Method and device for aggregation modeling of large wind power plant

Abstract

The invention relates to the technical field of power systems, in particular to a method and device for aggregation modeling of a large wind power plant. The method comprises the following steps: acquiring data of a target wind power plant, and extracting key parameters required for constructing a wind power plant aggregation model in the data; constructing a physical equivalence model of the target wind power plant according to the key parameters; introducing a controller model and a protection system model of the fan into the physical equivalent model to obtain a physical model framework; on a physical model framework, a neural network is introduced to carry out data-driven modeling, physical mechanism constraints are fused as regular terms, and a wind power plant aggregation model fusing physics and data is obtained. The method can solve the problems that the wind power plant aggregation equivalent model is dynamically inconsistent with a real wind power plant and the aggregation precision is poor.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method and apparatus for aggregate modeling of large-scale wind farms. Background Technology

[0002] In recent years, the global wind power industry has experienced rapid development, with installed capacity continuously expanding and wind farm capacity constantly increasing. As large-scale onshore and offshore wind farms are gradually connected to the grid, the proportion of wind power in the power system is increasing. This trend places higher demands on the safe and stable operation of the power grid. Therefore, accurate and efficient wind farm modeling and simulation methods have become an urgent need to support engineering applications such as power system steady-state analysis, fault simulation, and dispatch control. Currently, various technical approaches exist in the field of wind farm simulation modeling. On one hand, there are detailed doubly-fed induction generator (DFIG) wind turbine models. These models involve multi-timescale dynamic processes of electromagnetics and electromechanical systems, including complex control components. In large-scale wind farms, each turbine is modeled in detail. On the other hand, traditional wind farm aggregation equivalent modeling methods are also widely used. Most of these methods are either based on mechanistic simplification or on non-mechanistic identification based on specific operating conditions. Meanwhile, with the development of real-time data acquisition technologies such as wide-area measurement systems and phasor measurement units, a large amount of on-site wind farm data is now available. However, in large-scale wind farms, the detailed modeling of each doubly-fed induction generator (DFIG) wind turbine results in a massive model size and enormous computational burden, making it difficult to run in real-time on a digital simulation platform. Traditional wind farm aggregation equivalent modeling methods, due to their reliance on specific operating modes or disturbance conditions, struggle to obtain universally applicable equivalent models. This can lead to inconsistencies between the dynamic behavior of the aggregation model and that of the real wind farm, resulting in distorted dynamic response, low aggregation accuracy, and, in particular, an inability to simultaneously account for electromagnetic and electromechanical transient characteristics. Furthermore, wind farm output characteristics are influenced by multiple factors such as wind speed, wind direction, turbine aging, and control strategies, making it difficult for traditional models to reflect the actual wind farm conditions in a timely manner. Summary of the Invention

[0003] This invention provides a method and apparatus for large-scale wind farm aggregation modeling to solve the problems of inconsistency between the aggregated equivalent model of a wind farm and the dynamics of the actual wind farm, as well as poor aggregation accuracy.

[0004] In a first aspect, embodiments of the present invention provide a method for aggregated modeling of large-scale wind farms, including: Acquire data from the target wind farm and extract the key parameters required to construct a wind farm aggregation model from the data; Based on the key parameters, a physical equivalent model of the target wind farm is constructed; The controller model and protection system model of the wind turbine are introduced into the physical equivalent model to obtain the physical model framework; Based on the physical model framework, a neural network is introduced for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain a wind farm aggregation model that integrates physics and data.

[0005] In one possible implementation, after obtaining the wind farm aggregation model that integrates physics and data, the following is also included: The wind farm aggregation model is deployed to a real-time digital simulation platform to achieve dynamic real-time simulation of the wind farm.

[0006] In one possible implementation, a physical equivalent model of the target wind farm is constructed based on the key parameters, including: Based on the key parameters, and following the principle of keeping the characteristics of the wind farm to the external power grid unchanged, an electromagnetic-electromechanical model of the wind turbine is established. Multiple wind turbine electromagnetic-electromechanical models are simplified into an equivalent impedance network by connecting them to the step-up substation via a box-type transformer and collector lines. A physical equivalent model of the target wind farm is obtained by using a network.

[0007] In one possible implementation, the wind turbine electromagnetic-electromechanical model includes the electrical and mechanical equations of the generator; The electrical equations use the dq coordinate system to represent the dynamics of the stator and rotor circuits. The stator voltage equation is as follows: ; in, This represents the stator d-axis voltage in the dq coordinate system. This represents the stator q-axis voltage in the dq coordinate system. Indicates stator resistance. This represents the stator d-axis current in the dq coordinate system. This represents the stator q-axis current in the dq coordinate system. This represents the stator d-axis flux linkage in the dq coordinate system. This represents the stator q-axis flux linkage in the dq coordinate system. Indicates time, Indicates the synchronization angular frequency; The mechanical equation is: ; in, Represents the equivalent moment of inertia. Indicates the angular velocity of the fan rotor. This indicates the driving torque that the wind turbine obtains from the wind. Indicates electromagnetic output torque. This indicates the aerodynamic power of the wind turbine. Indicates air density, Indicates the swept area. Indicates the power factor. Indicates the tip speed ratio. Indicates the pitch angle. Indicates the incoming air velocity; The driving torque is: ; This indicates the driving torque.

[0008] In one possible implementation, the controller model of the wind turbine is a dual PWM converter control system model; the dual PWM converter control system model includes a Rotor-side converter control model and a Grid-side converter control model. The rotor-side converter control model adopts stator flux-oriented vector control, and adjusts the active power output by controlling the rotor q-axis current and the reactive power output by controlling the rotor d-axis current. The active power output on the stator side is determined by the stator voltage and q-axis current, and the reactive power output is determined by the d-axis current. The control model of the grid-side converter maintains a constant DC bus voltage and provides auxiliary reactive power support, and uses an outer voltage loop and an inner current loop to control the DC voltage and AC current.

[0009] In one possible implementation, the protection system model is a Crowbar protection mechanism model; The Crowbar protection mechanism model is as follows: if a fault is detected, a preset resistor is triggered to short-circuit the rotor winding and the rotor converter control is turned off; if the fault is eliminated, the rotor converter control is restarted.

[0010] In one possible implementation, the neural network is a long short-term memory network; Based on the aforementioned physical model framework, a neural network is introduced for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain a wind farm aggregation model that integrates physics and data, including: Key features affecting the dynamic output of the wind farm are identified from the key parameters, and these key features are used as the input of the long short-term memory network. The response variables of the wind farm aggregation model are used as the output of the long short-term memory network to train the wind farm aggregation model. During the training process, physical mechanism constraints are fused as regularization terms, and gradient descent optimization algorithm is used to adjust the internal parameters of the long short-term memory network. The loss function includes the sum of squares of the differences between the predicted and target values, and a penalty term for the part that violates the laws of physics.

[0011] In one possible implementation, the wind farm aggregation model is as follows: ; in, This represents the total active power output after the wind farm is aggregated. This represents the total reactive power output after the wind farm is aggregated. This represents the active power output of the wind farm calculated using a physical equivalent model. This represents the reactive power output of a wind farm calculated using a physical equivalent model. This represents the dynamic correction of active power in a wind farm caused by grid transients. This represents the dynamic correction of reactive power caused by grid transients in wind farms.

[0012] Secondly, embodiments of the present invention provide an apparatus for aggregate modeling of large-scale wind farms, comprising: The acquisition module is used to acquire data from the target wind farm and extract the key parameters required to construct the wind farm aggregation model from the data. The model building module is used to construct a physical equivalent model of the target wind farm based on the key parameters. The model building module is also used to introduce the wind turbine's controller model and protection system model into the physical equivalent model to obtain the physical model framework; The aggregation module is used to introduce a neural network for data-driven modeling on the physical model framework, and to fuse physical mechanism constraints as regularization terms to obtain a wind farm aggregation model that integrates physics and data.

[0013] Thirdly, embodiments of the present invention provide a terminal including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for aggregate modeling of large wind farms as described in the first aspect or any possible implementation thereof.

