power plant is improved, and therefore the urgent demand of a power system for rapid, accurate and cooperative control under the access of high-proportion new energy is better met.

Description

Technical Field

[0001] This invention relates to the field of tuning control technology, and more specifically, to a tuning control method for coordinated voltage and frequency in wind farms. Background Technology

[0002] As wind power penetration in power systems continues to increase, its impact on grid voltage and frequency stability is becoming increasingly significant. Traditionally, voltage and frequency control are often designed and optimized independently. However, in grids with a high proportion of renewable energy integration, this "island-like" control strategy may lead to key defects such as response lag, poor coordination, and insufficient adaptability. Moreover, traditional control strategies often rely on fixed droop coefficients or simple adjustment mechanisms based on local feedback, lacking the ability to predict future wind farm conditions. This makes it difficult to cope with rapid voltage and frequency disturbances caused by drastic fluctuations in wind power output, resulting in delayed control actions and a high risk of chain reactions. At the same time, most methods only optimize for a single objective (such as voltage stability or frequency regulation), lacking consideration for multi-objective coordinated control. This makes it difficult to effectively balance the contradictions between voltage deviation, frequency fluctuation, and unit output allocation, easily leading to control conflicts or resource waste.

[0003] Furthermore, existing optimization algorithms, such as traditional genetic algorithms or particle swarm optimization algorithms, often suffer from slow convergence speed, susceptibility to local optima, and insufficient population diversity when dealing with high-dimensional, nonlinear, and multi-constrained wind farm collaborative control problems, making it difficult to guarantee the global optimality and robustness of the solution. More importantly, most methods do not fully consider the dynamic coupling relationship between units within the wind farm and the impact of communication network topology on the transmission of control commands. The lack of global coordination in the control process leads to uneven regulation effects and limited overall performance. Therefore, existing optimization control methods often exhibit poor adaptability, low control accuracy, and insufficient stability when facing complex and changing operating conditions, making it difficult to meet the urgent needs of power systems with high proportions of renewable energy integration for fast, accurate, and collaborative control.

[0004] Therefore, designing a voltage and frequency coordinated tuning and control method is crucial for improving the grid connection performance of wind farms and supporting the stable operation of the power grid.

[0005] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0006] To address the problems in related technologies, this invention proposes an optimized control method for coordinated voltage and frequency control of wind farms, in order to overcome the aforementioned technical problems existing in the prior art.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows: A method for coordinated voltage and frequency tuning control of a wind farm, comprising the following steps: S1. Obtain real-time operation data of the wind farm, preprocess the real-time operation data of the wind farm, and use a cellular automata model to predict voltage fluctuations and frequency disturbances based on the preprocessed real-time operation data of the wind farm and the pre-acquired power grid data to obtain the prediction state matrix. S2. Solve the multi-objective optimization problem and constraints based on the predicted state matrix and pre-constructed multi-objective optimization problem, and obtain the optimal droop coefficient vector based on the preset rules. S3. Use the optimal droop coefficient vector as the control reference, and combine it with the real-time acquired grid connection point voltage, frequency and communication network topology data to perform coordinated voltage and frequency tuning.

[0008] Furthermore, in order to capture the complex dynamic interactions within the wind farm and with the external power grid, and to simulate the propagation of voltage and frequency among different turbines under the generation of a predictive state matrix, avoiding potential voltage collapse or frequency instability events, real-time wind farm operation data is acquired. This data is preprocessed, and based on the preprocessed real-time wind farm operation data combined with pre-acquired power grid data, a cellular automata model is used to predict voltage fluctuations and frequency disturbances, resulting in a predictive state matrix including: S11. Obtain real-time operation data of the wind farm, and perform filtering and noise reduction processing on the real-time operation data of the wind farm to obtain the filtered and noise-reduced real-time operation data of the wind farm. S12. Perform time alignment processing on the filtered and denoised real-time wind farm operation data to obtain pre-processed real-time wind farm operation data. S13. The cellular automata model is initialized based on the preprocessed real-time operation data of the wind farm. The initialized cellular automata model is combined with the pre-acquired power grid data to predict voltage fluctuations and frequency disturbances, and the predicted state matrix is ​​obtained. The pre-acquired grid data includes the pre-acquired wind farm power network topology and the pre-acquired grid voltage and frequency data.

[0009] Furthermore, the cellular automata model is initialized based on the preprocessed real-time wind farm operation data. The initialized cellular automata model, combined with pre-acquired grid data, is then used to predict voltage fluctuations and frequency disturbances, resulting in a predicted state matrix including: S131. Based on the preprocessed real-time operation data of the wind farm, a spatial grid of the wind farm is constructed using a cellular automata model. The state vector of each grid point is initialized through the constructed spatial grid of the wind farm to obtain the initial grid point state matrix. S132. Based on the initialized grid state matrix and the pre-acquired wind farm electrical network topology, construct the neighborhood relationship between grid points, and use the weighted electrical distance method of the inverse of line impedance to quantify the coupling strength between grid points. Integrate the neighborhood relationship between grid points and the coupling strength to obtain a structured neighborhood relationship matrix. S133. Using the neighborhood relation matrix and the initialized grid state matrix, combined with the pre-acquired grid voltage and frequency data, construct the grid state transition function, and predict voltage fluctuations and frequency disturbances based on the grid state transition function to obtain the predicted state matrix. The pre-acquired grid voltage and frequency data include the pre-acquired grid reactive voltage sensitivity and the pre-acquired grid power frequency fluctuation.

[0010] Furthermore, a grid-based state transition function is constructed using the neighborhood relation matrix and the initialized grid-point state matrix, combined with pre-acquired grid voltage and frequency data. Voltage fluctuations and frequency disturbances are then predicted based on this grid-based state transition function, resulting in a predicted state matrix including: S1331. Based on the neighborhood relation matrix and the initialized grid state matrix, voltage update rules and frequency response rules are constructed by combining the pre-acquired power grid reactive voltage sensitivity and power grid frequency fluctuation degree, and the voltage update rules and frequency response rules are integrated to obtain a local evolution rule set. S1332. Obtain the mapping relationship between the grid state and time based on the local evolution rule set, and construct the state transition function through the mapping relationship; S1333. Use the state transition function to perform iterative simulation prediction of voltage fluctuation and frequency disturbance, obtain the prediction deviation of voltage and frequency, and generate the prediction state matrix based on the prediction deviation of voltage and frequency.

