Method and device for primary frequency modulation of wind power plant
By using LSTM models and multi-objective solution functions, combined with electromagnetic torque and rotor kinetic energy regulation, the problems of adaptability and synergistic optimization in the primary frequency regulation method of wind farms are solved, thereby improving both economy and control accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA RESOURCES NEW ENERGY INVESTMENT CO LTD SHANXI BRANCH
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-08
AI Technical Summary
In existing technologies, the primary frequency regulation method for wind farms has poor adaptability to the fluctuation of wind power output and the time-varying nature of load, and its ability to coordinate and optimize economic dispatch objectives and frequency regulation performance is insufficient, making it difficult to achieve dynamic balance of multiple objectives.
A prediction model for wind power output and grid frequency changes is constructed using a long short-term memory network (LSTM). Combined with a multi-objective solution function, the primary frequency regulation of the wind farm is achieved through electromagnetic torque regulation and rotor kinetic energy regulation, thereby optimizing economic efficiency and control accuracy.
While ensuring economic efficiency, the control accuracy of primary frequency regulation in wind farms has been improved, adapting to dynamic processes under complex operating conditions and achieving synergistic optimization of economic dispatch objectives and frequency regulation performance.
Smart Images
Figure FT_1 
Figure FT_2 
Figure SMS_1
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system control technology, and in particular to a method and apparatus for primary frequency regulation of a wind farm. Background Technology
[0002] With the increasing penetration rate of wind power, the participation of wind farms in grid frequency regulation has become a key requirement for ensuring the frequency stability of the power system. Primary frequency regulation, as the first line of defense for grid frequency stability, is an inertial response at the millisecond to second level. Its goal is to suppress frequency abrupt changes and slow down the rate of frequency change by rapidly releasing / absorbing rotor kinetic energy, thereby maintaining system frequency stability.
[0003] Currently, primary frequency regulation in wind farms is mostly based on traditional numerical optimization methods, such as dynamic programming and genetic algorithms, which have the following shortcomings: First, traditional simulation methods are poorly adaptable to the fluctuations in wind power output and the time-varying nature of loads, making it difficult to characterize the dynamic process of frequency regulation under complex operating conditions; Second, the ability to coordinate the optimization of economic dispatch objectives and frequency regulation performance is insufficient, often focusing solely on minimizing cost or frequency deviation, without achieving a dynamic balance of multiple objectives.
[0004] Based on this, the present invention proposes a method and apparatus for primary frequency regulation of wind farms to solve the above-mentioned technical problems. Summary of the Invention
[0005] This invention describes a method and apparatus for primary frequency regulation in wind farms, which can meet the control accuracy requirements of primary frequency regulation in wind farms while ensuring economic efficiency.
[0006] According to a first aspect, the present invention provides a method for primary frequency regulation of a wind farm, comprising: Acquire historical data on primary frequency regulation of the wind farm; wherein, the historical data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data, and primary frequency regulation trigger signal data; The historical data is input into the wind power output and grid frequency change prediction model to obtain the wind power output prediction value and grid frequency change amplitude prediction value for a preset short period of time in the future; wherein, the wind power output prediction value includes the maximum active power support and the predicted reference output, and the wind power output and grid frequency change prediction model is constructed based on a long short-term memory network; The wind power output prediction value, the power grid frequency change amplitude prediction value, and the preset multi-objective solution function are used to obtain the corresponding wind farm frequency regulation control parameters; wherein, the wind farm frequency regulation control parameters include electromagnetic torque adjustment amount and rotor kinetic energy adjustment amount; Based on the electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount, the wind farm is subjected to primary frequency regulation.
[0007] According to a second aspect, the present invention provides an apparatus for primary frequency regulation of a wind farm, comprising: The acquisition unit is configured to acquire historical data of primary frequency regulation of the wind farm; wherein, the historical data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data, and primary frequency regulation trigger signal data; The first data processing unit is configured to input the historical data into the wind power output and grid frequency change prediction model to obtain the wind power output prediction value and grid frequency change amplitude prediction value for a preset short period of time in the future; wherein, the wind power output prediction value includes the maximum active power support and the predicted reference output, and the wind power output and grid frequency change prediction model is constructed based on a long short-term memory network; The second data processing unit is configured to obtain the corresponding wind farm frequency regulation control parameters by combining the predicted wind power output value, the predicted magnitude of the grid frequency change, and a preset multi-objective solution function; wherein the wind farm frequency regulation control parameters include electromagnetic torque adjustment amount and rotor kinetic energy adjustment amount. The third data processing unit is configured to perform primary frequency regulation on the wind farm based on the electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount.
