A multi-scale coupled wind farm high-precision electrical model simulation method and system

Abstract

The application discloses a kind of multi-scale coupling's wind farm high-precision electrical model simulation method and system.Method includes: acquisition pre-processing wind farm multi-source heterogeneous data;Component-level refinement model and cluster model are constructed, and multi-time scale coupling is realized using event-triggered multi-rate simulation;Based on reinforcement learning, model precision and efficiency are adaptively optimized;Complex working condition scenario library is constructed and parallel simulation calculation is executed;Simulation results are analyzed and standardized model verification is carried out.System includes data acquisition and preprocessing, multi-scale model construction, adaptive model optimization, simulation calculation, result analysis and model verification and man-machine interaction module.The application realizes wind farm full-time scale high-precision simulation, precision and efficiency are considered, and is suitable for wind farm planning design, operation control and power system stability analysis.

Description

Technical Field

[0001] This invention relates to the field of power system simulation technology, specifically a high-precision electrical model simulation method and system for multi-scale coupled wind farms. Background Technology

[0002] With the acceleration of global energy structure transformation, wind power, as an important component of clean and renewable energy, has seen its installed capacity continue to grow rapidly. However, large-scale wind power grid connection has brought severe challenges to the safe and stable operation of the power system: wind farm output has strong randomness and volatility, the widespread application of power electronic equipment leads to strong nonlinear characteristics in the system, and the fault response mechanism under extreme conditions is complex and variable.

[0003] Electrical model simulation of wind farms is a core technical means to analyze wind power grid connection characteristics, optimize control strategies, and conduct stability assessments. Currently, the mainstream simulation methods are mainly divided into three categories: First, electromechanical transient simulation based on aggregated equivalents, which equates the wind farm to a few equivalent machines, offering high computational efficiency but neglecting the differences between units and the rapid dynamic characteristics of power electronic equipment, making it difficult to meet the analysis needs of high-proportion renewable energy grids; second, refined simulation based on full electromagnetic transients, which can accurately simulate the actions of power electronic switches, but the computational load grows exponentially, making it unsuitable for the simulation analysis of large-scale wind farm clusters; and third, data-driven simulation methods, which rely on historical data to train models, exhibiting poor generalization ability and lacking physical interpretability.

[0004] The core problems with existing technologies include:

[0005] 1. The problem of multi-timescale coupling: Wind farms cover multi-timescale processes, from millisecond-level power electronic switching dynamics to minute-level wind speed changes. Existing simulation methods are difficult to achieve the organic integration of models of different timescales. They either sacrifice accuracy for speed or are too computationally intensive to be applied in engineering.

[0006] 2. Conflict between model accuracy and computational efficiency: Refined models have high accuracy but low computational efficiency, while simplified models are fast but have large errors. There is a lack of adaptive modeling methods that can achieve the optimal balance between accuracy and efficiency.

[0007] 3. Insufficient simulation capability for complex operating conditions: The simulation capability for complex operating conditions such as extreme wind speed fluctuations, cascading failures, and multi-field coupling is limited, and it cannot accurately reproduce the dynamic response of wind farms in actual scenarios.

[0008] 4. Inadequate model validation system: The lack of standardized model validation processes and quantitative evaluation indicators makes it difficult to guarantee the reliability of simulation results.

[0009] Therefore, developing a simulation method and system for wind farm electrical models that can achieve multi-scale coupling, combine high precision and high efficiency, and simulate complex operating conditions has become an urgent need to ensure the safe and stable operation of high-proportion new energy power systems. Summary of the Invention

[0010] This invention aims to address the problems of existing wind farm electrical model simulation methods, such as difficulties in multi-timescale coupling, difficulty in balancing accuracy and efficiency, insufficient simulation capabilities for complex operating conditions, and imperfect model verification systems. It provides a high-precision electrical model simulation method and system for wind farms with multi-scale coupling, so as to achieve high-precision simulation analysis of wind farms across all time scales and operating conditions, and provide reliable technical support for the planning, operation, and control of power systems.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A high-precision electrical model simulation method for multi-scale coupled wind farms includes the following steps:

[0013] Step 1: Collect multi-source data from the wind farm, and perform data cleaning, data standardization, and feature engineering to obtain a standardized dataset. The multi-source data includes meteorological data, unit operation data, power grid operation data, and fault recording data.

[0014] Step 2: Based on the standardized dataset, construct a component-level refined model including wind turbine, generator, converter, and transmission system, as well as a wind farm cluster model based on dynamic clustering. Then, use an event-triggered multi-rate simulation method to couple the electromagnetic transient scale, electromechanical transient scale, and medium- and long-term dynamic scale to generate a multi-scale coupled model.

[0015] Step 3: The accuracy and complexity of the multi-scale coupled model are evaluated, and an adaptive optimization method based on reinforcement learning is adopted to automatically adjust the model complexity and parameters with the goal of achieving the best combination of model accuracy and computational efficiency, thereby generating an optimized adaptive model.

[0016] Step 4: Based on the optimized adaptive model, select or configure working conditions from the constructed complex working condition scenario library, and perform parallel simulation calculations using a simulation algorithm that matches the time scale to generate simulation result data;

[0017] Step 5: Perform time-domain, frequency-domain, and modal analysis on the simulation results data, and standardize the model accuracy through static verification, dynamic verification, and extreme condition verification. Correct the model parameters and update the model based on the verification results.

[0018] Furthermore, the component-level refined model in step 2 includes:

[0019] A wind turbine aerodynamic model based on blade element momentum theory and considering tip loss, hub loss, and dynamic stall; a doubly-fed induction generator model based on the dq coordinate system and considering stator winding, rotor winding, inductance, and flux coupling; a permanent magnet synchronous generator model considering permanent magnet flux, stator winding resistance, and inductance; a converter model considering switching losses, dead zone effect, and parasitic parameters; and a transmission system model considering gearbox stiffness, damping, and rotational inertia.

[0020] The wind farm cluster model includes: a unit aggregation equivalent model that uses a dynamic clustering algorithm to perform dynamic clustering based on the unit output characteristics and wind speed response curves, and determines equivalent parameters by combining capacity weighting method and least squares method parameter identification; a collector network model that uses a π-type equivalent circuit and includes temperature and altitude correction; and a reactive power compensation device model that includes dual closed-loop control of voltage outer loop and current inner loop and reactive power coordinated control.

[0021] Furthermore, in step 3:

[0022] The model accuracy evaluation is based on a multi-dimensional evaluation index system that includes steady-state accuracy index, transient accuracy index and frequency characteristic index.

[0023] The model complexity analysis uses computational cost, memory usage, and simulation time as evaluation metrics.

[0024] The adaptive optimization method based on reinforcement learning employs a deep deterministic policy gradient algorithm. Its state space is defined by the current model accuracy, complexity, and operating conditions; its action space is defined by the model's simplification / refinement operations and parameter adjustments; and the reward function R is designed as follows:

[0025] ;

[0026] Where R is the reward value, α and β are weighting coefficients, E is the error of the current model, Emax is the maximum permissible error, Tref is the reference simulation time, and T is the current simulation time; the accuracy weight and efficiency weight are dynamically adjusted according to the working condition complexity index.

[0027] Furthermore, in step 4:

[0028] The complex operating condition scenario library includes extreme weather conditions, power grid failure conditions, unit failure conditions, and multi-field coupling conditions.

[0029] The simulation algorithms include: electromagnetic transient simulation using the implicit trapezoidal integration method with a step size of 10μs to 100μs; electromechanical transient simulation using the improved Euler method or Runge-Kutta method with a step size of 10ms to 100ms; and medium- and long-term dynamic simulation using the Euler method or trapezoidal integration method with a step size of 1s to 10s.

[0030] The parallel simulation computation is based on a distributed computing architecture. It uses a region decomposition method to divide the wind farm into multiple sub-regions and realizes data interaction and synchronization between sub-regions through a message passing interface.