[0014] This invention provides a method and apparatus for aggregated modeling of large-scale wind farms. The method involves acquiring data from a target wind farm and extracting key parameters required for constructing an aggregated wind farm model. Based on these parameters, a physical equivalent model of the target wind farm is constructed. A controller model and a protection system model for the wind turbine are introduced into the physical equivalent model to obtain a physical model framework. A neural network is then introduced into this physical model framework for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain an aggregated wind farm model that integrates physics and data. Through the deep fusion of the physical model and data-driven approach, the dynamic response characteristics of large-scale wind farms can be reproduced. Compared to traditional equivalent models, this method exhibits higher aggregation accuracy and generalization ability under different operating conditions, and its dynamic characteristics are highly consistent with those of real wind farms. Attached Figure Description

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

[0016] Figure 1 This is a flowchart illustrating the implementation of the method for aggregate modeling of large-scale wind farms provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the physical equivalent model of a wind farm provided in an embodiment of the present invention; Figure 3 (1) is a schematic diagram of the positive sequence voltage waveform provided in the embodiment of the present invention; Figure 3 (2) is a schematic diagram of the active power waveform provided in the embodiment of the present invention; Figure 3 (3) is a schematic diagram of reactive power waveform provided in the embodiment of the present invention; Figure 4 This is a schematic diagram of general active and reactive power waveforms provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the device for large-scale wind farm aggregation modeling provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a device for large-scale wind farm aggregation modeling provided in another embodiment of the present invention; Figure 7 This is a schematic diagram of the terminal provided in an embodiment of the present invention. Detailed Implementation

[0017] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0019] Figure 1 A flowchart illustrating the implementation of a method for aggregate modeling of large-scale wind farms, as provided in this embodiment of the invention, is detailed below: Step 101: Obtain data from the target wind farm and extract the key parameters required to build the wind farm aggregation model from the data.

[0020] Optionally, the data acquired from the target wind farm includes: operational data and wide-area measurement information. The operational data may include historical data from Supervisory Control and Data Acquisition (SCADA) of each wind turbine in the wind farm and wind speed and direction data provided by the meteorological tower. The wide-area measurement information may include dynamic phasor data of grid connection point voltage and current provided by the Wide Area Measurement System (WAMS) or the Phasor Measurement Unit (PMU).

[0021] SCADA is an automation system used for industrial process control and monitoring. It is widely used in fields such as power, water conservancy, oil and gas, and transportation to realize remote monitoring, data acquisition, and control functions for distributed equipment.

[0022] Both WAMS and PMU are key technological devices or systems used for real-time monitoring and analysis in power systems. PMU is a core component of WAMS, which measures voltage, current, and other data of the power system synchronously with high precision. WAMS integrates data from multiple PMUs to achieve real-time monitoring, dynamic analysis, and stability control of the operating status of a large-scale power network.

[0023] Optionally, the acquired data needs to be preprocessed and cleaned to provide a data foundation for building a highly accurate and reliable wind farm aggregation model. Here, we select representative operating conditions and fault scenarios and obtain the dynamic response curves of the total active and reactive power of the wind farm under these scenarios. The selected representative operating conditions can include: normal operation and sudden wind speed changes, and the faults can include: three-phase short-circuit faults, single-phase grounding faults, etc.

[0024] The extracted data is then processed by removing outliers, synchronizing the time scale, and considering the impact of the wind turbine wake effect on the effective wind speed of each unit, forming training and validation datasets for modeling.

[0025] In one embodiment, we describe in detail two scenarios: an offshore wind farm modeling scenario and an onshore centralized wind farm modeling scenario.

[0026] Specifically, in the offshore wind farm modeling scenario, a) measured and predicted wind resource data are collected, such as wind speed and direction data obtained from weather towers and lidar. Layout information on wind turbine arrangement and spacing is acquired for wake effect analysis. SCADA operation data for each turbine is also collected, including power and status data. Simultaneously, the topology and parameters of the offshore power collection system's collection lines are obtained, such as length, impedance, and capacitance. During preprocessing, data quality checks are performed considering the characteristics of the marine environment, such as removing abnormal instrument data caused by sea salt or humidity, to ensure the accuracy and reliability of wind speed and direction data. For wake effect-related data, preliminary wake influence coefficients can be calculated based on the turbine layout and prevailing wind direction, providing a basis for subsequent modeling.

[0027] b. In the scenario of modeling a centralized onshore wind farm, collect large-scale wind resource data covering complex terrain, including time series of wind speed and direction measured by multiple meteorological towers or lidar in different wind zones, as well as topographic information and atmospheric stability parameters. To integrate with the large-scale topographic wind speed model, topographic flow field simulation or numerical weather prediction results can be introduced to obtain the wind speed distribution of each region. Data from multiple wind zones needs to undergo wind speed consistency processing: wind speed data at different heights and locations are normalized to a unified height or reference condition, for example, by converting logarithmic wind profiles to the turbine hub height, and obtaining the overall effective wind speed field of the wind farm through spatial interpolation or weighted averaging. During the cleaning process, abnormal peak values ​​and missing data during downtime and maintenance are removed from the meteorological data to ensure the continuity and reliability of wind speed data for each wind zone.

[0028] After data collection and cleaning, key parameters required for the wind farm aggregation model are extracted based on the processed and effective data. Through analysis and calculation of wind turbine, wind farm, and power grid data, equivalent parameters for each component of the model are determined. For different application scenarios, key parameter extraction includes: a. In the offshore wind farm modeling scenario: Based on the operating data and design parameters of the offshore wind turbine cluster, calculate the equivalent key parameters of the aggregated wind turbine units. This includes summing the total installed capacity of the wind farm and the inertia of each turbine to determine the rated capacity and equivalent moment of inertia of the aggregated model; extracting the average air density and power coefficient curves based on the power characteristic curves of the turbines and the measured wind speed distribution to characterize the overall wind energy utilization performance. Regarding the wake effect, adjust the effective wind speed of each turbine using the wake influence coefficient calculated above, and then obtain the overall equivalent wind speed of the wind farm using a weighted average method; alternatively, group the turbines according to the degree of wake influence, extract representative wind speeds and outputs for each group, thereby assigning parameters that consider wake attenuation to the wind farm aggregated model.

[0029] Regarding the key parameters of the collector cable, the impedance and capacitance of each branch cable in the offshore collector system are combined according to the principle of equivalence, such as converting them into a cable of equal length or equivalent. This network model ensures that the internal grid parameters of the wind farm aggregation model, such as equivalent resistance, reactance, and capacitance, are consistent with the original system across the main frequency range. Optionally, iterative adjustments to the equivalent cable parameters can be made to ensure that the model's short-circuit current and other characteristics match those of the actual wind farm.

[0030] b. In the modeling scenario of centralized onshore wind farms: When extracting the aggregated model parameters of onshore wind farms, the influence of complex terrain and multiple wind zones should be fully considered. Based on the statistical distribution of wind speed in different regions, the equivalent wind speed of the entire wind farm or the characteristic wind speed values ​​of sub-regions should be calculated, and the equivalent output power parameters, such as the power corresponding to the annual average wind speed and the maximum output, should be obtained in combination with the wind turbine power curves. For the case of multiple wind zones, the equivalent wind speed and power can be calculated separately for each sub-region, and then the total equivalent wind speed and power of the entire farm can be synthesized according to the regional installed capacity weight; at the same time, unified key control parameters, such as cut-in wind speed, cut-out wind speed and rated wind speed, should be extracted to maintain the consistency of wind turbine control throughout the farm. In terms of mechanical and electrical parameters, the total equivalent moment of inertia and damping coefficient of all wind turbines should be aggregated, and the main parameters of the step-up transformer and collector lines, such as impedance and turns ratio, should be obtained. If the geographical span is large, the equivalent impedance of multiple feeders can be calculated according to the principle of equivalence. By identifying parameters from historical operational data, theoretically calculated parameter values ​​can be corrected. For example, the equivalent inertia or damping can be adjusted based on the measured dynamic response to improve model accuracy.

[0031] Step 102: Construct a physical equivalent model of the target wind farm based on the key parameters.

[0032] After obtaining the key parameters, a dynamic real-time aggregation model of a large-scale wind farm can be constructed based on these parameters. The model design follows the principle of keeping the characteristics of the wind farm to the external power grid equivalent, that is, while simplifying the multiple internal wind turbines and the collection network, ensuring that the static and dynamic characteristics such as power and voltage at the common grid connection point are consistent with the original wind farm.

[0033] In one embodiment, constructing a physical equivalent model of the target wind farm based on key parameters may include: establishing a wind turbine electromagnetic-electromechanical model based on the key parameters and following the principle of maintaining the equivalent characteristics of the wind farm's external power grid; simplifying multiple wind turbine electromagnetic-electromechanical models into an equivalent impedance by converging them into a substation network via a box-type transformer and collector lines. A network model is used to obtain the physical equivalent model of the target wind farm. The physical equivalent model of the target wind farm reflects the impedance characteristics between the wind turbine grid connection point and the point of common coupling, providing electrical connection constraints for the subsequent wind farm aggregation model.

