Cross-scale digital twinning mixed operation method and system for network construction type wind turbine generator

Abstract

The invention provides a cross-scale digital twinning mixed operation method and system for a network construction type wind turbine generator, and relates to the technical field of wind turbine generators, a network construction type wind turbine generator digital twinning model comprising a refined converter model and an average value converter model is built, and a direct current bus current is introduced to construct a dynamic threshold triggering mechanism; when the instantaneous change rate exceeds a dynamic threshold value, the converter is triggered to be switched from an average value model to a refined model, and state synchronous compensation is carried out; and when the instantaneous change rate is lower than a dynamic threshold value and exceeds a preset duration, triggering switching from the refined model to the average value model, and performing state average value reinjection. Therefore, the hybrid modeling method based on multiple time scales realizes model switching through a dynamic threshold triggering mechanism, ensures data continuity and smooth transition based on a state synchronous compensation and state mean value reinjection strategy, improves calculation efficiency, and reduces tracking errors.

Description

Technical Field

[0001] The present application relates to the technical field of wind turbines, and more specifically, to a cross-scale digital twin hybrid operation method and system for a grid-type wind turbine. Background Art

[0002] As the penetration of renewable energy power systems continues to increase, digital twin technology is becoming increasingly important in wind turbine condition monitoring and fault warning. In the construction of digital twins for permanent magnet direct-drive wind turbines, the converter, as the core power conversion unit, has a direct impact on the reliability of transient response analysis.

[0003] Currently, practical projects often use an average value model to model the generator-side and grid-side converters of permanent magnet direct-drive wind turbines. This model ignores the switching behavior of power devices and represents the converter's dynamic characteristics through equivalent voltage or current sources. While this method significantly improves simulation speed by ignoring the high-frequency behavior of switching devices, it also lacks the ability to capture the dynamic characteristics of transient processes such as grid voltage sags and DC bus oscillations. This is particularly true in the virtual synchronous control of grid-connected converters, where harmonic characteristics and electromagnetic transients at the switching frequency level cannot be accurately reflected. To accurately simulate electrical characteristics during real-time simulation, a refined converter model incorporating the physical characteristics of IGBT / MOSFET switching devices is used. Converter electromagnetic transients are simulated with microsecond steps, and real-time simulation is achieved through FPGA hardware acceleration. However, when the unit is normally connected to the grid, the refined converter model requires solving differential algebraic equations within each switching cycle. The real-time simulation of a single converter requires more than 8 cores of CPU resources, resulting in a delay of more than 100ms in the digital twin of the entire wind turbine system, which cannot meet the real-time interaction requirements and will cause a waste of computing resources. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a cross-scale digital twin hybrid operation method and system for grid-type wind turbines, which realizes model switching through a dynamic threshold trigger mechanism based on a multi-time-scale hybrid modeling method, and ensures data continuity and smooth transition based on state synchronization compensation and state mean reinjection strategies, thereby improving computing efficiency and reducing tracking errors.

[0005] The embodiment of the present application provides a cross-scale digital twin hybrid operation method for a grid-type wind turbine, comprising the following steps:

[0006] Build a digital twin model of a grid-connected wind turbine, including a refined converter model and an average converter model;

[0007] Extract the synchronous rotating coordinate system component of the grid voltage and calculate its instantaneous rate of change. In addition, introduce the DC bus current to construct a dynamic threshold trigger mechanism.

[0008] Switching between the average value converter model and the refined converter model is performed based on the dynamic threshold trigger mechanism.

[0009] In some embodiments, the refined converter model is constructed based on a physical topology, including IGBT switching devices, DC bus capacitors, and PWM modulation; the average value converter model is constructed based on an equivalent circuit abstraction that ignores high-frequency switching dynamics.

[0010] In some embodiments, the instantaneous rate of change of the dq-axis components of the synchronous rotating coordinate system is calculated by the following formula:

[0011]

[0012] Where, v d (t), v q (t) is the d-axis and q-axis components of the grid voltage after decoupling at time t.