[0031] Furthermore, in step 5:

[0032] The modal analysis includes the calculation of eigenvalues ​​of nonlinear factors and uses the damping ratio as a stability margin quantification index. When the damping ratio is greater than 0.1, the system is considered stable, and when the damping ratio is less than 0.05, a stability warning is issued.

[0033] The static verification is performed by comparing the voltage, power, and power factor errors between the simulation results and the actual operating data.

[0034] The dynamic verification is carried out by comparing the transient response characteristics of simulation results and fault recording data, and the root mean square error is used as the waveform similarity evaluation index. When the root mean square error is less than 0.05, the similarity is judged to be no less than 95%. The verification samples cover 20 wind speed conditions and 5 fault types, and the number of samples is no less than 50 groups.

[0035] The model correction and update include adjusting model parameters using parameter identification methods and establishing an update mechanism that regularly collects new running data to retrain and validate the model.

[0036] A high-precision electrical model simulation system for multi-scale coupled wind farms, used to execute the method described above, includes:

[0037] The data acquisition and preprocessing module is used to acquire multi-source data from the wind farm and perform data cleaning, data standardization and feature engineering to obtain a standardized dataset. The multi-source data includes meteorological data, unit operation data, power grid operation data and fault recording data.

[0038] The multi-scale model building module is used to construct a component-level refined model including wind turbine, generator, converter, and transmission system, as well as a wind farm cluster model based on dynamic clustering, based on the standardized dataset. It also uses an event-triggered multi-rate simulation method to couple the electromagnetic transient scale, electromechanical transient scale, and medium- and long-term dynamic scale to generate a multi-scale coupled model.

[0039] The adaptive model optimization module is used to evaluate the accuracy and complexity of the multi-scale coupled model, and adopts an adaptive optimization method based on reinforcement learning to automatically adjust the model complexity and parameters with the goal of achieving the best combination of model accuracy and computational efficiency, thereby generating an optimized adaptive model.

[0040] The simulation calculation module is used to select or configure working conditions from the constructed complex working condition scenario library based on the optimized adaptive model, and perform parallel simulation calculations using a simulation algorithm that matches the time scale to generate simulation result data.

[0041] The results analysis and model verification module is used to perform time-domain, frequency-domain and modal analysis on the simulation results data, and to perform standardized verification of the model accuracy through static verification, dynamic verification and extreme working condition verification, and to correct the model parameters and update the model based on the verification results.

[0042] The human-computer interaction module is connected to the data acquisition and preprocessing module, the multi-scale model construction module, the adaptive model optimization module, the simulation calculation module, and the result analysis and model verification module, respectively, and is used to provide an interactive interface for data query, model setting, simulation control, and result display.

[0043] Furthermore, the multi-scale model construction module includes:

[0044] The component modeling unit is used to create detailed models of wind turbines, generators, converters, and transmission systems. The converter model includes dead zone effect and switching loss models, and the transmission system model includes a flexible shaft nonlinear torsion model.

[0045] The cluster modeling unit is used to establish a wind farm cluster model that includes a unit aggregation equivalent model, a power collection network model, and a reactive power compensation device model using dynamic clustering and adaptive aggregation methods.

[0046] The multi-scale coupling unit is used to achieve coupling of electromagnetic transients, electromechanical transients and medium- to long-term dynamic scales using an event-triggered multi-rate simulation method. It includes a step-size gradual smooth transition mechanism and a Kalman filter error propagation control mechanism for scale switching.

[0047] Furthermore, the adaptive model optimization module includes:

[0048] The accuracy evaluation unit is used to evaluate the model accuracy based on steady-state accuracy indicators, transient accuracy indicators, and frequency characteristic indicators.

[0049] The complexity analysis unit is used to evaluate the computational cost, memory usage, and simulation time metrics of the model.

[0050] The adaptive optimization unit is used to automatically adjust the model complexity and parameters by employing a deep deterministic gradient algorithm, combined with a priority experience replay mechanism and a working condition adaptive learning rate, with the goal of achieving the optimal balance between model accuracy and computational efficiency.

[0051] Furthermore, the simulation calculation module includes:

[0052] The working condition scenario construction unit is used to build a complex working condition scenario library that includes extreme weather conditions, power grid fault conditions, unit fault conditions and multi-field coupling conditions. The extreme weather conditions include typhoon, thunderstorm, low temperature icing and sandstorm conditions. The unit fault conditions include converter IGBT open circuit fault, bearing wear fault and gearbox tooth breakage fault.

[0053] The simulation algorithm configuration unit is used to select and configure electromagnetic transient, electromechanical transient, or medium- to long-term dynamic simulation algorithms and step sizes according to the time scale and operating conditions.

[0054] Parallel computing units are used to perform parallel simulation calculations of multiple sub-regions based on a distributed computing architecture and the region decomposition method, through a message passing interface.

[0055] Furthermore, the result analysis and model validation module includes:

[0056] The results analysis unit is used to perform time-domain analysis, frequency-domain analysis, and modal analysis, including nonlinear eigenvalue calculation and damping ratio evaluation, on the simulation results data.

[0057] The model verification unit is used to perform static verification, dynamic verification, and extreme condition verification. The dynamic verification uses root mean square error as the waveform similarity evaluation index.

[0058] The model update unit is used to adjust the model parameters according to the validation results using the parameter identification method, and to establish an update mechanism for regularly collecting new running data for retraining and validation.

[0059] The human-computer interaction module supports data import and export, model version management, and verification report generation.

[0060] The present invention has the following beneficial effects:

[0061] 1. Multi-scale coupled simulation: By dividing different time scales and adopting an event-triggered multi-rate simulation method, the organic integration of wind farm processes from millisecond to hourly scales is realized, solving the problem that existing methods cannot take into account both fast and slow dynamic processes, and enabling more realistic simulation of the actual operating characteristics of wind farms.

[0062] 2. Adaptive Accuracy-Efficiency Balance: An adaptive model optimization method based on reinforcement learning is proposed, which can automatically adjust the complexity and parameters of the model according to the operating conditions and simulation requirements, significantly improving computational efficiency while ensuring simulation accuracy, thus resolving the contradiction between accuracy and efficiency.

[0063] 3. Complex operating condition simulation capability: A complex operating condition scenario library covering multiple dimensions such as meteorology, power grid, and generating units has been built, which can simulate the dynamic response of wind farms under complex operating conditions such as extreme wind speed fluctuations, cascading failures, and multi-field coupling, providing more comprehensive technical support for the safety and stability analysis of power systems.

[0064] 4. Standardized model verification system: A multi-dimensional accuracy evaluation index system and a standardized model verification process have been established. Through static verification, dynamic verification and extreme working condition verification, the credibility and reliability of simulation results are ensured.

[0065] 5. High engineering application value: The system has good versatility and scalability, and can adapt to the simulation needs of wind farms of different types and scales. It can provide a scientific basis for the planning and design, operation and management, control strategy optimization and power system stability assessment of wind farms, and has significant engineering application value and economic benefits. Attached Figure Description

[0066] Figure 1 This is a flowchart of the multi-scale coupled high-precision electrical model simulation method for wind farms according to the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Please see Figure 1 This invention provides a high-precision electrical model simulation method for multi-scale coupled wind farms, comprising the following steps:

[0069] Step 1: Acquisition and Preprocessing of Multi-Source Heterogeneous Data from Wind Farms

[0070] 1.1 Multi-source data acquisition: Construct a multi-source data acquisition system for the wind farm. The types of data acquired include:

[0071] Meteorological data: Wind speed, wind direction, temperature, air pressure, humidity and other data are collected through wind measurement towers and weather stations installed in the wind farm at a frequency of 1Hz;

[0072] Unit operation data: The SCADA system of the wind turbines collects data such as output power, voltage, current, speed, pitch angle, and converter switching status of each unit at a frequency of 10Hz.