[0008] Thirdly, embodiments of this specification also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of this specification.
[0009] Fourthly, embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.
[0010] According to the method and apparatus for primary frequency regulation of a wind farm provided by the present invention, firstly, historical data on primary frequency regulation of the wind farm is collected. The collected data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data (i.e., the difference between the actual frequency and the rated frequency), and primary frequency regulation trigger signal data (including trigger time, trigger threshold, response duration, etc.). Subsequently, the historical data is input into a wind power output and grid frequency mutation prediction model constructed based on a Long Short-Term Memory (LSTM) network. The LSTM model outputs predicted wind power output and predicted grid frequency mutation amplitude for a preset short time period (adapting to the millisecond-level response requirements of primary frequency regulation, typically 0-2 seconds). The predicted wind power output is further refined into the maximum active power support (the peak frequency regulation power that the wind farm can provide) and the predicted baseline output (the base output under normal operating conditions). Based on this, the predicted wind power output and the predicted amplitude of grid frequency fluctuations are substituted into a preset multi-objective solution function. By quantifying and balancing the primary frequency regulation operation cost of the wind farm, the rotor kinetic energy loss cost, and the grid frequency stability requirements, the optimal wind farm frequency regulation control parameters are obtained. These parameters include electromagnetic torque regulation and rotor kinetic energy regulation. The electromagnetic torque regulation is used to control the active power output of the turbines, while the rotor kinetic energy regulation is used to schedule rotor energy storage to respond to frequency regulation demands, achieving a synergistic optimization of economy and control accuracy. Finally, based on the obtained electromagnetic torque regulation and rotor kinetic energy regulation, real-time control commands are issued to each turbine through the wind farm SCADA control system to complete the primary frequency regulation response of the wind farm. Thus, this invention can meet the control accuracy requirements of primary frequency regulation of wind farms while ensuring economic efficiency. Attached Figure Description
[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 A flowchart illustrating a method for primary frequency regulation in a wind farm according to one embodiment is shown; Figure 2 A schematic block diagram of a wind farm primary frequency regulation apparatus according to one embodiment is shown. Detailed Implementation
[0013] The solution provided by the present invention will now be described with reference to the accompanying drawings.
[0014] Figure 1A flowchart illustrating a method for primary frequency regulation in a wind farm according to one embodiment is shown. It will be understood that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 1 As shown, the method includes: Step 100: Obtain historical data on primary frequency regulation of the wind farm; the historical data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data, and primary frequency regulation trigger signal data. Step 102: Input historical data into the wind power output and grid frequency change prediction model to obtain the wind power output prediction value and grid frequency change amplitude prediction value for the future preset short period of time; wherein, the wind power output prediction value includes the maximum active power support and the predicted baseline output, and the wind power output and grid frequency change prediction model is constructed based on the long short-term memory network; Step 104: Combine the predicted wind power output, the predicted amplitude of grid frequency fluctuations, and the preset multi-objective solution function to obtain the corresponding wind farm frequency regulation control parameters; wherein, the wind farm frequency regulation control parameters include electromagnetic torque adjustment and rotor kinetic energy adjustment. Step 106: Based on the electromagnetic torque adjustment and rotor kinetic energy adjustment, perform primary frequency regulation on the wind farm.
[0015] In this embodiment, firstly, historical data on primary frequency regulation of the wind farm is collected. This data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data (i.e., the difference between the actual frequency and the rated frequency), and primary frequency regulation trigger signal data (including trigger time, trigger threshold, response time, etc.). Subsequently, the historical data is input into a wind power output and grid frequency mutation prediction model constructed based on a Long Short-Term Memory (LSTM) network. The LSTM model outputs predicted wind power output and grid frequency mutation amplitude for a preset short time period (adapting to the millisecond-level response requirements of primary frequency regulation, typically 0-2 seconds). The predicted wind power output is further refined into the maximum active power support (the peak frequency regulation power that the wind farm can provide) and the predicted baseline output (the basic output under normal operating conditions). Based on this, the predicted wind power output and grid frequency mutation amplitude are substituted into a preset multi-objective solution function. By quantifying and balancing the wind farm's primary frequency regulation operating cost, rotor kinetic energy loss cost, and grid frequency stability requirements, the optimal wind farm frequency regulation control parameters are obtained. The wind farm frequency regulation control parameters include electromagnetic torque regulation and rotor kinetic energy regulation. The electromagnetic torque regulation is used to control the active power output of the turbines, while the rotor kinetic energy regulation is used to schedule rotor energy storage to respond to frequency regulation demands, achieving a synergistic optimization of economy and control accuracy. Finally, based on the solved electromagnetic torque regulation and rotor kinetic energy regulation, real-time control commands are issued to each turbine through the wind farm SCADA control system to complete the primary frequency regulation response of the wind farm. Thus, this invention can meet the control accuracy requirements of primary frequency regulation in wind farms while ensuring economic efficiency.