[0034] The physical equivalent model of the target wind farm includes a single wind turbine model and a power collection network equivalent model. See also... Figure 2The diagram shows a physical equivalent model of a wind farm. An electromagnetic-electromechanical model of the wind turbine is established, using a doubly-fed induction generator (DFIG) as an example. This model includes the electrical and mechanical equations of the generator; the electrical equations use a dq coordinate system to represent the dynamics of the stator and rotor circuits, and the stator voltage equation is as follows: ; in, This represents the stator d-axis voltage in the dq coordinate system. This represents the stator q-axis voltage in the dq coordinate system. Indicates stator resistance. This represents the stator d-axis current in the dq coordinate system. This represents the stator q-axis current in the dq coordinate system. This represents the stator d-axis flux linkage in the dq coordinate system. This represents the stator q-axis flux linkage in the dq coordinate system. Indicates time, Indicates the synchronization angular frequency; The rotor equation is similar.

[0035] The mechanical part uses a second-order rotational equation: The mechanical equation is: ; in, Represents the equivalent moment of inertia. Indicates the angular velocity of the fan rotor. This indicates the driving torque that the wind turbine obtains from the wind. Indicates electromagnetic output torque. This indicates the aerodynamic power of the wind turbine. Indicates air density, Indicates the swept area. Indicates the power factor. Indicates the tip speed ratio. Indicates the pitch angle. Indicates the incoming air velocity; The driving torque is: ; This indicates the driving torque.

[0036] The following sections describe in detail the construction of physical equivalent models of target wind farms for different application scenarios.

[0037] a. In the scenario of offshore wind farm modeling: Establish a physical equivalent model of an offshore wind farm considering wake effect and collector cable network. Group the wind turbines according to the degree of wake influence, merge the wind turbines in the same group into an equivalent wind turbine unit to make the wind speed and output characteristics within the group approximately the same; multiple equivalent wind turbines are connected in parallel to the wind farm point of common coupling (PCC) through an equivalent collector network to form a multi-machine equivalent model. If a single-machine equivalent model is adopted, by setting the equivalent wind speed input and adjusting the model parameters, this single-machine model can represent the overall field comprehensive effect, including the effective wind speed attenuation caused by wake.

[0038] Regarding the model of the collector system, equalize several feeders and step-up substations inside the offshore wind farm. For example, use type equivalent circuit to represent the main submarine cable (including cable capacitance) and busbar, and connect an equivalent step-up transformer in series, so that the impedance characteristics of the physical equivalent model at the electrical port are consistent with the actual wind farm. The physical equivalent model also reserves a remote measurement and control interface, sets analog and digital measurement points to output key quantities such as total active power, total reactive power, voltage, etc., and accepts remote control commands, such as active power output setting, reactive power voltage and other control commands, so as to be docked with the offshore wind farm monitoring system and dispatching control center through the communication link during engineering deployment.

[0039] b. In the scenario of onshore centralized wind farm modeling: When constructing the aggregated model structure of an onshore large-scale wind farm, considering the geographical span and internal differences of the wind farm, a hierarchical aggregation modeling strategy can be adopted. For multiple wind zones with significant wind speed differences, an equivalent wind turbine sub-model can be established for each region. Each sub-model receives the equivalent wind speed input of the region and contains the inertia and power characteristics of the wind turbines in the corresponding region; these sub-models are aggregated to the PCC through an equivalent collector network to achieve zonal aggregation. If the wind conditions inside the wind farm are relatively uniform, it can be simplified to a single equivalent wind turbine model, with the comprehensive wind speed of the entire wind farm as the driving input.

[0040] The collector system model usually includes multiple feeders and collection stations in the onshore scenario. In the model, the main feeders are summarized into an equivalent line through topological merging, and type or T-type network can be used to represent its equivalent impedance and capacitance to ground, and an equivalent step-up transformer is connected, with the high voltage side connected to the common network node. In this physical equivalent model, monitoring and control interfaces are also set, such as the positions of the substation power factor control unit, main control system interface, etc., and are connected to the actual control system through standardized signal ports, so that the model can be integrated into the simulation environment of the on-site control link or the superior dispatching system.

[0041] Step 103, introduce the controller model and protection system model of the wind turbine into the physical equivalent model to obtain the physical model framework.

[0042] After establishing the physical equivalent model of the aggregated model, the control and protection logic of the wind farm is integrated into this physical equivalent model. This enables the physical equivalent model to not only simulate electrical and mechanical dynamics but also to correctly respond to control commands and execute regulation functions, thus fully characterizing the dynamic characteristics of the wind farm. By combining the control algorithms at the turbine and site levels with the model, the behavior of the physical equivalent model is ensured to be consistent with the actual operation and control of the wind farm.

[0043] In one embodiment, taking a doubly-fed induction generator (DFIG) wind turbine as an example, a dual-PWM converter control system model for the DFIG wind turbine is established. The controller model of the wind turbine is a dual-PWM converter control system model; the dual-PWM converter control system model includes a Rotor-side converter control model and a Grid-side converter control model. The rotor-side converter control model employs stator flux-oriented vector control, incorporating dual closed-loop regulation of rotor current for both active and reactive power output. Active power output is regulated by controlling the rotor q-axis current (i.e., adjusting the electromagnetic torque), while reactive power output is regulated by controlling the rotor d-axis current (i.e., adjusting the stator voltage or power factor). For example, the output voltage command for the rotor current q-axis regulator can be: ; in, This represents the rotor q-axis reference current, given by the active power command. This represents the measured current along the rotor's q-axis. , These represent the PI control parameters.

[0044] The active power output on the stator side is determined by the stator voltage and the q-axis current. For example, the active power output on the stator side can be approximated as: The reactive power output is determined by the d-axis current, reflecting the characteristic of the doubly-fed generator unit to independently regulate active and reactive power. The control model of the grid-side converter maintains a constant DC bus voltage and provides auxiliary reactive power support, and uses an outer voltage loop and an inner current loop to control the DC voltage and AC current.

[0045] Refer to Figures 3(1)-3(3) for the active and reactive power control strategies of the doubly fed wind turbine controller and Figure 4 The diagram shown illustrates the general active and reactive power waveforms, focusing on the dynamic response of a doubly-fed induction generator (DFIG) to active and reactive power during grid voltage dips (faults), and incorporating low-voltage ride-through control.

[0046] Among them, Figure 3 (1) is a schematic diagram of positive sequence voltage waveform, simulating the grid voltage dip fault, that is, the voltage drop segment in the middle, comparing the grid connection point voltage changes of a fully powered doubly fed wind turbine and a 1 / 3 rated power wind turbine; the key feature is that the voltage drops briefly during the fault (such as due to a short circuit fault), and the voltage recovers after the fault is cleared.

[0047] Figure 3(2) is a schematic diagram of the active power waveform. When the full-power doubly fed wind turbine is faulted, the voltage drop causes a sudden change in the stator flux linkage. The rotor-side converter needs to limit the current to avoid overcurrent. The active power drops rapidly and may even reverse for a short time, reflecting transient energy exchange. During the fault period, the active power remains at a low level. After the fault is cleared, the active power gradually recovers to the rated value. The 1 / 3 rated power wind turbine has a low active power output and a large current margin during the fault, so the active power fluctuation is smoother. During the recovery phase, the active power gradually returns to 1 / 3 of the rated value.

[0048] Figure 3 (3) is a schematic diagram of reactive power waveform. When a fully powered doubly fed wind turbine experiences a fault, the LVRT strategy is triggered: the converter prioritizes injecting reactive power support voltage into the grid, causing a rapid surge in reactive power; during the fault duration, high reactive power output is maintained (continuous support voltage); after the voltage recovers, reactive power gradually decreases and returns to steady state, such as Q≈0 at unity power factor. For a 1 / 3 rated power wind turbine, under light load, the converter already has sufficient reactive power margin, and the reactive power increment is small during a fault. Since it can output a lot of reactive power normally, there is no need for significant adjustments; the overall reactive power response is smoother.

[0049] Figure 4 The system is divided into three phases based on time: steady-state, fault, and recovery. The steady-state phase indicates a normal power grid with stable active power (P) output (full capacity or 1 / 3 of rated capacity), and reactive power (Q) adjusted according to grid demand; for example, with a unity power factor, Q≈0. During the fault phase, voltage drops trigger LVRT control: active power (P) drops rapidly due to voltage drops and converter current limiting, even experiencing brief reverse cycles, reflecting transient energy exchange; reactive power (Q) is injected to support the voltage, causing Q to rise rapidly. During the recovery phase, the fault is cleared, and the voltage recovers. Active power (P) gradually recovers to its steady-state value, i.e., from full capacity to rated capacity, and from light load to 1 / 3 of rated capacity; reactive power (Q) gradually falls back to steady-state as the voltage recovers, at which point no further large-scale voltage support is needed.

[0050] From Figure 3(1)-Figure 3(3) and Figure 4 It can be seen that the Crowbar protection mechanism model is established based on the power response of the wind turbine during the fault process: to simulate the low-voltage ride-through behavior of the wind turbine, Crowbar protection logic is added to the physical equivalent model. In one embodiment, the protection system model is a Crowbar protection mechanism model; The Crowbar protection mechanism model is as follows: if a fault is detected, a preset resistor is triggered to short-circuit the rotor winding and the rotor converter control is turned off; if the fault is cleared, the rotor converter control is restarted.