[0073] Power grid operation data: Data such as voltage, frequency, power factor, and line power flow at the grid connection point are collected through the power grid dispatching system at a frequency of 5Hz;

[0074] Fault recording data: Transient data under abnormal operating conditions such as power grid faults and unit faults are collected by a fault recorder, with a sampling frequency of 10kHz.

[0075] 1.2 Data Cleaning: Data cleaning is performed using a combination of rule-based and machine learning methods.

[0076] Outlier detection: The 3σ principle is used to identify data points that deviate from the normal range, and the Isolation Forest algorithm is used to detect outlier samples in the data;

[0077] Missing value imputation: For continuous data, linear interpolation or time series prediction methods based on LSTM neural networks are used for imputation; for discrete data, the mode imputation method is used.

[0078] Duplicate value handling: Identify and delete duplicate data through data hash comparison.

[0079] 1.3 Data Standardization: To eliminate the differences in units between different data, the Z-score standardization method is used to process the data. The formula is as follows:

[0080]

[0081] Where Xstd represents the standardized data, x represents the original data, μ represents the mean of the original data, and σ represents the standard deviation of the original data.

[0082] 1.4 Feature Engineering: Extracting features related to the electrical characteristics of the wind farm from the preprocessed data, including:

[0083] Time-domain characteristics: mean, variance, peak value, trough value, root mean square, etc.

[0084] Frequency domain features: Spectral components, amplitude, phase, etc., are extracted using Fast Fourier Transform (FFT);

[0085] Time-frequency domain features: Time-frequency distribution features at different scales are extracted using wavelet transform;

[0086] Statistical characteristics: skewness, kurtosis, correlation coefficient, etc.

[0087] Step 2: Construction of a multi-scale model of the wind farm

[0088] 2.1 Construction of Component-Level Refined Model

[0089] Wind turbine aerodynamic model: A wind turbine model considering tip losses, hub losses, and dynamic stall is established using blade element momentum theory (BEM) combined with a modified Glauert formula to calculate wind energy capture efficiency and aerodynamic torque. The model formulas are as follows:

[0090] Among them, C p λ is the wind energy utilization coefficient, β is the tip speed ratio, and β is the blade pitch angle. i The leaf tip speed ratio is corrected for the induction factor.

[0091] Generator Models: Detailed models are established for different types of wind turbine units:

[0092] Doubly fed induction generator (DFIG): A mathematical model based on the dq coordinate system is established, considering the resistance, inductance, and magnetic flux coupling relationship of the stator and rotor windings. The formulas are as follows: Among them, u ds u qs u dr u qr The d-axis and q-axis components of the stator and rotor voltages, respectively, i ds i qs i dr i q These are the d-axis and q-axis components of the stator and rotor currents, respectively, R s R r The stator and rotor resistances are respectively, L s L r These are the stator and rotor inductances, ω s For synchronous angular velocity, ω r ω is the rotor angular velocity.

[0093] Permanent Magnet Synchronous Generator (PMSG): A mathematical model considering the permanent magnet flux linkage, stator winding resistance, and inductance is established, as shown in the following formula:

[0094] Among them, u d u q Let i be the d-axis and q-axis components of the stator voltage. d i q R represents the d-axis and q-axis components of the stator current. s L is the stator resistance. d L q Let ω be the d-axis and q-axis components of the stator inductance. e Let ψ be the electric angular velocity. f It is a permanent magnet flux linkage.

[0095] Converter Model: A power electronic converter model based on switching functions is established, considering switching losses, dead-zone effects, and parasitic parameters. For back-to-back converters, models of the generator-side converter and the grid-side converter are established separately, and bidirectional power flow and reactive power regulation are achieved through control strategies.

[0096] Transmission system model: A lumped mass block model is adopted, considering the stiffness, damping, and moment of inertia of the gearbox, to simulate the torsional vibration characteristics of the transmission system. The formulas are as follows: Where Jt is the moment of inertia of the transmission system, ωt is the angular velocity of the transmission system, Ta is the aerodynamic torque, Tg is the electromagnetic torque of the generator, Bt is the damping coefficient, Kt is the stiffness coefficient, and θt is the torsional angular displacement.

[0097] 2.2 Wind Farm Cluster Model Construction

[0098] Equivalent Unit Aggregation Model: An adaptive aggregation method based on dynamic clustering is proposed. According to the operating characteristics, electrical distance, and wind speed correlation of the units, the wind farm is divided into several equivalent units. The parameters of the equivalent units are determined by a combination of capacity weighting and parameter identification, as shown in the following formula: Among them, X eq S is an equivalent parameter. i Let X be the capacity of the i-th unit. i Let be the original parameters of the i-th unit, and n be the number of units.

[0099] Collector network model: A π-type equivalent circuit model is used to establish the collector line model, considering the resistance, inductance, capacitance, and conductance parameters of the line. Based on the actual topology of the wind farm, the node admittance matrix of the collector network is established for power flow calculation and transient analysis.

[0100] Reactive power compensation device model: Establish models of reactive power compensation devices such as capacitor banks, static var compensators (SVC), and static var generators (SVG) to simulate their reactive power regulation characteristics and voltage support capabilities.

[0101] 2.3 Multi-scale model coupling

[0102] Time scale division: The dynamic process of a wind farm is divided into three time scales:

[0103] Electromagnetic transient scale: The time scale is in the millisecond range, mainly simulating the rapid dynamic response of power electronic switch action, fault transient process, etc.

[0104] Electromechanical transient scale: The time scale is on the order of seconds, mainly simulating the medium-speed dynamic response of the unit, such as power regulation and speed change;

[0105] Medium- to long-term dynamic scale: The time scale is from minutes to hours, mainly simulating slow dynamic processes such as wind speed changes and unit start-up and shutdown.

[0106] Multi-scale coupling strategy: An event-triggered multi-rate simulation method is employed to couple models at different time scales. When a rapid dynamic event occurs in the system (such as a fault or switching action), the simulation automatically switches to the electromagnetic transient scale; when the system is in a steady state or slow dynamic process, the simulation uses the electromechanical transient or medium-to-long-term dynamic scale. Data interaction and synchronization between models at different scales are achieved through an interface conversion module.

[0107] Step 3: Adaptive Model Optimization

[0108] 3.1 Model Accuracy Evaluation: Establish a multi-dimensional model accuracy evaluation index system, including:

[0109] Steady-state accuracy indicators: voltage deviation, power deviation, power factor deviation, etc.

[0110] Transient accuracy metrics include: transient response time, overshoot, and decay rate.

[0111] Frequency response indicators: amplitude frequency response error, phase frequency response error, etc.

[0112] The accuracy of the model is evaluated by comparing the simulation results with actual operating data or high-precision simulation data and calculating the values ​​of various evaluation indicators.

[0113] 3.2 Model Complexity Analysis: The complexity of the model is evaluated using metrics such as computational cost, memory usage, and simulation time. A Pareto front is established to assess the trade-off between accuracy and complexity for different models.

[0114] 3.3 Adaptive Model Optimization: An adaptive model optimization method based on reinforcement learning is proposed, aiming to achieve a comprehensive optimization of model accuracy and computational efficiency by automatically adjusting the model's complexity and parameters. The specific steps are as follows:

[0115] State space definition: The current model accuracy, complexity, and operating conditions are used as the state space;

[0116] Action space definition: The simplification / refinement operations and parameter adjustments of the model are defined as the action space;

[0117] Reward Function Design: Design a reward function that comprehensively considers accuracy and efficiency, as shown in the following formula: ;

[0118] Where R is the reward value, α and β are weight coefficients, and E is the error of the current model. max For the maximum permissible error, T refThe simulation time is used as a reference, and T is the current simulation time.

[0119] Reinforcement learning training: The Deep Deterministic Policy Gradient (DDPG) algorithm is used for training to learn the optimal model adjustment strategy and achieve an adaptive balance between accuracy and efficiency.