[0016] In one embodiment of the present invention, the multi-objective solution function is constructed using the following formula:
[0017] In the formula, To find the function value of a multi-objective solution function, As the first weighting coefficient, For the cost of one frequency regulation operation, This is the second weighting coefficient. Cost of rotor kinetic energy loss, The third weighting coefficient, It is the fourth weighting coefficient. This is a safety penalty item for the crew. This is the rotor kinetic energy adjustment amount. This is the electromagnetic torque adjustment amount. The unit active power support energy consumption cost coefficient. For the total active power support, The maximum supporting power loss coefficient, For maximum active power support, For the final rotational speed, For the unit's transmission efficiency, To predict the baseline output, For frequency power response coefficient, This is the predicted value of the amplitude of the power grid frequency change. The cost coefficient per unit of kinetic energy loss. The initial rotational speed, This represents the rotor's moment of inertia.
[0018] In this embodiment, the invention innovatively combines economy (operating cost + rotor kinetic energy loss cost), stability (maximum frequency deviation of the power grid), and safety (risk of unit parameter exceeding limits) to adapt to the subjective priority requirements of power grid dispatching. It can be dynamically adjusted according to the real-time operating conditions of the power grid (e.g., increasing the stability weight when frequency fluctuations are severe). Increase safety weight when the unit approaches the safety threshold. This function truly achieves dynamic equilibrium optimization across multiple objectives, rather than simply superimposing them. Traditional optimization functions often perform ex-post optimization based on real-time operating conditions, resulting in a response lag behind frequency regulation requirements, and the correlation between control variables and optimization objectives lacks quantitative support. This function combines the output of an LSTM prediction model, using the predicted wind power output and the predicted amplitude of grid frequency fluctuations within the next 0-2 seconds as inputs. This upgrades optimization from "responding to current operating conditions" to "predicting future scenarios and optimizing in advance," avoiding the lag inherent in traditional solutions from the outset. Simultaneously, by using quantitative formulas to directly correlate electromagnetic torque adjustment and rotor kinetic energy adjustment with optimization objectives, the optimization results can directly guide engineering execution, significantly improving control accuracy.
[0019] In one embodiment of the present invention, the unit safety penalty item is determined by the following formula:
[0020] In the formula, This is the penalty coefficient for exceeding the first speed limit. This is the first rotor safe speed boundary. This is the penalty coefficient for exceeding the second speed limit. This is the safe speed boundary for the second rotor. This is the first rotor kinetic energy adjustment amount. This is the first rotor kinetic energy adjustment boundary. The current rotor kinetic energy, This is the second rotor kinetic energy adjustment amount. This is the boundary for adjusting the kinetic energy of the second rotor.
[0021] In this embodiment, traditional optimization often treats unit safety as an external constraint, eliminating solutions upon reaching the boundary, which can easily lead to an excessively small feasible solution space and optimization deadlock. This invention innovatively uses a penalty term to apply the safety objective. The embedded function significantly increases the penalty term when the rotor speed and electromagnetic torque approach or exceed the safety threshold, directly boosting the function value of the multi-objective solution function and forcing the optimization process to proactively avoid safety risks. This objective-based constraint design ensures the engineering feasibility of the optimization results while avoiding the impact of traditional external constraints on the optimization efficiency.
[0022] In one embodiment of the present invention, the predicted wind power output, the predicted magnitude of grid frequency fluctuations, and a preset multi-objective solution function are used to obtain the corresponding wind farm frequency regulation control parameters, including: The electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount are used as the particle group, and fifty particles are randomly generated to form the initial population. The maximum number of iterations is set to one hundred. The predicted wind power output and the predicted magnitude of grid frequency fluctuations are substituted into the multi-objective solution function for iteration. For each iteration, the particle velocity and position are adjusted by dynamic inertia weight, while retaining the best particle of each generation. When the number of iterations reaches one hundred or the function value of the multi-objective solution function converges, the iteration stops and the final optimal particle is saved as the frequency regulation control parameter of the wind farm.
