Digital twinning-based inverter transformer loss dynamic adaptation method

Abstract

The invention discloses an inverter transformer loss dynamic adaptation method based on digital twinning, and relates to the technical field of power electronics. The method comprises the following steps: S1, constructing a multi-physical field digital twinborn model integrating magnetic core loss, winding loss, parasitic parameters and a thermal-magnetic coupling sub-model; s2, initial operation parameters are optimized and determined for topology such as LLC resonance and dual active bridges; s3, acquiring signals such as magnetic flux density in real time through a sensor, synchronizing to model update loss and parasitic parameters, and executing dynamic regulation and control when deviating from a preset range; and S4, recording full-life-cycle data, constructing a machine learning time sequence loss degradation prediction model, and dynamically adjusting parameters and strategies in combination with load fluctuation characteristics to realize full-working-condition adaptive optimization. The method solves the problems that a traditional method is low in prediction precision, rigid in adaptation, lagged in regulation and control and the like, transformer loss prediction accuracy and operation efficiency are improved, system reliability is enhanced, and the service life is prolonged.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, specifically to a method for dynamic adaptation of inverter transformer losses based on digital twins. Background Technology

[0002] Inverters are core components of power electronic systems such as renewable energy generation, electric vehicles, and data centers. The performance of their internal isolation transformers directly determines the efficiency, reliability, and lifespan of the entire system. Transformer losses, primarily core losses and winding losses, are key factors affecting inverter efficiency, and these losses dynamically change with switching frequency, load conditions, ambient temperature, and component aging. Therefore, accurate prediction and dynamic optimization of transformer losses are crucial for improving system energy efficiency and economy.

[0003] Currently, the technical solutions for controlling and optimizing inverter transformer losses mainly have the following limitations: Firstly, there is the static design method based on fixed parameters. This method pre-sets and fixes transformer parameters according to typical operating conditions during the design phase. However, this method cannot adapt to real-world load fluctuations, temperature changes, and component aging during operation, resulting in poor adaptability.

[0004] Secondly, simplified loss prediction models are used. Calculations often rely on empirical formulas for single physical fields, failing to adequately consider skin effects and proximity effects at high frequencies, and particularly neglecting the coupling effects between multiple physical fields such as magneto-thermal and electro-magnetic fields. For example, temperature rise alters core characteristics, thus affecting losses; this dynamic feedback process is difficult to reflect in existing models, leading to insufficient prediction accuracy.

[0005] Third, it relies on simple feedback control lacking deep data fusion. Direct feedback control using sensor-collected operational data results in a simplistic control method. This approach lacks modeling of the complex dynamic mapping relationship between losses, parasitic parameters, and temperature, leading to lag in response and an inability to achieve globally optimal control.

[0006] Fourth, topology-specific adaptation schemes lack versatility. Different topologies, such as LLC and dual active bridges, require separate design of parameters and control logic, lacking a universal, model-optimized adaptation framework.

[0007] In recent years, digital twin technology has provided a new approach for the real-time mapping and simulation of physical objects. However, in the field of inverter transformers, most existing digital twin applications remain at the level of condition monitoring and visualization, that is, constructing a one-way, shallow virtual mapping, failing to achieve closed-loop dynamic interaction and proactive decision optimization between the model and the physical entity. This makes it unable to support the aforementioned requirements for accurate prediction, real-time control, and lifecycle management.

[0008] In summary, the existing technology has the following systemic shortcomings: 1. The loss prediction model has low accuracy and does not integrate multi-physics coupling and parasitic parameter effects; 2. The parameter adaptation mechanism is rigid and lacks intelligent optimization capabilities based on different topological characteristics; 3. The real-time control strategy is superficial and cannot achieve dynamic closed-loop calibration of loss and parameters based on high-fidelity models; 4. Full-cycle optimization is lacking and fails to incorporate device aging trends for forward-looking strategy adjustments. Summary of the Invention

[0009] The purpose of this invention is to provide a method for dynamic adaptation of inverter transformer losses based on digital twins, so as to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic adaptation of inverter transformer losses based on digital twins, comprising the following steps: S1. Construct and run a multi-physics digital twin model of the transformer used in the inverter. The digital twin model integrates the core loss sub-model, winding loss sub-model, parasitic parameter sub-model, and thermo-magnetic coupling sub-model. The core loss sub-model and the winding loss sub-model both receive the basic operating parameters of the transformer and output the dynamic loss values ​​of the core and the winding, respectively. The basic operating parameters of the transformer include the core material properties, switching frequency, magnetic flux density and ambient temperature. The parasitic parameter sub-model is used to extract the leakage inductance and distributed capacitance of the transformer; The thermo-magnetic coupling sub-model is used to establish a dynamic mapping relationship between loss values ​​and temperature field; S2. Based on the digital twin model, optimize the preset topology type of the target inverter to determine the initial operating parameters. The initial operating parameters include magnetizing inductance, leakage inductance, winding turns ratio, and core pre-magnetization parameters, which are used to adapt to the operating requirements of the preset topology type. The preset topology type includes LLC resonant topology and dual active bridge topology. S3. The operating status signal of the transformer is acquired in real time through sensors and transmitted to the digital twin model; the digital twin model updates the loss distribution and parasitic parameters according to the operating status signal; the operating status signal includes magnetic flux density, ambient temperature and winding current; When the updated loss distribution or parasitic parameters deviate from the preset range, dynamic adjustment is performed to suppress loss and calibrate parameters. S4. Record the operating data of the transformer throughout its entire life cycle using the digital twin model, and construct a loss degradation prediction model based on the operating data; combine the output results of the prediction model with the load fluctuation characteristics to dynamically adjust the operating parameters and control strategies of the transformer to achieve adaptive optimization under all operating conditions. The full lifecycle operation data includes historical loss data, temperature drift records, and parameter control records.

[0011] Preferably, the core loss sub-model is constructed by integrating an improved core loss calculation algorithm with a three-dimensional finite element analysis algorithm. The transformer's basic operating parameters are input, and the core dynamic loss value is output to achieve core dynamic loss prediction with a preset accuracy.

[0012] Preferably, the winding loss sub-model introduces skin effect and proximity effect correction coefficients, and is constructed by combining conductor structure parameters and winding process parameters to output the winding dynamic loss value; the conductor structure parameters include conductor diameter and cross-sectional shape, and the winding process parameters include winding spacing and number of layers.

[0013] Preferably, when the preset topology type is an LLC resonant topology, the optimization goal is to match the resonant frequency and optimize the ratio of magnetizing inductance to leakage inductance; when the preset topology type is a dual active bridge topology, the optimization goal is to improve bidirectional energy transmission efficiency and determine the winding turns ratio and the suitable range of leakage inductance.

[0014] Preferably, the real-time acquisition method includes real-time acquisition of magnetic flux density, ambient temperature and winding current, and the acquired operating status signal is synchronously transmitted to the multi-physics digital twin model through a data transmission link.

[0015] Preferably, the dynamic control operation includes core air gap length adjustment, winding effective turns switching, cooling system linkage control, and excitation current adjustment, which are used to achieve magnetoresistive optimization, leakage inductance calibration, and heat loss suppression, respectively.

[0016] Preferably, the adjustment of the air gap length of the magnetic core adopts a micro piezoelectric drive method, responding to the instructions issued by the multi-physics digital twin model, and adjusting the air gap length based on the preset step size determined by the magnetic core material characteristics and real-time loss data, thereby suppressing magnetic core loss.

[0017] Preferably, the loss degradation prediction model is constructed based on a machine learning time series data analysis algorithm, which takes the full life cycle operation data as input and outputs the core aging trend and winding performance degradation results.

[0018] Preferably, the full-condition adaptive optimization further includes constructing a parameter control strategy library, which is established based on load fluctuation characteristics and includes adaptive parameter combinations and control logic under light load, full load and overload conditions.

