Management method of intelligent power electronic converter

Abstract

The invention relates to the technical field of power electronics, discloses a management method of an intelligent power electronic converter, and aims to solve the problem that in the prior art, operation performance and long-term reliability cannot be collaboratively optimized. According to the method, voltage, current, temperature and vibration data are collected in real time through a multi-mode sensor, and a high-fidelity digital twinborn model fusing electromagnetism, thermodynamics and component aging mechanisms is constructed and calibrated online; based on the model, a multi-objective optimization function including efficiency, electric energy quality, residual life and accumulated damage is established in a prediction time domain, a Pareto optimal control parameter set is solved by adopting a multi-objective evolutionary algorithm, and an optimal solution is selected according to a dynamic utility function to generate a driving instruction. The system comprises a state sensing module, a digital twin modeling module, a prediction optimization module, a strategy decision module and a driving execution module, full-dimension self-adaptive closed-loop management of performance and service life is achieved, and the intelligent level, economical efficiency and reliability of the converter are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of power electronics technology, specifically relating to a management method for an intelligent power electronic converter. Background Technology

[0002] With the rapid development of smart grids and renewable energy systems, power electronic converters, as core devices for power conversion and control, are increasingly becoming crucial for ensuring system stability in terms of operational reliability and fault diagnosis accuracy. Modern converters generally integrate multi-source sensors and embedded processing units to achieve real-time monitoring of key parameters such as current, temperature, and drive signals.

[0003] In addition, traditional fault diagnosis methods usually use fixed thresholds or single criteria for judgment, lacking the ability to dynamically assess the confidence of multi-source sensor data under noise interference, making it difficult to distinguish between real faults and external interference, resulting in a high false alarm rate, which seriously affects system availability and operation and maintenance efficiency.

[0004] In existing technologies, although some solutions have attempted to introduce fuzzy logic or machine learning models for fault identification to compensate for the above-mentioned shortcomings, they generally suffer from high computational latency, large resource consumption, and tight coupling with the underlying hardware, making it difficult to complete closed-loop decisions on a microsecond timescale. More importantly, most systems have not completely decoupled the diagnostic module from the control layer, making the diagnostic results susceptible to disturbances by control strategies, and making it impossible to independently verify the effectiveness of the diagnostic logic without interrupting operation.

[0005] Therefore, there is an urgent need for an intelligent management method and system that is deeply integrated into the hardware monitoring layer and supports dynamic weighting and adaptive threshold generation, so as to fundamentally solve the problem of false alarms caused by power grid noise interference. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a management method for intelligent power electronic converters that can overcome the defects of existing power electronic converter management strategies, such as the disconnect between operating performance and long-term reliability and the inability to optimize them in a coordinated manner.

[0007] To address the aforementioned technical problems, this invention provides a management method for intelligent power electronic converters. The core concept involves constructing a high-fidelity digital twin model coupled with multiple physics fields, which is mapped and evolves synchronously with the physical converter in real time. This model not only accurately reproduces the instantaneous electromagnetic transients and thermodynamic behavior of the converter but also incorporates a deep aging and degradation model of key components based on physical failure mechanisms. Based on this digital twin model, a multi-objective optimization function is constructed within the predicted future time domain, integrating performance indicators such as immediate operating efficiency and output power quality with reliability indicators such as the expected remaining lifespan and cumulative damage of key components. By solving this multi-objective optimization function online, a set of Pareto optimal control parameters that maximizes the overall utility of the converter in its current and future states is obtained. The optimal solution is then selected from these parameters to generate real-time drive commands, thereby achieving comprehensive, forward-looking, and adaptive closed-loop management of the converter from performance to lifespan.

[0008] According to one aspect of the present invention, a management method for an intelligent power electronic converter is provided, comprising the following steps: By deploying a multi-modal sensor array inside the power electronic converter, the converter's multi-dimensional, multi-timescale operating status data stream is acquired in real time. The operating status data stream includes: nanosecond-resolution voltage waveform data acquired by a wide-bandgap semiconductor voltage divider arm connected in parallel with the power switching devices; megahertz-bandwidth current waveform data acquired by a planar Rogowski coil integrated on the bus and bridge arm paths; millisecond-refresh-rate two-dimensional temperature distribution map data acquired by a miniature thermocouple array or infrared thermal imaging sensor array covering the power modules, DC bus capacitors, and magnetic components; and broadband vibration signal data reflecting mechanical stress and vibration state acquired by a piezoelectric accelerometer mounted on the converter housing or main circuit board.