[0013] In some embodiments, the dynamic threshold is calculated using the following formula:

[0014] V th (t) = μ Δv +3σ Δv +α×|i dc (t)|

[0015] Where μ Δv is the mean value of the instantaneous change rate of the dq axis voltage, σ Δv is the standard deviation of the mean value of the instantaneous change rate of the dq axis voltage, α is the weight coefficient, i dc is the DC bus current;

[0016] The mean and standard deviation of the instantaneous rate of change of the dq axis voltage are calculated using the following formula:

[0017]

[0018] Where Δv dq (t) is the instantaneous rate of change of the dq-axis voltage at time t, and N is the number of data points.

[0019] In some embodiments, the switching between the average value converter model and the refined converter model based on the dynamic threshold trigger mechanism includes the following steps:

[0020] When the instantaneous change rate exceeds a dynamic threshold, triggering a switch from the average value converter model to the refined converter model, and performing state synchronization compensation during the switch;

[0021] When the instantaneous change rate is lower than a dynamic threshold and exceeds a preset time period, switching from the refined converter model to the average converter model is triggered, and state mean re-injection is performed during the switching.

[0022] In some embodiments, state synchronization compensation is performed in the following manner, including the following steps:

[0023] performing filtering preprocessing on the DC voltage output by the average value converter model;

[0024] The pre-processed DC voltage is decomposed into multiple layers using Daubechies 4 wavelet to separate the high- and low-frequency signal components layer by layer through low-pass and high-pass filter groups to obtain high-frequency detail coefficients.

[0025] The high-frequency detail coefficients of the selected layer are synthesized into a time-domain voltage ripple signal by using a wavelet reconstruction algorithm, and are mapped to the refined converter model through a differential operation.

[0026] In some embodiments, state mean re-injection is performed in the following manner, including the following steps:

[0027] Calculating the mean value of the key state quantity required by the average value converter model at the switching moment;

[0028] The calculated mean value is injected into the mean value converter model.

[0029] In some embodiments, the key state quantity includes a DC voltage and a dq axis current, and the mean value of the key state quantity at the switching moment is calculated by the following formula:

[0030]

[0031] Where Δt is the sampling interval and N is the number of samples.

[0032] In some embodiments, a system is also provided, including an FPGA, a DSP and a CPU, wherein the FPGA deploys the refined converter model, the DSP deploys the dynamic threshold trigger mechanism, and the CPU deploys the average value converter model, and performs real-time closed-loop interaction and data synchronization through LVDS / PCIe to execute the steps of any of the above-mentioned cross-scale digital twin hybrid operation methods of the grid-type wind turbine.

[0033] In some embodiments, the FPGA transmits grid data to the DSP via an LVDS differential interface with a transmission delay of <50ns; the CPU interacts with the FPGA via a PCIe interface with a transmission delay of <5μs; and the DSP executes the dynamic threshold trigger mechanism with a sampling step of 10μs.

[0034] The present application describes a cross-scale digital twin hybrid operation method and system for a grid-type wind turbine. The method comprises building a digital twin model of a grid-type wind turbine including a refined converter model and an average converter model. The method extracts the synchronous rotating coordinate system component of the grid voltage and calculates its instantaneous rate of change. Furthermore, the method introduces the DC bus current to construct a dynamic threshold trigger mechanism. The method switches between the average converter model and the refined converter model based on the dynamic threshold trigger mechanism. When the instantaneous rate of change exceeds the dynamic threshold, the method triggers the switch from the average converter model to the refined converter model, and performs state synchronization compensation during the switch. When the instantaneous rate of change is lower than the dynamic threshold and exceeds a preset duration, the method triggers the switch from the refined converter model to the average converter model, and performs state mean re-injection during the switch. This method utilizes the simulation accuracy of the refined converter model and the simulation efficiency of the average converter model to fundamentally ensure the real-time and accuracy requirements of digital twin modeling. In addition, a dynamic threshold is used to trigger model switching, which solves the problems of system switching lag and false triggering. The continuity and stability of the data before and after the model switching are maintained through state synchronization compensation and state mean re-injection, ensuring bidirectional and disturbance-free switching of the converter model during system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 A flow chart of a cross-scale digital twin hybrid operation method for a grid-type wind turbine generator system according to an embodiment of the present application is shown;

[0037] Figure 2 A flowchart of state synchronization compensation according to an embodiment of the present application is shown;