[0023] In this embodiment, the optimization variables and population initialization rules are first defined. The electromagnetic torque adjustment and rotor kinetic energy adjustment are combined to form particles in the algorithm, with each particle corresponding to a group of frequency regulation control schemes. To ensure the comprehensiveness of the optimization, 50 particles are randomly generated within the preset constraint boundaries (including constraints such as power support, unit operation, and response speed) to form the initial population. At the same time, the maximum number of iterations is set to 100 to balance global search capability and solution efficiency, avoiding insufficient optimization due to insufficient iterations or increased computational costs due to excessive iterations. Subsequently, the iterative optimization process is started, substituting the predicted wind power output (including maximum active power support and predicted baseline output) and the predicted amplitude of grid frequency fluctuations into the multi-objective solution function to calculate the fitness value (F) corresponding to each particle, thereby measuring the comprehensive performance of each control scheme in terms of economy, stability, and safety. To improve optimization accuracy and convergence speed, a dynamic inertia weight strategy is adopted to adjust the velocity and position of particles during the iteration process: a larger inertia weight is set in the early stage of iteration to enhance the global search capability of particles and avoid getting trapped in local optima; in the later stage of iteration, the inertia weight is gradually reduced to improve the local fine-grained search capability of particles and accelerate convergence to the optimal solution. Simultaneously, an elite retention mechanism is introduced, retaining the particle with the best fitness value after each iteration to prevent the optimal solution from being lost during iteration updates and to ensure the correctness of the optimization direction. Finally, convergence judgment and optimal parameter output are performed, and the iteration process and the fitness value changes of the multi-objective solution function are monitored in real time: when the number of iterations reaches a preset 100, or the fluctuation range of the optimal fitness value for 10 consecutive generations is ≤10⁻ 6When the function value tends to converge to a stable state, the iteration is terminated immediately. The final optimal particle saved at this time, and its corresponding electromagnetic torque adjustment and rotor kinetic energy adjustment, are the optimal wind farm frequency regulation control parameters.
[0024] In one embodiment of the present invention, when the absolute value of the predicted amplitude of the power grid frequency change is greater than the preset amplitude of the frequency change, the following operation is preferentially performed: Based on the initial virtual inertia coefficient, determine the first virtual inertial control additional power command; The first virtual inertial control additional power command is input into the wind turbine simulation model to obtain simulation data, which includes the simulation full-time frequency change rate, the simulation full-time rotor speed, and the simulation full-time frequency deviation. The simulation data is substituted into the first objective function for optimization to obtain the optimal virtual inertia coefficient corresponding to the minimum value of the first objective function. Based on the optimal virtual inertia coefficient, determine the optimal virtual inertial control additional power command; The wind turbine is controlled based on the optimal virtual inertial control additional power command.
[0025] In this embodiment, when the absolute value of the predicted amplitude of the grid frequency mutation is greater than the preset amplitude of the frequency mutation, it indicates the existence of a rate drop disturbance event or a frequency rise disturbance event. Therefore, an initial virtual inertia coefficient (within a range of 2~6s) is determined, and a first virtual inertial control additional power command is determined. Subsequently, the first virtual inertial control additional power command is imported into a preset wind turbine simulation model. The simulation model replicates the rotor dynamics characteristics of the wind turbine, the converter control logic, and the equivalent inertial response of the grid. It uses a simulation step size of 0.01s, covering the entire time period (0~t1, t1=0.5s), and outputs simulation data: the simulated full-time frequency change rate (reflecting the frequency disturbance suppression effect), the simulated full-time rotor speed (characterizing the unit's operational safety), and the simulated full-time frequency deviation (quantifying the degree of frequency deviation from the rated value). Then, the above three sets of simulation data are substituted into the first objective function for multi-constraint optimization. The first objective function has two objectives: optimal suppression of frequency change rate and minimum rotor speed fluctuation (the correlation between speed fluctuation and the total cost of wind turbine losses; minimizing speed fluctuation can minimize the total operating cost). Iterative solutions using the Particle Swarm Optimization (PSO) algorithm are employed to obtain the optimal virtual inertia coefficient corresponding to the minimum value of the first objective function. Finally, based on the optimized virtual inertia coefficient, the optimal virtual inertia control additional power command is calculated, ensuring that the command maximizes suppression of frequency change rate while strictly controlling the rotor speed within a safe range. The optimal virtual inertia control additional power command is sent to the wind turbine converter controller to adjust the release or absorption rhythm of rotor kinetic energy, achieving a balance between unit operating safety and dynamic frequency suppression.
[0026] In one embodiment of the present invention, the first virtual inertial control additional power command is determined by the following formula:
[0027] In the formula, Add power commands to the first virtual inertial control. The initial virtual inertia coefficient, The initial frequency change rate, This is the derivative of the initial rate of change of frequency.
[0028] In this embodiment, the solution equations for the optimal virtual inertial control additional power command and the first virtual inertial control additional power command are the same and will not be repeated here.