[0009] Based on the real-time acquired operational status data stream, a high-fidelity digital twin model is constructed and continuously calibrated online. This high-fidelity digital twin model integrates an electromagnetic field model, a thermodynamic model, and a multi-component physical failure and aging model within a unified computational framework. The electromagnetic field model is established using the finite element method or state-space averaging method to accurately simulate the voltage stress, current ripple, and electromagnetic interference characteristics of the converter under different switching frequencies and modulation strategies. The thermodynamic model, based on Fourier's law of heat conduction, establishes a three-dimensional thermal resistance network along the entire path from the power device chip junction temperature to the substrate, heat sink, and environment, used to calculate the thermal resistance based on the electromagnetic field model. The calculated power loss is analyzed in real time to determine the transient temperature field distribution inside the converter. The multi-component physical failure and aging model is established for key components. Specifically, for insulated gate bipolar transistors or silicon carbide metal oxide semiconductor field-effect transistors, a bond wire fatigue model based on the Norris-Lansberg equation and a solder layer fatigue model based on the coffin-Manson model are established. For DC bus-supported thin film capacitors, a dielectric aging model based on Arrhenius's law and an electrode metallization layer evaporation model considering self-healing effects are established. For high-frequency inductors, a core loss and magnetostriction aging model based on the extended Steinmetz formula is established.

[0010] The continuous online calibration process specifically involves: using the real-time acquired operating status data stream as the observation input; using the state variables inside the high-fidelity digital twin model, including junction temperature, equivalent series resistance, thermal resistance, etc., as the state vector to be estimated; and applying the extended Kalman filter algorithm or the unscented Kalman filter algorithm to recursively correct and update the internal parameters of the digital twin model, ensuring that the deviation between the model state and the physical entity state is constrained within a preset error threshold, which is within five percent.

[0011] Based on the calibrated high-fidelity digital twin model, a multi-objective optimization function is constructed within a preset prediction time domain. The objective vector of the multi-objective optimization function includes at least two mutually balancing sub-objective functions: a performance index sub-objective function and a reliability index sub-objective function. The performance index sub-objective function is composed of a weighted sum of the converter system efficiency function and the output voltage ripple function. The system efficiency is calculated by simulating the energy loss and transmission energy of a complete switching cycle in the digital twin model, and the output voltage ripple is obtained by performing a fast Fourier transform on the simulated output voltage. The reliability index sub-objective function is composed of the maximum value function of the expected remaining service life consumption of all modeled key components in the converter within the prediction time domain. The remaining service life consumption of each component is calculated by integrating the future temperature cycle profile and electrical stress profile predicted by the digital twin model based on its corresponding physical failure and aging model.

[0012] A multi-objective evolutionary algorithm is employed to solve the multi-objective optimization function in a rolling manner to generate a set of Pareto optimal control parameters. Specifically, the multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm, which encodes the adjustable control parameters of the converter, including switching frequency, dead time, modulation ratio, and fan speed or water pump flow rate of the active thermal management system, into chromosome individuals. By performing non-dominated sorting and crowding calculation on the population, and applying genetic operators such as selection, crossover, and mutation, the population is iteratively evolved until it converges to a Pareto front that covers the entire performance and reliability trade-off space. Each solution on this front corresponds to a set of control parameters, representing an optimal trade-off between performance and reliability.

[0013] According to a preset system-level operation strategy, a set of optimal control parameters is dynamically selected from the Pareto optimal control parameter set. The system-level operation strategy defines the trade-off preference between performance and reliability, specifically embodied in a utility function defined on the Pareto front. This utility function dynamically adjusts its weight coefficients based on external instructions, such as grid dispatch instructions or instructions from the upper-level energy management system. For example, during peak electricity price periods or when the system requires rapid dynamic response, the utility function assigns higher weights to performance indicators, thereby selecting a solution closer to the high-performance region on the Pareto front. When the system is in steady-state operation or needs to ensure long-term service, higher weights are assigned to reliability indicators, selecting a solution closer to the high-reliability region.