[0019] Preferably, the multiphysics digital twin model adopts a real-time data synchronization mechanism, which realizes the uploading of operating status signals and the issuance of dynamic control commands based on a high-frequency data transmission link, ensuring the temporal consistency of loss prediction and control.

[0020] Compared with the prior art, the beneficial effects of the present invention are: by constructing a multi-physics digital twin model and updating loss distribution and parasitic parameters in real time, performing dynamic control operations, recording full life cycle data to construct a prediction model, and dynamically adjusting parameters in combination with load fluctuation characteristics, the present invention achieves full-condition adaptive optimization, which has the advantages of improving the accuracy of inverter transformer loss prediction, realizing dynamic parameter adaptation and optimization, enhancing system adaptability and reliability, and extending transformer life. Attached Figure Description

[0021] Figure 1 This is a flowchart of the dynamic adaptation method according to an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 This application proposes a dynamic adaptation method for inverter transformer losses based on digital twins. By constructing a multiphysics digital twin model integrating core loss sub-models, winding loss sub-models, parasitic parameter sub-models, and thermo-magnetic coupling sub-models, it achieves accurate mapping and prediction of transformer operating states. Furthermore, this method optimizes the initial operating parameters for a preset topology type of the target inverter and dynamically adjusts them based on real-time operating state signals acquired by sensors. Finally, by recording operating data throughout the transformer's entire lifecycle, a loss degradation prediction model is constructed to achieve adaptive optimization under all operating conditions.

[0024] For ease of understanding, the following explains some key terms in this embodiment: A digital twin model is a real-time, dynamic, and high-fidelity mapping of a physical entity in virtual space. This model can integrate multi-source data to simulate the behavior, state, and performance of physical entities, and supports analysis, prediction, and optimization of these entities. In this method, the digital twin model serves as the core carrier, used to simulate the operating characteristics of the inverter transformer.

[0025] The core loss sub-model is a module in the digital twin model specifically designed for calculating transformer core losses. This sub-model receives basic transformer operating parameters and outputs dynamic core loss values ​​based on factors such as core material properties, magnetic flux density, and switching frequency.

[0026] The winding loss sub-model is a module in the digital twin model specifically used to calculate transformer winding losses. This sub-model receives basic transformer operating parameters and outputs dynamic winding loss values ​​based on factors such as winding current, conductor resistance, and temperature.

[0027] The parasitic parameter sub-model refers to the module in the digital twin model used to extract internal parasitic parameters of the transformer. These parasitic parameters mainly include leakage inductance and distributed capacitance, which affect the performance of the transformer under high-frequency operating conditions.

[0028] The thermo-magnetic coupling sub-model is a module in the digital twin model used to establish the dynamic mapping relationship between loss values ​​and the temperature field. This sub-model can simulate how the heat generated by internal losses in a transformer affects the temperature distribution, and how temperature changes, in turn, affect the properties and losses of magnetic materials.

[0029] Transformer operating parameters are the input data used to describe the basic operating state and material properties of a transformer. These parameters include core material properties, switching frequency, magnetic flux density, and ambient temperature, and are key bases for calculating core and winding losses.

[0030] Initial operating parameters refer to the configuration parameters determined through optimization using a digital twin model before the transformer is put into operation. These parameters include magnetizing inductance, leakage inductance, winding turns ratio, and core pre-magnetization parameters, designed to enable the transformer to reach its operating state under a specific topology.

[0031] Operating status signals refer to the actual operating data of the transformer collected in real time by sensors. These signals include magnetic flux density, ambient temperature, and winding current, which are transmitted to the digital twin model to update the model's status in real time and trigger regulation.

[0032] Full lifecycle operation data refers to all operational information recorded throughout the entire process of a transformer from commissioning to decommissioning. This data includes historical loss data, temperature drift records, and parameter control records, providing data support for loss degradation prediction and long-term optimization.

[0033] A loss degradation prediction model is a model built based on full lifecycle operating data to predict the degradation trend of transformer losses over time. This model can assess the impact of core aging and winding performance degradation on losses.

[0034] This method achieves dynamic adaptation of inverter transformer losses through the following steps: In step S1, a multiphysics digital twin model of the inverter transformer is constructed and put into operation. This digital twin model is designed to integrate a core loss sub-model, a winding loss sub-model, a parasitic parameter sub-model, and a thermo-magnetic coupling sub-model. The core loss and winding loss sub-models are configured to receive fundamental transformer operating parameters and output dynamic core loss and winding loss values, respectively. Fundamental transformer operating parameters may include core material properties, switching frequency, magnetic flux density, and ambient temperature. The core loss sub-model can calculate core losses at different magnetic flux densities and frequencies based on the classical Steinmetz equation or its modified form, combined with the BH curve data of the core material. The winding loss sub-model can calculate ohmic losses based on the DC resistance and real-time current of the winding. The parasitic parameter sub-model is used to extract the transformer's leakage inductance and distributed capacitance, through simplified modeling of the transformer geometry and calculations combined with electromagnetic field theory. The thermo-magnetic coupling sub-model is used to establish a dynamic mapping relationship between loss values ​​and temperature field. By constructing a thermal network model inside the transformer, the losses of each part are used as heat sources to calculate the temperature distribution and correct the magnetic material parameters according to temperature changes.

[0035] In step S2, the initial operating parameters are determined based on the aforementioned digital twin model, targeting the preset topology type of the inverter. These initial operating parameters may include magnetizing inductance, leakage inductance, winding turns ratio, and core pre-magnetization parameters. These parameters are used to adapt to the operating requirements of the preset topology type. The preset topology type may include LLC resonant topology and dual active bridge topology. When the preset topology type is LLC resonant topology, the digital twin model can simulate the resonant characteristics and power transfer efficiency under different combinations of magnetizing inductance and leakage inductance. Through iterative simulation or parameter scanning, it seeks the parameter combination that matches the resonant frequency with the switching frequency and minimizes losses. When the preset topology type is dual active bridge topology, the digital twin model can simulate the bidirectional energy transfer efficiency and current stress under different combinations of winding turns ratio and leakage inductance, thereby determining the initial operating parameters.

[0036] In step S3, the transformer's operating status signal is acquired in real time via sensors and transmitted to the digital twin model. The operating status signal may include magnetic flux density, ambient temperature, and winding current. The digital twin model updates its internal loss distribution and parasitic parameters based on the received operating status signal. When the updated loss distribution or parasitic parameters deviate from a preset range, dynamic control operations are performed to suppress losses and calibrate parameters. Sensors can use Hall effect sensors to measure winding current in real time, thermistors to measure ambient temperature, and flux density to measure via a fluxmeter or an induction coil integrated into the magnetic core. When the digital twin model detects that leakage inductance has drifted due to some reason, such as mechanical vibration, leading to increased winding losses, it can trigger a control command to adjust the inverter's switching frequency or duty cycle to change the transformer's operating point, thereby suppressing losses.

[0037] In step S4, the operating data of the transformer throughout its entire lifecycle is recorded using the aforementioned digital twin model, and a loss degradation prediction model is constructed based on this data. The lifecycle operating data can include historical loss data, temperature drift records, and parameter control records. Combining the output of the loss degradation prediction model with load fluctuation characteristics, the transformer's operating parameters and control strategies are dynamically adjusted to achieve adaptive optimization across all operating conditions. The digital twin model can continuously record the transformer's loss data under different loads, temperatures, and operating times, as well as detailed records of each parameter control. Based on this historical data, the loss degradation prediction model can use statistical regression analysis or time series analysis methods to predict the aging trends of the core and windings and their impact on losses. When the prediction model indicates that core aging will lead to increased losses, operating parameters are adjusted in advance, such as slightly reducing the switching frequency or adjusting the core pre-magnetization parameters, to slow down the aging process or reduce losses caused by aging. Simultaneously, combined with the analysis of load fluctuation characteristics, load changes are predicted, and control strategies are adjusted in advance to ensure that the transformer maintains efficient operation under different conditions such as light load, full load, and overload.