[0029] In one embodiment of the present invention, the wind turbine simulation model is constructed using the following equation:
[0030] In the formula, The moment of inertia of the wind turbine rotor. The mechanical torque of the wind turbine. The electromagnetic torque of the wind turbine generator. This refers to the damping torque of the wind turbine. The electromagnetic power of the wind turbine. To simulate rotor speed throughout the entire time period, This is the unit's basic output power. The equivalent inertia of the power grid, This is the difference in grid power. For power fluctuations in grid load, To simulate frequency deviation over the entire time period, This is the actual value of the timing frequency. The rated frequency of the power grid. To simulate the frequency change rate over the entire time period.
[0031] In this embodiment, Rotor motion equations are used to simulate the rotor dynamics characteristics of wind turbine generators. This refers to the rotor inertia of the wind turbine (a fixed mechanical parameter). The mechanical torque of the wind turbine is dynamically determined by the real-time wind speed and blade pitch angle. The electromagnetic torque of the wind turbine generator. The electromagnetic torque serves as the damping torque for the wind turbine (to suppress sudden speed changes). When the first virtual inertial control additional power command is input, the electromagnetic torque adjusts according to the power change, thereby driving the rotor speed to dynamically fluctuate and simulating the release or absorption of rotor kinetic energy. The power grid power difference simulation equation is as follows: Similarly, after inputting the first virtual inertial control additional power command, the power grid power difference changes and is adjusted to simulate the frequency change rate. After solving for the frequency change rate over the entire simulation period, the frequency change rate at time t is obtained through first-order numerical integration (such as the Euler method). , Substitute into the following equation This allows us to obtain the actual value of the timing frequency, and then solve for the frequency deviation throughout the simulation.
[0032] In one embodiment of the present invention, the first objective function is constructed by the following formula:
[0033] In the formula, The first objective function value, This is the frequency deviation penalty coefficient. This refers to the runtime of the first frequency modulation control mode. The first preset weighting coefficient, This is the second preset weighting coefficient. The cumulative penalty coefficient for speed fluctuations. The rated speed of the unit, To simulate the frequency change rate over the entire time period.
[0034] In this embodiment, the present invention innovatively nonlinearly couples the frequency dynamic suppression term with the rotor safety protection term through an exponential term. Strengthening the penalty for large frequency deviations (i.e., the frequency deviation increases slowly at small values and rapidly at large values throughout the simulation time period), and the quadratic integral term. By capturing cumulative speed fluctuations, it can quickly suppress frequency change rate to mitigate grid impacts while also considering rotor mechanical impacts and long-term stability, achieving a dynamic balance between disturbance response speed and unit operational safety (increasing unit lifespan and improving economic efficiency). Breaking through the rigid limitations of traditional fixed-weight designs, the weights are dynamically allocated based on initial frequency deviations, following the logic that the more severe the disturbance, the higher the priority for frequency stability.
[0035] Through formula , To achieve adaptive adjustment of the initial frequency deviation from disturbances, this design avoids energy waste caused by excessive suppression under small disturbance conditions while prioritizing grid frequency stability under large disturbance conditions, adapting to scenarios with varying frequency disturbance intensities. A frequency deviation penalty coefficient of 5Hz is used. -1 Offline simulation verification showed that 5Hz -1 It can effectively amplify the penalty weight when the frequency deviation is >0.1Hz, strengthen the frequency regulation under large deviation conditions, and avoid power fluctuations caused by excessive suppression under small deviation conditions, thus adapting to my country's power grid frequency deviation control standards. (Cumulative penalty coefficient for speed fluctuation) is determined in combination with the mechanical characteristics of the rotor of a 1.5MW wind turbine. This value can accurately capture the cumulative effect of small speed fluctuations, avoid long-term fluctuations from causing wear on components such as bearings and gearboxes, and at the same time, it does not affect the reasonable speed fluctuation under normal inertial response.
[0036] In one embodiment of the present invention, the operating parameters of the wind turbine include rotor speed, electromagnetic torque, pitch angle and real-time active power output.
[0037] In one embodiment of the present invention, the wind power output and grid frequency change prediction model is trained in the following manner: The preprocessed standardized time series dataset is divided into training, validation, and test sets, and the Adam optimizer is used for model training. The training set is used for model training, and the test set is used for model testing. After training is completed, the validation set is fed into the model. If the predicted wind power output or the predicted magnitude of grid frequency fluctuations does not meet the preset error, the number of hidden layer nodes and the learning rate are adjusted and the model is retrained until a qualified prediction model is obtained. In the training process of the wind power output and grid frequency change prediction model, an early shutdown strategy is adopted.
