A solid state transformer predictive maintenance method based on model parameter self-correction
By constructing a multi-level parameterized model and combining it with RLS and APF algorithms for self-calibration, the problem that the predictive maintenance model for solid-state transformers in the existing technology cannot adapt to time-varying dynamic characteristics is solved, and accurate condition assessment and predictive maintenance of SST are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- EAGLERISE MAGNETOELECTRIC TECH (JI AN) CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-23
AI Technical Summary
In existing predictive maintenance methods for solid-state transformers (SSTs), digital twin models are mostly static or offline calibrated, which cannot adapt to the time-varying dynamic characteristics of SSTs during actual operation, resulting in low reliability of the state assessment results output by the model.
A multi-level parameterized model of SST is constructed. Health feature vectors are extracted by collecting time-series operation data. The model parameters are self-calibrated by combining recursive least squares (RLS) and adaptive particle filter (APF) algorithms. Long short-term memory network (LSTM) is used to predict the remaining service life of components and generate early warning signals.
This improves the reliability of the state assessment results output by the SST multi-level parameterized model, enhances the accuracy and effectiveness of predictive maintenance decisions, and enables adaptive tracking and accurate prediction of the SST aging process.
Smart Images
Figure CN122264758A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of predictive maintenance technology for solid-state transformers, specifically a predictive maintenance method for solid-state transformers based on model parameter self-calibration. Background Technology
[0002] Solid-state transformers (SSTs), as core energy routers in future smart grids, directly impact grid reliability and power quality. Therefore, predictive maintenance of SSTs is of paramount importance. Current predictive maintenance methods for SSTs typically involve constructing digital twin models to simulate their dynamic behavior under different operating conditions, and then using the model outputs to conduct health status assessments and fault predictions. However, most digital twin models used for predictive maintenance of SSTs are static or offline calibrated models, making it difficult to effectively track the dynamic characteristics changes of SSTs caused by time-varying factors such as component aging and temperature drift during actual operation. As operating time increases, model accuracy gradually decreases, leading to reduced reliability of the status assessment results based on the outputs, thus affecting the accuracy and effectiveness of predictive maintenance decisions. Summary of the Invention
[0003] To address the aforementioned shortcomings, this invention proposes a predictive maintenance method for solid-state transformers (SSTs) based on model parameter self-calibration. The aim is to solve the problem that the digital twin models used in existing SST predictive maintenance methods are mostly static or offline calibrated models, which cannot adapt to the time-varying dynamic characteristics of SSTs during actual operation, resulting in low reliability of the state assessment results output by the models.
[0004] To achieve this objective, the present invention adopts the following technical solution: A predictive maintenance method for solid-state transformers based on model parameter self-calibration includes the following steps: Step S1: Collect timing operation data of the solid-state transformer (SST); Step S2: Based on the time-series running data of SST, extract health feature vectors related to the degradation process of SST components; Step S3: Construct a multi-level parameterized model of SST, and input the health feature vector into the multi-level parameterized model of SST for training to obtain an effective multi-level parameterized model of SST. Step S4: Input the real-time collected SST time series running data into the effective SST multi-level parameterized model for simulation calculation, and output the predicted value of SST running status; Step S5: Calculate the error value based on the predicted value of SST running status and the real-time collected SST time series running data; Step S6: Based on the error value, the algorithm combining Recursive Least Squares (RLS) and Adaptive Particle Filter (APF) is used to self-calibrate the parameters in the effective SST multi-level parameterized model to obtain the corrected model parameters. Step S7: Obtain the nominal values of the corrected model parameters, and calculate the health indicators of each component of SST based on the corrected model parameters and their nominal values. Step S8: Input the health index sequence of each component of SST into the preset Long Short-Term Memory (LSTM) network model for calculation, and output the remaining service life of each component of SST. Step S9: Determine whether the health index of any component in the SST is less than the first preset threshold, or whether the remaining service life of any component in the SST is less than the second preset threshold. If yes, generate an early warning signal and generate a corresponding predictive maintenance decision based on the early warning signal; otherwise, do not perform the decision generation operation.