[0038] Figure 3 A flow chart showing state mean re-injection in an embodiment of the present application is shown;

[0039] Figure 4 A schematic diagram of the structure of the system according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of illustration and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps without logical context can be reversed or implemented simultaneously. In addition, those skilled in the art, under the guidance of the contents of this application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0041] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0042] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0043] In view of the technical problems raised by the background technology, the present application provides a cross-scale digital twin hybrid operation method and system for grid-type wind turbines, which can realize model switching through a dynamic threshold trigger mechanism, solve the problems of system switching lag and false triggering, and ensure data continuity and smooth transition based on state synchronization compensation and state mean re-injection strategy, improve computing efficiency, and reduce tracking errors.

[0044] See the instructions attached Figure 1 The present application provides a cross-scale digital twin hybrid operation method for a grid-type wind turbine, comprising the following steps:

[0045] S1. Build a digital twin model of a grid-type wind turbine generator system, including a refined converter model and an average converter model;

[0046] S2. Extract the synchronous rotating coordinate system component of the grid voltage and calculate its instantaneous rate of change. In addition, introduce the DC bus current to construct a dynamic threshold trigger mechanism.

[0047] S3. Switching between the average value converter model and the refined converter model based on the dynamic threshold trigger mechanism;

[0048] S4. When the instantaneous change rate exceeds a dynamic threshold, triggering a switch from the average value converter model to the refined converter model, and performing state synchronization compensation during the switch;

[0049] S5. When the instantaneous change rate is lower than a dynamic threshold and exceeds a preset time period, triggering a switch from the refined converter model to the average converter model, and performing state mean back-injection during the switch.

[0050] Specifically, in step S1, the grid-connected wind turbine, as a key supporting unit of the new power system, features virtual synchronous machine control technology that simulates the dynamic characteristics of traditional synchronous generators. This allows for the autonomous construction of grid voltage and frequency, making it particularly suitable for scenarios with weak grids or high proportions of renewable energy access. The digital twin model of the grid-connected wind turbine includes a rotor drive train, a permanent magnet synchronous generator, and a full-power converter system. The converter, as the core link in energy conversion and grid interaction, employs two differentiated modeling approaches to accommodate different simulation requirements.

[0051] The refined converter model is constructed in detail based on the physical topology and includes IGBT switching devices, DC bus capacitors, and PWM modulation links. It fully reproduces the current ripple, dead zone effect, and device loss characteristics at the switching frequency (typical value 2kHz), making it suitable for scenarios such as fault ride-through and harmonic resonance analysis that require capturing microsecond transient responses. The average value converter model uses equivalent circuit abstraction to replace the switching branch with an ideal controlled voltage source, ignoring high-frequency switching dynamics but retaining the DC voltage dynamic equation and virtual synchronous control loop. It can efficiently simulate medium-speed dynamic processes such as unit inertia response, primary frequency modulation, and power angle stability on a millisecond time scale.

[0052] The collaborative application of the refined converter model and the average converter model covers full-scale simulation requirements from device-level transient to system-level steady-state. The refined converter model focuses on the evaluation of internal converter switching losses, accurate prediction of short-circuit current, and setting of protection settings, providing a high-fidelity analysis tool for extreme working conditions such as lightning strikes and asymmetric grounding. The average converter model supports grid-level stability research, such as multi-machine parallel oscillation suppression, black start collaborative control, and other long-cycle strategy verification. Its computational efficiency is more than 10 times higher than that of the detailed model, meeting the real-time requirements of the digital twin system.

[0053] In step S2 and step S3, a dynamic threshold trigger mechanism is set. By analyzing the statistical characteristics of the grid signal in real time, the trigger threshold is adaptively adjusted to achieve seamless switching between the refined converter model and the average value converter model. When a sudden change in grid voltage or frequency exceeding the limit is detected, the average value converter model quickly locates the fault period and triggers the refined converter model to intervene, forming a hybrid simulation mode of "wide-area monitoring-local focusing", providing an integrated digital mapping foundation for the multi-level interaction of "control-equipment-grid" of the grid-type unit.