[0038] This method, by constructing a multiphysics digital twin model, achieves accurate prediction and dynamic adaptation of inverter transformer losses, effectively solving the problems of low prediction accuracy, rigid parameter adaptation, insufficient real-time control, and lack of full-cycle optimization in traditional solutions. As a result, the transformer can maintain efficient and reliable operation under different loads, temperatures, and aging conditions, improving system energy efficiency and service life.

[0039] In some of the above-mentioned schemes of this application, a core loss sub-model is proposed to predict the dynamic loss value of the core. However, in its implementation, due to the use of simplified empirical formulas or single physical field models, the prediction accuracy is insufficient and cannot fully reflect the nonlinear characteristics, temperature coupling effect and spatial unevenness of the core material under high-frequency switching conditions, resulting in a large deviation in loss calculation and making it difficult to support the accuracy of subsequent dynamic adaptation optimization.

[0040] In response, this application further proposes a core loss sub-model that integrates an improved core loss calculation algorithm with a three-dimensional finite element analysis algorithm. It takes the basic operating parameters of the transformer as input and outputs the dynamic loss value of the core, thereby achieving a core dynamic loss prediction with a preset accuracy.

[0041] Specifically, the improved core loss calculation algorithm aims to more accurately capture the loss characteristics of magnetic cores at high frequencies, over a wide temperature range, and with varying magnetic flux densities, particularly considering nonlinearity, frequency dependence, and temperature dependence. This algorithm can be based on improved models of the Steinmetz equation, such as the generalized Steinmetz equation or the modified Steinmetz equation. These models, by introducing dynamic adjustment terms for the frequency and magnetic flux density exponents, can better fit the actual loss curve. Furthermore, the algorithm can also be based on hysteresis loop modeling methods, such as the Jiles-Atherton model or the Preisach model, which more accurately reflect hysteresis loss by simulating domain wall motion and magnetization processes.

[0042] The three-dimensional finite element method (FEM) is a numerical computation approach used to solve complex geometries and multiphysics coupling problems. It can accurately simulate the magnetic field distribution, eddy current distribution, and resulting losses within a magnetic core, and can also perform coupled analysis with the thermal field. This algorithm can be used to model and simulate using commercial finite element software, establishing a three-dimensional geometric model of the magnetic core, defining material properties, applying an excitation source, and solving the magnetic field equations to obtain the magnetic flux density distribution, eddy current density distribution, and corresponding losses. Alternatively, the algorithm can be custom-developed based on an open-source finite element library. By discretizing the geometric region of the magnetic core, constructing a finite element mesh, and using an iterative solver to solve Maxwell's equations, high-precision calculations of the magnetic field and losses can be achieved.

[0043] The fusion model refers to combining the advantages of improved core loss calculation algorithms with three-dimensional finite element analysis algorithms to form a more comprehensive and accurate core loss prediction model. This fusion can be implemented sequentially, where the improved core loss calculation algorithm provides an initial or macroscopic loss estimate, and then the three-dimensional finite element analysis algorithm performs local refinement and precise spatial distribution calculations based on this. Alternatively, it can be implemented in parallel, where the two algorithms run independently, and then the results are integrated through data exchange or weighted averaging. The finite element analysis provides the magnetic field distribution, and the improved algorithm uses this distribution to calculate local losses before integration.

[0044] By fusing an improved core loss calculation algorithm with a three-dimensional finite element analysis algorithm, the core loss sub-model overcomes the shortcomings of traditional simplified models in prediction accuracy. The improved core loss calculation algorithm more accurately captures the nonlinear hysteresis and eddy current loss characteristics of the core material under high-frequency switching conditions, while the three-dimensional finite element analysis algorithm can finely simulate the complex magnetic field distribution, eddy current effects, and resulting spatial distribution of losses within the core, effectively considering the influence of temperature on magnetic properties. This fusion approach allows the core loss sub-model to fully reflect the nonlinear characteristics, temperature coupling effects, and spatial unevenness of the core material under high-frequency switching conditions, improving the prediction accuracy of dynamic core loss values. High-precision loss prediction provides a more reliable input for the digital twin model, effectively supporting subsequent dynamic adaptation optimization, operating parameter adjustment, and control strategy formulation, ensuring the accuracy and effectiveness of the entire inverter transformer loss dynamic adaptation method.

[0045] In some of the embodiments described above in this application, a winding loss sub-model is proposed to output the dynamic loss value of the winding. However, in its implementation, the skin effect and proximity effect under high-frequency operation are not fully considered, and the influence of conductor structure and winding process parameters is ignored, resulting in insufficient loss prediction accuracy and inability to accurately reflect the dynamic changes under real operating conditions.

[0046] To address this, this application further proposes a winding loss sub-model that incorporates correction coefficients for the skin effect and proximity effect, and constructs it by combining conductor structure parameters and winding process parameters to output the dynamic loss value of the winding; wherein, the conductor structure parameters include conductor diameter and cross-sectional shape, and the winding process parameters include winding spacing and number of layers.

[0047] The skin effect and proximity effect correction coefficients introduced into the winding loss sub-model aim to quantify and compensate for the additional losses caused by the uneven distribution of high-frequency current in the conductor. The skin effect refers to the phenomenon that current tends to flow towards the conductor surface at high frequencies, while the proximity effect refers to the interaction of currents in adjacent conductors, leading to further distortion of the current distribution. The introduction of these correction coefficients allows the winding loss sub-model to more accurately reflect the actual losses of the transformer during high-frequency operation. These correction coefficients can be obtained through analytical calculation methods, such as formula derivation based on the Dowell model or its improved models, calculating the ratio of AC resistance to DC resistance according to frequency, conductor size, and winding geometry; or they can be obtained through numerical simulation methods, such as using three-dimensional finite element analysis (FEA) software to perform electromagnetic field simulations of the winding under different frequency and current conditions, extracting loss data, and fitting the correction coefficients.

[0048] Conductor structure parameters are key inputs for constructing the winding loss sub-model, affecting the winding's resistance characteristics and heat dissipation capacity. Conductor diameter refers to the cross-sectional dimension of the conductor forming the winding; it can be expressed using standard wire gauges or directly in millimeters. Conductor cross-sectional shape refers to the geometry of the conductor's cross-section; common shapes include circular, rectangular, or flat, as well as Litz wire, which is composed of multiple strands of fine wire twisted together. These parameters determine the conductor's DC resistance and, together with skin effect and proximity effect correction factors, determine the AC resistance.

[0049] The winding process parameters describe the specific arrangement of the windings in the transformer, affecting electromagnetic coupling, parasitic parameters, and loss distribution. Winding spacing refers to the physical distance between adjacent turns or layers, influencing the magnetic field distribution and the strength of the proximity effect; methods include close winding, spaced winding, or segmented winding. The number of layers refers to the number of conductor layers stacked vertically; multi-layer winding increases inter-layer capacitance and inter-layer loss. The precise inclusion of these parameters allows the winding loss sub-model to more accurately simulate the electromagnetic behavior of actual windings.

[0050] Through the above technical solution, the winding loss sub-model can fully consider the additional losses caused by the skin effect and proximity effect under high-frequency operation. Combined with conductor structure parameters and winding process parameters, it constructs a more accurate and comprehensive winding loss calculation model. This improves the prediction accuracy of dynamic winding losses, enabling it to more realistically reflect the dynamic loss changes of the transformer under different operating conditions. This high-fidelity loss prediction capability provides reliable input data for the aforementioned digital twin model, thus enabling a more accurate assessment of the transformer's operating status. This supports subsequent dynamic control operations and full-condition adaptive optimization, effectively solving the prediction bias problem caused by the neglect of these key factors in traditional models, thereby improving the efficiency and reliability of the entire inverter system.

[0051] In some of the solutions described above in this application, optimization is proposed to determine the initial operating parameters. However, in this process, since no specific optimization target is specified for different topology types, the optimization process may not accurately match the needs of a specific topology, resulting in suboptimal parameter adaptation and affecting system efficiency. Therefore, the technical problem to be solved by this application is: how to specify specific optimization targets for different topology types, such as LLC resonant topology and dual active bridge topology, in order to optimize parameter adaptation and improve system efficiency.