[0011] Furthermore, the expression for the local evolution rule set is: ; In the formula, Δ V i Indicates the first i The change in the terminal voltage of the typhoon generator unit; X ij Indicates the connection unit i and j The line reactance; N ( i ) indicates the generator set i The neighborhood set; j Indicates the relationship with the first i Adjacent typhoon generator units with direct electrical connection j ; Q j Indicates the first j The reactive power output of the typhoon generator set; Qi Indicates the first i The reactive power output of the typhoon generator set; Z ij Indicates the connection unit i and j Total line impedance; H eff This represents the equivalent inertia of a wind farm. P j Indicates the first j The active power output of the typhoon generator; Δ f i Indicates the first i The change in the frequency of the typhoon generator grid connection point; P i ref Indicates the first i Active power setting value of typhoon generator unit.

[0012] Furthermore, in order to formulate more flexible and efficient control strategies, and to achieve precise control by dynamically adjusting control parameters according to different operating scenarios under the multi-objective optimization solution of the differential evolution genetic fusion algorithm, the optimal droop coefficient vector is obtained by combining the predicted state matrix with the pre-constructed multi-objective optimization problem and constraints, and by combining preset rules. S21. Based on the predicted state matrix and the pre-constructed multi-objective optimization problem, and combined with the pre-constructed constraints, the optimization problem model is generated. S22. Based on the optimization problem model, the hyperparameters of the differential evolution genetic fusion algorithm are initialized, and the Pareto front is approximated by the hyperparameters initialized by non-dominated sorting to obtain the Pareto optimal solution set. S23. The optimal droop coefficient vector is obtained by performing comprehensive performance optimization screening on the Pareto optimal solution set using preset rules.

[0013] Furthermore, based on the optimization problem model, the differential evolution genetic fusion algorithm is initialized with hyperparameters, and the Pareto front is approximated using non-dominated sorting to guide the initialization of hyperparameters, resulting in the Pareto optimal solution set, including: S221. Extract the search space features of the optimization problem based on the optimization problem model, and initialize the hyperparameters of the differential evolution genetic fusion algorithm based on the search space features; S222. Construct a feasible solution generation mechanism based on the constraints of the optimization problem model, and use the feasible solution generation mechanism in combination with the initialized hyperparameters to generate the hyperparameter running boundary. S223. Using differential evolution and genetic algorithms, combined with the initialized hyperparameters and the hyperparameter running boundary, a population generation operation is performed to obtain an evolutionary population. Then, Pareto optimal solutions are identified from the evolutionary population through non-dominated sorting to obtain a Pareto optimal solution set.

[0014] Furthermore, differential evolution and genetic algorithms are combined with initialized hyperparameters and their operational boundaries to generate a population, resulting in an evolutionary population. Pareto optimal solutions are then identified from this population using non-dominated sorting, yielding a set of Pareto optimal solutions including: S2231. The mutation and crossover selection mechanism of differential evolution is used in combination with the initialized hyperparameters and the hyperparameter running boundary to perform a global search operation on the population to obtain the first generation population. S2232. Based on the selection, crossover and mutation mechanism of the genetic algorithm, the first generation population is screened to obtain the offspring population, and the first generation population is preserved using the elite preservation strategy to obtain the preserved population. The offspring population and the preserved population are integrated to obtain the evolved population. S2233. Based on the Pareto dominance relation, non-dominated ordination is used to stratify individuals in the evolutionary population, and the distribution degree of the ordination results is evaluated by crowding distance. The distribution degree evaluation results guide the population to approach the Pareto front and obtain the candidate solution set. S2234. Remove duplicate individuals from the candidate solution set to obtain the removed candidate solution set. Then, identify the Pareto optimal solution from the removed candidate solution set through cluster analysis to obtain the Pareto optimal solution set.

[0015] Furthermore, to maintain good control performance and enable rapid response and timely adjustment of control strategies in dynamically changing environments under the guidance of the pigeon flock biomimetic navigation algorithm, thereby maintaining the stable operation of the wind farm, the optimal droop coefficient vector is used as the control reference. Combined with real-time acquired grid connection point voltage, frequency, and communication network topology data, voltage and frequency are coordinated for optimization, including: S31. Use the optimal droop coefficient vector as the control reference to set the initial droop control parameters, and obtain the grid connection point voltage, frequency and communication network topology data in real time. S32. Based on the real-time acquired grid connection point voltage, frequency, and communication network topology data, as well as the initial droop control parameters, the pigeon flock bionic navigation algorithm is used to correct the voltage and frequency deviations, thereby obtaining optimized droop coefficient adjustment data. S33. Based on the optimized droop coefficient adjustment data, generate wind turbine control commands, and realize the coordinated optimization of wind farm voltage and frequency through control commands.

[0016] Furthermore, based on real-time acquired grid connection point voltage, frequency, and communication network topology data, as well as initial droop control parameters, a pigeon flock biomimetic navigation algorithm is used to correct the voltage and frequency deviations, resulting in optimized droop coefficient adjustment data, including: S321. The pigeon flock bionic navigation algorithm is initialized based on the real-time acquired grid connection point voltage, frequency and communication network topology data and the initial droop control parameters to obtain the initialized pigeon flock bionic navigation algorithm. S322. Calculate the weighted sum of squares of voltage and frequency deviations using the initialized pigeon flock biomimetic navigation algorithm, and evaluate fitness based on the weighted sum of squares. Adjust the information interaction intensity based on the fitness evaluation results and the communication network topology data. S323. Based on the adjusted information interaction intensity, the pigeon flock navigation behavior is simulated through position and velocity update rules, and the optimal deviation correction direction is iteratively searched to obtain optimized droop coefficient adjustment data.

[0017] The beneficial effects of this invention are as follows: 1. This invention uses a cellular automata model for prediction, effectively simulating the nonlinear coupling and spatiotemporal evolution characteristics between various wind turbines in a wind farm. It is particularly suitable for complex and variable operating conditions such as strong randomness in wind power output and uneven spatiotemporal distribution, improving the rapid and coordinated adjustment capability of wind farm voltage and frequency, and significantly enhancing the stability of voltage and frequency. Simultaneously, it can solve for the optimal droop coefficient vector under constraints, further realizing adaptive tuning of control parameters, overcoming the shortcomings of uneven response and blind spots in traditional fixed droop control, and enhancing the control accuracy and adaptability of the wind farm. Finally, combined with the communication topology, it dynamically implements coordinated voltage and frequency tuning, balancing global coordination and local response speed, effectively compensating for the shortcomings of traditional control strategies such as slow response, coarse adjustment, and lack of coordination, better meeting the urgent needs of power systems for rapid, accurate, and coordinated control under high-proportion renewable energy access.