[0038] In this embodiment, the Adam optimizer is used to iteratively train the prediction model. The Adam optimizer combines the advantages of momentum gradient descent and adaptive learning rate adjustment, effectively improving training efficiency and convergence stability. During training, the learning rate is set to 0.001 and the batch size to 32, balancing training speed and parameter update accuracy. Simultaneously, to avoid a decline in generalization ability due to overtraining, an early stopping strategy is introduced during training. The loss change of the validation set is monitored in real time. When the validation set loss does not decrease for 10 consecutive rounds, training is immediately terminated to lock the model into its optimal training state and prevent overfitting. After training, the model performance is verified and parameters are optimized: the validation set data is input into the trained model to obtain the corresponding wind power output prediction value and grid frequency fluctuation amplitude prediction value, which are then compared with the actual values to calculate the prediction error. If the error does not meet the preset standards (requiring wind power output prediction error ≤5% and grid frequency fluctuation amplitude prediction error ≤3%), the model training is deemed unqualified, and the model hyperparameters need to be adjusted and retrained. Adjustments include the number of hidden layer nodes (increasing or decreasing nodes based on the error magnitude to balance fitting ability and computational efficiency) and the learning rate (appropriately reducing the learning rate for finer optimization when the error is too large). Through multiple hyperparameter adjustments and iterative training, until the model's prediction error on the validation set meets the preset requirements, the final performance verification is performed using the test set. After confirming that the model's prediction accuracy and stability meet the standards, a qualified wind power output and grid frequency fluctuation prediction model is finally obtained.
[0039] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0040] According to another embodiment, the present invention provides a device for primary frequency regulation of a wind farm. Figure 2 A schematic block diagram of a wind farm primary frequency regulation apparatus according to one embodiment is shown. It will be understood that this apparatus can be implemented using any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the device includes: an acquisition unit 200, a first data processing unit 202, a second data processing unit 204, and a third data processing unit 206. The main functions of each component are as follows: The acquisition unit 200 is configured to acquire historical data of primary frequency regulation of the wind farm; wherein, the historical data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data, and primary frequency regulation trigger signal data; The first data processing unit 202 is configured to input the historical data into the wind power output and grid frequency change prediction model to obtain the wind power output prediction value and grid frequency change amplitude prediction value for a preset short period of time in the future; wherein, the wind power output prediction value includes the maximum active power support and the predicted reference output, and the wind power output and grid frequency change prediction model is constructed based on a long short-term memory network. The second data processing unit 204 is configured to obtain the corresponding wind farm frequency regulation control parameters by taking the predicted wind power output value, the predicted value of the grid frequency change amplitude, and the preset multi-objective solution function; wherein, the wind farm frequency regulation control parameters include electromagnetic torque adjustment amount and rotor kinetic energy adjustment amount; The third data processing unit 206 is configured to perform primary frequency regulation on the wind farm based on the electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount.
[0041] In one embodiment of the present invention, the multi-objective solution function is constructed by the following formula:
[0042] In the formula, To find the function value of a multi-objective solution function, As the first weighting coefficient, For the cost of one frequency regulation operation, This is the second weighting coefficient. Cost of rotor kinetic energy loss, The third weighting coefficient, It is the fourth weighting coefficient. This is a safety penalty item for the crew. This is the rotor kinetic energy adjustment amount. This is the electromagnetic torque adjustment amount. The unit active power support energy consumption cost coefficient. For the total active power support, The maximum supporting power loss coefficient, The maximum active power support is... For the final rotational speed, For the unit's transmission efficiency, Output power to the predicted reference. For frequency power response coefficient, This is the predicted value of the amplitude of the power grid frequency change. The cost coefficient per unit of kinetic energy loss. The initial rotational speed, This represents the rotor's moment of inertia.
[0043] In one embodiment of the present invention, the unit safety penalty item is determined by the following formula:
[0044] In the formula, This is the penalty coefficient for exceeding the first speed limit. This is the first rotor safe speed boundary. This is the penalty coefficient for exceeding the second speed limit. This is the safe speed boundary for the second rotor. This is the first rotor kinetic energy adjustment amount. This is the first rotor kinetic energy adjustment boundary. The current rotor kinetic energy, This is the second rotor kinetic energy adjustment amount. This is the boundary for adjusting the kinetic energy of the second rotor.