[0051] A fault can be detected using a conditional function: when When a grid fault causes the rotor current to exceed a set threshold, or when the DC bus voltage exceeds a set threshold, the Crowbar circuit is instantaneously activated, connecting a power-dissipating resistor to the rotor side to short-circuit the rotor windings, thus reducing the total impedance of the rotor circuit to [value missing]. Simultaneously, the control trigger of the rotor-side converter (RSC) is turned off to protect the RSC from overload. In the model, this is manifested as switching the equivalent circuit of the rotor winding: after triggering, the equivalent damping of the rotor circuit increases, and the rotor current decays rapidly.

[0052] After the fault is cleared, when the rotor current drops below the safe value and the DC voltage returns to normal, the Crowbar is deactivated and the RSC is put back into operation to restore normal control. The modeling of the above controllers and protections ensures that the behavior of the aggregated model during the fault dynamic process is consistent with that of the real wind turbine, especially reflecting the abrupt changes and recovery process of active and reactive power output characteristics during the fault.

[0053] The control logic of the controller model and protection system model is described in detail below for different application scenarios.

[0054] a. In the offshore wind farm modeling scenario: The turbine control and centralized control strategies of the offshore wind farm are integrated into the physical equivalent model. This includes representing the key control loops of a single turbine (such as pitch angle control, generator torque control, and low voltage ride-through protection) in the physical equivalent model in an equivalent manner. This allows the physical equivalent model to produce a regulatory response similar to that of an actual wind turbine group when encountering wind speed changes or grid disturbances in the simulation. Furthermore, station-level power and voltage control logic is introduced, such as the wind farm's active power output limit controller and reactive power voltage regulator. These control modules receive dispatch or grid commands, such as active power output setpoints, power factor, or voltage setpoints, and act on the physical equivalent model through a remote monitoring and control interface, thereby simulating the entire offshore wind farm's response to commands. With the cooperation of the remote monitoring and control interface, the model can interact with the actual monitoring system, receiving remote control signals from the onshore control center, such as emergency shutdown and power reduction commands. The model's internal control logic adjusts the output power and reactive power support, making the simulation behavior consistent with that of a real wind farm. When necessary, an environmental auxiliary control module can also be connected, such as a weather forecast-based advanced control strategy to improve the model's adaptability to special marine environments.

[0055] b. In the modeling scenario of a centralized onshore wind farm, control strategies for each wind turbine and the entire farm are embedded in the physical equivalent model of the onshore wind farm to simulate the coordinated operation of multiple wind turbines under centralized control. The single-unit control section covers aspects such as speed regulation, pitch control, and converter control for doubly-fed or direct-drive wind turbines, and is implemented in the aggregated model as an equivalent controller to ensure that the dynamic characteristics of the model's output power changing with wind speed are consistent with the real situation. At the farm level, systems such as automatic active power control and reactive power voltage control are integrated. For example, when the dispatcher issues a power command, the automatic control module in the aggregated model adjusts the output according to the equivalent ramp rate limit, so that the total power changes according to the command curve; when voltage support is required, the automatic control module in the model adjusts the equivalent reactive power output to maintain the bus voltage. In the case of multiple wind zones, coordinated control logic can also be introduced to balance the output of each zone to meet the overall control objectives, such as reducing the power of each zone proportionally to achieve power curtailment for the entire farm. All control logic is connected to the model through an interface consistent with the actual control system. After testing and calibration, its action threshold and control gain are consistent with the real controller, so that the model's response to various control commands approximates the behavior of the real wind field.

[0056] Step 104: On the physical model framework, introduce a neural network for data-driven modeling, and use physical mechanism constraints as regularization terms to obtain a wind farm aggregation model that integrates physics and data.

[0057] Based on the aforementioned physical model framework, a neural network model is introduced to perform data-driven modeling of the complex dynamics of wind farms, thereby improving the accuracy and generalization ability of the aggregate model. Optionally, a Long Short-Term Memory (LSTM) network can be used as the neural network structure. LSTM is good at capturing long-term dependencies in time series and is suitable for simulating the dynamic time-varying characteristics of wind farm output.

[0058] In one embodiment, a neural network is introduced into the physical model framework for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain a wind farm aggregation model that integrates physics and data, including: Key features affecting the dynamic output of wind farms are identified from the key parameters. These key features are used as inputs to the long short-term memory network (LSM) and the response variables of the wind farm aggregation model are used as outputs to train the LSM network. During training, physical mechanism constraints are fused as regularization terms, and gradient descent optimization algorithms are used to adjust the internal parameters of the LSM network. The loss function includes the sum of squares of the differences between the predicted and target values, and a penalty term for the part that violates the laws of physics.

[0059] Optionally, the training data in the training dataset obtained in step 101 can be used to select key features that affect the dynamic output of the wind farm as input to the LSTM, such as wind condition features of representative wind speed, wind direction or equivalent wind speed sequence; grid features, such as voltage amplitude and phase at the grid connection point, or the rate of change of voltage amplitude and phase; historical states, such as the time delay term of active or reactive power output of the wind farm.

[0060] The output of the LSTM network is set as the response variable of the wind farm aggregation model, which is the change in total active power and reactive power at the next time step.

[0061] Through the aforementioned input-output design, LSTM learns the dynamic mapping relationship between wind farms and changes in wind conditions and grid disturbances. Furthermore, during LSTM training, physical constraints are incorporated as regularization terms. This involves transforming the physical laws, principles, or empirical rules of the research field into mathematical constraints, which are then integrated into the model training through regularization terms. This forces the model output or parameter updates to conform to physical logic, thereby improving the model's physical consistency.

[0062] During training, optimization algorithms such as gradient descent are used to adjust the internal parameters of the LSTM. The adjusted internal parameters include the input gate, forget gate, output gate, and cell state weights, so that the model gradually approximates the real dynamic characteristics of the wind farm.

[0063] The loss function used during training not only considers the sum of squares of the differences between the predicted and target values, but also includes a penalty term for parts that violate physical laws. This penalizes parts that exceed the wind turbine's power limit or fail to meet power balance constraints, in order to reduce unreasonable fluctuations in the neural network's prediction results.

[0064] The part that violates the laws of physics refers to the situation in the neural network prediction results that does not conform to the physical laws, engineering constraints or mechanism characteristics related to wind farm operation. These "parts that violate the laws of physics" are included in the loss function through penalty terms, which can force the neural network prediction results to conform to the physical operation laws of wind farms, avoid unreasonable outputs that do not conform to engineering reality, such as negative power, power fluctuations exceeding the rated power, etc., thereby improving the physical consistency and reliability of the aggregation model.

[0065] Optional, for example, penalties for violating the laws of physics could include the following: 1. Power conservation constraint means that the input mechanical power and the output electrical power are conserved, which corresponds to the active power or reactive power balance equation; If the power conservation constraint is violated, the residuals of the power balance equations are added to the loss function in the solution of power flow using physical information neural networks or data-driven methods.

[0066] 2. Inertia and mechanical dynamics constraints; If the constraints of inertia and mechanical dynamics are violated, the residuals of the differential equations will be added to the loss function when the inertial dynamics equations are incorporated into the neural network as physical constraints.

[0067] 3. Voltage operating constraints; If the voltage operating constraints are violated, the deviation from the aforementioned voltage control relationship can be used as a regularization term during neural network training. For example, calculate for each sample... and will Add a loss function to make the network output satisfy... and The correspondence, where, Indicates voltage deviation. This represents the measured voltage value. This represents the voltage prediction value output by the neural network. This represents the reactive power weighting coefficient. This represents the predicted reactive power output from the neural network. This represents the measured value of reactive power.

[0068] Understandably, in power system flow calculations, such relationships are often added as additional equations to the solver. Simulation software ensures that reactive power output adjusts the voltage according to a given droop curve by adding a special equation at the regulated bus.

[0069] 4. The equivalent node voltage of the wind farm must be maintained within the specified range in order to keep the error of the reference voltage close to zero. This range can be from 0.95 pu to 1.05 pu. If the equivalent node voltage of the wind farm exceeds the above range, a penalty will be imposed. This penalty can be a set penalty parameter.

[0070] 5. Electromagnetic transient response constraints, i.e., constraints based on the differential or difference equations of the electromagnetic components; If the above constraints are violated, the residuals of the circuit's dynamic equations will be added to the loss function.