[0018] 2. This invention uses a cellular automata model to capture the complex dynamic interactions within a wind farm and with the external power grid. By simulating the propagation of voltage and frequency between different turbines, it generates a predictive state matrix, thus avoiding potential voltage collapse or frequency instability events. Furthermore, the flexibility and scalability of the cellular automata model make it applicable to wind farms of different sizes and topologies, and it has broad application prospects.

[0019] 3. This invention employs a differential evolutionary genetic fusion algorithm for multi-objective optimization, comprehensively considering multiple conflicting objective functions and introducing strict constraints to ensure the feasibility and security of the solution. It combines the global search capability of the differential evolutionary algorithm with the local exploitation accuracy of the genetic algorithm, significantly improving the convergence speed and solution quality. In particular, the elite retention strategy ensures the continuation of the best individuals in each generation, preventing the loss of high-quality solutions. Furthermore, it can dynamically adjust control parameters according to different operating scenarios, achieving precise regulation and providing decision-makers with more choices, thus facilitating the development of more flexible and efficient control strategies.

[0020] 4. This invention utilizes a pigeon flock biomimetic navigation algorithm, which enables rapid response and timely adjustment of control strategies in dynamically changing environments, thereby maintaining the stable operation of the wind farm. Furthermore, the robustness and efficiency of the pigeon flock biomimetic navigation algorithm provide strong protection for the wind farm, maintaining good control performance even under complex operating conditions. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a wind farm voltage and frequency coordinated optimization control method according to an embodiment of the present invention. Detailed Implementation

[0023] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0024] According to an embodiment of the present invention, a method for coordinated tuning and control of voltage and frequency in wind farms is provided.

[0025] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the wind farm voltage and frequency coordinated optimization control method according to an embodiment of the present invention includes the following steps: S1. Obtain real-time operation data of the wind farm, preprocess the real-time operation data of the wind farm, and use a cellular automata model to predict voltage fluctuations and frequency disturbances based on the preprocessed real-time operation data of the wind farm and the pre-acquired power grid data to obtain the predicted state matrix.

[0026] Specifically, real-time operation data of wind farms is acquired, preprocessed, and then, based on the preprocessed wind farm operation data and pre-acquired grid data, a cellular automata model is used to predict voltage fluctuations and frequency disturbances, resulting in a predicted state matrix including: S11. Obtain real-time operation data of the wind farm, and perform filtering and noise reduction processing on the real-time operation data of the wind farm to obtain the filtered and noise-reduced real-time operation data of the wind farm. S12. Perform time alignment processing on the filtered and denoised real-time wind farm operation data to obtain pre-processed real-time wind farm operation data. S13. The cellular automata model is initialized based on the preprocessed real-time operation data of the wind farm. The initialized cellular automata model is combined with the pre-acquired power grid data to predict voltage fluctuations and frequency disturbances, and the predicted state matrix is ​​obtained. The pre-acquired grid data includes the pre-acquired wind farm power network topology and pre-acquired grid voltage and frequency data; it can simulate the propagation process of voltage and frequency between different units under the action of generating a predictive state matrix, and avoid potential voltage collapse or frequency instability events.

[0027] Specifically, the cellular automata model is initialized based on preprocessed real-time wind farm operation data. This initialized cellular automata model, combined with pre-acquired grid data, is then used to predict voltage fluctuations and frequency disturbances, resulting in a predicted state matrix including: S131. Based on the preprocessed real-time operation data of the wind farm, a spatial grid of the wind farm is constructed using a cellular automata model. The state vector of each grid point is initialized through the constructed spatial grid of the wind farm to obtain the initial grid point state matrix. S132. Based on the initialized grid state matrix and the pre-acquired wind farm electrical network topology, construct the neighborhood relationship between grid points, and use the weighted electrical distance method of the inverse of line impedance to quantify the coupling strength between grid points. Integrate the neighborhood relationship between grid points and the coupling strength to obtain a structured neighborhood relationship matrix. S133. Using the neighborhood relation matrix and the initialized grid state matrix, combined with the pre-acquired grid voltage and frequency data, construct the grid state transition function, and predict voltage fluctuations and frequency disturbances based on the grid state transition function to obtain the predicted state matrix. The pre-acquired grid voltage and frequency data include the pre-acquired grid reactive voltage sensitivity and the pre-acquired grid power frequency fluctuation.

[0028] Specifically, a grid-based state transition function is constructed using a neighborhood relation matrix and an initialized grid-based state matrix, combined with pre-acquired grid voltage and frequency data. Voltage fluctuations and frequency disturbances are then predicted based on this grid-based state transition function, resulting in a predicted state matrix including: S1331. Based on the neighborhood relation matrix and the initialized grid state matrix, voltage update rules and frequency response rules are constructed by combining the pre-acquired power grid reactive voltage sensitivity and power grid frequency fluctuation degree, and the voltage update rules and frequency response rules are integrated to obtain a local evolution rule set. S1332. Obtain the mapping relationship between the grid state and time based on the local evolution rule set, and construct the state transition function through the mapping relationship; S1333. Use the state transition function to perform iterative simulation prediction of voltage fluctuation and frequency disturbance, obtain the prediction deviation of voltage and frequency, and generate the prediction state matrix based on the prediction deviation of voltage and frequency.

[0029] Specifically, the expression for the local evolution rule set is: ; In the formula, Δ V i Indicates the first i The change in the terminal voltage of the typhoon generator unit; X ij Indicates the connection unit i and j The line reactance; N ( i ) indicates the generator set i The neighborhood set; j Indicates the relationship with the first i Adjacent typhoon generator units with direct electrical connection j ; Q j Indicates the first j The reactive power output of the typhoon generator set; Q i Indicates the first i The reactive power output of the typhoon generator set; Z ij Indicates the connection unit i and j Total line impedance; H eff This represents the equivalent inertia of a wind farm. P j Indicates the first j The active power output of the typhoon generator; Δ f i Indicates the first iThe change in the frequency of the typhoon generator grid connection point; P i ref Indicates the first i Active power setting value of typhoon generator unit.