[0014] The selected optimal control parameters are converted into specific hardware drive instructions and applied to the underlying drive circuit and thermal management actuator of the power electronic converter through the control interface. Specifically, the optimal switching frequency and modulation ratio parameters are input to the pulse width modulation module in the digital signal processor to generate a high-precision gate drive signal sequence. The optimal dead time parameter is loaded into the dead time control register of the gate driver in real time. The optimal fan speed or water pump flow rate parameter is sent to the controller of the thermal management system through the serial communication interface, thus forming a complete closed-loop intelligent management loop from sensing, modeling, prediction, optimization to execution.

[0015] In summary, this application includes at least one of the following beneficial technical effects: First, this invention constructs a high-fidelity digital twin model that couples multiple physical fields such as electricity, heat, and aging. For the first time, it integrates the instantaneous operating performance and long-term service reliability of power electronic converters into a unified, accurate, and predictable mathematical framework for integrated modeling. This fundamentally solves the technical problem of the separation of performance control and health management in the prior art, which leads to the loss of one for the other.

[0016] Second, this invention adopts a prediction-based multi-objective optimization method, which transforms control decision-making from passively responding to the current state to actively planning the future state. By solving the Pareto optimal solution set, it provides converter managers with a decision-making basis for making quantitative and transparent trade-offs between performance and lifespan, realizing a control paradigm shift from "best effort" to "global optimality".

[0017] Third, the management system constructed by this invention forms a complete autonomous intelligent loop from multimodal deep perception to real-time synchronization of digital twins, then to predictive optimization decision-making, and finally to precise closed-loop execution. This enables the converter to have unprecedented self-adaptive and self-optimizing capabilities, and can dynamically adjust its operating mode according to its own health status and external task requirements. In this way, while ensuring the completion of tasks, it maximizes the comprehensive value of its entire life cycle, and significantly improves the intelligence level, operating economy and task reliability of power electronic equipment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the management method for the intelligent power electronic converter proposed in this invention; Figure 2 This is a schematic diagram of the multi-physics coupled high-fidelity digital twin model and multi-objective optimization function in this invention; Figure 3 This is a schematic diagram of the management system for the intelligent power electronic converter proposed in this invention. Detailed Implementation

[0019] This invention provides a management method for intelligent power electronic converters. Its core lies in constructing a high-fidelity digital twin model coupled with multiple physics fields, which is mapped and evolves synchronously with the physical converter in real time. Based on this model, predictive multi-objective optimization control that integrates operational performance and long-term reliability is implemented. The specific implementation steps of this method will be described in detail below to ensure the full disclosure and feasibility of the technical solution.

[0020] Reference Figures 1 to 3 The management method for intelligent power electronic converters begins with step S1, which is as follows: A multi-modal sensor array deployed inside the power electronic converter is used to collect and acquire multi-dimensional, multi-timescale operational status data streams of the converter in real time. This data stream covers multiple physical dimensions, including electromagnetic, thermal, and mechanical aspects.

[0021] Specifically, the voltage waveform data with nanosecond resolution is acquired by a wide-bandgap semiconductor voltage divider arm connected in parallel with the power switching device. This voltage divider arm is made of silicon carbide or gallium nitride material and has high bandwidth and low parasitic capacitance characteristics, which can accurately capture voltage overshoot and oscillation during the switching transient process.

[0022] The planar Rogowski coil, integrated into the busbar and bridge arm paths, acquires current waveform data with a bandwidth of megahertz. This coil is manufactured using flexible printed circuit board technology and features low insertion loss and high common-mode rejection ratio, accurately reproducing the rising and falling edge details of high-frequency switching current.

[0023] Two-dimensional temperature distribution data with millisecond refresh rates are obtained by using miniature thermocouple arrays or infrared thermal imaging sensor arrays covering the surfaces of power modules, DC bus capacitors, and magnetic components. The miniature thermocouple arrays are arranged at micrometer-level intervals on the surfaces of key heat sources, while the infrared thermal imaging sensors provide the temperature distribution across the entire field in a non-contact manner. The data from both are fused after time alignment to form a thermal field observation with high spatiotemporal resolution.