[0071] After training, an integrated wind farm model combining physics and data is obtained. In this model, the physics module provides basic mechanistic constraints, such as power limits and inertial response, while the LSTM network learns microscopic dynamic characteristics and unknown nonlinear relationships based on data, thus achieving accurate fitting. Formally, the model can be understood as the superposition of the physical model output and the neural network correction terms. The integrated wind farm model can be: ; in, This represents the total active power output after the wind farm is aggregated. This represents the total reactive power output after the wind farm is aggregated. This represents the active power output of the wind farm calculated using a physical equivalent model. This represents the reactive power output of a wind farm calculated using a physical equivalent model. This represents the dynamic correction of active power in a wind farm caused by grid transients. This represents the dynamic correction of reactive power caused by grid transients in wind farms.

[0072] In this way, the data-driven model and the physical model are organically integrated, and the aggregate dynamic behavior of the wind farm can be accurately characterized during both fault periods and steady-state conditions.

[0073] In one embodiment, after obtaining the wind farm aggregation model that integrates physics and data, the process may further include: deploying the wind farm aggregation model to a real-time digital simulation platform to achieve dynamic real-time simulation of the wind farm.

[0074] On the one hand, the differential equations of the physical model are solved by programming on a real-time digital simulation platform. Small-step numerical integration is used to capture electromagnetic transient details. On the other hand, the weights obtained from offline training of the LSTM neural network model are embedded into the model components of the real-time digital simulation platform, and a fast inference algorithm is used to ensure that the neural network calculation is completed within each simulation step.

[0075] Through the combination of the aforementioned hardware and software, the real-time digital simulation platform can accept external inputs in real time, such as real-time grid connection point voltage data or scheduling command changes from the phasor measurement unit (PMU), and drive the aggregated model to generate wind farm response output.

[0076] During simulation, by integrating real-time measurement data, the model can synchronize its state with the actual wind farm, thus forming a digital twin system: the state of the real wind farm is transmitted to the simulation model via a Wide Area Measurement System (WAMS), and the model output is continuously compared with the real measurements. This allows for adjustments to the parameters of the wind farm aggregation model or the weights of the neural network based on deviations, achieving bidirectional feedback for virtual-real calibration. This real-time simulation and digital twin capability ensures that the wind farm aggregation model maintains high fidelity in actual operation. When hardware-in-the-loop testing is required, this wind farm aggregation model can also serve as a real-time test platform for the wind farm to be integrated into relay protection, secondary control, and other systems, evaluating the impact of control strategies on the wind farm.

[0077] Optionally, the data required to connect the wind farm aggregation model to the real-time digital simulation platform includes the model's initial operating points, such as initial values ​​of output, voltage, and current under typical operating conditions, as well as the parameter configuration files required for simulation. The aforementioned offline extracted turbine and network parameters need to be converted to a format acceptable to the real-time simulation platform and imported into the platform environment. During this process, the data undergoes format and unit consistency checks, such as ensuring unit uniformity and that electrical parameters are adapted to the simulation timeline. Furthermore, the interface signal range data required for hardware-in-the-loop processing is collected, such as voltage transformer / PT and current transformer / CT ratios, to prepare for subsequent joint commissioning. All input data undergo validity verification, noise removal, and ensure that each data point read by the simulation platform accurately represents the actual system.

[0078] To accurately represent wind farm characteristics in real-time simulation, the extracted key parameters need further adjustment and encapsulation. First, according to the model types and component library formats supported by the real-time digital simulator, aggregated wind turbine parameters, such as equivalent inductance, resistance, inertia, and controller gain, are input into the simulation modeling tool. Some parameters are then fine-tuned to address potential numerical errors caused by simulation timing, such as slightly increasing the damping coefficient to ensure numerical stability. Simultaneously, the extracted collector network parameters are divided according to the requirements of the real-time digital simulation platform. The entire wind farm equivalent network can be decomposed into several sub-module parameters for operation on different computational nodes. For the hardware-in-the-loop testing section, reference values ​​and ranges for each interface parameter are determined. For example, the voltage and current ratings of the equivalent wind farm are provided to the measurement and control interface module to ensure correct conversion between simulation and hardware. Through these processes, a parameter set that can be directly called from the real-time digital simulation platform system is formed and undergoes static checks before uploading to ensure consistency with the actual wind farm operating points.

[0079] When building a physical equivalent model of a wind farm on a real-time digital simulation platform, the structure of the physical equivalent model needs to be converted into a form that can run in real time on the hardware of the real-time digital simulation platform. Using real-time simulation modeling tools, the wind turbine, power electronic converter, and equivalent network components are aggregated and built into a modular circuit model according to the extracted parameters. Control logic is embedded as an independent module, such as the wind turbine controller and wind farm power controller modules. For the multi-processor architecture of the real-time digital simulation platform, the model is divided into several sub-modules, such as a motor and control module and a grid interface module. Data is exchanged during simulation through high-speed interconnect interfaces to achieve parallel computing, meet real-time requirements, and improve data processing speed. The model structure is simplified as much as possible to reduce unnecessary branches and states; for example, insignificant oscillation modes are appropriately damped approximated to reduce the computational burden.

[0080] Each module connects to the I / O interface of the real-time digital simulation platform to connect to external hardware or software signals. For example, the wind speed signal input module acquires wind speed data from external files or a real-time sensor simulator, the control command input module connects to the hardware control console, and the measurement output module sends PCC voltage, current, and other data to the monitoring system. After the entire model is assembled on the real-time digital simulation platform, a complete simulation run verifies the correctness of the structure and the correct matching of the interfaces of each module, laying the foundation for subsequent real-time integration and debugging.

[0081] In a real-time simulation environment, interfacing the actual wind farm control system with the physical equivalent model is a crucial step. One approach is to integrate the actual wind turbine or wind farm controller hardware into a hardware-in-the-loop (HIL) test. For example, a programmable logic controller (PLC) for the plant control or a wind turbine converter control board can be connected to a real-time digital simulation platform via I / O interfaces. The physical equivalent model, calculated in real-time by the HIL, provides analog measurement signals, such as voltage, current, and speed, to the controller, which then feeds back control commands to the model for execution. This physical controller intervention verifies the compatibility between the physical equivalent model and the controller. Another approach is to implement all control logic in software on the HIL platform: utilizing the platform's built-in control blocks or porting the actual control algorithm's program code, such as C / C++ or IEC 61131-3 logic, into the simulation model, allowing the model and control to execute synchronously on the same platform. Regardless of the approach, it is essential to ensure that the execution cycle of the control loop is consistent with the simulation step size to avoid delays or calculation overshoot. After the control logic is integrated, typical control actions are tested in real-time simulation, such as switching control modes and protection actions under extreme conditions, to ensure that the equivalent control in the model reflects the function and limits of the real control system. Thus, the physical equivalent model possesses closed-loop operation capability and can correctly interact with control signals in real-time simulation.

[0082] In one embodiment, the constructed wind farm aggregation model can be verified in multiple scenarios using detailed models or measured data.

[0083] In simulation tests, the differences in response between the aggregated model and the original detailed model / actual test under different disturbances such as three-phase short circuit, single-phase fault, turbine trip, and sudden wind speed change are compared and analyzed.

[0084] During the model validation phase, historical operating data from an actual wind farm were compared with the detailed simulation model. Typical disturbance scenarios, such as three-phase short circuits, sudden wind speed changes, and turbine tripping, were selected for testing. Results showed that under sudden wind speed changes, the total active power change predicted by the aggregated model highly matched the SCADA measured data, with a root mean square error of 2.47% and a relative error peak of no more than 4%. During the simulation of a three-phase short circuit fault, the wind farm aggregated model accurately reproduced the power drop and recovery process caused by Crowbar triggering, with an active power response error of less than 3.2% and a maximum electromagnetic transient current error controlled within 5.1%, reflecting the model's accuracy in fault dynamic characteristics. In steady-state and small-disturbance tests, the relative error between the model output and the detailed model remained within 1.6%, and there was no significant phase shift in the dynamic response process, demonstrating that the wind farm aggregated model constructed in this application possesses good dynamic consistency and high-precision fitting capabilities under various operating conditions. The test results above show that the method for large-scale wind farm aggregation modeling proposed in this application not only has efficiency advantages in simplifying engineering calculations, but also has significant advantages in maintaining high-fidelity dynamic response, which can provide solid support for wind farm grid connection simulation and scheduling control.

[0085] Wind farm aggregation models can be widely used in engineering applications such as power system simulation analysis and dispatch control optimization. They replace detailed models of hundreds of wind turbines in large-scale power grid transient stability calculations on real-time power system simulation platforms, significantly reducing the simulation load while ensuring accuracy. At the dispatch control level, the model can be embedded in energy management systems to predict the wind farm's response under different dispatch commands or fault conditions in real time, providing decision support for dispatchers. Because the model integrates actual measurement data, it can be updated in real time according to wind condition changes and unit status, improving the accuracy of wind farm output prediction and sensitivity to abnormal operating conditions. This helps optimize wind farm active power output planning and reactive power / voltage control strategies, improving the economic efficiency and safety margin of power grid operation.