[0030] Specifically, key operating parameters of each wind turbine in the wind farm are acquired in real time through wind farm monitoring and data acquisition, and synchronous phasor measurement. These parameters include active power, reactive power, terminal voltage, grid connection frequency, wind speed, power factor, and other multi-dimensional data, forming the original real-time wind farm operating data. To remove noise, outliers, or time misalignments caused by communication delays in the original real-time wind farm operating data, preprocessing is required to ensure the accuracy of subsequent modeling. Digital filtering technology can be used to denoise the acquired real-time wind farm operating data, effectively filtering out high-frequency interference and measurement errors while retaining valid signals reflecting the actual physical processes, thus obtaining filtered and denoised real-time wind farm operating data. The denoised real-time wind farm operating data is then time-aligned using interpolation to unify data from different wind turbines and sampling frequencies onto the same time base, ensuring consistency and comparability of all data in the time dimension. This ultimately forms high-quality preprocessed real-time wind farm operating data, providing reliable input for subsequent modeling. The Cellular Automaton (CA) model is initialized based on preprocessed real-time wind farm operation data. This CA model abstracts the wind farm as a discrete spatial network composed of multiple grid points, with each grid point corresponding to a wind turbine. The state vector of each grid point is initialized based on the preprocessed real-time wind farm operation data. The state vector typically contains variables such as voltage, frequency, and active or reactive power at the current moment, forming the initial grid point state matrix. Combined with the pre-acquired wind farm electrical network topology (which includes line connections, transformer configurations, etc.), the neighborhood relationships between grid points are constructed, i.e., the direct electrical neighbors of each turbine are determined. N ( iTo further quantify the dynamic coupling strength between units, a weighted electrical distance method based on the reciprocal of line impedance is introduced, where a smaller line impedance indicates a tighter electrical connection and a stronger coupling strength. Neighborhood relationships and coupling weights are integrated to form a structured neighborhood relationship matrix, used to describe the interaction structure within the wind farm. A grid-based state transition function is constructed to achieve dynamic prediction; this function is established based on a combination of physical mechanisms and data-driven approaches. Voltage update rules and frequency response rules are constructed by combining pre-acquired grid reactive voltage sensitivity and power frequency fluctuation (specifically, inertial response characteristics). These two rules are integrated to form a local evolution rule set, serving as the core mechanism for state transition. Based on this rule set, a mapping relationship between grid state and time evolution is established, i.e., the grid-based state transition function, describing how the current state affects the state at the next moment. Finally, this state transition function is used to perform multi-step iterative simulations of the wind farm, predicting the voltage and frequency deviations of each unit at future moments, and all predicted deviations are summarized to form a predicted state matrix. This matrix not only captures the dynamic propagation process between units within the wind farm, but also integrates the voltage and frequency fluctuation trends of the external power grid, providing a forward-looking decision-making basis for subsequent optimization control, effectively avoiding risks such as voltage collapse or frequency instability, and realizing refined modeling and prediction of the dynamic behavior of complex power systems.

[0031] Specifically, in practical applications, wind farm monitoring and data acquisition (SCADA) and synchronous phasor measurement (PMU) equipment collect raw data such as active power (kW), reactive power (kvar), terminal voltage (kV), grid connection frequency (Hz), and wind speed (m / s) of each wind turbine in real time with a sampling period of 100ms. For example, a 200MW wind farm contains 50 4MW turbines, and each turbine uploads a set of measured values ​​containing the above parameters every 100ms, forming a raw data stream. Due to the complex electromagnetic environment on site, some data may contain glitches and jumps, such as a turbine voltage experiencing instantaneous fluctuations of ±0.05pu around 1.02pu, or reactive power values ​​being missing due to communication interruptions. To address this, wavelet denoising (db4 wavelet basis, 3-level decomposition) was employed to filter voltage and frequency signals, effectively suppressing high-frequency noise. Moving average filtering (window length of 5 points, i.e., 500ms) was used for active and reactive power sequences to smooth power fluctuations. Linear interpolation was used to complete missing data, ensuring data integrity and ultimately obtaining smooth, continuous filtered and denoised operating data. Since the data upload cycle for some older units is 500ms, while that for newer units is 100ms, time alignment was necessary. Using 100ms as a unified time reference, cubic spline interpolation was used before and after the 500ms sampled unit data to generate estimates for the four intermediate time points, ensuring that all 50 units had synchronized observations every 100ms, forming a time-aligned preprocessed dataset. This dataset was used for initializing the cellular automata model. The geographical coordinates or electrical connections of the wind farm are mapped to a 5×10 two-dimensional grid, with each grid point corresponding to one turbine. Its state vector is initialized as a vector composed of the measured active power, reactive power, turbine terminal voltage, and grid connection frequency at the current moment, forming a 50×4 initial grid point state matrix. The electrical neighbors (busbar connections) of each turbine are determined according to the wind farm's primary wiring diagram, constructing a neighborhood set. Furthermore, based on line parameters (e.g., a 2km long 35kV line with a unit impedance of 0.17+), the process is repeated. j ×0.35Ω / km, then Z ij =0.34+ j (×0.7Ω), calculate coupling weight w ij =1 / | Z ij | ≈ 1.28, the smaller the impedance, the greater the weight, forming a 50×50 structured neighborhood relation matrix. This is combined with the regional power grid reactive power and voltage sensitivity (e.g., 0.08 pu / Mvar) and equivalent inertia provided by dispatching. H eff=4.5s, construct local evolution rules, where Δt=0.1s, and integrate them into a set of state update rules. Establish the state transition function formula. S i ( t +1)= S i ( t )+Δ S i The state transition function is the core function in cellular automata models used to predict the future states of each wind turbine in a wind farm. This formula describes the state transition function of the first wind turbine. i How to determine the status of the typhoon generator units from the current moment? t Evolving to the next moment t +1. Among them, S i ( t ) indicates the first i At the current moment, the typhoon turbine unit t The state vector. This vector typically contains multiple physical quantities reflecting the unit's operating state, such as terminal voltage ( V i ( t )), Grid connection frequency ( f i ( t )), active power ( P i ( t ()) and reactive power ( Q i ( t The state vector can be defined as follows: S i ( t )=[ V i ( t ), f i ( t ), P i ( t ), Q i ( t )] T . S i ( t +1) indicates the first i The typhoon turbine unit at the next moment t The predicted state vector of +1 has the same structure as S i ( t The value is the same. This value is calculated by adding a state change variable to the current state, reflecting the continuous evolution of the wind farm's state. ΔS i Indicates the first i The state change vector of a typhoon generator unit within a unit time step. It determines the increment of the state vector during time progression and is the core driving force for state transition. This state transition function quantifies the physical mechanisms (the relationship between voltage and reactive power, and frequency and active power) into mathematical expressions, and, combined with real-time operating data and grid parameters, achieves dynamic simulation of the spatiotemporal propagation process of voltage and frequency within the wind farm, providing a forward-looking predictive basis for subsequent coordinated optimization control. Finally, a 5-step iteration (predicting the next 500ms) is performed to output the state change vector of each unit within a unit time step. t The predicted voltage and frequency deviations at time +1 are summarized into a 50×2 predicted state matrix, which is used for subsequent optimized control to achieve accurate prediction based on real data.