[0052] In this regard, this application further proposes that when the preset topology type is LLC resonant topology, the optimization goal is to match the resonant frequency and optimize the ratio of magnetizing inductance to leakage inductance; when the preset topology type is dual active bridge topology, the optimization goal is to improve bidirectional energy transmission efficiency and determine the adaptation range of winding turns ratio and leakage inductance.

[0053] When the preset topology type is LLC resonant topology, the optimization goal is to match the resonant frequency. LLC resonant topology is a type of resonant converter widely used in DC-DC converters, and its operating principle relies on the resonant characteristics of the resonant cavity. Matching the resonant frequency is crucial to ensuring the efficient operation of the LLC resonant converter, as deviation from the resonant frequency leads to increased switching losses and decreased efficiency. The optimization process can be carried out through simulation analysis, experimental testing, or combined with a digital twin model. In a digital twin model, the transformer's response at different frequencies is simulated by adjusting parameters such as magnetizing inductance, leakage inductance, and resonant capacitance, and losses are calculated to find the parameter combination that matches the resonant frequency with the operating frequency. Another approach is to derive the parameters backwards through frequency sweep testing on a physical prototype, combined with loss measurement data. Simultaneously, the ratio of magnetizing inductance to leakage inductance is optimized. Magnetizing inductance and leakage inductance are key transformer parameters, jointly determining the resonant characteristics, gain characteristics, and soft-switching range of the LLC resonant converter. Optimizing their ratio is essential for achieving wide-range, efficient operation. This ratio can be optimized through various methods. One approach is to derive the ratio range based on theoretical analysis and empirical formulas, combined with the gain curve and soft-switching conditions of the LLC topology. Another approach utilizes a digital twin model for parameter scanning, iteratively calculating the efficiency and losses under different magnetizing inductance to leakage inductance ratios within a preset load and input voltage range, thereby determining the ratio.

[0054] When the preset topology is a dual active bridge (DAB) topology, the optimization goal is to improve bidirectional energy transfer efficiency. The dual active bridge (DAB) topology is a commonly used topology for high power density, bidirectional DC-DC conversion, its core advantage being the ability to achieve bidirectional energy flow. Improving bidirectional energy transfer efficiency is the primary goal of DAB topology design, as it directly relates to the overall energy efficiency and economy of the system. Optimization for improving bidirectional energy transfer efficiency can be achieved by optimizing the phase shift angle control strategy and reducing switching and conduction losses. In a digital twin model, the energy transfer process under different phase shift angles and transformer parameter combinations is simulated, losses are calculated, and the most efficient parameter combination under different power flow directions and load conditions is found. Another approach is to measure efficiency under different power transfer directions and loads using an experimental platform, and adjust transformer parameters based on the measurement results. Simultaneously, the winding turns ratio and leakage inductance range are determined. The winding turns ratio is a key parameter determining the voltage transformation ratio of the DAB topology, while leakage inductance plays the role of the energy transfer medium in the DAB topology, its magnitude directly affecting power transfer capability, soft-switching range, and circulating current losses. Determining the compatibility range of both is crucial for maximizing efficiency and ensuring reliable operation while meeting voltage transformation requirements. The winding turns ratio is typically determined based on input and output voltage requirements. The compatibility range of the leakage inductance can be analyzed in detail using a digital twin model. By varying the leakage inductance value in the model, its impact on the soft-switching conditions, peak current, and efficiency of the DAB topology is simulated, thus determining a leakage inductance range that guarantees soft switching while limiting circulating current losses. Another approach is to combine the mathematical model of the DAB topology and use analytical or numerical methods to calculate the leakage inductance range that satisfies both efficiency and soft-switching conditions under different loads and voltages.

[0055] Through the above technical solutions, this application specifies specific optimization targets for different inverter topology types, thereby solving the problem of inaccurate parameter adaptation and achieving targeted optimization. When the preset topology type is LLC resonant topology, by matching the resonant frequency and optimizing the ratio of magnetizing inductance to leakage inductance, the resonant circuit is ensured to operate stably and efficiently at a specific frequency, effectively avoiding efficiency reduction caused by frequency detuning, and precisely controlling the resonant characteristics to reduce energy loss. When the preset topology type is dual active bridge topology, by improving bidirectional energy transfer efficiency and determining the adaptation range of winding turns ratio and leakage inductance, this application optimizes the bidirectional energy flow characteristics of this topology, ensuring efficient energy conversion at the input and output ends and reducing losses caused by parasitic effects. This method of dynamically adjusting the optimization target according to the topology type enhances the versatility and adaptation accuracy of the method, enabling the optimization process based on the digital twin model to more accurately provide optimal initial operating parameters for a specific topology, thereby improving the overall operating efficiency and reliability of the inverter transformer.

[0056] In some of the solutions mentioned above in this application, a real-time acquisition method is proposed to obtain the operating status signal of the transformer and transmit it to the digital twin model. However, in this process, due to the lack of specific signal type or unclear transmission method, incomplete data and transmission delay may occur, affecting the accuracy of real-time model updates and dynamic control, and failing to guarantee the timing consistency of loss prediction and control.

[0057] In this regard, this application further proposes that the real-time acquisition method includes real-time acquisition of magnetic flux density, ambient temperature and winding current, and the acquired operating status signal is synchronously transmitted to the multi-physics digital twin model through a data transmission link.

[0058] Real-time data acquisition refers to the continuous or periodic acquisition of instantaneous values ​​of key transformer operating parameters during transformer operation by deploying appropriate sensors or measuring devices. Magnetic flux density is a major factor affecting core losses; its real-time acquisition can be achieved through Hall effect sensors, fluxgate sensors integrated inside or outside the core, or methods based on winding voltage integration. Ambient temperature is an important parameter affecting transformer heat dissipation and thermo-magnetic coupling effects; it is monitored in real-time using thermistors, thermocouples, or infrared temperature sensors. Winding current directly relates to the dynamic changes in winding losses and parasitic parameters; its real-time acquisition is achieved through current transformers, Hall effect current sensors, or shunts. Real-time acquisition of these parameters is fundamental for the accurate reflection of the physical entity's state by the digital twin model.

[0059] Furthermore, the data transmission link refers to the communication channel connecting the sensors on the physical transformer side and the computing platform where the digital twin model resides. This link can be implemented using wired or wireless methods. Wired transmission links, such as Ethernet, CAN bus, and RS485, feature high bandwidth and low latency, making them suitable for scenarios with large data volumes and high real-time requirements. Wireless transmission links, such as Wi-Fi, Bluetooth, LoRa, or 5G cellular networks, offer advantages in flexible deployment and low cost, making them suitable for distributed or difficult-to-wire environments. The synchronous transmission mechanism ensures that the different types of operational status signals collected are consistent in time; that is, all signals are collected within the same sampling period and transmitted to the digital twin model almost simultaneously. This can be achieved through a unified clock synchronization protocol such as NTP or PTP, or by embedding timestamps in data packets and having the receiving end perform time calibration, to avoid deviations in model state updates due to inconsistent data arrival times.