[0086] After thorough verification, the final step is to deploy the wind farm aggregation model in actual engineering projects and integrate it with existing monitoring and dispatching systems. This stage involves migrating the wind farm aggregation model from the development environment to the operational environment, including hardware and software installation, communication configuration, and trial operation, ensuring that the wind farm aggregation model continuously and stably supports wind farm control and grid dispatching. Deployment and integration may include the following depending on the application scenario: a. In the offshore wind farm modeling scenario: A validated wind farm aggregation model is deployed into the control and monitoring system of the offshore wind farm as a tool for real-time auxiliary analysis and control decision-making. Typically, the model's runtime environment, such as an industrial computer or edge computing device, is installed on a server in the onshore control center and connected to the offshore wind farm's SCADA system via a dedicated communication line. A remote monitoring and control interface is configured, enabling the wind farm aggregation model to subscribe to real-time wind farm data, such as turbine status, meteorological data, voltage, and current, and to receive remote control commands. For example, the wind farm aggregation model periodically acquires the real-time wind speed and power of each turbine, summarizing them into the overall dynamic status of the wind farm. When the offshore dispatch control center issues active power dispatch commands or reactive power support requirements, the wind farm aggregation model calculates the wind farm response and provides operational suggestions based on the previously embedded control logic. Key outputs from the wind farm aggregation model, such as power output predictions for the next 5 minutes and current available reserve capacity, are presented to operators through the SCADA interface to assist their decision-making. During deployment, it is crucial to ensure the reliability and network security of the communication link with the offshore wind farm to prevent data delays or packet loss from affecting the model's real-time performance. Once deployed, the level of model participation in regulation can be gradually increased during actual operation, for example, gradually transitioning from initial monitoring and assessment to partial automatic control, so as to give full play to the role of the wind farm aggregation model in improving the operational efficiency and safety of offshore wind farms.

[0087] b. In the scenario of centralized onshore wind farm modeling: Deploy a wind farm aggregation model at the onshore wind farm or its superior control center to enhance the monitoring and control capabilities of large-scale wind farm clusters. Since onshore wind farms are usually connected to the regional power grid dispatch, the deployment location of the wind farm aggregation model can be selected in the data center of the power grid dispatch sub-center, utilizing the high-speed communication network between the sub-center and the wind farm to acquire data. Install the necessary software modules for the wind farm aggregation model, such as wind farm digital simulation services, into the dispatch automation system platform, and connect the wind farm aggregation model to the dispatch data bus, enabling it to access real-time measurements, such as station output, bus voltage, and dispatch commands. For wind power bases with multiple wind zones, a separate wind farm aggregation model can be deployed for each station, and the outputs of these models can be aggregated at the dispatch end to comprehensively evaluate the regional wind power output level. In the initial stage of the wind farm aggregation model's online deployment, it mainly operates as a decision support tool, such as providing dispatchers with short-term forecasts of wind farm output and evaluations of power control schemes. Once the wind farm aggregation model demonstrates sufficient accuracy and stability, closed-loop applications can be explored. For example, the dispatch AGC system can optimize the power allocation scheme for the next time period based on model predictions and directly issue control commands with minimal human-machine interaction to the wind farms. The deployment and integration of the wind farm aggregation model requires strict verification of permissions and security settings to ensure that the dispatch system can only read information provided by the wind farm aggregation model or control wind farms with authorization, and will not interfere with existing controls due to model failures. By gradually promoting the deployment of this model within the scope of centralized onshore wind farms, unified dynamic management of high-penetration wind farm clusters can be achieved, improving the dispatch system's control over renewable energy.

[0088] Optionally, historical and real-time data from the power dispatch center's existing wide-area measurement system and station monitoring system are collected, including wind farm output records under different meteorological conditions, dynamic voltage and frequency data measured by PMUs, and dispatch instructions and response records. This data is then processed and cleaned, removing abnormal records caused by communication delays or data loss, and filling in missing information to establish a database for predictive model training and validation. Specifically, electrical state changes at wind farm grid connection points are obtained through PMU / WAMS data to assist in model parameter calibration and predictive algorithm development. The data preparation phase also includes establishing a data interface to enable real-time data from the dispatch center, such as real-time wind speed, station output, and system frequency, to be transmitted online to the aggregation model module, and to verify the accuracy and stability of the interface. This provides a high-quality data foundation for subsequent dispatch prediction and control optimization.

[0089] In dispatching applications, parameter extraction also involves specific parameters used in the wind farm aggregation model for prediction and control. Based on historical operating data, a characteristic curve from wind speed to output, i.e., the wind farm power curve, and the influence coefficient of wind direction changes on power are fitted as parameters for the dispatching prediction module. Using dynamic measurement data from WAMS's PMU, dynamic parameters such as the wind farm's equivalent inertia constant and primary frequency regulation coefficient can be estimated to reflect the wind farm's ability to support the system frequency. These dynamic parameters are obtained by analyzing the wind farm's power-frequency response curve during fault disturbances and are applied to the dynamic safety analysis model at the dispatching end. In addition, wind farm dispatching operation constraint parameters, such as ramp rate limits and minimum technical output, are extracted to consider the actual capacity of the wind farm when optimizing dispatching calculations. All extracted parameters are verified and stored in the model library on the dispatching side. Through an interface combined with real-time data, the aggregation model is updated online, for example, by periodically correcting the power curve or inertia parameters based on new data to ensure the accuracy of prediction and control.

[0090] When deploying an aggregated model on a power dispatching platform, it needs to be integrated into the analysis and control system of the dispatching center. The model structure can exist as a functional module of the dispatch automation system, for example, adding a "wind farm dynamic digital simulation" module to the energy management system or embedding a digital twin model of the wind farm in the dispatch control system. Depending on the dispatching requirements, the aggregated model can be implemented using a continuous state-space model or a discrete-time algorithm to ensure real-time operation on the dispatching server and synchronization with the actual system. The model's structural interface needs to interface with the power grid dispatching model: in the power grid topology, the wind farm aggregated model is treated as a special generation unit connected to the corresponding node, and the active and reactive power outputs of the model directly participate in dispatching calculations, such as power flow distribution and estimated frequency regulation effects. Simultaneously, through a data bus or middleware, the model obtains real-time external inputs, such as the latest wind speed forecast, current wind farm output, system frequency, and voltage, and sends the predicted output and regulation response curves calculated by the model back to the dispatching decision module. To achieve this, a standard data interaction interface needs to be defined to ensure that each port of the model is correctly connected to the database of the dispatching center's SCADA / WAMS and other systems. The model architecture deployment also considers fault tolerance and security mechanisms, such as running shadow models redundantly on the scheduling server to ensure that when one model fails, the backup model can seamlessly take over, thus reliably integrating into the power dispatching and operation system in engineering.

[0091] When deploying the aggregated model in the dispatch center, it is necessary to integrate the model with dispatch commands and control algorithms in conjunction with the dispatch control process. The model should contain a control module reflecting the response characteristics of the wind farm, such as a simplified model simulating AGC / AVC control of a wind farm: when the dispatch center issues an active power output adjustment plan, the model generates a future output change curve, i.e., a dispatch response curve, through its internal control logic, reflecting the process of the wind farm gradually adjusting to the target power. If the system frequency deviates, the model calculates the power change of the wind farm participating in the frequency response based on the equivalent primary frequency regulation coefficient, reflecting the wind farm's support role in frequency control. In the dispatch optimization process, the model can be bidirectionally coupled with the dispatch algorithm: on the one hand, it receives wind farm output references or constraints given by the dispatch optimization program; on the other hand, it calculates feasible output trajectories or reserve capacity through control logic and feeds them back to the optimization program. In implementation, these control modules are embedded into the dispatch center's software platform through standard interfaces, for example, as components of the dispatch decision support system, participating in simulation calculations together with other power source models and load models. Through linkage with real-time dispatch data, the model can automatically execute dispatch commands and submit the results for dispatchers' reference, realizing intelligent control simulation of the wind farm. The integration of this control logic ensures that the aggregation model can not only predict wind farm trends, but also simulate the dynamic response of wind farms under control decisions at the dispatch level, providing an effective basis for grid regulation.