[0032] S2. Based on the predicted state matrix, multi-objective optimization is performed in combination with the pre-constructed multi-objective optimization problem and constraints, and the optimal droop coefficient vector is obtained in combination with preset rules.

[0033] Specifically, based on the predicted state matrix and a pre-constructed multi-objective optimization problem and constraints, multi-objective optimization is performed, and the optimal droop coefficient vector is obtained by combining preset rules, including: S21. Based on the predicted state matrix and the pre-constructed multi-objective optimization problem, and combined with the pre-constructed constraints, the optimization problem model is generated. S22. Based on the optimization problem model, the hyperparameters of the differential evolution genetic fusion algorithm are initialized, and the Pareto front is approximated by the hyperparameters initialized by non-dominated sorting to obtain the Pareto optimal solution set.

[0034] Specifically, based on the optimization problem model, the hyperparameters of the differential evolution genetic fusion algorithm are initialized, and the Pareto front is approximated using non-dominated sorting to guide the initialization of the hyperparameters, resulting in the Pareto optimal solution set, including: S221. Extract the search space features of the optimization problem based on the optimization problem model, and initialize the hyperparameters of the differential evolution genetic fusion algorithm based on the search space features; S222. Construct a feasible solution generation mechanism based on the constraints of the optimization problem model, and use the feasible solution generation mechanism in combination with the initialized hyperparameters to generate the hyperparameter running boundary. S223. Using differential evolution and genetic algorithms, combined with the initialized hyperparameters and the hyperparameter running boundary, a population generation operation is performed to obtain an evolutionary population. Then, Pareto optimal solutions are identified from the evolutionary population through non-dominated sorting to obtain a Pareto optimal solution set.

[0035] Specifically, differential evolution and genetic algorithms are used in combination with initialized hyperparameters and their operational boundaries to generate a population, resulting in an evolutionary population. Pareto optimal solutions are then identified from this population using non-dominated sorting. The resulting Pareto optimal solution set includes: S2231. The mutation and crossover selection mechanism of differential evolution is used in combination with the initialized hyperparameters and the hyperparameter running boundary to perform a global search operation on the population to obtain the first generation population. S2232. Based on the selection, crossover and mutation mechanism of the genetic algorithm, the first generation population is screened to obtain the offspring population, and the first generation population is preserved using the elite preservation strategy to obtain the preserved population. The offspring population and the preserved population are integrated to obtain the evolved population. S2233. Based on the Pareto dominance relation, non-dominated ordination is used to stratify individuals in the evolutionary population, and the distribution degree of the ordination results is evaluated by crowding distance. The distribution degree evaluation results guide the population to approach the Pareto front and obtain the candidate solution set. S2234. Remove duplicate individuals from the candidate solution set to obtain the removed candidate solution set. Then, identify the Pareto optimal solution from the removed candidate solution set through cluster analysis to obtain the Pareto optimal solution set.

[0036] S23. By using preset rules to perform comprehensive performance optimization screening of the Pareto optimal solution set and the droop coefficient vector, the optimal droop coefficient vector is obtained. Under the action of the differential evolution genetic fusion algorithm for multi-objective optimization, the control parameters can be dynamically adjusted according to different operating scenarios, thereby achieving precise control.

[0037] Specifically, the output predicted state matrix, containing predicted voltage and frequency deviations for each wind turbine at multiple future time points, serves as a key input for evaluating control performance. Based on this, a pre-built multi-objective optimization framework is used. This framework typically includes two conflicting objective functions: minimizing the sum of squared voltage deviations across the entire wind farm to improve voltage stability; and minimizing the sum of absolute frequency deviations to enhance frequency regulation capability. Furthermore, a third objective can be introduced, such as minimizing the imbalance in active power output among turbines to extend equipment lifespan. Simultaneously, a pre-built set of constraints is used, including upper and lower bound constraints on variables (such as active power and frequency droop coefficients). k pi Between 0.005 pu / MW and 0.02 pu / MW, reactive power and voltage droop factor k qiA complete constrained optimization problem is formed, encompassing factors such as a reactive power output between 0.01 pu / Mvar and 0.05 pu / Mvar, a maximum reactive power output of the converter less than or equal to 1.2 Mvar, a response time constraint of less than or equal to 200 ms due to communication delay, wind farm constraints (e.g., total reactive power not exceeding converter capacity), and dynamic response time constraints based on communication delay. The predicted state matrix is ​​used as input data for the objective function. Combining the aforementioned objective function and constraints, the final optimization problem model is generated. This model simultaneously optimizes multiple objective functions while satisfying all equality and inequality constraints. Based on this model, search space features are extracted, including the dimension of decision variables (i.e., the total number of droop coefficients, e.g., 100-dimensional variables for 50 units), the upper and lower bounds of variables, and the constraint type (linear or nonlinear). These features are then used to initialize the hyperparameters of the differential evolutionary genetic fusion algorithm, such as setting the population size. NP =100, crossover probability CR =0.8, variation factor FThe algorithm is set to 0.5, the selection strategy is binary tournament, the crossover operator is SBX, the mutation operator is polynomial mutation, and the maximum number of generations is set to 200. Both the crossover and mutation distribution indices are 20. A feasible solution generation mechanism is constructed based on the constraints in the optimization problem model to ensure that both the initial population and subsequently generated individuals satisfy all boundary and constraint conditions. Specifically, either a "repair method" or a "projection method" is used to handle out-of-bounds variables. For complex constraints, a penalty function mechanism is introduced, adding a penalty term for constraint violations to the objective function. Using this mechanism, combined with the upper and lower bounds of variables, the actual operating boundary of hyperparameters is determined. For example, the search range of the droop coefficient is strictly limited to the upper and lower bounds of the variables to ensure that the algorithm searches within the legal space. The "mutation-crossover-selection" mechanism of differential evolution (DE) is used to perform a global search within the hyperparameter operating boundary. First, 100 individuals are randomly initialized, each representing a set of droop coefficient vectors. Then, in each generation, a mutation operation is performed on each target individual to generate a difference vector. Next, a crossover operation is performed to mix the parent generation and the mutated vector genes according to probability. Finally, through a selection operation, non-dominated solutions are retained, and the generation update is completed, ultimately obtaining an initial population with good global exploration capabilities. Further optimization is achieved by introducing a genetic algorithm (GA): GA-based selection operations (such as binary tournaments) select high-fitness individuals from the initial population as parents; crossover operations (such as SBX crossover) generate the offspring population; local perturbation is enhanced through polynomial mutation to obtain a new offspring population; simultaneously, an elitism strategy is applied, directly replicating several optimal individuals with high crowding in the first non-dominated ordination layer of the current generation to the next generation, forming a reserve population; finally, the offspring population and the reserve population are merged to form a new generation of evolutionary population, ensuring the inheritance of high-quality genes. Pareto optimal solutions are identified for the evolutionary population: NSGA-II non-dominated ordination is used to stratify all individuals in the population according to their dominance relationship, with the first layer being Pareto front candidates; simultaneously, the crowding distance of each individual is calculated to assess its sparse distribution in the target space; the non-dominated level and crowding jointly guide the selection operation, prioritizing the retention of individuals with high levels and sparse distribution, guiding the population towards the Pareto front while maintaining solution set diversity, forming a candidate solution set after multiple generations of iteration. Post-processing of the candidate solution set: First, duplicate or highly similar individuals are removed (based on Euclidean distance threshold) to obtain a deduplicated candidate solution set; then, cluster analysis is used to divide the solution set into several clusters, which are divided into 5 clusters in this embodiment, each cluster representing a control strategy mode; the most representative individual (such as the one closest to the cluster center) is selected from each cluster, and finally a set of evenly distributed and highly representative Pareto optimal solutions is output.In practical applications, preference rules based on the operating scenario can be adopted (such as prioritizing frequency stability under high wind speeds) to select the single optimal droop coefficient vector that best matches the current operating conditions from the Pareto front as the final output, which is used to guide the generation of subsequent control commands and realize closed-loop decision-making from multi-objective optimization to precise control.