[0120] Step 4: Simulation and Calculation of Complex Working Conditions

[0121] 4.1 Construction of Complex Operating Condition Scenarios: A library of complex operating condition scenarios covering multiple dimensions, including meteorology, power grid, and generating units, is constructed, including:

[0122] Extreme weather conditions: typhoons, thunderstorms, freezing temperatures, sudden changes in wind speed, etc.

[0123] Power grid fault conditions: three-phase short circuit, two-phase short circuit, single-phase ground fault, voltage drop, frequency fluctuation, etc.

[0124] Unit failure conditions: generator failure, converter failure, transmission system failure, blade failure, etc.

[0125] Multi-field coupling conditions: wind-solar-storage joint operation, interaction between wind farm and power grid, etc.

[0126] 4.2 Simulation Algorithm Selection and Configuration: Select appropriate simulation algorithms based on different operating conditions and time scales.

[0127] Electromagnetic transient simulation: The implicit trapezoidal integration method is adopted, and the step size is set to 10μs - 100μs, which is suitable for the simulation of fast transient processes;

[0128] Electromechanical transient simulation: The improved Euler method or Runge-Kutta method is used, with a step size of 10ms - 100ms, which is suitable for simulation of medium-speed dynamic processes;

[0129] Medium- and long-term dynamic simulation: Using the Euler method or trapezoidal integration method, with a step size of 1s-10s, it is suitable for simulating slow dynamic processes and is used for scenarios involving long-term wind speed changes and power prediction.

[0130] Adaptive step size adjustment: Develop an automatic step size switching module that automatically switches to the electromagnetic transient step size when a fault occurs, switches to the electromechanical transient step size 10 seconds after the fault is recovered, and switches to the long-term simulation step size after 30 minutes of steady-state operation.

[0131] 4.3 Parallel Simulation Computation: Based on a distributed computing architecture, parallel simulation of the wind farm model is achieved. The wind farm is divided into multiple sub-regions using a region decomposition method, with each sub-region assigned to a computing node for parallel computation. Data interaction and synchronization between computing nodes are achieved through a message passing interface (MPI), improving simulation efficiency.

[0132] Step 5: Simulation Result Analysis and Model Validation

[0133] 5.1 Simulation Result Analysis: The simulation results are analyzed in depth using methods such as time domain analysis, frequency domain analysis, and modal analysis.

[0134] Time-domain analysis: Plot time-domain curves of parameters such as power, voltage, current, and speed, and analyze their dynamic response characteristics and steady-state characteristics;

[0135] Frequency domain analysis: Converting time-domain signals into frequency-domain signals using FFT or wavelet transform to analyze the frequency components and spectral characteristics of the signals;

[0136] Modal analysis: Through eigenvalue calculation and mode shape analysis, the oscillation modes and damping characteristics of the system are identified, and the stability of the system is evaluated. 5.2 Model validation: A standardized model validation process is established, including:

[0137] Static verification: Compare the simulation results with actual operating data under steady-state conditions for parameters such as voltage, power, and power factor to verify the static accuracy of the model;

[0138] Dynamic verification: Compare the simulation results with the fault recording data under transient operating conditions to verify the dynamic accuracy of the model;

[0139] Extreme operating condition verification: Simulations are performed under extreme operating conditions, and the simulation results are compared with theoretical analysis or experimental data to verify the robustness of the model.

[0140] Power fluctuation: The power fluctuation limit is ±5% (1-minute average power), supplemented by power fluctuation frequency distribution analysis (using histograms to show the proportion of different fluctuation ranges).

[0141] Waveform similarity: The root mean square error (RMSE) was used as the evaluation index. When RMSE < 0.05, the similarity was judged to be ≥ 95%. The verification samples covered 20 wind speed conditions and 5 fault types, with a sample size of ≥ 50 groups.

[0142] Modal analysis: Eigenvalue calculations are performed by incorporating nonlinear factors (such as converter dead zone and flexible shaft torsion). The damping ratio is used as a quantitative index for stability margin. The system is considered stable when the damping ratio is >0.1, and a stability warning is issued when the damping ratio is <0.05.

[0143] 5.3 Model Correction and Update: Based on the validation results, the model is corrected and updated. Parameter identification methods are used to adjust model parameters and reduce simulation errors. A model update mechanism is established to periodically collect new runtime data, retrain and validate the model, and ensure its timeliness and accuracy.

[0144] This invention provides a high-precision electrical model simulation system for multi-scale coupled wind farms, corresponding to the above-described method, comprising:

[0145] The data acquisition and preprocessing module is used to collect multi-source heterogeneous data from the wind farm and perform preprocessing operations such as cleaning, standardization, and feature engineering. This module includes a data acquisition unit, a data cleaning unit, a data standardization unit, and a feature engineering unit. The data acquisition unit communicates with sensors, SCADA systems, and power grid dispatching systems in the wind farm to collect multi-source data in real time. The data cleaning unit uses rule-based and machine learning methods to remove outliers, fill in missing values, and delete duplicate values. The data standardization unit uses the Z-score method to standardize the data. The feature engineering unit extracts time-domain, frequency-domain, time-frequency-domain, and statistical features from the preprocessed data.

[0146] The multi-scale model building module is used to construct component-level refined models, cluster models, and multi-scale coupled models of wind farms. This module includes component modeling units, cluster modeling units, and multi-scale coupling units. The component modeling unit establishes refined models of components such as wind turbines, generators, converters, and transmission systems; the cluster modeling unit uses dynamic clustering and adaptive aggregation methods to build a wind farm cluster model; and the multi-scale coupling unit organically integrates models at different time scales to establish a multi-scale coupled model.

[0147] The adaptive model optimization module is used to perform accuracy evaluation, complexity analysis, and adaptive optimization on the multi-scale coupled model. This module includes an accuracy evaluation unit, a complexity analysis unit, and an adaptive optimization unit. The accuracy evaluation unit establishes a multi-dimensional accuracy evaluation index system to evaluate the model's accuracy; the complexity analysis unit evaluates the model's computational cost, memory usage, simulation time, and other complexity indicators; the adaptive optimization unit uses reinforcement learning algorithms to achieve an adaptive balance between accuracy and efficiency.

[0148] The simulation calculation module is used to select or configure working conditions from a constructed library of complex working conditions for the optimized adaptive model, and to perform parallel simulation calculations using simulation algorithms matched to the time scale, generating simulation result data. This module includes a working condition scenario construction unit, a simulation algorithm configuration unit, and a parallel computing unit. The working condition scenario construction unit builds a library of complex working conditions covering multiple dimensions; the simulation algorithm configuration unit selects appropriate simulation algorithms and configures parameters according to the working condition scenario and time scale; the parallel computing unit implements parallel simulation of the model based on a distributed computing architecture, improving computational efficiency.

[0149] The results analysis and model verification module is used to analyze simulation results and verify the model. This module includes a results analysis unit, a model verification unit, and a model update unit. The results analysis unit uses time-domain, frequency-domain, and modal analysis methods to conduct in-depth analysis of the simulation results; the model verification unit establishes a standardized verification process to verify the static accuracy, dynamic accuracy, and robustness of the model; the model update unit corrects and updates the model based on the verification results to ensure the timeliness and accuracy of the model.

[0150] The human-computer interaction module is used to enable interaction between the user and the system. This module provides a user-friendly interface and supports functions such as data querying, model setup, simulation control, and result display. Users can input relevant wind farm parameters, select operating scenarios, start simulation calculations, and view simulation results and analysis reports through the interface. The human-computer interaction module also supports data import and export, facilitating further analysis and processing by users.