[0045] In one embodiment of the present invention, the second data processing unit 204 is configured to perform the following operations: The electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount are used as the particle group, and fifty particles are randomly generated to form the initial population. The maximum number of iterations is set to one hundred. The predicted wind power output and the predicted magnitude of the grid frequency fluctuation are substituted into the multi-objective solution function for iteration. For each iteration, the particle velocity and position are adjusted by dynamic inertia weight, while retaining the optimal particle of each generation. When the number of iterations reaches one hundred or the function value of the multi-objective solution function converges, the iteration stops and the final optimal particle is saved as the frequency regulation control parameter of the wind farm.
[0046] In one embodiment of the present invention, the device further includes a fourth data processing unit. When the absolute value of the predicted amplitude of the power grid frequency change is greater than a preset frequency change amplitude, the fourth data processing unit preferentially performs the following operation: Based on the initial virtual inertia coefficient, determine the first virtual inertial control additional power command; The first virtual inertial control additional power command is input into the wind turbine simulation model to obtain simulation data, which includes the simulation full-time frequency change rate, the simulation full-time rotor speed and the simulation full-time frequency deviation. The simulation data is then substituted into the first objective function for optimization to obtain the optimal virtual inertia coefficient corresponding to the minimum value of the first objective function. Based on the optimal virtual inertia coefficient, determine the optimal virtual inertial control additional power command; The wind turbine is controlled based on the optimal virtual inertial control additional power command.
[0047] In one embodiment of the present invention, the first virtual inertial control additional power command is determined by the following formula:
[0048] In the formula, Add a power command to the first virtual inertial control. The initial virtual inertia coefficient, The initial frequency change rate, This is the derivative of the initial rate of change of frequency.
[0049] In one embodiment of the present invention, the wind turbine simulation model is constructed using the following equations:
[0050] In the formula, The moment of inertia of the wind turbine rotor. The mechanical torque of the wind turbine. The electromagnetic torque of the wind turbine generator. This refers to the damping torque of the wind turbine. The electromagnetic power of the wind turbine. The simulated full-time rotor speed is... This is the unit's basic output power. The equivalent inertia of the power grid, This is the difference in grid power. For power fluctuations in grid load, The frequency deviation over the entire simulation period is... This is the actual value of the timing frequency. The rated frequency of the power grid. The frequency change rate over the entire simulation period is given.
[0051] In one embodiment of the present invention, the first objective function is constructed by the following formula:
[0052] In the formula, The first objective function value, This is the frequency deviation penalty coefficient. This refers to the runtime of the first frequency modulation control mode. The first preset weighting coefficient, This is the second preset weighting coefficient. The cumulative penalty coefficient for speed fluctuations. The rated speed of the unit, The frequency change rate over the entire simulation period is given.
[0053] In one embodiment of the present invention, the operating parameters of the wind turbine include rotor speed, electromagnetic torque, pitch angle and real-time active power output.
[0054] In one embodiment of the present invention, the device further includes a fifth processing unit, the fifth processing unit being configured to perform the following operations: The preprocessed standardized time series dataset is divided into a training set, a validation set, and a test set, and the Adam optimizer is used for model training; wherein, the training set is used for model training, and the test set is used for model testing; After training is completed, the validation set is fed into the model. If the predicted wind power output or the predicted magnitude of the grid frequency change does not meet the preset error, the number of hidden layer nodes and the learning rate are adjusted and the model is retrained until a qualified prediction model is obtained. In the training process of the wind power output and grid frequency change prediction model, an early shutdown strategy is adopted.
[0055] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.
[0056] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.
[0057] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0058] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.
[0059] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for primary frequency regulation in a wind farm, characterized in that, include: Acquire historical data on primary frequency regulation of the wind farm; wherein, the historical data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data, and primary frequency regulation trigger signal data; The historical data is input into the wind power output and grid frequency change prediction model to obtain the wind power output prediction value and grid frequency change amplitude prediction value for a preset short period of time in the future; wherein, the wind power output prediction value includes the maximum active power support and the predicted reference output, and the wind power output and grid frequency change prediction model is constructed based on a long short-term memory network; The wind power output prediction value, the power grid frequency change amplitude prediction value, and the preset multi-objective solution function are used to obtain the corresponding wind farm frequency regulation control parameters; wherein, the wind farm frequency regulation control parameters include electromagnetic torque adjustment amount and rotor kinetic energy adjustment amount; Based on the electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount, the wind farm is subjected to primary frequency regulation.