[0054] In this application, the synchronous rotating coordinate system (dq axis) component of the grid voltage is selected as the analysis object, and its instantaneous change rate is:

[0055]

[0056] Where, v d (t), v q (t) is the direct axis (d axis) and quadrature axis (q axis) components of the grid voltage after decoupling at time t. In order to enhance the model's sensitivity to fault current, the DC bus current is introduced as an auxiliary variable in the calculation of the dynamic threshold. The threshold at time t is:

[0057] V th (t) = μ Δv +3σ Δv +α×|i dc (t)|

[0058] Where μ Δv is the mean value of the instantaneous change rate of the dq axis voltage, σ Δv is the standard deviation of the mean value of the instantaneous change rate of the dq axis voltage, α is the weight coefficient, i dc is the DC bus current. Where:

[0059]

[0060] Where Δv dq (t) is the instantaneous rate of change of the dq axis voltage at time t, N is the number of data points, and the window length T w With the sampling frequency f s In steady state, the system trigger flag is 0, and the converter operates under the average value model. dq (t)>V th At (t), the trigger flag is 1, and the system converter switches to the refined model.

[0061] In step S4, when the voltage change rate exceeds the dynamic threshold V thWhen switching models, the design of state synchronization compensation at the moment of switching is the core link to ensure the confidence of digital twin system simulation. This is because the average value model only contains macroscopic state quantities such as low-frequency-dominated DC voltage and dq-axis current, while the refined model needs to additionally track the high-frequency current transients of the IGBT switch branch (such as the ripple caused by 2kHz PWM modulation) and the charging and discharging details of the capacitor. If switched directly, the initial state of the refined model will lack high-frequency components, causing DC bus voltage jumps and power angle instability of the virtual synchronous control loop. More seriously, the mismatch of the initial value of the switch branch current will generate non-physical impact currents on the AC side, destroying the authenticity of the simulation results.

[0062] The wavelet compensation algorithm combines time-frequency localization capabilities with adaptive resolution characteristics through multi-scale decomposition: its high-frequency detail coefficients can accurately capture the microsecond current ripple characteristics corresponding to switching transients (2-5kHz frequency band), while the low-frequency approximation coefficients retain the millisecond-level dynamic response trend of virtual synchronous control, thereby completing the missing initial value of the switch branch current at the switching moment and dynamically correcting the DC voltage ripple offset through the time-varying threshold.

[0063] The Daubechies 4 (db4) wavelet can accurately locate transient signals in the time domain and avoid spectrum leakage caused by boundary effects. Its second-order vanishing moment characteristic can automatically filter out linear trend items in signals such as Vdc, focusing on extracting high-frequency dynamic components such as PWM switching. In this application, the db4 algorithm is selected to decompose and compensate the characteristic signal state, see the attached manual. Figure 2 , including the following steps:

[0064] S401, performing Savitzky-Golay filtering preprocessing on the DC voltage output by the average value converter model to eliminate noise interference;

[0065] S402, using Daubechies 4 wavelet to perform n-layer decomposition on the preprocessed DC voltage, separating high-frequency and low-frequency signal components layer by layer through a low-pass and high-pass filter group to obtain high-frequency detail coefficients;

[0066] S403 , synthesizing the high-frequency detail coefficients of the selected layer into a time-domain voltage ripple signal using a wavelet reconstruction algorithm, and mapping the signal to the refined converter model through a differential operation.

[0067] The DB4 wavelet filter consists of a low-pass filter and a high-pass filter, and its coefficients are determined by the Daubechies orthogonal wavelet construction method:

[0068]

[0069] h[3]~h[7]=0

[0070] g[n]=(-1) n h[7-n],n=0,1,…,7

[0071] Where h[n] and g[n] are the coefficients of the low-pass filter and high-pass filter, respectively. To eliminate the spectrum aliasing caused by downsampling, the filter coefficients need to be reconstructed to meet the biorthogonal condition.

[0072] The decomposition process divides the signal into low-frequency approximation coefficients and high-frequency detail coefficients through convolution and downsampling, and filters and samples the high and low frequency signals:

[0073]

[0074] Where: a j [k] and d j [k] represents the low-frequency and high-frequency signals of the jth layer, and N represents the signal length.