[0024] Wideband vibration signal data reflecting mechanical stress and vibration state is acquired by a piezoelectric accelerometer mounted on the converter housing or main circuit board. The frequency response range of the sensor covers 10 Hz to 10 kHz and is used to monitor structural vibrations caused by electromagnetic forces, thermal expansion or external impacts.

[0025] All sensor data are synchronously acquired via a high-precision analog-to-digital converter at a sampling rate of no less than 10 MHz, and then timestamped to form a unified multimodal operating status data stream.

[0026] After completing step S1, proceed to step S2: based on the real-time collected operational status data stream, construct and continuously calibrate a high-fidelity digital twin model online.

[0027] This model integrates an electromagnetic field model, a thermodynamic model, and a multi-component physical failure and aging model within a unified computational framework. The electromagnetic field model is established using the state-space averaging method, abstracting the converter's main circuit topology as a linear time-varying system controlled by switching functions. Its state variables include inductor current, capacitor voltage, and bus voltage. The model input is a sequence of gate drive signals, and the outputs are the branch currents and node voltages.

[0028] This model can accurately simulate the voltage stress, current ripple, and common-mode electromagnetic interference characteristics generated under different switching frequencies and pulse width modulation strategies. The thermodynamic model is based on Fourier's law of heat conduction and constructs a three-dimensional thermal resistance network throughout the entire path from the junction temperature of the power device chip to the ceramic substrate, to the heat sink base, and finally to the ambient air.

[0029] The network consists of multiple series and parallel thermal resistance-thermal capacity units, each corresponding to a physical layer or interface. Its parameters are determined through a combination of finite element thermal simulation and experimental calibration. The input to the thermodynamic model is the power of each loss source calculated by the electromagnetic field model, including conduction loss, switching loss, and drive loss. The output is the transient temperature at each critical location.

[0030] Multi-component physical failure and aging models were established for three types of core components: For insulated-gate bipolar transistors (IGBTs) or silicon carbide metal-oxide-semiconductor field-effect transistors (SFETs), a bond wire fatigue model based on the Norris-Lansberg equation was established, and the cumulative damage was measured. Temperature cycle amplitude The number of iterations N is determined by the loop count, and the expression is: ,in and is a material constant.

[0031] Simultaneously, a fatigue model of the solder layer based on the coffin-Manson model was established, and its damage amount was determined. ,in The range of plastic strain is determined by the mismatch of thermal expansion coefficients and temperature gradient. n_solder is the toughness coefficient of the solder material, and C_solder is the fatigue index of the solder material.

[0032] For DC bus-supported thin-film capacitors, a dielectric aging model based on Arrhenius's law is established. The relationship between its lifetime L_cap and operating temperature T and electric field strength E is as follows: ,in , , For material parameters, Boltzmann's constant; An electrode metallization layer evaporation model considering the self-healing effect was established, and the rate of increase of its equivalent series resistance with time was: ,in, dt represents the minute change in the equivalent series resistance of the capacitor, where dt is the time interval. Let be the evaporation rate constant. It is the activation energy for evaporation; For high-frequency inductors, a core loss model based on the extended model of the Steinmetz formula is established, and its unit volume loss is... Where f is the frequency, B is the magnetic flux density amplitude, and α and β are material indices; and a magnetostrictive aging model is established, in which the mechanical stress σ_mag is proportional to the magnetic flux density change rate dB / dt, and long-term action leads to winding insulation fatigue.

[0033] The continuous online calibration process is as follows: The operating status data stream acquired in step S1 is used as the observation input. The state variables within the digital twin model, including the junction temperature T_j of the power devices, the equivalent series resistance ESR of the capacitors, and the thermal resistance R_th, are used as the state vector x to be estimated. The extended Kalman filter algorithm is applied to recursively correct the model parameters. The state prediction equation of the extended Kalman filter is: , where f is the nonlinear state transition function of the digital twin model, u is the control input, k represents the time step, and k-1 represents the previous or preceding time step; The observation equation is Where h is the observation function and v_k is the observation noise. The filter linearizes the system by calculating the Jacobian matrix and updates the state estimate and covariance matrix each time new data arrives, ensuring that the deviation between the model output and the physical entity observations is constrained to within 5%. The calibrated model state remains highly synchronized with the physical entity, providing a reliable basis for subsequent predictions.