[0038] S3. Use the optimal droop coefficient vector as the control reference, and combine it with the real-time acquired grid connection point voltage, frequency and communication network topology data to perform coordinated voltage and frequency tuning.

[0039] Specifically, the optimal droop coefficient vector is used as the control benchmark, and voltage and frequency are coordinated and optimized by combining real-time acquired grid connection point voltage, frequency, and communication network topology data. S31. Use the optimal droop coefficient vector as the control reference to set the initial droop control parameters, and obtain the grid connection point voltage, frequency and communication network topology data in real time. S32. Based on the real-time acquired grid connection point voltage, frequency, and communication network topology data, as well as the initial droop control parameters, the pigeon flock bionic navigation algorithm is used to correct the voltage and frequency deviations, thereby obtaining optimized droop coefficient adjustment data.

[0040] Specifically, based on real-time acquired grid connection point voltage, frequency, and communication network topology data, as well as initial droop control parameters, a pigeon flock biomimetic navigation algorithm is used to correct voltage and frequency deviations, resulting in optimized droop coefficient adjustment data, including: S321. The pigeon flock bionic navigation algorithm is initialized based on the real-time acquired grid connection point voltage, frequency and communication network topology data and the initial droop control parameters to obtain the initialized pigeon flock bionic navigation algorithm. S322. Calculate the weighted sum of squares of voltage and frequency deviations using the initialized pigeon flock biomimetic navigation algorithm, and evaluate fitness based on the weighted sum of squares. Adjust the information interaction intensity based on the fitness evaluation results and the communication network topology data. S323. Based on the adjusted information interaction intensity, the pigeon flock navigation behavior is simulated through position and velocity update rules, and the optimal deviation correction direction is iteratively searched to obtain optimized droop coefficient adjustment data.

[0041] S33. Based on the optimized droop coefficient adjustment data, wind turbine control commands are generated, and the voltage and frequency of the wind farm are coordinated and optimized through the control commands. Under the action of the pigeon flock bionic navigation algorithm, it can achieve rapid response and timely adjustment of control strategies in a dynamically changing environment, thereby maintaining the stable operation of the wind farm.

[0042] Specifically, the optimal droop coefficient vector is used as the initial control reference. For example, in a wind farm with 50 wind turbines of 4MW each, the active power droop coefficient of the first turbine is 0.013 pu / MW, and the reactive power droop coefficient is 0.035 pu / Mvar. The remaining turbines are similarly set, forming the initial droop control parameter set. This optimal vector is sent to the converter controllers of each wind turbine as the initial value of the current control law. At the same time, the voltage (kV), frequency (Hz), and communication network topology data (such as the status, delay, and bandwidth of communication links between turbines) at the grid connection points of each turbine are collected in real time through SCADA and PMU. For example, if the communication delay between turbines in a certain area increases from the normal 50ms to 150ms, it indicates that information transmission is obstructed. Based on the aforementioned real-time data and initial droop parameters, the pigeon flock bionic navigation algorithm is initialized. Each "pigeon" is encoded as a droop coefficient adjustment vector, representing the fine-tuning amount of the current droop coefficient. The initial flock size is set to 50, and the position of each pigeon is randomly generated within the range [-0.002, 0.002], representing an adjustment space of ±15% of the original droop coefficient. The velocity vector is initially zero. Simultaneously, an information interaction graph is constructed based on the communication network topology, defining the information transmission weights between pigeons. If the unit... m and nIf communication is normal and latency is low, the information transmission weight is 1. If the link is interrupted or latency is too high, the information transmission weight is 0.3, which is used to adjust the information fusion strength later. Fitness is evaluated using the initialized pigeon flock bionic navigation (PIO) algorithm: for each pigeon, its corresponding adjustment vector is superimposed on the initial droop coefficient to obtain the droop parameter for actual application; for example, for a certain pigeon, the adjusted value is 0.013 + 0.0012 = 0.0142. Then, based on the current grid connection point voltage and frequency measurements, voltage deviation and frequency deviation are calculated, and a fitness function is constructed using a linear weighting method, where voltage deviation has a weight of 0.6 and frequency deviation has a weight of 0.4, reflecting the higher priority of voltage control. If a pigeon's fitness is 0.045, which is better than other individuals, its fitness is higher. Simultaneously, the information interaction strength is adjusted according to the communication topology data: during the location update process, the pigeon's navigation direction is not only influenced by its own optimal memory but also by the guidance of the best individual among its neighbors; the influence weight is determined by the information transmission weight, forming a weighted information fusion mechanism. Based on the adjusted information interaction intensity, the PIO position and velocity update rules are executed for iterative search: position update is divided into two stages, namely map and compass operator and landmark operator. After a certain number of iterations (e.g., 50 generations), the algorithm switches to the landmark operator to simulate the behavior of a pigeon approaching a target, enhancing local search capabilities. Throughout the iteration process, each generation dynamically adjusts the influence range of the optimal neighbor solution according to the communication weight, ensuring that a certain degree of cooperation can still be maintained in communication-restricted areas. After 100 iterations, the algorithm converges and outputs the position vector corresponding to the optimal pigeon, which is the optimized droop coefficient adjustment data. For example, the active power droop coefficient is +0.0011, and the reactive power droop coefficient is -0.0008, indicating that the active power droop gain needs to be slightly increased to improve the frequency response, and the reactive power droop needs to be reduced to alleviate voltage fluctuations. Finally, this adjusted data is superimposed with the original optimal droop coefficient to generate the final wind turbine control command, which is then sent to each turbine converter through the communication network to update its droop control law in real time. The system continues to monitor voltage and frequency response. If the deviation further decreases (e.g., the average voltage deviation decreases from 0.03 pu to 0.01 pu, and the frequency deviation decreases from 0.15 Hz to 0.05 Hz), it indicates that the optimization is effective; otherwise, a new round of PIO optimization can be triggered.