[0092] The aggregation model is validated on the dispatching side, focusing primarily on prediction accuracy and the feasibility of the control scheme. First, historical wind speed and power output data for a given period are input into the aggregation model's prediction module on the dispatching side to generate wind farm power predictions for the corresponding period. These predictions are compared with actual power output, and error indices, such as root mean square error and maximum deviation, are calculated to assess the model's prediction accuracy. If large errors are found under certain specific conditions, the reasons are analyzed, which may be due to the model not considering special meteorological changes or parameter aging. Correction factors or model enhancements are then introduced to improve prediction performance. Second, in terms of control optimization, the model is used in a simulated dispatching process: for example, a daily dispatching plan is selected, and the model outputs dispatching response curves and frequency regulation response curves based on real-time wind speed and frequency data. The dispatching center then uses the information provided by the model to perform grid safety checks and optimization adjustments. The differences between the dispatching schemes with and without this model are compared to verify the effectiveness of the model's support in optimizing control. During this process, the data interaction between the model and the dispatch automation system is checked for accuracy, ensuring that predicted values ​​and response curves are accurately transmitted to the dispatching control interface. After the above tests and verifications, the reliability and effectiveness of the aggregation model in the scheduling center have been confirmed, and it can be put into actual operation.

[0093] A dynamic aggregation model of wind farms has been officially launched at the power dispatch center, providing real-time decision support and automatic adjustment methods for power grid dispatch operations. The model is typically deployed in the dispatch technical support system or a new energy cloud platform, achieving integration with actual dispatch operations through data interface with the dispatch master station system. During each dispatch cycle, the model obtains the latest wind farm operating status and system conditions from WAMS and SCADA, such as current wind farm output, key node voltage and frequency, and short-term wind forecasts. The model instantly calculates and updates the wind farm output prediction curve and adjustability margin, and generates corresponding dispatch response curves and control schemes when dispatch commands are issued. Dispatchers can view the suggested output curves and frequency response capabilities provided by the model on the energy management system interface and make decisions based on safety constraints. When an emergency frequency or voltage event occurs in the power grid, the model can also automatically assess the wind farm's response potential, allowing dispatch to take rapid measures. After necessary trial operation and training, the dispatch center can partially adopt the model's output as the basis for automatic dispatch; for example, in secondary frequency regulation control, the wind farm auxiliary frequency regulation power reference value calculated by the model can be introduced to improve frequency stability. Throughout the deployment process, emphasis was placed on interface compatibility with existing dispatching tools, such as load forecasting and security analysis software, to ensure the model integrates into the dispatching ecosystem. Simultaneously, a comprehensive operation and maintenance plan was developed, and model parameters and performance were regularly verified to ensure long-term reliable operation. Once deployed, the power dispatch center can utilize this dynamic aggregation model to achieve more accurate forecasting and optimized control of large-scale wind farms, significantly improving the grid's ability to accommodate and control renewable energy.

[0094] This invention acquires data from a target wind farm and extracts key parameters required for constructing a wind farm aggregation model. Based on these parameters, a physical equivalent model of the target wind farm is constructed. A controller model and a protection system model for the wind turbine are introduced into the physical equivalent model to obtain a physical model framework. A neural network is then introduced into this physical model framework for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain a wind farm aggregation model that integrates physics and data. This embodiment, through deep fusion of physical model and data-driven approaches, can realistically reproduce the dynamic response characteristics of large-scale wind farms. Compared to traditional equivalent models, it exhibits higher fitting accuracy and generalization ability under different operating conditions, and its dynamic characteristics are highly consistent with those of real wind farms.

[0095] In this embodiment of the invention, regarding electromagnetic transients, the wind farm aggregation model considers details such as converter control and Crowbar protection, significantly improving the accuracy of fault simulation.

[0096] This invention is applicable to the aggregated modeling of various large-scale wind farms, including onshore centralized wind farms and offshore wind farms. The physical constraints integrated into the model have clear engineering implications and can be extended and adjusted according to different wind turbine types and site topologies. Furthermore, the neural network learns data features and adaptively captures the unique dynamics of different wind farms, thus exhibiting good versatility and being portable to wind farms of different locations and sizes.

[0097] The wind farm aggregation model constructed in this embodiment of the invention has a simplified structure and efficient algorithm, and can be deployed on a real-time simulation platform to achieve real-time digital simulation of wind farms. Compared with full-detail simulation modeling each wind turbine, the computational scale is significantly reduced. This allows the simulation step size and total time to meet real-time requirements while ensuring accuracy, supporting applications such as online dynamic safety assessment and hardware-in-the-loop testing.

[0098] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0099] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0100] Figure 5 A schematic diagram of a device for aggregate modeling of large-scale wind farms provided by an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 5 As shown, the device 5 for large-scale wind farm aggregation modeling includes: an acquisition module 51, a model building module 52, and an aggregation module 53.

[0101] The acquisition module 51 is used to acquire data from the target wind farm and extract the key parameters required to build the wind farm aggregation model from the data; Model building module 52 is used to build a physical equivalent model of the target wind farm based on key parameters; The model building module 52 is also used to introduce the wind turbine's controller model and protection system model into the physical equivalent model to obtain the physical model framework; The aggregation module 53 is used to introduce a neural network for data-driven modeling on the physical model framework, and to fuse physical mechanism constraints as regularization terms to obtain a wind farm aggregation model that integrates physics and data.

[0102] In one possible implementation, see Figure 6 As shown, after obtaining the wind farm aggregation model that integrates physics and data, the processing module 54 is used for: Deploy the wind farm aggregation model onto a real-time digital simulation platform to achieve dynamic real-time simulation of the wind farm.

[0103] In one possible implementation, when the model building module 52 constructs a physical equivalent model of the target wind farm based on key parameters, it is used for: Based on key parameters and following the principle of keeping the characteristics of the wind farm to the external power grid constant, an electromagnetic-electromechanical model of the wind turbine is established. Multiple wind turbine electromagnetic-electromechanical models are simplified into an equivalent impedance network by connecting them to the step-up substation via a box-type transformer and collector lines. A physical equivalent model of the target wind farm is obtained by using a network.

[0104] In one possible implementation, the wind turbine electromagnetic-electromechanical model includes the electrical and mechanical equations of the generator; The electrical equations use the dq coordinate system to represent the dynamics of the stator and rotor circuits. The stator voltage equation is as follows: ; in, This represents the stator d-axis voltage in the dq coordinate system. This represents the stator q-axis voltage in the dq coordinate system. Indicates stator resistance. This represents the stator d-axis current in the dq coordinate system. This represents the stator q-axis current in the dq coordinate system. This represents the stator d-axis flux linkage in the dq coordinate system. This represents the stator q-axis flux linkage in the dq coordinate system. Indicates time, Indicates the synchronization angular frequency; The mechanical equation is: ; in, Represents the equivalent moment of inertia. Indicates the angular velocity of the fan rotor. This indicates the driving torque that the wind turbine obtains from the wind. Indicates electromagnetic output torque. This indicates the aerodynamic power of the wind turbine. Indicates air density, Indicates the swept area. Indicates the power factor. Indicates the tip speed ratio. Indicates the pitch angle. Indicates the incoming air velocity; The driving torque is: ; This indicates the driving torque.

[0105] In one possible implementation, the wind turbine's controller model is a dual PWM converter control system model; the dual PWM converter control system model includes a Rotor-side converter control model and a Grid-side converter control model. The rotor-side converter control model adopts stator flux-oriented vector control, and adjusts the active power output by controlling the rotor q-axis current and the reactive power output by controlling the rotor d-axis current. The active power output on the stator side is determined by the stator voltage and q-axis current, and the reactive power output is determined by the d-axis current. The control model of the grid-side converter maintains a constant DC bus voltage and provides auxiliary reactive power support, and uses an outer voltage loop and an inner current loop to control the DC voltage and AC current.

[0106] In one possible implementation, the protection system model is the Crowbar protection mechanism model; The Crowbar protection mechanism model is as follows: if a fault is detected, a preset resistor is triggered to short-circuit the rotor winding and the rotor converter control is turned off; if the fault is cleared, the rotor converter control is restarted.

[0107] In one possible implementation, the neural network is a long short-term memory network; Within the physical model framework, aggregation module 53 introduces a neural network for data-driven modeling and incorporates physical mechanism constraints as a regularization term. When obtaining the wind farm aggregation model that integrates physics and data, it is used for: Key features affecting the dynamic output of wind farms are identified from the key parameters. These key features are used as inputs to the long short-term memory network (LSM) and the response variables of the wind farm aggregation model are used as outputs to train the LSM network. During training, physical mechanism constraints are fused as regularization terms, and gradient descent optimization algorithms are used to adjust the internal parameters of the LSM network. The loss function includes the sum of squares of the differences between the predicted and target values, and a penalty term for the part that violates the laws of physics.

[0108] In one possible implementation, the wind farm aggregation model is as follows: ; in, This represents the total active power output after the wind farm is aggregated. This represents the total reactive power output after the wind farm is aggregated. This represents the active power output of the wind farm calculated using a physical equivalent model. This represents the reactive power output of a wind farm calculated using a physical equivalent model. This represents the dynamic correction of active power in a wind farm caused by grid transients. This represents the dynamic correction of reactive power caused by grid transients in wind farms.