[0034] After completing step S2, proceed to step S3: Based on the calibrated high-fidelity digital twin model, construct a multi-objective optimization function within the preset prediction time domain T_horizon. The objective vector of this function includes the performance index sub-objective function J_perf and the reliability index sub-objective function J_reli.

[0035] The performance index sub-objective function is defined as follows: η_sim is the system efficiency obtained from the digital twin model simulation, which is obtained by calculating the ratio of output energy to input energy in a complete switching cycle; V_ripple_rms is the effective value of the output voltage ripple, which is calculated by integrating the fundamental and upper harmonic components after performing a fast Fourier transform on the simulated output voltage sequence; w_η and w_ripple are normalized weighting coefficients that satisfy w_η + w_ripple = 1.

[0036] The sub-objective function of the reliability index is defined as follows: Where D_igbt is the cumulative fatigue damage of the insulated-gate bipolar transistor in the prediction time domain, obtained by integrating the bonding wire and solder layer model; D_cap is the dielectric aging damage of the capacitor, defined as the ratio of actual lifetime consumption to rated lifetime in the prediction time domain; and D_ind is the cumulative damage of the inductor's core and windings. This multi-objective optimization function aims to simultaneously minimize performance loss and reliability risk.

[0037] After step S3 is completed, step S4 is executed: a non-dominated sorting genetic algorithm is used to solve the multi-objective optimization function in a rolling manner to generate a set of Pareto optimal control parameters.

[0038] The algorithm encodes the adjustable control parameters of the converter as individual chromosomes, specifically including the switching frequency f_sw, dead time t_dead, modulation ratio m, and fan speed n_fan of the active thermal management system. The initial population size is set to one hundred, and the maximum number of iterations is fifty. In each generation, the control parameters corresponding to each individual are first simulated in the digital twin model for T_horizon time, and their J_perf and J_reli values ​​are calculated; then, the population is non-dominated and sorted, and the individuals are assigned to different frontier levels.

[0039] Next, the crowding distance of individuals within the same front is calculated to maintain solution diversity; finally, a new generation of population is generated through tournament selection, simulated binary crossover, and polynomial mutation. After the algorithm converges, the output is a Pareto front covering the performance-reliability trade-off space, where each solution on the front corresponds to a set of control parameters, representing an optimal compromise.

[0040] Then, step S5 is executed: based on the preset system-level operating strategy, a set of optimal control parameters is dynamically selected from the Pareto optimal control parameter set.

[0041] This system-level operational strategy is embodied in a utility function. The weighting coefficient α is dynamically set by external commands. When a grid dispatch command is received requiring the converter to operate at maximum efficiency, α is set to 0.9, the utility function is performance-biased, and the solution with the minimum J_perf on the Pareto front is selected.

[0042] When the upper-level energy management system instructs that the equipment lifespan be extended, α is set to 0.3, the utility function is biased towards reliability, and the solution with the minimum J_reli is selected. Without specific instructions, α defaults to 0.6, achieving balanced optimization. The selection process involves traversing all solutions on the Pareto front, calculating their utility function values, and selecting the one with the minimum U as the optimal control parameter.

[0043] Finally, step S6 is executed: the selected optimal control parameters are converted into specific hardware drive instructions, which are then applied to the underlying drive circuit and thermal management actuator of the power electronic converter through the control interface. Specifically, the optimal switching frequency f_sw and modulation ratio m are input to the pulse width modulation module in the digital signal processor. This module generates a high-precision gate drive signal sequence based on the carrier frequency and modulation waveform. Its dead time is loaded in real time into the gate driver's dedicated control register by the optimal dead time t_dead parameter, ensuring that the upper and lower bridge arms do not shoot-through.

[0044] The optimal fan speed parameter n_fan is sent to the microcontroller of the thermal management system via a serial peripheral interface. The microcontroller then adjusts the duty cycle of the pulse width modulation signal to drive the fan motor to the target speed. The generation and issuance of the above instructions are completed within each control cycle, forming a complete closed-loop intelligent management circuit from sensing, modeling, prediction, optimization to execution.