[0005] Preferably, in step S3, the multi-level parameterization model of SST includes the device-level parameterization model of SST and the circuit-level parameterization model of SST. SST's device-level parameterization model includes a power MOSFET switching model, a capacitor RC degradation model, and an inductor loss model. The power MOSFET switching model includes a switching transient mathematical model during the turn-on phase and a switching transient mathematical model during the turn-off phase. The specific mathematical expression of the transient switching mathematical model during the conduction phase is as follows: ; in, This represents the instantaneous current between the drain and source stages during the conduction phase; Indicates the DC bus voltage; This represents the gate drive resistance; t represents the time variable. Indicates the conduction time constant; The specific mathematical expression for the transient mathematical model of the switching phase during the turn-off phase is as follows: ; in, This represents the instantaneous voltage between the drain and source during the turn-off phase. This represents the equivalent resistance between the drain and source of the power MOSFET when it is turned on. Indicates the output capacitance of the power MOSFET; The specific mathematical expression for the capacitor RC degradation model is as follows: ; ; in, This indicates the instantaneous capacitance of an electrolytic capacitor. This indicates the nominal capacitance of an electrolytic capacitor in its initial state. Aging rate coefficient representing capacitance; This indicates the cumulative operating time of the electrolytic capacitor; This indicates the amplitude of temperature fluctuations during operation; This represents the equivalent series resistance of an electrolytic capacitor. This represents the nominal equivalent series resistance of an electrolytic capacitor in its initial state. The aging rate coefficient represents the equivalent series resistance. This represents the effective value of the ripple current flowing through the capacitor; The specific mathematical expression for the inductor loss model is as follows: ; ; ; in, This represents the total power loss of the inductor; This represents the copper loss power of the inductor. This represents the iron loss power of the inductor; This represents the effective value of the ripple current flowing through the inductor winding; Indicates the inductor winding at the reference temperature DC resistance below; This indicates the temperature coefficient of resistance of the copper conductor; T represents the current operating temperature of the inductor winding. Indicates the hysteresis loss coefficient; The eddy current loss coefficient is represented by f; the alternating magnetic field frequency of the inductor during operation is represented by f. Indicates the magnitude of magnetic flux density; This represents the material constant of the magnetic core.
[0006] Preferably, the circuit-level parameterization model of SST includes a topology-based time-domain simulation model, wherein the state equations of the topology-based time-domain simulation model are as follows: ; in, Represents the state variables of the SST system; This represents the input vector of the SST system; and Both represent the SST system matrix; This represents a parameter vector.
[0007] Preferably, in step S5, the specific formula for calculating the error value is as follows: ; in, This represents the error value at time k; This represents the SST time-series running data collected in real time at time k; This represents the time step k, based on the parameter vector to be corrected. The predicted value of the SST running status output by the effective SST multi-level parameterized model.
[0008] Preferably, step S6 specifically includes the following sub-steps: Step S61: Update the parameter vectors in the effective SST multi-level parameterized model using the RLS algorithm. ,in, The specific update formula is as follows: ; ; ; in, This represents the parameter vector at time k; This represents the Kalman gain at time k; This represents the error value at time k; Let the covariance matrix at time k be represented. Indicates the forgetting factor; This represents the sensitivity vector at time k; Step S62: Use the APF algorithm to update the... Perform global sampling on the parameter space to find the global optimum. Correction results.
[0009] Preferably, in step S7, the specific calculation formulas for the health indicators of each component of the SST are as follows: ; in, This represents the health index of the i-th component in the SST at time k; n represents the total number of corrected model parameters for the i-th component in the SST. This represents the weight corresponding to the j-th corrected model parameter of the i-th component in SST; This represents the normalization function corresponding to the j-th corrected model parameter of the i-th component in SST; This represents the j-th corrected model parameter of the i-th component in the SST at time k; This represents the nominal value of the j-th corrected model parameter of the i-th component in SST.