[0151] The following is a specific example to illustrate this:

[0152] Taking a real offshore wind farm as an example, this wind farm is equipped with 100 5MW permanent magnet synchronous wind turbines. The power from the wind turbines is collected by a 35kV collection line to a 220kV substation and then connected to the power grid. The following are the specific implementation steps of the method of this invention:

[0153] Step 1: Acquisition and Preprocessing of Multi-Source Heterogeneous Data from Wind Farms

[0154] 1.1 Multi-source data acquisition: Ten meteorological towers were installed within the wind farm to collect meteorological data such as wind speed, wind direction, and temperature at a frequency of 1 Hz; operational data such as output power, voltage, current, speed, and pitch angle were collected through the SCADA system of each turbine at a frequency of 10 Hz; grid data such as voltage, frequency, and power factor at the grid connection point were collected through the grid dispatch system at a frequency of 5 Hz; fault recorders were installed at the substation and the outlet of each turbine to collect transient fault data at a sampling frequency of 10 kHz. Multi-source data was continuously collected for three months, yielding approximately 1.2 TB of data.

[0155] 1.2 Data Cleaning: First, the 3σ principle was used to identify approximately 0.5% of outlier data points. Then, the Isolation Forest algorithm was used to further detect approximately 0.2% of outlier samples, which were then deleted. For missing values, LSTM neural network time series prediction was used to fill in continuous data, and mode imputation was used to fill in discrete data. After imputation, the data integrity reached over 99.8%. Approximately 0.1% of duplicate data was identified through data hash comparison and deleted.

[0156] 1.3 Data Standardization: The Z-score standardization method was used to process the cleaned data, converting data with different dimensions into standard normal distribution data with a mean of 0 and a standard deviation of 1, eliminating the difference in dimensions and facilitating subsequent model training and analysis.

[0157] 1.4 Feature Engineering: Time-domain features (mean, variance, peak value, etc.), frequency-domain features (spectral components, amplitude, etc.), time-frequency-domain features (wavelet transform time-frequency distribution), and statistical features (skewness, kurtosis, etc.) were extracted from the standardized data, totaling 32 features, which were used for subsequent model building and optimization.

[0158] Step 2: Construction of a multi-scale model of the wind farm

[0159] 2.1 Construction of Component-Level Refined Model

[0160] Wind turbine aerodynamic model: A wind turbine model was established using blade element momentum theory combined with a modified Glauert formula. Based on the airfoil parameters and aerodynamic characteristics of the blades, the relationship curves between the wind energy utilization coefficient and the tip speed ratio and blade pitch angle were calculated. By comparing with wind tunnel test data, the error of the model's wind energy utilization coefficient was within 2%.

[0161] Doubly fed induction generator (DFIG) model: The DFIG model includes stator / rotor winding equivalent circuits, rotor-side converter (RSC) and grid-side converter (GSC) sub-modules; the rotor resistance, leakage inductance and stator parameters of the DFIG are mapped to the stator parameters of the permanent magnet synchronous generator (PMSG) based on the rated capacity, power factor and other indicators of the unit. ( This is a model correction factor, ranging from 0.8 to 1.2.

[0162] Permanent Magnet Synchronous Generator Model: A mathematical model of a permanent magnet synchronous generator based on the dq coordinate system was established. Based on the generator's nameplate parameters and design data, parameters such as stator resistance, inductance, and permanent magnet flux linkage were determined. Simulation calculations showed that the generator's output power, voltage, current, and other parameters had an error of less than 3% compared to actual operating data.

[0163] Converter Model: A back-to-back converter model based on switching functions is established, considering switching losses and dead-zone effects. A control strategy is implemented to achieve bidirectional power flow and reactive power regulation, with a converter response time of less than 20ms and harmonic distortion rate of less than 3%. Dead-zone Effect: The dead-zone time ranges from 0.5 to 2μs. The dead-zone voltage compensation formula is as follows: ( This is the DC bus voltage. (Switching cycle); Switching loss: On-state loss is modeled using on-state voltage drop-current fitting. The turn-off loss is based on a piecewise function established by the loss curve provided by the manufacturer, covering different current and voltage operating conditions.

[0164] Transmission system model: A lumped mass block model is used to establish the transmission system model, considering the gearbox's stiffness, damping, and moment of inertia. The gearbox is a 3-stage planetary gear structure, and the moment of inertia distribution of each mass block is based on the actual weight ratio of the unit (hub:high-speed shaft:low-speed shaft = 6:3:1); the flexible shaft uses a nonlinear modeling option, employing a torsional spring-damping model. ( To increase torsional stiffness, (where is the damping coefficient), simulating the impact of shaft torsional vibration on unit output. Through simulation calculations, the error between the torsional vibration characteristics of the transmission system and the actual test results is within 5%.

[0165] Adaptive adaptation function: The system develops a model selection module. After the user inputs the unit model / parameters, the system automatically matches the corresponding model and adaptively adjusts the model accuracy according to the operating conditions (such as wind speed range and grid voltage level). For example, it simplifies the converter switching model under low wind speed conditions to improve simulation efficiency.

[0166] 2.2 Wind Farm Cluster Model Construction

[0167] Unit Aggregated Equivalent Model: Using a dynamic clustering method, 100 units are divided into 10 equivalent units based on wind speed correlation and electrical distance. The parameters of the equivalent units are determined by a combination of capacity weighting and parameter identification. The power error between the aggregated model and the full model under steady-state conditions is within 2%, and the transient response characteristics are basically consistent.

[0168] Cluster Model Parameter Method Description

[0169] Dynamic clustering: The K-means clustering algorithm is specified, with unit output characteristics and wind speed response curves as clustering features, and the silhouette coefficient (with a value range of 0-1, the closer to 1 the better the clustering effect) is used as the effectiveness evaluation index.

[0170] Capacity weighting and parameter identification:

[0171] The capacity-weighted weight is the ratio of a single unit's capacity to the total cluster capacity. ;

[0172] Parameter identification process: Cluster the units into groups; use the sum of squared errors between the output and voltage of the group as the objective function, and use the least squares method to iteratively identify the equivalent unit parameters; the convergence criterion for iteration is that the error is <1% for three consecutive iterations.

[0173] Collector network environment correction: explicitly adopt single π-type equivalent circuit, and supplement parameter correction formula:

[0174] Temperature correction: ( For the temperature coefficient of a conductor, copper is taken as 0.00393 / ℃. (at 20℃)

[0175] Altitude correction: When the altitude is >1000m, the line insulation strength correction factor is applied. (h represents altitude).

[0176] Collector network model: A 35kV collector line model is established using a π-type equivalent circuit model. Based on the conductor type and length, the resistance, inductance, capacitance, and conductance parameters of the line are calculated. The nodal admittance matrix of the collector network is established for power flow calculation and transient analysis. The error between the power flow calculation results and actual operating data is within 1%.

[0177] Reactive power compensation device model: An SVG (Static Var Generator) is installed at the substation to establish an SVG model, simulating its reactive power regulation characteristics and voltage support capability. The SVG's response time is less than 10ms, and its reactive power regulation range is ±50Mvar. Reactive power compensation device control strategy and coordination mechanism control strategy model: A dual closed-loop control model of voltage outer loop + current inner loop is supplemented. The voltage outer loop uses a PI controller, and the parameter tuning adopts the Ziegler-Nichols method; the current inner loop uses a PR controller to achieve zero steady-state error tracking of reactive power commands. The phase-locked loop (PLL) adopts a second-order generalized integral (SOGI) structure, improving voltage distortion resistance by 30%.

[0178] Reactive power coordinated control: Establish a priority mechanism for reactive power control between SVG and generator unit: When the generator terminal voltage deviation is <±2%, the generator unit adjusts reactive power autonomously; when the deviation is ≥±2%, the SVG responds first and the generator unit assists in adjustment; under fault conditions, the SVG and the generator unit output reactive power synchronously to support the operation, with a response time ≤50ms.

[0179] 2.3 Multi-scale model coupling

[0180] Time scale division: The dynamic process of the wind farm is divided into electromagnetic transient scale (milliseconds), electromechanical transient scale (seconds), and medium-to-long-term dynamic scale (minutes). Event triggering conditions: Grid fault triggering: When the voltage drops below 70% of the rated voltage for a duration of ≥20ms, the electromagnetic transient scale simulation is triggered; after the voltage recovers to above 90% of the rated voltage for a duration of ≥100ms, the simulation switches to the electromechanical transient scale.