[0045] Furthermore, the multi-component physical failure and aging model includes an online identification sub-model for the equivalent series resistance of the DC bus capacitors. This sub-model injects a sinusoidal disturbance voltage signal with a frequency of 10 kHz and an amplitude of one-thousandth of the rated voltage into the DC bus during normal converter operation intervals.

[0046] Voltage and current responses caused by disturbances are synchronously acquired using a wide-bandgap semiconductor voltage divider arm and a planar Rogowski coil. Utilizing the principle of a lock-in amplifier, the current signal and a reference signal are subjected to phase-sensitive detection to extract in-phase and quadrature components, thereby calculating the complex impedance Z=V / I of the capacitor at that frequency. The equivalent series resistance ESR is the real part of the impedance. This ESR value is fed back to the digital twin model in real time to update the state variables of the capacitor aging model, improving the accuracy of health state estimation.

[0047] Furthermore, the predictive multi-objective optimization module dynamically adjusts the length of its prediction time domain, T_horizon, during rolling solution. The system continuously monitors the rate of change of input voltage and output current. When the rate of change of input voltage exceeds 10% per millisecond, or the step amplitude of load current exceeds 30% of the rated value, it is determined to be a drastic change in operating conditions, and T_horizon is automatically shortened from the default ten seconds to one second to reduce computational load and improve control response speed. When the system runs continuously for more than five minutes and the rate of change of all state variables is lower than the system's preset threshold, it is determined to be in steady-state operation, and T_horizon is automatically extended to thirty seconds to obtain more accurate long-term lifetime prediction and generate a better global control strategy.

[0048] As one embodiment of the present invention, the system-level operation strategy includes a self-learning mechanism. This mechanism allocates a dedicated storage area in the embedded computing unit to record historical data for each optimization cycle, including the Pareto front at that time, the selected control parameters, the actual operating efficiency, the measured temperature changes, and subsequent fault warning information.

[0049] Using this data, the system periodically runs a reinforcement learning algorithm based on Q-learning. Its state space is a feature vector representing operating conditions, its action space consists of discrete values ​​of the weight coefficient α, and its reward function is defined as a comprehensive score of actual performance, including task completion, energy efficiency, and health maintenance. Through continuous trial and error and experience replay, the algorithm autonomously optimizes the selection strategy of α, enabling the system to gradually adapt to specific application scenarios without human intervention, thus achieving continuous evolution of the control strategy.

[0050] Finally, corresponding to the methods described above, such as Figure 3 As shown, this invention also provides a management system for an intelligent power electronic converter. The system includes a multimodal state perception module, whose hardware comprises a wide-bandgap semiconductor voltage divider arm, a planar Rogowski coil, a miniature thermocouple array or infrared thermal imaging sensor array, and a piezoelectric accelerometer, used for data acquisition in step S1. A digital twin modeling and synchronization module, which is an embedded computing unit, internally contains software programs for electromagnetic field models, thermodynamic models, and multi-component physical failure and aging models, and runs an extended Kalman filter algorithm routine, used for model construction and calibration in step S2. A predictive multi-objective optimization module, which is the parallel processing core within the embedded computing unit, contains multi-objective optimization function definitions and a non-dominated sorting genetic algorithm program, used for steps S3 and S4. A control strategy decision and instruction generation module, which is a logic processing unit within the embedded computing unit, stores effective functions and connects to an external strategy instruction interface, used for step S5. A driver execution interface module, which includes a digital signal processor, gate driver circuits, and a serial communication interface, used for instruction issuance in step S6. The modules are interconnected via a high-speed internal bus, forming a highly integrated closed-loop intelligent management system.