[0075] During the signal reconstruction process, in order to compensate for the data compression caused by step downsampling, zero insertion is performed on the decomposed high and low frequency signal sequences:

[0076]

[0077] The complete high and low frequency signals are reconstructed twice through the filter to complete the signal decomposition

[0078]

[0079] In addition, in step S5, when the system operates under the refined model, the voltage change rate is less than the dynamic threshold V th And the continuous stable operation exceeds the preset time, such as 10ms, the converter switches to the average value model. In order to ensure that there is no jump in the signals at both ends of the converter at the switching moment and do not affect the current steady-state performance of the system, this application adopts the state mean re-injection method to directly inherit the real-time operation data of the refined model. Figure 3 , the state mean is re-injected in the following way, including the following steps:

[0080] S501, calculating the mean value of the key state quantity required by the average value converter model at the switching time;

[0081] S502: Inject the calculated mean value into the average value converter model.

[0082] First, calculate the mean values of the three state quantities (DC voltage and dq axis current) required by the average value model at the switching time:

[0083]

[0084] Where Δt is the sampling interval and N is the number of samples. At time t0 when the switchback condition is met, the mean value is injected into the mean value model:

[0085]

[0086] Where: Indicates the last moment before the switch, when the refined model is still running. Indicates the first moment after the switch, the average value model takes over the operation.

[0087] In one embodiment, the db4 algorithm is selected to decompose and compensate the DC side voltage and AC side current signals at the switching moment:

[0088] (1) DC side state compensation design. First, the DC voltage output by the average value model is preprocessed by Savitzky-Golay filtering to eliminate noise interference. Then, the Daubechies 4 wavelet is used to perform n-layer decomposition. The high-frequency and low-frequency signal components are separated layer by layer through low-pass and high-pass filter groups. Then, the high-frequency detail coefficients of the selected layer are synthesized into a time-domain voltage ripple signal using the wavelet reconstruction algorithm, and mapped to the refined model through differential operation. The specific form is:

[0089] 1. Signal decomposition and feature extraction:

[0090] In the data preprocessing stage, the DC voltage Vdc of the average model is subjected to Savitzky-Golay filtering to remove noise. The voltage value at time t after filtering is expressed as:

[0091]

[0092] Where: w k is the filter weight coefficient, V dc_avg is the DC voltage value of the t-average model at time t. Perform n-layer DB4 decomposition on the filtered voltage signal to extract the high-frequency components:

[0093]

[0094] Where: a j (n) and d j (n) are the low-frequency and high-frequency signals of the jth layer obtained by decomposition, h(k) and g(k) are the coefficients of the low-pass filter and high-pass filter respectively, where the initial layer is The high-frequency components of the ef layer obtained by decomposition are selected to reconstruct the high-frequency signal, and the reconstructed voltage signal is obtained as follows:

[0095]

[0096] ψ j,n (t) = 2j / 2 ψ(2 j tn)

[0097] Where: V hf (t) is the high-frequency ripple signal obtained after reconstruction, corresponding to the voltage fluctuation caused by the operation of the refined converter switching model, ψ j,n (t) is the wavelet basis function, the purpose is to transform the discrete high frequency signal d j (n) Convert to a continuous-time signal and reconstruct the high-frequency components.

[0098] 2. Mapping high-frequency components to a refined model:

[0099] Taking into account the influence of high-frequency DC voltage signal on the charging and discharging of DC bus capacitor, the capacitor current is calculated as:

[0100]

[0101] Where C represents the DC bus capacitance. Using differential operation to map to the initial value of the capacitor current in the refined model, the corrected initial value of the DC bus voltage is obtained:

[0102]

[0103] Where: T w Indicates the integration window length, which aims to cover the complete high-frequency dynamic cycle caused by the IGBT switching action. When the switching frequency is 2kHz, the corresponding integration window length is 0.5ms.

[0104] (2) AC side state compensation design. The three-phase current of the average value model is converted to the αβ coordinate system to decouple the fundamental component and perform Daubechies 4 wavelet packet decomposition. The high-frequency harmonic components are screened out by the node energy threshold and mapped to the six IGBT branch currents based on the switching function matrix of the converter topology. At the same time, a dynamic attenuation factor is introduced to perform exponential integral correction on the historical high-frequency current. The current attenuation characteristics are compensated by combining the inductor and resistor parameters. Finally, the low-frequency fundamental component and the high-frequency compensation term are synchronously injected into the initial value of the switch branch of the refined model through the state expansion matrix.