[0043] In summary, by utilizing the above-mentioned technical solution of this invention, the present invention, through prediction using a cellular automata model, can effectively simulate the nonlinear coupling and spatiotemporal evolution characteristics between various wind turbine units in a wind farm. It is particularly suitable for complex and variable operating conditions such as strong randomness in wind power output and uneven spatiotemporal distribution, improving the rapid and coordinated adjustment capability of wind farm voltage and frequency, and significantly enhancing the stability of voltage and frequency. Simultaneously, it can solve for the optimal droop coefficient vector under constraints, further realizing adaptive tuning of control parameters, overcoming the shortcomings of uneven response and blind zone in traditional fixed droop control, and enhancing wind power efficiency. The invention improves the control precision and adaptability of the electric field; finally, by combining the communication topology, it dynamically implements coordinated voltage and frequency tuning, taking into account both global coordination and local response speed. This effectively compensates for the shortcomings of traditional control strategies, such as slow response, coarse adjustment, and lack of coordination, significantly improving the speed, accuracy, and coordination of control. This better meets the urgent needs of the power system for fast, accurate, and coordinated control under a high proportion of renewable energy integration. Through a cellular automata model, this invention can capture the complex dynamic interaction relationships within the wind farm and between it and the external power grid. By simulating the propagation process of voltage and frequency between different units, it generates… This invention uses a predictive state matrix to avoid potential voltage collapse or frequency instability events. Furthermore, the flexibility and scalability of the cellular automata model make it applicable to wind farms of different sizes and topologies, offering broad application prospects. The invention employs a differential evolutionary-genetic fusion algorithm for multi-objective optimization, comprehensively considering multiple conflicting objective functions and introducing strict constraints to ensure the feasibility and safety of the solution. It combines the global search capability of the differential evolutionary algorithm with the local exploitation accuracy of the genetic algorithm, significantly improving the convergence speed and solution quality. In particular, the elite retention strategy ensures the continuation of the best individuals in each generation, preventing the loss of high-quality solutions. It can dynamically adjust control parameters according to different operating scenarios, achieving precise control and providing decision-makers with more choices, facilitating the development of more flexible and efficient control strategies. The invention also utilizes a pigeon flock biomimetic navigation algorithm, enabling rapid response and timely adjustment of control strategies in dynamically changing environments, thereby maintaining the stable operation of the wind farm. Furthermore, the robustness and efficiency of the pigeon flock biomimetic navigation algorithm provide strong protection for the wind farm, maintaining good control performance even under complex operating conditions.

[0044] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for tuning the voltage and frequency coordination of a wind farm, characterized in that, The method comprises the following steps: S1, obtaining real-time operation data of a wind farm, preprocessing the real-time operation data of the wind farm, and using a cellular automaton model to predict voltage fluctuation and frequency disturbance based on the preprocessed real-time operation data of the wind farm in combination with pre-acquired power grid data to obtain a prediction state matrix; S2, performing multi-objective optimization solving according to the prediction state matrix in combination with a pre-constructed multi-objective optimization problem and constraint condition, and acquiring an optimal droop coefficient vector in combination with a pre-set rule; S3, taking the optimal droop coefficient vector as a control reference, and performing coordinated optimization of voltage and frequency in combination with real-time acquired point-of-connection voltage, frequency and communication network topology data.

2. The method of claim 1, wherein, The obtaining of the real-time operation data of the wind farm, the preprocessing of the real-time operation data of the wind farm, the prediction of voltage fluctuation and frequency disturbance based on the preprocessed real-time operation data of the wind farm in combination with the pre-acquired power grid data using the cellular automaton model, and the obtaining of the prediction state matrix comprise: S11, obtaining real-time operation data of a wind farm, and performing filtering and denoising processing on the real-time operation data of the wind farm to obtain filtered and denoised real-time operation data of the wind farm; S12, performing time alignment processing on the filtered and denoised real-time operation data of the wind farm to obtain preprocessed real-time operation data of the wind farm; S13, initializing the cellular automaton model based on the preprocessed real-time operation data of the wind farm, and using the initialized cellular automaton model to predict voltage fluctuation and frequency disturbance in combination with pre-acquired power grid data to obtain a prediction state matrix; The pre-acquired power grid data comprise pre-acquired wind farm network topology and pre-acquired power grid voltage and frequency data.

3. The method of claim 2, wherein, The initialization of the cellular automaton model based on the preprocessed real-time operation data of the wind farm, the prediction of voltage fluctuation and frequency disturbance using the initialized cellular automaton model in combination with the pre-acquired power grid data, and the obtaining of the prediction state matrix comprise: S131, constructing a wind farm spatial grid using the cellular automaton model according to the preprocessed real-time operation data of the wind farm, initializing a state vector of each grid point through the constructed wind farm spatial grid to obtain an initialized grid point state matrix; S132, constructing a grid point neighborhood relationship based on the initialized grid point state matrix and the pre-acquired wind farm network topology, and quantifying the coupling strength between grid points using a weighted electrical distance method of line impedance reciprocal to integrate the grid point neighborhood relationship and the coupling strength to obtain a structured neighborhood relationship matrix; S133, constructing a grid point state transition function using the neighborhood relationship matrix and the initialized grid point state matrix in combination with the pre-acquired power grid voltage and frequency data, and predicting voltage fluctuation and frequency disturbance based on the grid point state transition function to obtain a prediction state matrix; The pre-acquired power grid voltage and frequency data comprise pre-acquired power grid reactive voltage sensitivity and pre-acquired power grid power frequency fluctuation degree.