[0181] Converter switching action trigger: When the converter switching frequency fluctuation exceeds 10% of the rated value, the electromagnetic transient fine model is automatically activated; under steady-state conditions, it switches to the average model.

[0182] Scale switching boundary: The switching lag time between electromagnetic transient and electromechanical transient is defined as 50ms, and the switching condition between electromechanical transient and long-term simulation is 30s without voltage / power sudden change.

[0183] Multi-scale coupling strategy: An event-triggered multi-rate simulation method is adopted. When the system experiences rapid dynamic events such as grid faults or converter switching operations, the simulation automatically switches to the electromagnetic transient scale. When the system is in a steady state or the wind speed changes slowly, the electromechanical transient or medium-to-long-term dynamic scale is used for simulation. An interface conversion module enables data interaction and synchronization between models of different scales, ensuring the continuity and consistency of simulation results. Data interpolation and synchronization: Cubic spline interpolation is specified for data conversion between different scales, with a synchronization accuracy requirement of ≤1ms. Hardware clock synchronization (PTP protocol) is used to ensure the consistency of timestamps across the multi-scale modules.

[0184] Error propagation control: A Kalman filter algorithm is introduced to correct the data propagation error between scales. The error accumulation threshold is set to 2%, and data calibration is automatically triggered when the threshold is exceeded.

[0185] Smooth transition mechanism: In the 10 power frequency cycles before scale switching, the simulation step size is gradually adjusted (e.g., from 50ms to 50μs) to avoid abrupt changes in simulation results.

[0186] Step 3: Adaptive Model Optimization

[0187] 3.1 Model Accuracy Evaluation: A multi-dimensional accuracy evaluation index system was established, including steady-state accuracy indicators (voltage deviation, power deviation, power factor deviation), transient accuracy indicators (transient response time, overshoot, attenuation rate), and frequency characteristic indicators (amplitude-frequency characteristic error, phase-frequency characteristic error). By comparing the simulation results with actual operating data, the calculated steady-state voltage deviation of the model was 1.2%, the power deviation was 1.8%, and the transient response time error was 5%, all of which met the engineering requirements.

[0188] 3.2 Model Complexity Analysis: The complexity of the model was evaluated, including computational cost, memory usage, and simulation time. The simulation time of the full model was approximately 10 times the actual time, while the simulation time of the aggregated model was approximately 0.5 times the actual time, indicating a significant reduction in model complexity.

[0189] 3.3 Adaptive Model Optimization: Reinforcement learning algorithms are used to adaptively optimize the model, aiming for a comprehensive balance between accuracy and efficiency. After training, the model can automatically adjust the aggregation level and parameters according to the operating conditions. While ensuring accuracy, the simulation efficiency is improved by more than 4 times, achieving the optimal balance between accuracy and efficiency.

[0190] Algorithm selection and improvement: The Proximal Policy Optimization (PPO) algorithm is adopted, and the Priority Experience Replay (PER) mechanism is added to improve the training efficiency in complex working conditions. The algorithm improvement is to introduce a working condition adaptive learning rate, which is reduced to 0.0001 when the working condition is complex and increased to 0.001 when the working condition is steady.

[0191] Dynamic weight adjustment: Establishing a working condition complexity index ( For voltage fluctuation amplitude, (where C is the rate of change of wind speed). When C > 0.5, the accuracy weight is set to 0.7 and the efficiency weight to 0.3; when C ≤ 0.5, the accuracy weight is 0.3 and the efficiency weight is 0.7.

[0192] Training dataset: The dataset consists of 1000 sets, including 600 sets of actual wind farm operation data (covering different seasons and wind speed ranges) and 400 sets of simulation-generated fault / extreme operating condition data; the dataset labels include three categories of indicators: unit output, voltage stability, and simulation efficiency.

[0193] Adaptive adjustment logic condition complexity determination: steady-state condition is defined as wind speed fluctuation <2m / min, voltage fluctuation <2%, and unit output fluctuation <1%; transient condition is defined as sudden wind speed change ≥5m / s, voltage drop ≥10%, or unit failure.

[0194] Aggregation / deaggregation trigger: Under steady-state conditions, the cluster aggregation degree is 80% (aggregating 100 units into 20 equivalent units); under transient conditions, it automatically deaggregates to a single unit model.

[0195] Parameter adjustment constraints: The parameter deviation of the equivalent unit after aggregation is ≤10% (compared with the parameters of a single unit); the convergence criterion is that the output error between the aggregation model and the deaggregation model is <3% within three consecutive simulation cycles.

[0196] Step 4: Simulation and Calculation of Complex Working Conditions

[0197] 4.1 Construction of Complex Operating Scenarios: Extreme weather conditions (typhoon wind speed of 15 m / s, sudden wind speed change from 8 m / s to 18 m / s), power grid fault conditions (three-phase short-circuit fault lasting 0.1 s, voltage drop to 60% of rated voltage), and generator unit fault conditions (generator stator winding short-circuit fault) are constructed. The extreme weather conditions include: Low-temperature icing: A correlation model between blade icing thickness and aerodynamic efficiency is established; aerodynamic efficiency decreases by 20% when icing thickness is 5 mm and by 35% when icing thickness is 10 mm; Sandstorms: Simulating sand and dust particles abrading the blade surface, reducing the lift coefficient by 10%-15% and increasing the drag coefficient by 5%-10%, and incorporating the impact of sand and dust on converter heat dissipation performance (heat dissipation efficiency decreases by 8%). Unit failure scenarios: New complex operating conditions such as IGBT open circuit fault in converter (output current distortion rate ≥20%), bearing wear fault (introducing random torque disturbance, fluctuation amplitude ≤5% of rated torque), and gearbox tooth breakage fault (torque mutation amplitude ≥30%).

[0198] 4.2 Simulation Algorithm Selection and Configuration: For grid fault and unit fault conditions, an electromagnetic transient simulation algorithm is adopted with a step size of 50μs; for extreme weather conditions, an electromechanical transient simulation algorithm is adopted with a step size of 50ms; for medium- and long-term wind speed change conditions, a medium- and long-term dynamic simulation algorithm is adopted with a step size of 5s.

[0199] 4.3 Parallel Simulation Computation: Based on a distributed computing architecture, the wind farm is divided into 10 sub-regions, with each sub-region assigned a computing node for parallel computation. Data interaction and synchronization between computing nodes are achieved through MPI, resulting in parallel computing efficiency more than 8 times higher than serial computing, enabling simulation calculations under complex operating conditions to be completed in a shorter time.

[0200] Step 5: Simulation Result Analysis and Model Validation

[0201] 5.1 Simulation Results Analysis: Time-domain analysis was used to plot the time-domain curves of power, voltage, and current parameters of the wind farm under complex operating conditions, analyzing its dynamic response characteristics. Under typhoon conditions, the power output of the wind farm fluctuated significantly, but through control strategy adjustments, the power fluctuation range was controlled within ±10%. Under grid fault conditions, the wind farm maintained grid-connected operation, and the power output quickly returned to normal levels after the fault was cleared. Frequency-domain analysis was used to perform FFT analysis on the power signal to identify the main frequency components and assess the power quality of the wind farm. Modal analysis was used to calculate the system's eigenvalues ​​and mode shapes, evaluating the system's stability. The results show that the wind farm has good stability under complex operating conditions.

[0202] 5.2 Model Validation: Static validation was performed, comparing the simulation results with actual operating data under steady-state conditions for parameters such as voltage, power, and power factor. The errors were all within 2%. Dynamic validation was performed, comparing the simulation results with fault recording data under transient conditions for transient response characteristics. The waveform similarity reached over 95%. Extreme condition validation was performed, and the simulation results under typhoon conditions were basically consistent with the theoretical analysis and field test data, verifying the robustness of the model.