[0105] 1. Signal decomposition and feature extraction:

[0106] Convert the three-phase current on the AC side of the average value model into the αβ coordinate system:

[0107]

[0108] Where: i α (t) and i β (t) is the αβ axis current value, i a (t), i b (t) and ic (t) is the three-phase current value. α (t) and i β (t) Perform db4 wavelet packet decomposition, the method is the same as the DC side wavelet packet decomposition, and obtain the αβ high frequency component signal i α_hf (t) and i β_hf (t).

[0109] 2. High-frequency components are mapped to switching currents:

[0110] The αβ high-frequency components are mapped to 6 IGBT branches through the topology matrix:

[0111]

[0112] Modify the dynamic compensation term of the initial switch current according to the obtained historical high-frequency signal component:

[0113]

[0114] Where: R and L represent the equivalent resistance and filter inductance of the AC side of the converter.

[0115] (3) Refine the model initial state injection:

[0116] The DC side and current side compensation terms are integrated to obtain the state expansion matrix, and the average value model state X avg =[V dc ,i d ,i q ] T Extension to refined model X det =[V dc ,i d ,i q ,i sw1 ,i sw2 ,i sw3 ,i sw4 ,i sw5 ,i sw6 ] T get:

[0117]

[0118] Where: γ1-γ 18 Indicates V dc 、i d (t) and i q (t) for the switch branch current i sw1 -i sw6 The steady-state coupling coefficient, η1-η6 represents the dynamic weight of the high-frequency signal component on the switch branch current. The initial state of the model after switching is expressed as:

[0119]

[0120]

[0121] Where: Input u(t)=[V hf (t),i sw1 (t),i sw2 (t),i sw3 (t),i sw4 (t),i sw5 (t),i sw6 (t)] T , A and B matrices are the state space matrices of the three-term two-level converter refined model.

[0122] When the system operates under the refined model, the voltage change rate is less than the dynamic threshold V th And it continues to run stably for more than 10ms, and the converter switches to the average value model. To ensure that there is no jump in the signals at both ends of the converter at the switching moment and do not affect the current steady-state performance of the system, this method adopts the state mean re-injection method, directly inheriting the real-time operation data of the refined model. First, the mean of the three state quantities required by the average value model at the switching moment is calculated:

[0123]

[0124] Where: Δt is the sampling interval, N is the number of samples. At the time point t0 that meets the switchback condition, the mean value is injected into the mean value model:

[0125]

[0126] Where: Indicates the last moment before the switch, when the refined model is still running. Indicates the first moment after the switch, the average value model takes over the operation.

[0127] The present application provides a cross-scale digital twin hybrid operation method for a grid-type wind turbine, which is based on a multi-time-scale hybrid modeling method and realizes intelligent switching between models through a dynamic threshold trigger mechanism. When the system state deviates by more than the dynamic threshold, it automatically triggers the switch from the average value model to the refined model to ensure high-precision characterization under transient conditions. Otherwise, it switches to the average value model to release computing power. If the parameters are directly replaced during the model switching process, it is easy to cause state jumps and data faults. The state synchronization compensation strategy compensates for the time scale difference through an interpolation algorithm when the model switches, dynamically corrects the initial state of the refined model, and avoids numerical oscillations caused by model switching; when switching to the average value model, the state mean reinjection strategy is adopted to reinject the transient characteristics of the refined model into the average value model in the form of a weighted mean, retaining key dynamic information and achieving a smooth transition.

[0128] Based on the same concept of the present invention, as shown in the attached specification Figure 4 As shown, a system provided by an embodiment of the present application includes an FPGA, a DSP and a CPU. The FPGA deploys the refined converter model described in the above-mentioned method for cross-scale digital twin hybrid operation of a meshed wind turbine generator set, the DSP deploys the dynamic threshold trigger mechanism described in the above-mentioned method for cross-scale digital twin hybrid operation of a meshed wind turbine generator set, and the CPU deploys the average value converter model described in the above-mentioned method for cross-scale digital twin hybrid operation of a meshed wind turbine generator set, and performs real-time closed-loop interaction and data synchronization through LVDS / PCIe.