[0010] Preferably, in step S8, the specific calculation formula for the remaining service life of each component of the SST is as follows: ; in, This represents the remaining service life of the i-th component in the SST at time k. Denotes the infimum function; Indicates the time offset; Indicates the failure threshold of health indicators; This represents the historical health index of the i-th component in SST from the initial time to the k-th time.
[0011] The technical solution provided by this invention may include the following beneficial effects: This scheme first constructs and trains a multi-level parameterized model of the SST to obtain an effective multi-level parameterized model of the SST. Then, an algorithm combining RLS and APF is used to perform online self-calibration of the parameters of the effective multi-level parameterized model of the SST. Next, the health indicators of each component of the SST are calculated based on the calibrated model parameters and their nominal values. Then, the health indicator sequences of each component of the SST are input into an LSTM model for calculation, and the remaining service life of each component of the SST is output. Finally, it is determined whether the health indicator or remaining service life of any component in the SST reaches the corresponding preset threshold. If it does, an early warning signal is triggered, and targeted predictive maintenance decisions are generated accordingly. Compared with existing SST predictive maintenance schemes based on static or offline calibrated digital twin models, this scheme uses a continuously self-calibrating multi-level parameterized model of the SST, which can adaptively evolve with the aging of the physical entity of the SST, improving the reliability of the state assessment results output by the multi-level parameterized model of the SST, thereby improving the accuracy and effectiveness of predictive maintenance decisions. Attached Figure Description
[0012] Figure 1 This is a flowchart illustrating the steps of a predictive maintenance method for solid-state transformers based on model parameter self-calibration. Detailed Implementation
[0013] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0014] A predictive maintenance method for solid-state transformers based on model parameter self-calibration includes the following steps: Step S1: Collect timing operation data of the solid-state transformer (SST); Step S2: Based on the time-series running data of SST, extract health feature vectors related to the degradation process of SST components; Step S3: Construct a multi-level parameterized model of SST, and input the health feature vector into the multi-level parameterized model of SST for training to obtain an effective multi-level parameterized model of SST. Step S4: Input the real-time collected SST time series running data into the effective SST multi-level parameterized model for simulation calculation, and output the predicted value of SST running status; Step S5: Calculate the error value based on the predicted value of SST running status and the real-time collected SST time series running data; Step S6: Based on the error value, the algorithm combining Recursive Least Squares (RLS) and Adaptive Particle Filter (APF) is used to self-calibrate the parameters in the effective SST multi-level parameterized model to obtain the corrected model parameters. Step S7: Obtain the nominal values of the corrected model parameters, and calculate the health indicators of each component of SST based on the corrected model parameters and their nominal values. Step S8: Input the health index sequence of each component of SST into the preset Long Short-Term Memory (LSTM) network model for calculation, and output the remaining service life of each component of SST. Step S9: Determine whether the health index of any component in the SST is less than the first preset threshold, or whether the remaining service life of any component in the SST is less than the second preset threshold. If yes, generate an early warning signal and generate a corresponding predictive maintenance decision based on the early warning signal; otherwise, do not perform the decision generation operation.