[0051] 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 present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0052] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A management method for an intelligent power electronic converter, characterized in that, include: By deploying a multi-modal sensor array inside the power electronic converter, the converter's multi-dimensional operating status data stream is collected and acquired in real time; Based on real-time acquired operational status data streams, a high-fidelity digital twin model that couples and integrates electromagnetic field model, thermodynamic model and multi-component physical failure and aging model is constructed and continuously calibrated online. During the continuous online calibration process, the extended Kalman filter algorithm or the unscented Kalman filter algorithm is applied to recursively correct and update the internal parameters of the digital twin model to ensure that the deviation between the model state and the physical entity state is constrained within the preset error threshold. Based on the calibrated high-fidelity digital twin model, a multi-objective optimization function is constructed within a preset prediction time domain. The objective vector of the multi-objective optimization function includes a performance index sub-objective function and a reliability index sub-objective function. The reliability index sub-objective function is based on the expected remaining service life consumption of key components within the prediction time domain calculated by a multi-component physical failure and aging model. A multi-objective evolutionary algorithm is used to solve the multi-objective optimization function in a rolling manner to generate a set of Pareto optimal control parameters; Based on the preset system-level operation strategy, a set of optimal control parameters is dynamically selected from the Pareto optimal control parameter set. The selected optimal control parameters are converted into specific hardware drive instructions and applied to the underlying drive circuit and thermal management actuator of the power electronic converter through the control interface.

2. The management method for an intelligent power electronic converter according to claim 1, characterized in that, The continuous online calibration process includes: The real-time acquired operational status data stream is used as the observation input, and the state variables inside the high-fidelity digital twin model are used as the state vector to be estimated; the state variables include junction temperature, equivalent series resistance, and thermal resistance; The extended Kalman filter algorithm or the unscented Kalman filter algorithm is applied to recursively correct and update the internal parameters of the digital twin model, ensuring that the deviation between the model state and the physical entity state is constrained to within five percent.

3. The management method for an intelligent power electronic converter according to claim 2, characterized in that, The performance index sub-objective function is defined as follows: ,in The system efficiency obtained from simulation of the digital twin model. This is the effective value of the output voltage ripple. and These are normalized weighting coefficients; the sub-objective function of the reliability index is defined as follows: ,in This represents the cumulative fatigue damage of an insulated-gate bipolar transistor. This represents the amount of dielectric damage caused by aging in the capacitor. This represents the cumulative damage to the inductor's core and windings.

4. The management method for an intelligent power electronic converter according to claim 3, characterized in that, The multi-objective evolutionary algorithm is specifically a non-dominated sorting genetic algorithm, and the solution process of the non-dominated sorting genetic algorithm includes: The switching frequency, dead time, modulation ratio, and fan speed or water pump flow rate are encoded as chromosome individuals; In the digital twin model, the control parameters corresponding to each individual are simulated in the prediction time domain, and the sub-objective function values ​​of its performance index and reliability index are calculated. The population is subjected to non-dominated sorting and crowding calculation, and selection, crossover, and mutation genetic operators are applied to iteratively evolve the population until it converges to the Pareto front of the space that covers the entire trade-off between performance and reliability.

5. The management method for an intelligent power electronic converter according to claim 4, characterized in that, The utility function in the system-level operation strategy is defined as follows: The weighting coefficient α is dynamically set according to external instructions; when a power grid dispatch instruction is received requiring high-efficiency operation, α is set to 0.9; when an instruction is received from the upper-level energy management system requiring extended lifespan, α is set to 0.3; and when there are no special instructions, α defaults to 0.

6.

6. The management method for an intelligent power electronic converter according to claim 5, characterized in that, The process of translating optimal control parameters into hardware driver instructions includes: The optimal switching frequency and modulation ratio parameters are input into the pulse width modulation module in the digital signal processor to generate a high-precision gate drive signal sequence. The optimal dead time parameters are loaded into the dead time control register of the gate driver in real time. The optimal fan speed or water pump flow rate parameters are sent to the controller of the thermal management system via a serial communication interface.

7. The management method for an intelligent power electronic converter according to claim 6, characterized in that, The multi-component physical failure and aging model further includes an online identification sub-model for the equivalent series resistance of the DC bus capacitor. The online identification sub-model injects a small disturbance signal of a specific frequency into the converter and uses the principle of a lock-in amplifier to extract the real part of the capacitor's impedance at that frequency from the collected voltage and current data as the equivalent series resistance, and updates the capacitor health status parameters in the digital twin model in real time.

8. The management method for an intelligent power electronic converter according to claim 7, characterized in that, The length of the prediction time domain is dynamically adjustable; when a drastic change in the input power or load is detected, the prediction time domain is automatically shortened to improve the response speed; when the system enters a long period of stable operation, the prediction time domain is automatically extended to obtain a more accurate long-term life prediction.

Images