[0203] 5.3 Model Correction and Update: Based on the validation results, some parameters of the model are adjusted to further reduce simulation errors. A model update mechanism is established, with new operational data collected quarterly to retrain and validate the model, ensuring its timeliness and accuracy.

[0204] Based on the above methods, a high-precision electrical model simulation system for multi-scale coupled wind farms was developed. This system adopts a client / server architecture, with a high-performance computing cluster on the server side and a Windows operating system on the client side. The following are the main module functions and implementation details of the system:

[0205] Data acquisition and preprocessing module

[0206] This module, developed in Java, communicates with sensors, SCADA systems, and power grid dispatching systems in wind farms via OPC and MQTT protocols to collect multi-source data in real time. Data cleaning, standardization, and feature engineering functions are implemented in Python, utilizing data analysis libraries such as Pandas, NumPy, and Scikit-learn for data processing. The module runs stably and can collect and process multi-source heterogeneous data accurately and in real time.

[0207] Multi-scale model building module

[0208] This module, developed in C++, integrates the Matlab / Simulink model library for building refined component-level models of wind farms. The cluster model is implemented in Python, incorporating machine learning algorithms for dynamic clustering and adaptive aggregation. The multi-scale coupling module, developed in C, enables the organic fusion and data interaction of models across different time scales. The module provides a visual model editing interface, allowing users to quickly build and modify wind farm models.

[0209] Adaptive Model Optimization Module

[0210] This module is developed in Python and integrates deep learning frameworks such as TensorFlow and PyTorch to achieve adaptive model optimization based on reinforcement learning. Accuracy evaluation and complexity analysis functions are implemented in Matlab, using scripts to calculate various evaluation metrics. The module provides a visualization interface for model optimization, allowing users to view the accuracy and complexity change curves, as well as the changes in reward values ​​during the optimization process.

[0211] Simulation Calculation Module

[0212] This module is developed in C++ and uses MPI for distributed parallel computing. The operating scenarios are stored in XML format, and users can edit and select scenarios through the interface. The simulation algorithm configuration module provides a variety of simulation algorithms for users to choose from and supports parameter configuration. The parallel computing module can automatically allocate computing tasks, achieving efficient parallel simulation computing and significantly shortening simulation time.

[0213] Results Analysis and Model Validation Module

[0214] This module, developed in Python, integrates visualization libraries such as Matplotlib and Seaborn to perform time-domain, frequency-domain, and modal analysis of simulation results. The model validation module implements a standardized validation process, automatically calculating various validation metrics and generating validation reports. The model update module automatically adjusts model parameters based on validation results and supports model version management and updates.

[0215] Human-computer interaction module

[0216] This module is developed using the Qt framework and provides a user-friendly interface. The interface is simple and clear, easy and quick to operate, and supports functions such as data querying, model setting, simulation control, and result display. Users can input relevant wind farm parameters, select operating scenarios, start simulation calculations, and view simulation results and analysis reports through the interface. The human-computer interaction module also supports data import and export, facilitating further analysis and processing by users.

[0217] After actual operation testing, the system can accurately and efficiently complete multi-scale coupled simulation calculations of wind farms under complex operating conditions, providing strong technical support for the operation management of wind farms and the stability assessment of power systems.

[0218] This invention has the following features and effects:

[0219] 1. Multi-scale coupled simulation: This invention divides the time scale into three time scales: electromagnetic transient, electromechanical transient and medium-to-long-term dynamic, and adopts an event-triggered multi-rate simulation method to achieve organic coupling. This solves the problem that existing methods cannot take into account both fast and slow dynamic processes, and can realistically simulate the full-time scale operation characteristics of wind farms from milliseconds to hours.

[0220] 2. Adaptive Accuracy-Efficiency Balance: This invention proposes an adaptive model optimization method based on reinforcement learning, which aims to achieve the optimal balance between model accuracy and computational efficiency. It automatically adjusts model complexity and parameters, significantly improving computational efficiency while ensuring simulation accuracy, thus resolving the contradiction between accuracy and efficiency in traditional methods.

[0221] 3. Complex operating condition simulation capability: This invention constructs a complex operating condition scenario library covering multiple dimensions such as meteorology, power grid, and generating units. It can simulate the dynamic response of wind farms under complex operating conditions such as typhoons, short circuit faults, equipment failures, and multi-field coupling, providing comprehensive technical support for power system safety and stability analysis.

[0222] 4. Standardized Model Validation System: This invention establishes a multi-dimensional accuracy evaluation index system and a standardized model validation process. Through static validation, dynamic validation, and extreme working condition validation, the credibility and reliability of simulation results are ensured.

[0223] The multi-scale coupled high-precision electrical model simulation method and system for wind farms of the present invention has significant industrial applicability and can be widely applied to fields such as wind farm planning and design, operation management, control strategy optimization, and power system stability analysis.

[0224] During the wind farm planning and design phase, the simulation method and system of this invention can be used to simulate and analyze different wind farm layouts, turbine selections, and power collection network topologies, evaluating the technical and economic indicators of various schemes and providing a scientific basis for wind farm planning and design. By simulating the operating characteristics of wind farms under different wind speeds and grid conditions, the site selection and turbine layout of wind farms can be optimized, improving wind energy capture efficiency and power generation benefits.

[0225] During the wind farm operation and management phase, the simulation method and system of this invention can monitor the wind farm's operating status in real time and evaluate and analyze its operational performance. By simulating the dynamic response of the wind farm under complex operating conditions, potential faults and operational problems can be identified in a timely manner, allowing for proactive measures to address them and prevent failures. Furthermore, the simulation results can be used to optimize the wind farm's operating strategy, adjust the turbine output and reactive power compensation device switching, thereby improving the wind farm's operating efficiency and economic benefits.

[0226] In terms of control strategy optimization, the simulation method and system of this invention can provide a test platform for the development and optimization of control strategies for wind farms. By testing different control algorithms and parameters in the simulation model, the performance and effectiveness of the control strategy can be evaluated, the optimal control strategy can be selected, and the power regulation capability, voltage support capability, and fault ride-through capability of the wind farm can be improved.

[0227] In the field of power system stability analysis, the simulation method and system of this invention can be used to evaluate the impact of wind farm integration on power system stability. By establishing a power system model that includes wind farms, the dynamic response of the system under conditions such as grid faults and load fluctuations is simulated, and the fault ride-through capability, voltage stability, and frequency stability of wind farms are analyzed. This provides a scientific basis for the planning, operation, and control of power systems, ensuring the safe and stable operation of power systems.

[0228] In summary, the multi-scale coupled high-precision electrical model simulation method and system for wind farms of the present invention has significant theoretical and engineering application value, and can provide strong technical support for the development of wind farms and the safe and stable operation of power systems.

[0229] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A high-precision electrical model simulation method for multi-scale coupled wind farms, characterized in that, Includes the following steps: Step 1: Collect multi-source data from the wind farm, and perform data cleaning, data standardization, and feature engineering to obtain a standardized dataset. The multi-source data includes meteorological data, unit operation data, power grid operation data, and fault recording data. Step 2: Based on the standardized dataset, construct a component-level refined model including wind turbine, generator, converter, and transmission system, as well as a wind farm cluster model based on dynamic clustering. Then, use an event-triggered multi-rate simulation method to couple the electromagnetic transient scale, electromechanical transient scale, and medium- and long-term dynamic scale to generate a multi-scale coupled model. Step 3: The accuracy and complexity of the multi-scale coupled model are evaluated, and an adaptive optimization method based on reinforcement learning is adopted to automatically adjust the model complexity and parameters with the goal of achieving the best combination of model accuracy and computational efficiency, thereby generating an optimized adaptive model. Step 4: Based on the optimized adaptive model, select or configure working conditions from the constructed complex working condition scenario library, and perform parallel simulation calculations using a simulation algorithm that matches the time scale to generate simulation result data; Step 5: Perform time-domain, frequency-domain, and modal analysis on the simulation results data, and standardize the model accuracy through static verification, dynamic verification, and extreme condition verification. Correct the model parameters and update the model based on the verification results.