[0060] The above technical solution enables real-time acquisition of magnetic flux density, ambient temperature, and winding current, selectively addressing key parameters directly affecting transformer losses. Magnetic flux density is closely correlated with the dynamic changes in core losses, ambient temperature affects the thermo-magnetic coupling effect, and winding current reflects the dynamic characteristics of winding losses and parasitic parameters. This comprehensive parameter acquisition avoids prediction biases caused by signal gaps in existing methods, ensuring the digital twin model acquires sufficiently complete and accurate physical entity operational information. Simultaneously, the acquired operational status signals are synchronously transmitted to the multiphysics digital twin model via a data transmission link, utilizing a synchronization mechanism to guarantee the timeliness and consistency of data transmission. This allows the digital twin model to receive and process operational status signals from the physical transformer in real time, updating its internal loss distribution and parasitic parameters based on this real-time, synchronous data. Based on the constructed multiphysics digital twin model, this real-time, synchronous data stream provides high-precision, time-consistent input to the core loss sub-model, winding loss sub-model, parasitic parameter sub-model, and thermo-magnetic coupling sub-model, enhancing the digital twin model's ability to perceive the transformer's operating status and the accuracy of loss prediction. When the model detects that the loss distribution or parasitic parameters deviate from the preset range based on this real-time data, it can immediately trigger dynamic control operations. Due to the synchronous nature of data transmission, the issuance of control commands can also respond promptly to the latest state of the physical entity, thereby effectively suppressing losses and calibrating parameters, avoiding the problems of control lag and poor performance caused by data delays. Ultimately, through this high-precision, real-time synchronous data interaction, this application can ensure the time consistency between loss prediction and dynamic control, improving the real-time performance, accuracy, and effectiveness of the inverter transformer loss dynamic adaptation method.

[0061] In some of the solutions mentioned above in this application, dynamic control operation is proposed to suppress losses and calibrate parameters. In this process, more specific control methods are needed to effectively optimize magnetoresistive resistance, calibrate leakage inductance and suppress heat loss, so as to avoid response lag and insufficient global optimization caused by a single control method.

[0062] In response, this application further proposes dynamic control operations including core air gap length adjustment, winding effective turns switching, cooling system linkage control, and excitation current adjustment, which are used to achieve reluctance optimization, leakage inductance calibration, and heat loss suppression, respectively.

[0063] Core air gap length adjustment refers to changing the size of the physical air gap in the transformer core, thereby altering the magnetic reluctance in the magnetic circuit. This is crucial for optimizing the transformer's magnetizing inductance, saturation characteristics, and core loss distribution. The methods for achieving this can include, but are not limited to: using mechanical structures such as precision screws, stepper motor-driven sliders, or wedges to achieve precise and reversible adjustment of the air gap length; or using electromagnetic force to drive a movable core section and finely adjust the air gap by controlling the current magnitude, achieving rapid response and high-precision control; or utilizing the deformation property of piezoelectric materials under an electric field to drive micro-components to adjust the air gap.

[0064] Effective turns switching refers to changing the number of turns involved in electromagnetic conversion by altering the connection method of the transformer windings. This directly affects the transformer's turns ratio, leakage inductance, and impedance characteristics, and is crucial for performance optimization to adapt to different topologies and operating conditions. Implementation methods can include, but are not limited to: pre-setting multiple taps on the windings and selecting different tap combinations via relays, MOSFET switches, or mechanical switches to change the effective turns; or designing a switch matrix composed of semiconductor switches to dynamically connect different sections of the windings in series or parallel by controlling the on / off state of these switches, achieving a more flexible turns configuration.

[0065] Cooling system linkage control refers to dynamically adjusting the operating parameters of the cooling system based on the real-time temperature, loss distribution, and other operating conditions of the transformer to maintain the transformer within its optimal operating temperature range and effectively suppress heat loss and temperature drift. This can be achieved, but is not limited to: adjusting the cooling fan speed in real time based on temperature feedback from temperature sensors using pulse width modulation or other control algorithms to provide adequate heat dissipation; or, for systems using liquid cooling, dynamically adjusting the coolant flow rate by regulating the coolant pump speed or controlling the valve opening to control the heat dissipation effect.

[0066] Excitation current adjustment refers to controlling the magnetic flux density in the core by changing the magnitude or waveform of the current applied to the transformer's excitation winding. This directly affects the core's saturation, core losses, and the transformer's excitation characteristics. The methods used can include, but are not limited to: superimposing an adjustable DC bias current onto the AC excitation current to change the core's operating point, thereby affecting the core's magnetization curve and losses; or using pulse width modulation (PWM) technology to adjust the effective value or waveform of the excitation voltage, thereby controlling the magnitude and frequency components of the excitation current.

[0067] Through the above technical solutions, this application effectively solves the limitations of single control methods in suppressing losses and calibrating parameters by providing diversified dynamic control methods. Adjusting the air gap length of the magnetic core can precisely optimize reluctance and reduce core losses; switching the effective number of turns in the winding can flexibly calibrate leakage inductance, ensuring parasitic parameters are within preset ranges; the coordinated control of the cooling system and the adjustment of the excitation current work together to effectively suppress heat loss and maintain the thermal balance of the transformer. Guided by a multi-physics digital twin model, these control operations can achieve real-time and refined intervention in the transformer's operating status, thereby improving the efficiency of loss suppression and the accuracy of parameter calibration, ensuring that the transformer maintains efficient and stable operation under various operating conditions, and avoiding the problems of response lag and insufficient global optimization caused by single control methods.

[0068] In some of the embodiments described above in this application, the adjustment of the air gap length of the magnetic core is proposed to suppress magnetic core loss. However, in the process of its implementation, the adjustment method may lack a precise control mechanism and real-time response capability, resulting in unreasonable air gap change step size or lag in response, which cannot effectively adapt to the characteristics of the magnetic core material and dynamic loss changes, thereby affecting the loss suppression effect.

[0069] In this regard, this application further proposes that the above-mentioned magnetic core air gap length adjustment adopts a micro piezoelectric drive method, responding to the instructions issued by the above-mentioned multi-physics digital twin model, and adjusting the air gap length based on the preset step size determined by the magnetic core material characteristics and real-time loss data, so as to suppress magnetic core loss.

[0070] The air gap length adjustment of the magnetic core adopts a micro piezoelectric drive method. The micro piezoelectric actuator is fixed to the movable part of the transformer core, and its telescopic end abuts against the fixed part of the magnetic core. The control end of the micro piezoelectric actuator is connected to a drive circuit. The drive circuit receives the digital adjustment command issued by the multiphysics digital twin model and converts it into an analog voltage signal applied to the micro piezoelectric actuator. The digital twin model is equipped with a step size calculation unit. The step size calculation unit determines the optimal air gap adjustment step size under the current operating condition by looking up a table or interpolation algorithm based on the pre-stored BH curve database of the magnetic core material and the real-time input dynamic loss value of the magnetic core.

[0071] Miniature piezoelectric actuation refers to a driving technology that utilizes the inverse piezoelectric effect of piezoelectric materials to induce minute deformations by applying voltage, thereby achieving precise displacement control. This method features fast response speed, high positioning accuracy, small size, and low power consumption, providing extremely high displacement resolution and rapid response capability. It employs piezoelectric ceramic stack actuators, precisely controlling their expansion and contraction by controlling the voltage across their terminals; or it uses a piezoelectric cantilever beam structure, controlling its bending deformation through voltage. The adjustment of the magnetic core air gap length is not independent but involves closed-loop interaction with the aforementioned multiphysics digital twin model, responding to its issued commands. The multiphysics digital twin model generates and issues specific adjustment commands in real time based on its internal simulation and optimization results. Commands can be digital signals, converted into analog voltage signals required by the piezoelectric actuator by the controller; or commands can directly contain the target air gap length value, implemented through closed-loop control within the drive system. Furthermore, the step size for air gap adjustment is not fixed but a preset step size dynamically determined based on the characteristics of the magnetic core material and real-time loss data. The magnetic core material characteristics refer to the inherent properties of the magnetic core, such as its permeability, saturation flux density, and Curie temperature. These characteristics affect the degree to which air gap adjustment influences reluctance and loss. Real-time loss data refers to the dynamic loss value of the magnetic core calculated in real time by the digital twin model or acquired in real time by sensors, serving as the basis for adjustment decisions. The preset step size refers to the minimum or optimal adjustment amount that can effectively suppress loss, calculated by the optimization algorithm within the digital twin model based on the magnetic core material characteristics and real-time loss data under specific operating conditions. When the loss deviates significantly, the step size can be appropriately increased; when it approaches the optimal value, the step size can be decreased for fine adjustment. By precisely adjusting the air gap length, the magnetic reluctance of the magnetic circuit can be changed, thereby affecting the flux density distribution and the magnetization state of the magnetic core, thus optimizing the core loss. Under certain operating conditions, appropriately increasing or decreasing the air gap can reduce the saturation of the magnetic core, reducing hysteresis loss and eddy current loss.