[0129] In one embodiment, considering the operational accuracy and real-time simulation requirements of different models, a three-level heterogeneous hardware platform based on FPGA, DSP, and CPU was constructed. Through a design strategy of interface feature adaptation, computational granularity grading, and hierarchical data flow management, high-precision real-time simulation of the entire frequency domain, from microsecond-level switching transients to millisecond-level electromechanical dynamics, was achieved. The FPGA layer deployed a refined power electronics topology model with a sampling time of 1μs, covering the electromagnetic characteristics of the generator, the switching process of the converter IGBT, the dynamics of the LCL filter, and the high-frequency coupling effects of the grid impedance network. The FPGA layer used an LVDS differential interface to transmit grid current / voltage data to the DSP. LVDS, with its strong common-mode interference immunity and high transmission rate, ensured high-frequency signal integrity and low-latency interaction (transmission delay <50ns) between the FPGA and DSP. Furthermore, closed-loop control of the switching frequency of 20kHz was achieved by receiving PWM commands issued by the DSP, avoiding the pulse loss problem caused by clock jitter in traditional SPI interfaces. The DSP layer runs the converter vector control algorithm, virtual synchronous generator control algorithm, dynamic threshold determination, and state synchronization compensation / reinjection modules with a 10μs sampling step. The HPU (Hardware Acceleration Processing Unit) is specifically designed for hard real-time interaction with the FPGA. With a built-in double-buffered DMA channel and a priority interrupt controller, the HPU completes threshold determination within 2μs and executes the state synchronization compensation algorithm in real time. It performs cubic spline interpolation on 1μs-level data uploaded from the FPGA to generate a continuous initial state sequence for the refined model, eliminating phase jumps during model switching. Furthermore, the HPU preloads state mean reinjection weight coefficients to perform a sliding window weighted average of transient features during the order reduction process, ensuring state continuity in the average model. The CPU layer runs the fan aerodynamic model, drive train model, and converter average model driven by aerodynamic load calculations based on blade element momentum theory at a 100μs step size. High-speed bidirectional data exchange is achieved with the FPGA via the PCIe interface. The PCIe 3.0x4 channel provides 8GT / s bandwidth, supporting the CPU to issue torque commands to the FPGA (transmission delay <5μs) and receive generator speed, rotor angle, and current signals, meeting the high throughput requirements of electromechanical-electromagnetic cross-scale coupled simulations. In addition, the PCIe link layer has a built-in CRC check and retransmission mechanism to ensure the integrity of data packets in long-term simulations and avoid synchronization problems between the digital twin and the physical system caused by communication errors. The three-level architecture achieves adaptive switching of model accuracy and dynamic allocation of hardware resources under complex working conditions through time scale decoupling and interface protocol optimization. Compared with traditional single computing unit solutions, it significantly reduces the discrete error of simulation steps and provides a hardware-level solution for balancing the real-time performance and fidelity of digital twin systems.

[0130] The hybrid modeling mechanism places higher demands on the heterogeneous collaborative capabilities of the hardware platform. Traditional single computing units are difficult to meet the needs of multi-model parallel scheduling. The multi-level collaborative architecture stores the entire system in modules with different sampling durations. The FPGA is responsible for high-frequency threshold detection and model switching logic control, the CPU is responsible for high-order numerical solutions of refined models, and the DSP is dedicated to fast iterative calculations of average value models. The dynamic allocation of hardware resources supports the real-time parallel operation of multi-time scale models. Compared with traditional methods, this method can effectively improve computing efficiency and reduce transient process tracking errors in the simulation of digital twin systems of grid-type wind turbines. At the same time, it reduces hardware resource usage, providing a high-precision, low-latency, and highly robust systematic solution for the engineering implementation of digital twin technology in complex industrial scenarios.