2. The method according to claim 1, characterized in that, The component-level refined model in step 2 includes: A wind turbine aerodynamic model based on blade element momentum theory and considering tip loss, hub loss, and dynamic stall; a doubly-fed induction generator model based on the dq coordinate system and considering stator winding, rotor winding, inductance, and flux coupling; a permanent magnet synchronous generator model considering permanent magnet flux, stator winding resistance, and inductance; a converter model considering switching losses, dead zone effect, and parasitic parameters; and a transmission system model considering gearbox stiffness, damping, and rotational inertia. The wind farm cluster model includes: a unit aggregation equivalent model that uses a dynamic clustering algorithm to perform dynamic clustering based on the unit output characteristics and wind speed response curves, and determines equivalent parameters by combining capacity weighting method and least squares method parameter identification; a collector network model that uses a π-type equivalent circuit and includes temperature and altitude correction; and a reactive power compensation device model that includes dual closed-loop control of voltage outer loop and current inner loop and reactive power coordinated control.

3. The method according to claim 1, characterized in that, In step 3: The model accuracy evaluation is based on a multi-dimensional evaluation index system that includes steady-state accuracy index, transient accuracy index and frequency characteristic index. The model complexity analysis uses computational cost, memory usage, and simulation time as evaluation metrics. The adaptive optimization method based on reinforcement learning employs a deep deterministic policy gradient algorithm. Its state space is defined by the current model accuracy, complexity, and operating conditions; its action space is defined by the model's simplification / refinement operations and parameter adjustments; and the reward function R is designed as follows: ； Where R is the reward value, α and β are weighting coefficients, E is the error of the current model, Emax is the maximum permissible error, Tref is the reference simulation time, and T is the current simulation time; the accuracy weight and efficiency weight are dynamically adjusted according to the working condition complexity index.

4. The method according to claim 1, characterized in that, In step 4: The complex operating condition scenario library includes extreme weather conditions, power grid failure conditions, unit failure conditions, and multi-field coupling conditions. The simulation algorithms include: electromagnetic transient simulation using the implicit trapezoidal integration method with a step size of 10μs to 100μs; electromechanical transient simulation using the improved Euler method or Runge-Kutta method with a step size of 10ms to 100ms; and medium- and long-term dynamic simulation using the Euler method or trapezoidal integration method with a step size of 1s to 10s. The parallel simulation computation is based on a distributed computing architecture. It uses a region decomposition method to divide the wind farm into multiple sub-regions and realizes data interaction and synchronization between sub-regions through a message passing interface.

5. The method according to claim 1, characterized in that, In step 5: The modal analysis includes the calculation of eigenvalues ​​of nonlinear factors and uses the damping ratio as a stability margin quantification index. When the damping ratio is greater than 0.1, the system is considered stable, and when the damping ratio is less than 0.05, a stability warning is issued. The static verification is performed by comparing the voltage, power, and power factor errors between the simulation results and the actual operating data. The dynamic verification is carried out by comparing the transient response characteristics of simulation results and fault recording data, and the root mean square error is used as the waveform similarity evaluation index. When the root mean square error is less than 0.05, the similarity is judged to be no less than 95%. The verification samples cover 20 wind speed conditions and 5 fault types, and the number of samples is no less than 50 groups. The model correction and update include adjusting model parameters using parameter identification methods and establishing an update mechanism that regularly collects new running data to retrain and validate the model.

6. A high-precision electrical model simulation system for multi-scale coupled wind farms, used to execute the method according to any one of claims 1 to 5, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-source data from the wind farm and perform data cleaning, data standardization and feature engineering to obtain a standardized dataset. The multi-source data includes meteorological data, unit operation data, power grid operation data and fault recording data. The multi-scale model building module is used to construct a component-level refined model including wind turbine, generator, converter, and transmission system, as well as a wind farm cluster model based on dynamic clustering, based on the standardized dataset. It also uses an event-triggered multi-rate simulation method to couple the electromagnetic transient scale, electromechanical transient scale, and medium- and long-term dynamic scale to generate a multi-scale coupled model. The adaptive model optimization module is used to evaluate the accuracy and complexity of the multi-scale coupled model, and adopts an adaptive optimization method based on reinforcement learning to automatically adjust the model complexity and parameters with the goal of achieving the best combination of model accuracy and computational efficiency, thereby generating an optimized adaptive model. The simulation calculation module is used to select or configure working conditions from the constructed complex working condition scenario library based on the optimized adaptive model, and perform parallel simulation calculations using a simulation algorithm that matches the time scale to generate simulation result data. The results analysis and model verification module is used to perform time-domain, frequency-domain and modal analysis on the simulation results data, and to perform standardized verification of the model accuracy through static verification, dynamic verification and extreme working condition verification, and to correct the model parameters and update the model based on the verification results. The human-computer interaction module is connected to the data acquisition and preprocessing module, the multi-scale model construction module, the adaptive model optimization module, the simulation calculation module, and the result analysis and model verification module, respectively, and is used to provide an interactive interface for data query, model setting, simulation control, and result display.

7. The system according to claim 6, characterized in that, The multi-scale model construction module includes: The component modeling unit is used to create detailed models of wind turbines, generators, converters, and transmission systems. The converter model includes dead zone effect and switching loss models, and the transmission system model includes a flexible shaft nonlinear torsion model. The cluster modeling unit is used to establish a wind farm cluster model that includes a unit aggregation equivalent model, a power collection network model, and a reactive power compensation device model using dynamic clustering and adaptive aggregation methods. The multi-scale coupling unit is used to achieve coupling of electromagnetic transients, electromechanical transients and medium- to long-term dynamic scales using an event-triggered multi-rate simulation method. It includes a step-size gradual smooth transition mechanism and a Kalman filter error propagation control mechanism for scale switching.

8. The system according to claim 6, characterized in that, The adaptive model optimization module includes: The accuracy evaluation unit is used to evaluate the model accuracy based on steady-state accuracy indicators, transient accuracy indicators, and frequency characteristic indicators. The complexity analysis unit is used to evaluate the computational cost, memory usage, and simulation time metrics of the model. The adaptive optimization unit is used to automatically adjust the model complexity and parameters by employing a deep deterministic gradient algorithm, combined with a priority experience replay mechanism and a working condition adaptive learning rate, with the goal of achieving the optimal balance between model accuracy and computational efficiency.

9. The system according to claim 6, characterized in that, The simulation calculation module includes: The working condition scenario construction unit is used to build a complex working condition scenario library that includes extreme weather conditions, power grid fault conditions, unit fault conditions and multi-field coupling conditions. The extreme weather conditions include typhoon, thunderstorm, low temperature icing and sandstorm conditions. The unit fault conditions include converter IGBT open circuit fault, bearing wear fault and gearbox tooth breakage fault. The simulation algorithm configuration unit is used to select and configure electromagnetic transient, electromechanical transient, or medium- to long-term dynamic simulation algorithms and step sizes according to the time scale and operating conditions. Parallel computing units are used to perform parallel simulation calculations of multiple sub-regions based on a distributed computing architecture and the region decomposition method, through a message passing interface.

10. The system according to claim 6, characterized in that, The results analysis and model validation module includes: The results analysis unit is used to perform time-domain analysis, frequency-domain analysis, and modal analysis, including nonlinear eigenvalue calculation and damping ratio evaluation, on the simulation results data. The model verification unit is used to perform static verification, dynamic verification, and extreme condition verification. The dynamic verification uses root mean square error as the waveform similarity evaluation index. The model update unit is used to adjust the model parameters according to the validation results using the parameter identification method, and to establish an update mechanism for regularly collecting new running data for retraining and validation. The human-computer interaction module supports data import and export, model version management, and verification report generation.

Images