[0072] Through the aforementioned technical solution, a micro-piezoelectric drive method is employed to achieve ultra-high precision and rapid response adjustment of the air gap length in the magnetic core, overcoming the insufficient precision and lag issues of traditional mechanical adjustment methods. Combined with the instructions issued by the aforementioned multi-physics digital twin model, this ensures that the air gap adjustment is based on a comprehensive analysis and optimization decision-making process considering the transformer's operating status, core material characteristics, and real-time loss data, avoiding blind or empirical adjustments. The preset step size determined based on core material characteristics and real-time loss data allows the air gap adjustment process to intelligently adapt to different operating conditions and core aging states, achieving refined and customized loss suppression. This closed-loop, intelligent, and high-precision air gap adjustment mechanism improves the effectiveness and real-time performance of core loss suppression, thereby further optimizing the overall operating efficiency and reliability of the inverter transformer.

[0073] In some of the embodiments described above in this application, a loss degradation prediction model is proposed to predict the aging trend of the magnetic core and the degradation of the winding performance. However, the model construction method is not clearly defined during its implementation, which may lead to insufficient prediction accuracy or failure to effectively capture time-series dynamic changes, thereby affecting the foresight and accuracy of the whole life cycle adaptive optimization.

[0074] In response, this application further proposes a loss degradation prediction model based on machine learning time series data analysis algorithm, which takes full life cycle operation data as input and outputs core aging trend and winding performance degradation results.

[0075] This loss degradation prediction model aims to quantitatively predict the performance degradation and loss increase trends of key components of inverter transformers, such as the magnetic core and windings, during long-term operation. Its core function is to identify and learn the intrinsic mechanisms and external influencing factors leading to loss degradation by analyzing historical operating data, thereby providing a forward-looking basis for transformer maintenance, operating parameter adjustment, and lifespan management.

[0076] The machine learning time series data analysis algorithm described above is a machine learning method specifically designed for processing and analyzing time-dependent data sequences. This algorithm can capture complex patterns in the data, such as long-term dependencies, periodicity, trends, and sudden events, thereby achieving accurate predictions of future states. The machine learning time series data analysis algorithm can be implemented in various ways. For example, it can employ architectures based on recurrent neural networks (RNNs), such as Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs), which excel at processing sequential data and remembering long-term information. Alternatively, it can use Transformer models based on attention mechanisms, which, through self-attention, can process sequential data in parallel and capture complex correlations between different time steps, thus exhibiting excellent performance when processing long-sequence data.

[0077] The full lifecycle operation data refers to various operation-related data continuously recorded and accumulated by a multiphysics digital twin model throughout the entire process of the inverter transformer from commissioning to decommissioning. This data comprehensively reflects the actual performance of the transformer under different operating conditions, environmental conditions, and control strategies. The full lifecycle operation data includes historical loss data, temperature drift records, and parameter control records. Historical loss data records the changes in core dynamic loss values ​​and winding dynamic loss values ​​over time; temperature drift records reflect the fluctuations in internal and external environmental temperatures of the transformer and their impact on performance; and parameter control records detail the adjustments made to transformer operating parameters and control strategies during operation and their corresponding effects.

[0078] Core aging trend refers to the predicted results of the degradation of transformer core material performance over time, obtained by analyzing full life-cycle operating data through a loss degradation prediction model. This aging may manifest as a decrease in magnetic permeability, an increase in hysteresis loss, and changes in eddy current loss, thereby affecting the overall efficiency and stability of the transformer. The output of this trend can be a performance curve changing over time, an aging degree index, or a predicted remaining life value.

[0079] Winding performance degradation results refer to the predicted results regarding the decline in the electrical and mechanical properties of transformer windings over time, obtained by analyzing full-lifecycle operating data using a loss degradation prediction model. This degradation may include increased winding resistance, decreased insulation performance, and weakened mechanical strength, leading to increased winding losses, localized overheating, and even short-circuit faults. The output of these results can be in the form of winding resistance change rate, insulation life prediction, or fault probability assessment.

[0080] The aforementioned technical solution concretizes the loss degradation prediction model into a model based on machine learning time-series data analysis algorithms, using full lifecycle operation data as input. This effectively addresses the shortcomings of traditional models in terms of prediction accuracy and dynamic time-series capture. Machine learning time-series data analysis algorithms can deeply mine the complex time dependencies, nonlinear relationships, and multi-factor coupling effects inherent in the full lifecycle operation data, thereby achieving accurate predictions of core aging trends and winding performance degradation. This high-precision prediction capability enables earlier and more accurate identification of potential performance degradation and fault risks during full-condition adaptive optimization, providing a forward-looking and reliable basis for dynamically adjusting transformer operating parameters and control strategies. When the system predicts that the core or winding is about to enter an accelerated aging stage, it can adjust the switching frequency, excitation current, or cooling strategy in advance to slow down the aging process, suppress losses, and extend the transformer's service life, improving the intelligence level and economic benefits of inverter transformer full lifecycle management.

[0081] In some of the solutions mentioned above in this application, full-condition adaptive optimization is proposed to dynamically adjust operating parameters and control strategies. However, in this process, due to the lack of predefined parameter combinations and control logic based on load fluctuation characteristics, the optimization efficiency is insufficient and the response is lagging when the load changes, making it impossible to quickly adapt to the loss suppression requirements under different operating conditions.

[0082] In this regard, this application further proposes that the full-condition adaptive optimization also includes the construction of a parameter control strategy library. The parameter control strategy library is established based on the load fluctuation characteristics and includes adaptive parameter combinations and control logic under light load, full load and overload conditions.

[0083] Building a parameter control strategy library refers to establishing a database or lookup table that pre-stores optimized parameters and control commands for different operating conditions and load characteristics. The purpose of this strategy library is to avoid performing complex real-time optimization calculations for every load change, thereby improving response speed and control efficiency. Through offline simulation, expert experience, or analysis of historical operating data, optimal combinations of operating parameters and corresponding control logic under different load conditions are pre-calculated and stored. Furthermore, this strategy library can also be in the form of a multidimensional lookup table, using load rate, ambient temperature, etc., as input indexes to directly output the corresponding optimized parameters and control commands; or it can employ a rule-based expert system, pre-setting a series of condition-action pairs, triggering the corresponding control strategy when specific load conditions are met.

[0084] The parameter control strategy library is built based on load fluctuation characteristics, meaning that the construction process of the strategy library fully considers the dynamic laws and characteristics of load changes in transformers during actual operation. Its function is to ensure that the control schemes in the strategy library can accurately match actual load changes, improving the targeting and effectiveness of control. By statistically analyzing historical load data, common load patterns, load change rates, load durations, and other characteristics are identified, and different load condition ranges are divided accordingly. For each range, corresponding control strategies are designed or optimized. Simultaneously, the typical load fluctuation curves and characteristics of the inverter can be analyzed in conjunction with the inverter's application scenarios, thereby enabling the targeted establishment of the strategy library.

[0085] The parameter control strategy library contains adaptive parameter combinations and control logic for light load, full load, and overload conditions. This is the specific content of the strategy library, which clarifies the preset schemes for different typical load conditions. Its function is to provide a comprehensive set of optimization strategies covering major operating scenarios, ensuring efficient loss suppression and parameter adaptation under various load conditions. For light load conditions, the strategy library may include parameter combinations to reduce switching frequency and adjust excitation current to reduce core losses, as well as corresponding low-power operation logic for the cooling system. For full load conditions, it may focus on optimizing winding current distribution and adjusting leakage inductance to balance winding losses and efficiency, in conjunction with efficient cooling strategies. For overload conditions, the strategy library may include control logic to limit output power, initiate emergency cooling, or adjust protective parameters to prevent device damage. These adaptive parameter combinations may include excitation inductance, leakage inductance, winding turns ratio, and core pre-magnetization parameters, while the control logic may include specific operation commands such as core air gap length adjustment, effective winding turns switching, cooling system linkage control, and excitation current adjustment.