2. The method according to claim 1, characterized in that, The multi-objective solution function is constructed using the following formula: In the formula, To find the function value of a multi-objective solution function, As the first weighting coefficient, For the cost of one frequency regulation operation, This is the second weighting coefficient. Cost of rotor kinetic energy loss, The third weighting coefficient, It is the fourth weighting coefficient. This is a safety penalty item for the crew. This is the rotor kinetic energy adjustment amount. This is the electromagnetic torque adjustment amount. The unit active power support energy consumption cost coefficient. For the total active power support, The maximum supporting power loss coefficient, The maximum active power support is... For the final rotational speed, For the unit's transmission efficiency, Output power to the predicted reference. For frequency power response coefficient, This is the predicted value of the amplitude of the power grid frequency change. The cost coefficient per unit of kinetic energy loss. The initial rotational speed, This represents the rotor's moment of inertia.
3. The method according to claim 2, characterized in that, The unit safety penalty items are determined by the following formula: In the formula, This is the penalty coefficient for exceeding the first speed limit. This is the first rotor safe speed boundary. This is the penalty coefficient for exceeding the second speed limit. This is the safe speed boundary for the second rotor. This is the first rotor kinetic energy adjustment amount. This is the first rotor kinetic energy adjustment boundary. The current rotor kinetic energy, This is the second rotor kinetic energy adjustment amount. This is the boundary for adjusting the kinetic energy of the second rotor.
4. The method according to claim 1, characterized in that, The step of obtaining the corresponding wind farm frequency regulation control parameters by combining the predicted wind power output value, the predicted magnitude of the grid frequency fluctuation, and a preset multi-objective solution function includes: The electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount are used as the particle group, and fifty particles are randomly generated to form the initial population. The maximum number of iterations is set to one hundred. The predicted wind power output and the predicted magnitude of the grid frequency fluctuation are substituted into the multi-objective solution function for iteration. For each iteration, the particle velocity and position are adjusted by dynamic inertia weight, while retaining the optimal particle of each generation. When the number of iterations reaches one hundred or the function value of the multi-objective solution function converges, the iteration stops and the final optimal particle is saved as the frequency regulation control parameter of the wind farm.
5. The method according to claim 1, characterized in that, When the absolute value of the predicted amplitude of the power grid frequency change is greater than the preset amplitude of the frequency change, the following operation will be performed first: Based on the initial virtual inertia coefficient, determine the first virtual inertial control additional power command; The first virtual inertial control additional power command is input into the wind turbine simulation model to obtain simulation data, which includes the simulation full-time frequency change rate, the simulation full-time rotor speed and the simulation full-time frequency deviation. The simulation data is then substituted into the first objective function for optimization to obtain the optimal virtual inertia coefficient corresponding to the minimum value of the first objective function. Based on the optimal virtual inertia coefficient, determine the optimal virtual inertial control additional power command; The wind turbine is controlled based on the optimal virtual inertial control additional power command.
6. The method according to claim 5, characterized in that, The first virtual inertial control additional power command is determined by the following formula: In the formula, Add a power command to the first virtual inertial control. The initial virtual inertia coefficient, The initial frequency change rate, This is the derivative of the initial rate of change of frequency.
7. The method according to claim 6, characterized in that, The wind turbine simulation model is constructed using the following equations: In the formula, The moment of inertia of the wind turbine rotor. The mechanical torque of the wind turbine. The electromagnetic torque of the wind turbine generator. This refers to the damping torque of the wind turbine. The electromagnetic power of the wind turbine. The simulated full-time rotor speed is... This is the unit's basic output power. The equivalent inertia of the power grid, This is the difference in grid power. For power fluctuations in grid load, The frequency deviation over the entire simulation period is... This is the actual value of the timing frequency. The rated frequency of the power grid. The frequency change rate over the entire simulation period is given.
8. A device for primary frequency regulation in a wind farm, characterized in that, include: The acquisition unit is configured to acquire historical data of primary frequency regulation of the wind farm; wherein, the historical data includes wind speed data, wind turbine operating parameters, grid frequency data, frequency deviation data, and primary frequency regulation trigger signal data; The first data processing unit is configured to input the historical data into the wind power output and grid frequency change prediction model to obtain the wind power output prediction value and grid frequency change amplitude prediction value for a preset short period of time in the future; wherein, the wind power output prediction value includes the maximum active power support and the predicted reference output, and the wind power output and grid frequency change prediction model is constructed based on a long short-term memory network; The second data processing unit is configured to obtain the corresponding wind farm frequency regulation control parameters by combining the predicted wind power output value, the predicted magnitude of the grid frequency change, and a preset multi-objective solution function; wherein the wind farm frequency regulation control parameters include electromagnetic torque adjustment amount and rotor kinetic energy adjustment amount. The third data processing unit is configured to perform primary frequency regulation on the wind farm based on the electromagnetic torque adjustment amount and the rotor kinetic energy adjustment amount.
9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.