4. The method of claim 3, wherein, The construction of the grid point state transition function using the neighborhood relationship matrix and the initialized grid point state matrix in combination with the pre-acquired power grid voltage and frequency data, the prediction of voltage fluctuation and frequency disturbance based on the grid point state transition function, and the obtaining of the prediction state matrix comprise: S1331, based on the neighborhood relationship matrix and the initialized grid point state matrix, combining the pre-acquired power grid reactive voltage sensitivity and the power grid power frequency fluctuation degree to construct a voltage update rule and a frequency response rule, and integrating the voltage update rule and the frequency response rule to obtain a local evolution rule set; S1332, obtaining a mapping relationship of the grid point state changing over time according to the local evolution rule set, and constructing a state transition function through the mapping relationship; S1333, using the state transition function to perform voltage fluctuation and frequency disturbance iterative simulation prediction to obtain a predicted deviation amount of voltage and frequency, and generating a predicted state matrix based on the predicted deviation amount of voltage and frequency.

5. The method of claim 4, wherein, An expression of the local evolution rule set is: ； In the formula, Δ V i Indicates the first i The change in the terminal voltage of the typhoon generator unit; X ij Indicates the connection unit i and j The line reactance; N ( i ) indicates the generator set i The neighborhood set; j Indicates the relationship with the first i Adjacent typhoon generator units with direct electrical connection j ; Q j Indicates the first j The reactive power output of the typhoon generator set; Q i Indicates the first i The reactive power output of the typhoon generator set; Z ij Indicates the connection unit i and j Total line impedance; H eff This represents the equivalent inertia of a wind farm. P j Indicates the first j The active power output of the typhoon generator set; Δ f i indicates the i the variation of the frequency of the grid point to which the wind turbine generator is connected i>P i ref indicates the active power setpoint of the wind turbine generator. i active power setpoint of the wind turbine generator.

6. The method of claim 1, wherein, The multi-objective optimization problem is solved according to the predicted state matrix, combining a pre-constructed multi-objective optimization problem and a constraint condition, and the optimal droop coefficient vector is obtained by combining a preset rule, which includes: S21, based on the predicted state matrix and the pre-constructed multi-objective optimization problem, and combining the pre-constructed constraint condition to generate an optimization problem, obtaining an optimization problem model; S22, initializing the hyperparameters of the differential evolution genetic fusion algorithm according to the optimization problem model, and using the non-dominated sorting guided initialization hyperparameters to approach the Pareto front, obtaining a Pareto optimal solution set; S23, performing optimal performance optimization screening of the droop coefficient vector on the Pareto optimal solution set through a preset rule, obtaining the optimal droop coefficient vector.

7. The method of claim 6, wherein, The hyperparameters of the differential evolution genetic fusion algorithm are initialized according to the optimization problem model, and the non-dominated sorting guided initialization hyperparameters are used to approach the Pareto front, obtaining a Pareto optimal solution set, which includes: S221, extracting the search space features of the optimization problem according to the optimization problem model, and initializing the hyperparameters of the differential evolution genetic fusion algorithm based on the search space features; S222, constructing a feasible solution generation mechanism based on the constraint condition of the optimization problem model, and generating the running boundary of the hyperparameters using the feasible solution generation mechanism combined with the initialized hyperparameters; S223, using the differential evolution and genetic algorithm to generate a population by combining the initialized hyperparameters and the running boundary of the hyperparameters, obtaining an evolution population, and identifying the Pareto optimal solution set by non-dominated sorting the evolution population.

8. The method of claim 7, wherein, The hyperparameters of the differential evolution genetic fusion algorithm are initialized according to the optimization problem model, and the non-dominated sorting guided initialization hyperparameters are used to approach the Pareto front, obtaining a Pareto optimal solution set, which includes: S2231, using the mutation crossover selection mechanism of the differential evolution to combine the initialized hyperparameters and the running boundary of the hyperparameters to perform global search operation on the population, obtaining a primary population; S2232, performing screening operation on the primary population based on the selection crossover mutation mechanism of the genetic algorithm, obtaining a child population, and performing reservation operation on the primary population using the elite reservation strategy, obtaining a reserved population, and integrating the child population and the reserved population to obtain an evolution population; S2233, the non-dominated sorting based on the Pareto dominance relationship is used to sort the individuals in the evolutionary population, and the crowding distance is used to evaluate the distribution degree of the sorting result, and the distribution degree evaluation result is used to guide the population to approach the Pareto front, and a candidate solution set is obtained; S2234, the repeated individual removal is performed on the candidate solution set to obtain a removed candidate solution set, and the removed candidate solution set is subjected to the Pareto optimal solution identification through the clustering analysis to obtain a Pareto optimal solution set.

9. The method of claim 1, wherein, The optimal droop coefficient vector is used as a control reference, and real-time acquisition of the grid-connected point voltage, frequency and communication network topology data is used for voltage and frequency collaborative optimization, which comprises the following steps: S31, the optimal droop coefficient vector is used as a control reference for initial droop control parameter setting, and real-time acquisition of the grid-connected point voltage, frequency and communication network topology data is performed; S32, based on the real-time acquisition of the grid-connected point voltage, frequency and communication network topology data and the initial droop control parameter, the pigeon bionic navigation algorithm is used to correct the voltage and frequency deviation to obtain optimized droop coefficient adjustment data; S33, the wind turbine control instruction is generated according to the optimized droop coefficient adjustment data, and the wind farm voltage and frequency collaborative optimization is realized through the control instruction.

10. The method of claim 9, wherein, The pigeon bionic navigation algorithm is used to correct the voltage and frequency deviation based on the real-time acquisition of the grid-connected point voltage, frequency and communication network topology data and the initial droop control parameter to obtain the optimized droop coefficient adjustment data, which comprises the following steps: S321, the pigeon bionic navigation algorithm is initialized based on the real-time acquisition of the grid-connected point voltage, frequency and communication network topology data and the initial droop control parameter to obtain the initialized pigeon bionic navigation algorithm; S322, the weighted sum of squares of the voltage and frequency deviation is calculated by using the initialized pigeon bionic navigation algorithm, and the fitness evaluation is performed according to the weighted sum of squares, and the information interaction intensity is adjusted based on the fitness evaluation result and the communication network topology data; S323, the pigeon navigation behavior is simulated and the optimal deviation correction direction is iteratively searched based on the adjusted information interaction intensity through the position and velocity update rule to obtain the optimized droop coefficient adjustment data.

Images