[0086] By constructing a parameter control strategy library in the full-condition adaptive optimization and establishing it based on load fluctuation characteristics, including adaptive parameter combinations and control logic for light load, full load, and overload conditions, this application can effectively solve the problems of low optimization efficiency and slow response during load fluctuations. When the load condition of the inverter transformer changes, the system does not need to perform complex real-time optimization calculations, but can directly retrieve and apply the most suitable adaptive parameter combination and control logic from the pre-established strategy library. Under light load, it quickly switches to a low-loss operation mode; under full load, it can quickly adjust to a high-efficiency operation state; and under overload, it promptly initiates protective control. The introduction of this preset strategy reduces the burden of real-time calculation and response delay, enabling the transformer to achieve fast and accurate loss suppression and parameter calibration under various load conditions, thereby improving the operating efficiency and stability under all conditions and extending the service life of the transformer.

[0087] In some of the embodiments described above in this application, a multiphysics digital twin model is proposed to achieve loss prediction and dynamic control. During its implementation, due to insufficient data synchronization mechanism, there is a delay in the uploading of operating status signals and the issuance of control commands, which causes the loss prediction results and control actions to be out of sync in time, affecting the overall optimization effect and system stability.

[0088] In this regard, this application further proposes that the multiphysics digital twin model adopts a real-time data synchronization mechanism, which realizes the uploading of operating status signals and the issuance of dynamic control commands based on a high-frequency data transmission link, so as to ensure the timing consistency of loss prediction and control.

[0089] A multiphysics digital twin model is a virtual model integrating core loss sub-models, winding loss sub-models, parasitic parameter sub-models, and thermo-magnetic coupling sub-models. It can reflect the real-time operating status and loss distribution of the physical transformer. In this scheme, this model is not only the core of data processing and decision-making but also the receiving end and command issuing end of the real-time data synchronization mechanism. Its accuracy and response speed directly affect the dynamic adaptation effect of the entire system. The real-time data synchronization mechanism aims to ensure extremely low latency and high time accuracy in data exchange between the digital twin model and the physical entity. A timestamp-based data packet protocol is adopted to ensure that each data point carries accurate time information, and timestamp alignment and sorting are performed at the receiving end to eliminate timing deviations during transmission. Alternatively, industrial communication protocols such as deterministic Ethernet or time-sensitive networking are used to reserve bandwidth and time slots for critical data streams, thereby ensuring the real-time nature and predictability of data transmission.

[0090] A high-frequency data transmission link refers to a physical or logical channel capable of supporting high-speed, low-latency data transmission. Optical fiber communication links utilize the high speed and strong anti-interference capabilities of optical signals to enable rapid uploading of operational status signals and immediate issuance of control commands. Alternatively, high-speed Ethernet combined with an optimized TCP / IP protocol stack, or a dedicated industrial fieldbus, can be used to reduce transmission latency and increase data throughput through hardware acceleration and protocol optimization.

[0091] The purpose of uploading operational status signals is to enable the digital twin model to acquire the latest operational data of the physical transformer in real time, including magnetic flux density, ambient temperature, and winding current, as input for model updates and loss prediction. These operational status signals can be acquired in real time by sensors integrated into the transformer body or its control unit and uploaded to the digital twin model server in the form of data streams or periodic data packets via the aforementioned high-frequency data transmission link. The purpose of issuing dynamic control commands is to promptly transmit the control strategies generated by the digital twin model based on prediction and optimization results to the actuators of the physical transformer to achieve loss suppression and parameter calibration. These dynamic control commands, including core air gap length adjustment commands, effective winding turns switching commands, cooling system linkage control commands, or excitation current adjustment commands, are generated by the digital twin model based on real-time analysis results and issued to the transformer's intelligent actuators or controllers via the high-frequency data transmission link in the form of control commands or parameter settings.

[0092] The above technical solution ensures the temporal consistency of loss prediction and control, guaranteeing a high degree of synchronization between the loss prediction results from the digital twin model and the actual operating state of the physical transformer. Furthermore, it ensures that control commands issued based on these predictions can be applied to the physical transformer promptly and accurately, avoiding prediction distortion or control lag due to time delays, which could negatively impact optimization effectiveness and system stability. This can be achieved by using a unified time base and synchronization protocol throughout the entire data acquisition, transmission, model calculation, and command execution chain. All devices and systems are synchronized with a high-precision clock source, ensuring accurate timestamps. Low-latency algorithms and high-performance computing platforms are employed during model calculation and decision-making to minimize processing time.

[0093] The above technical solution solves the problem of misalignment between prediction and control timing caused by data synchronization delay. Based on the aforementioned multiphysics digital twin model, this real-time synchronization mechanism enables the model to more accurately capture dynamic changes in transformer operation, such as the real-time evolution of core losses, winding losses, parasitic parameters, and thermo-magnetic coupling effects. When the updated loss distribution or parasitic parameters deviate from the preset range, the model can quickly judge and make decisions based on the latest, time-consistent data, and promptly issue control commands, such as adjusting the core air gap length, switching the effective number of winding turns, coordinating cooling system control, or adjusting the excitation current. This tight closed-loop feedback and rapid response capability greatly improves the accuracy of loss prediction and the timeliness and effectiveness of dynamic control, thereby ensuring the adaptive optimization effect of the transformer under all operating conditions and improving the overall efficiency, reliability, and service life of the inverter system.

[0094] The following example will provide a more detailed explanation of the above technical solution: In a large-scale photovoltaic power plant, the inverter is the core equipment, and the performance of its internal isolation transformer affects the overall efficiency and reliability of the power plant. Due to the dynamic changes in sunlight intensity, ambient temperature, and load demand throughout the day, the transformer faces complex operating conditions, making it difficult to accurately predict and effectively control losses. Traditional transformer design methods are usually based on static design with fixed parameters, which cannot adapt to load fluctuations, temperature changes, and component aging in actual operation. This leads to low efficiency and even accelerated aging of the transformer under off-design conditions. Furthermore, existing loss prediction models are often simplified, failing to fully consider the skin effect, proximity effect, and coupling effects between multiple physical fields such as magneto-thermal and electro-magnetic fields at high frequencies, resulting in insufficient prediction accuracy. Simple feedback control methods also lack modeling of the complex dynamic mapping relationship between losses, parasitic parameters, and temperature, resulting in lag and inability to achieve global control. There is also a lack of a universal, model-optimized adaptation framework for different inverter topologies.

[0095] To address the aforementioned issues, this solution provides a method for dynamic adaptation of inverter transformer losses based on digital twins.

[0096] First, a multiphysics digital twin model was constructed for the inverter transformer used in this photovoltaic power station. This model integrates a core loss sub-model, a winding loss sub-model, a parasitic parameter sub-model, and a thermo-magnetic coupling sub-model. The core loss sub-model is constructed by fusing an improved core loss calculation algorithm with a three-dimensional finite element analysis algorithm. It receives basic transformer operating parameters, including core material properties, switching frequency, magnetic flux density, and ambient temperature, and outputs dynamic core loss values, achieving a preset accuracy for core dynamic loss prediction. This overcomes the limitations of traditional single-physics empirical formulas and improves the accuracy of core loss prediction. The winding loss sub-model introduces skin effect and proximity effect correction coefficients, combined with conductor structure parameters and winding process parameters, and outputs dynamic winding loss values, solving the problem of inaccurate winding loss prediction caused by traditional models neglecting high-frequency effects. The parasitic parameter sub-model is used to extract the transformer's leakage inductance and distributed capacitance, providing a basis for subsequent parameter calibration. The thermo-magnetic coupling sub-model is used to establish a dynamic mapping relationship between loss values ​​and temperature field, realizing bidirectional feedback between magnetic loss and thermal effect, and making up for the shortcomings of existing models that cannot reflect the influence of temperature rise on magnetic core characteristics.