[0131] Finally, it should be noted that the above embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above embodiments within the technical scope disclosed in the present application, or replace some of the technical features therein with equivalents. However, these modifications, changes, or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A cross-scale digital twin hybrid operation method for a grid-type wind turbine, characterized in that: The method comprises the following steps: Build a digital twin model of a grid-connected wind turbine, including a refined converter model and an average converter model; Extract the synchronous rotating coordinate system component of the grid voltage and calculate its instantaneous rate of change. In addition, introduce the DC bus current to construct a dynamic threshold trigger mechanism. Switching between the average value converter model and the refined converter model is performed based on the dynamic threshold trigger mechanism.

2. The cross-scale digital twin hybrid operation method of a grid-type wind turbine according to claim 1 is characterized in that: The refined converter model is constructed based on physical topology, including IGBT switching devices, DC bus capacitors and PWM modulation; the average value converter model is constructed based on equivalent circuit abstraction that ignores high-frequency switching dynamics.

3. The cross-scale digital twin hybrid operation method of a grid-type wind turbine according to claim 1, characterized in that: The instantaneous rate of change of the dq-axis component of the synchronous rotating coordinate system is calculated by the following formula: Where, v d (t), v q (t) is the d-axis and q-axis components of the grid voltage after decoupling at time t.

4. The cross-scale digital twin hybrid operation method of a grid-type wind turbine according to claim 3 is characterized in that: The dynamic threshold is calculated using the following formula: V th (t)=μ Δv +3σ Δv +α×|i dc (t)| Where μ Δv is the mean value of the instantaneous change rate of the dq axis voltage, σ Δv is the standard deviation of the mean value of the instantaneous change rate of the dq axis voltage, α is the weight coefficient, i dc is the DC bus current; The mean and standard deviation of the instantaneous rate of change of the dq axis voltage are calculated using the following formula: Where Δv dq (t) is the instantaneous rate of change of the dq-axis voltage at time t, and N is the number of data points.

5. The cross-scale digital twin hybrid operation method of a grid-type wind turbine according to claim 1, characterized in that: The switching between the average value converter model and the refined converter model based on the dynamic threshold trigger mechanism includes the following steps: When the instantaneous change rate exceeds a dynamic threshold, triggering a switch from the average value converter model to the refined converter model, and performing state synchronization compensation during the switch; When the instantaneous change rate is lower than a dynamic threshold and exceeds a preset time period, switching from the refined converter model to the average converter model is triggered, and state mean re-injection is performed during the switching.

6. A cross-scale digital twin hybrid operation method for a grid-type wind turbine according to claim 5, characterized in that: State synchronization compensation is performed in the following manner, including the following steps: performing filtering preprocessing on the DC voltage output by the average value converter model; The pre-processed DC voltage is decomposed into multiple layers using Daubechies 4 wavelet to separate the high- and low-frequency signal components layer by layer through low-pass and high-pass filter groups to obtain high-frequency detail coefficients. The high-frequency detail coefficients of the selected layer are synthesized into a time-domain voltage ripple signal by using a wavelet reconstruction algorithm, and are mapped to the refined converter model through a differential operation.

7. The cross-scale digital twin hybrid operation method of a grid-type wind turbine according to claim 5, characterized in that: The state mean is re-injected in the following manner, including the following steps: Calculating the mean value of the key state quantity required by the average value converter model at the switching moment; The calculated mean value is injected into the mean value converter model.

8. The cross-scale digital twin hybrid operation method of a grid-type wind turbine according to claim 7, characterized in that: in, The key state quantities include DC voltage and dq axis current, and the mean value of the key state quantities at the switching moment is calculated by the following formula: Where Δt is the sampling interval and N is the number of samples.

9. A system, characterized in that: The invention comprises an FPGA, a DSP and a CPU, wherein the FPGA deploys the refined converter model described in claim 1, the DSP deploys the dynamic threshold trigger mechanism described in claim 1, and the CPU deploys the average value converter model described in claim 1, and performs real-time closed-loop interaction and data synchronization through LVDS / PCIe.

10. A system according to claim 9, characterized in that: The FPGA transmits grid data to the DSP via an LVDS differential interface, with a transmission delay of <50ns; the CPU interacts with the FPGA via a PCIe interface, with a transmission delay of <5μs; and the DSP executes the dynamic threshold trigger mechanism with a sampling step of 10μs.

Images