[0109] The aforementioned device for large-scale wind farm aggregation modeling acquires data from the target wind farm through an acquisition module and extracts key parameters required for constructing the wind farm aggregation model from the data. Based on the key parameters, the model building module constructs a physical equivalent model of the target wind farm, and introduces the controller model and protection system model of the wind turbine into the physical equivalent model to obtain a physical model framework. On the physical model framework, the aggregation module introduces a neural network for data-driven modeling and integrates physical mechanism constraints as regularization terms to obtain a wind farm aggregation model that integrates physics and data. This embodiment, through the deep integration of physical model and data-driven modeling, can realistically reproduce the dynamic response characteristics of large-scale wind farms. Compared with traditional equivalent models, it has higher fitting accuracy and generalization ability under different operating conditions, and its dynamic characteristics are highly consistent with those of real wind farms.

[0110] In this embodiment of the invention, regarding electromagnetic transients, the wind farm aggregation model considers details such as converter control and Crowbar protection, significantly improving the accuracy of fault simulation.

[0111] This invention is applicable to the aggregated modeling of various large-scale wind farms, including onshore centralized wind farms and offshore wind farms. The physical constraints integrated into the model have clear engineering implications and can be extended and adjusted according to different wind turbine types and site topologies. Furthermore, the neural network learns data features and adaptively captures the unique dynamics of different wind farms, thus exhibiting good versatility and being portable to wind farms of different locations and sizes.

[0112] The wind farm aggregation model constructed in this embodiment of the invention has a simplified structure and efficient algorithm, and can be deployed on a real-time simulation platform to achieve real-time digital simulation of wind farms. Compared with full-detail simulation modeling each wind turbine, the computational scale is significantly reduced. This allows the simulation step size and total time to meet real-time requirements while ensuring accuracy, supporting applications such as online dynamic safety assessment and hardware-in-the-loop testing.

[0113] Figure 7 This is a schematic diagram of a terminal provided in an embodiment of the present invention. Figure 7 As shown, the terminal 7 in this embodiment includes: a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and executable on the processor 70. When the processor 70 executes the computer program 72, it implements the steps in the above-described methods for aggregate modeling of large wind farms, for example... Figure 1 Steps 101 to 104 are shown. Alternatively, when the processor 70 executes the computer program 72, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 5 Or the functions of each module / unit shown in Figure 6.

[0114] For example, the computer program 72 can be divided into one or more modules / units, which are stored in the memory 71 and executed by the processor 70 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 72 in the terminal 7. For example, the computer program 72 can be divided into... Figure 5 Or the modules / units shown in Figure 6.

[0115] The terminal 7 may include, but is not limited to, a processor 70 and a memory 71. Those skilled in the art will understand that... Figure 7 This is merely an example of terminal 7 and does not constitute a limitation on terminal 7. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal may also include input / output devices, network access devices, buses, etc.

[0116] The processor 70 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0117] The memory 71 can be an internal storage unit of the terminal 7, such as a hard disk or memory of the terminal 7. The memory 71 can also be an external storage device of the terminal 7, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal 7. Furthermore, the memory 71 can include both internal storage units and external storage devices of the terminal 7. The memory 71 is used to store the computer program and other programs and data required by the terminal. The memory 71 can also be used to temporarily store data that has been output or will be output.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0119] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0120] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0121] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0122] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the above embodiments of the method for aggregate modeling of various large-scale wind farms. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0125] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for aggregated modeling of large-scale wind farms, characterized in that, include: Acquire data from the target wind farm and extract the key parameters required to construct a wind farm aggregation model from the data; Based on the key parameters, construct a physical equivalent model of the target wind farm; The controller model and protection system model of the wind turbine are introduced into the physical equivalent model to obtain the physical model framework; Based on the physical model framework, a neural network is introduced for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain a wind farm aggregation model that integrates physics and data.

2. The method for aggregated modeling of large-scale wind farms according to claim 1, characterized in that, After obtaining the integrated model of wind farm physics and data, the following is also included: The wind farm aggregation model is deployed to a real-time digital simulation platform to achieve dynamic real-time simulation of the wind farm.

3. The method for aggregated modeling of large-scale wind farms according to claim 1, characterized in that, Based on the key parameters, a physical equivalent model of the target wind farm is constructed, including: Based on the key parameters, and following the principle of keeping the characteristics of the wind farm to the external power grid unchanged, an electromagnetic-electromechanical model of the wind turbine is established. Multiple wind turbine electromagnetic-electromechanical models are simplified into an equivalent impedance network by connecting them to the step-up substation via a box-type transformer and collector lines. A physical equivalent model of the target wind farm is obtained by using a network.

4. The method for aggregated modeling of large-scale wind farms according to claim 3, characterized in that, The electromagnetic-electromechanical model of the wind turbine includes the electrical and mechanical equations of the generator; The electrical equations use the dq coordinate system to represent the dynamics of the stator and rotor circuits. The stator voltage equation is as follows: ; in, This represents the stator d-axis voltage in the dq coordinate system. This represents the stator q-axis voltage in the dq coordinate system. Indicates stator resistance. This represents the stator d-axis current in the dq coordinate system. This represents the stator q-axis current in the dq coordinate system. This represents the stator d-axis flux linkage in the dq coordinate system. This represents the stator q-axis flux linkage in the dq coordinate system. Indicates time, Indicates the synchronization angular frequency; The mechanical equation is: ; in, Represents the equivalent moment of inertia. Indicates the angular velocity of the fan rotor. This indicates the driving torque that the wind turbine obtains from the wind. Indicates electromagnetic output torque. This indicates the aerodynamic power of the wind turbine. Indicates air density, Indicates the swept area. Indicates the power factor. Indicates the tip speed ratio. Indicates the pitch angle. Indicates the incoming air velocity; The driving torque is: ; This indicates the driving torque.

5. The method for aggregated modeling of large-scale wind farms according to claim 1, characterized in that, The controller model of the wind turbine is a dual PWM converter control system model; the dual PWM converter control system model includes a Rotor-side converter control model and a Grid-side converter control model; The rotor-side converter control model adopts stator flux-oriented vector control, and adjusts the active power output by controlling the rotor q-axis current and the reactive power output by controlling the rotor d-axis current. The active power output on the stator side is determined by the stator voltage and q-axis current, and the reactive power output is determined by the d-axis current. The control model of the grid-side converter maintains a constant DC bus voltage and provides auxiliary reactive power support, and uses an outer voltage loop and an inner current loop to control the DC voltage and AC current.

6. The method for aggregated modeling of large-scale wind farms according to claim 5, characterized in that, The protection system model is the Crowbar protection mechanism model; The Crowbar protection mechanism model is as follows: if a fault is detected, a preset resistor is triggered to short-circuit the rotor winding and the rotor converter control is turned off; if the fault is eliminated, the rotor converter control is restarted.

7. The method for aggregated modeling of large-scale wind farms according to any one of claims 1-6, characterized in that, The neural network is a long short-term memory network; Based on the aforementioned physical model framework, a neural network is introduced for data-driven modeling, and physical mechanism constraints are fused as regularization terms to obtain a wind farm aggregation model that integrates physics and data, including: Key features affecting the dynamic output of the wind farm are identified from the key parameters, and these key features are used as the input of the long short-term memory network. The response variables of the wind farm aggregation model are used as the output of the long short-term memory network to train the wind farm aggregation model. During the training process, physical mechanism constraints are fused as regularization terms, and gradient descent optimization algorithm is used to adjust the internal parameters of the long short-term memory network. The loss function includes the sum of squares of the differences between the predicted and target values, as well as a penalty term for the part that violates the laws of physics.

8. The method for aggregate modeling of large-scale wind farms according to claim 7, characterized in that, The wind farm aggregation model is as follows: ； in, This represents the total active power output after the wind farm is aggregated. This represents the total reactive power output after the wind farm is aggregated. This represents the active power output of the wind farm calculated using a physical equivalent model. This represents the reactive power output of a wind farm calculated using a physical equivalent model. This represents the dynamic correction of active power in a wind farm caused by grid transients. This represents the dynamic correction of reactive power caused by grid transients in wind farms.

9. A device for aggregated modeling of large-scale wind farms, characterized in that, include: The acquisition module is used to acquire data from the target wind farm and extract the key parameters required to construct the wind farm aggregation model from the data. The model building module is used to construct a physical equivalent model of the target wind farm based on the key parameters. The model building module is also used to introduce the wind turbine's controller model and protection system model into the physical equivalent model to obtain the physical model framework; The aggregation module is used to introduce a neural network for data-driven modeling on the physical model framework, and to fuse physical mechanism constraints as regularization terms to obtain a wind farm aggregation model that integrates physics and data.

10. A terminal, comprising a memory and a processor, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the method for aggregate modeling of large wind farms as described in any one of claims 1 to 8.

Images