[0097] Secondly, for the LLC resonant topology used in this photovoltaic inverter, optimization is performed based on the aforementioned digital twin model to determine the initial operating parameters. When the preset topology type is LLC resonant topology, the optimization objective is to match the resonant frequency and optimize the ratio of magnetizing inductance to leakage inductance. Through the digital twin model, the transformer performance under different parameter combinations can be simulated quickly and accurately, thereby determining the magnetizing inductance, leakage inductance, winding turns ratio, and core pre-magnetization parameters to adapt to the operating requirements of the LLC resonant topology. This avoids the drawback of traditional static design methods requiring separate parameter design for different topologies, providing a universal model optimization framework.

[0098] During actual transformer operation, sensors acquire real-time operating status signals, including magnetic flux density, ambient temperature, and winding current. These acquired operating status signals are synchronously transmitted to a multiphysics digital twin model via a high-frequency data transmission link, ensuring the temporal consistency of loss prediction and control. The digital twin model updates the loss distribution based on the operating status signals and corrects parasitic parameters online based on historical parasitic parameter calibration data, current temperature drift records, and an electromagnetic force deformation model caused by winding current. When the updated loss distribution or parasitic parameters deviate from the preset range, the system performs dynamic control operations to suppress losses and calibrate parameters. Dynamic control operations include core air gap length adjustment, effective winding turns switching, cooling system linkage control, and excitation current adjustment. Core air gap length adjustment uses a micro-piezoelectric drive method, responding to commands issued by the multiphysics digital twin model. Based on the core material characteristics and real-time loss data, it adjusts the air gap length in a preset step size to achieve reluctance optimization, thereby suppressing core losses. The effective number of turns switching of the winding is used for leakage inductance calibration, while the linkage control of the cooling system and the adjustment of the excitation current are used for heat loss suppression. This closed-loop real-time control mechanism overcomes the shortcomings of traditional simple feedback control, such as slow response and inability to achieve global optimal control.

[0099] A multiphysics digital twin model records the transformer's operational data throughout its entire lifecycle, including historical loss data, temperature drift records, and parameter control records. Based on this full-lifecycle operational data, a loss degradation prediction model is constructed. This prediction model is built using machine learning time-series data analysis algorithms, taking the full-lifecycle operational data as input and outputting the core aging trend and winding performance degradation results. Combining the output results of the prediction model with the load fluctuation characteristics of the photovoltaic power station, the system dynamically adjusts the transformer's operating parameters and control strategies to achieve adaptive optimization under all operating conditions. This adaptive optimization also includes building a parameter control strategy library, which is based on load fluctuation characteristics and includes suitable parameter combinations and control logic for light load, full load, and overload conditions. This forward-looking prediction and optimization enables the transformer to adapt to component aging during long-term operation, avoiding the performance degradation problem caused by the lack of full-lifecycle optimization in traditional technologies, and improving the transformer's full-lifecycle energy efficiency and reliability.

[0100] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for dynamic adaptation of inverter transformer losses based on digital twins, characterized in that, Includes the following steps: S1. Construct and run a multi-physics digital twin model of the transformer used in the inverter. The digital twin model integrates the core loss sub-model, winding loss sub-model, parasitic parameter sub-model, and thermo-magnetic coupling sub-model. The core loss sub-model and the winding loss sub-model both receive the basic operating parameters of the transformer and output the dynamic loss values ​​of the core and the winding, respectively. The basic operating parameters of the transformer include the core material properties, switching frequency, magnetic flux density and ambient temperature. The parasitic parameter sub-model is used to extract the leakage inductance and distributed capacitance of the transformer; The thermo-magnetic coupling sub-model is used to establish a dynamic mapping relationship between loss values ​​and temperature field; S2. Based on the digital twin model, optimize the preset topology type of the target inverter to determine the initial operating parameters. The initial operating parameters include magnetizing inductance, leakage inductance, winding turns ratio, and core pre-magnetization parameters, which are used to adapt to the operating requirements of the preset topology type. The preset topology type includes LLC resonant topology and dual active bridge topology. S3. The operating status signal of the transformer is acquired in real time through sensors and transmitted to the digital twin model; the digital twin model updates the loss distribution and parasitic parameters according to the operating status signal; the operating status signal includes magnetic flux density, ambient temperature and winding current; When the updated loss distribution or parasitic parameters deviate from the preset range, dynamic adjustment is performed to suppress loss and calibrate parameters. S4. Record the operating data of the transformer throughout its entire life cycle using the digital twin model, and construct a loss degradation prediction model based on the operating data; combine the output results of the prediction model with the load fluctuation characteristics to dynamically adjust the operating parameters and control strategies of the transformer to achieve adaptive optimization under all operating conditions. The full lifecycle operation data includes historical loss data, temperature drift records, and parameter control records.

2. The method for dynamic adaptation of inverter transformer losses based on digital twins according to claim 1, characterized in that, The core loss sub-model is constructed by integrating an improved core loss calculation algorithm with a three-dimensional finite element analysis algorithm. It takes the basic operating parameters of the transformer as input and outputs the dynamic loss value of the core, thereby achieving a core dynamic loss prediction with a preset accuracy.

3. The inverter transformer loss dynamic adaptation method based on digital twin according to claim 1, characterized in that, The winding loss sub-model introduces skin effect and proximity effect correction coefficients, and is constructed by combining conductor structure parameters and winding process parameters to output the dynamic loss value of the winding; the conductor structure parameters include conductor diameter and cross-sectional shape, and the winding process parameters include winding spacing and number of layers.

4. The inverter transformer loss dynamic adaptation method based on digital twin according to claim 1, characterized in that, When the preset topology type is LLC resonant topology, the optimization goal is to match the resonant frequency and optimize the ratio of magnetizing inductance to leakage inductance; when the preset topology type is dual active bridge topology, the optimization goal is to improve bidirectional energy transmission efficiency and determine the appropriate range of winding turns ratio and leakage inductance.

5. The method for dynamic adaptation of inverter transformer losses based on digital twins according to claim 1, characterized in that, The real-time acquisition method includes real-time acquisition of magnetic flux density, ambient temperature and winding current. The acquired operating status signals are synchronously transmitted to the multi-physics digital twin model through a data transmission link.

6. The method for dynamic adaptation of inverter transformer losses based on digital twins according to claim 1, characterized in that, The dynamic control operations include core air gap length adjustment, winding effective turns switching, cooling system linkage control, and excitation current adjustment, which are used to achieve reluctance optimization, leakage inductance calibration, and heat loss suppression, respectively.

7. The inverter transformer loss dynamic adaptation method based on digital twin according to claim 6, characterized in that, The air gap length adjustment of the magnetic core adopts a micro piezoelectric drive method. In response to the command issued by the multi-physics digital twin model, the air gap length is adjusted based on the preset step size determined by the magnetic core material characteristics and real-time loss data to suppress magnetic core loss.

8. The method for dynamic adaptation of inverter transformer losses based on digital twins according to claim 1, characterized in that, The loss degradation prediction model is constructed based on machine learning time series data analysis algorithm. It takes the full life cycle operation data as input and outputs the core aging trend and winding performance degradation results.

9. The inverter transformer loss dynamic adaptation method based on digital twin according to claim 1, characterized in that, The full-condition adaptive optimization also includes the construction of a parameter control strategy library. The parameter control strategy library is based on the load fluctuation characteristics and includes adaptive parameter combinations and control logic under light load, full load and overload conditions.

10. The method for dynamic adaptation of inverter transformer losses based on digital twins according to claim 1, characterized in that, The multiphysics digital twin model adopts a real-time data synchronization mechanism, which realizes the uploading of operating status signals and the issuance of dynamic control commands based on a high-frequency data transmission link, ensuring the timing consistency of loss prediction and control.

Images