[0015] This scheme proposes a predictive maintenance method for solid-state transformers based on model parameter self-calibration, such as... Figure 1As shown, the first step is to collect the timing data of the solid-state transformer (SST). In this embodiment, collecting the timing data of the SST provides a complete and reliable data foundation for subsequent feature extraction and model training. Further, the timing data of the SST includes its voltage, current, and temperature. To ensure the capture of the dynamic characteristics of the SST, the acquisition rate of this timing data needs to be much higher than the fundamental frequency of the system. The second step is to extract health feature vectors related to the degradation process of the SST component based on the timing data of the SST. In this embodiment, by extracting health feature vectors related to the degradation process of the SST component from the timing data of the SST, interference data unrelated to the health status of the SST component can be effectively eliminated, reducing the complexity of subsequent model calculations. Further, the health feature vectors related to the degradation process of the SST component include energy loss during MOSFET switching, equivalent series resistance of capacitance, junction temperature fluctuation, and MOSFET switching time. The third step is to construct a multi-level parameterized model of the SST and input the health feature vector into the multi-level parameterized model of the SST for training, thereby obtaining an effective multi-level parameterized model of the SST. In this embodiment, the constructed multi-level parameterized model of the SST serves as a high-frequency synchronous digital twin, which can not only receive the same control signals as the physical solid-state transformer and realize synchronous simulation operation of the two, but also accurately map the device-level and circuit-level structural characteristics and operating rules of the SST. By training the multi-level parameterized model of the SST, it can fully learn the correlation between the degradation characteristics of SST components and the operating state, thereby improving the accuracy of subsequent operating state prediction. The fourth step is to input the real-time collected SST timing operation data into the effective multi-level parameterized model of the SST for simulation calculation and output the predicted value of the SST operating state. In this embodiment, by inputting the real-time collected SST timing operation data into the effective multi-level parameterized model of the SST for simulation calculation, the internal dynamic behavior of the solid-state transformer can be reproduced. The fifth step is to calculate the error value based on the predicted value of the SST running status and the real-time collected SST time series running data. In this embodiment, by comparing the predicted value of the SST running status and the real-time collected SST time series running data to calculate the error value, the deviation between the effective SST multi-level parameterized model and the SST physical entity can be accurately quantified, clearly reflecting the simulation accuracy of the effective SST multi-level parameterized model, and providing a clear correction basis for the subsequent self-calibration of model parameters.The sixth step is to perform self-calibration on the parameters in the effective SST multi-level parameterized model based on the error value, using an algorithm combining Recursive Least Squares (RLS) and Adaptive Particle Filtering (APF). This yields the calibrated model parameters. In this embodiment, the RLS algorithm can quickly track the time-varying changes in the parameters of the effective SST multi-level parameterized model, enabling rapid online parameter updates and adapting to real-time operating condition changes during SST operation. The APF algorithm can effectively handle parameter optimization problems in nonlinear and non-Gaussian scenarios, preventing the RLS algorithm from getting trapped in local optima and improving the robustness of parameter calibration. The seventh step is to obtain the nominal values of the calibrated model parameters and calculate the health indicators of each SST component based on these parameters and their nominal values. In this embodiment, the health indicators of each SST component are calculated by comparing the calibrated model parameters and their nominal values, achieving a quantitative assessment of the health status of each SST component. This provides a direct and accurate representation of the aging degree and health level of each SST component. The eighth step involves inputting the health indicator sequences of each component of the SST into a preset Long Short-Term Memory (LSTM) network model for computation, outputting the remaining lifespan of each component. In this embodiment, by inputting the health indicator sequences of each component into the LSTM model, the LSTM model can effectively capture the long-term evolution trend of the health indicator sequences of each component, providing reliable support for accurately predicting the remaining lifespan of each component. The ninth step involves determining whether the health indicator of any component in the SST is less than a first preset threshold, or whether the remaining lifespan of any component in the SST is less than a second preset threshold. If so, an early warning signal is generated, and a corresponding predictive maintenance decision is generated based on the early warning signal; otherwise, no decision generation operation is performed. In this embodiment, the first preset threshold is set to 0.85, and the second preset threshold is set to 1 hour. By determining whether the health indicator or remaining lifespan of any component in the SST is less than the corresponding preset threshold, the potential failure risks of the SST can be comprehensively and accurately identified, avoiding early warning omissions or false early warnings caused by single threshold judgments, thereby improving the accuracy and reliability of early warnings. By generating targeted predictive maintenance decisions based on early warning signals, on-demand and precise maintenance of the SST can be achieved.
[0016] This scheme first constructs and trains a multi-level parameterized model of the SST to obtain an effective multi-level parameterized model of the SST. Then, an algorithm combining RLS and APF is used to perform online self-calibration of the parameters of the effective multi-level parameterized model of the SST. Next, the health indicators of each component of the SST are calculated based on the calibrated model parameters and their nominal values. Then, the health indicator sequences of each component of the SST are input into an LSTM model for calculation, and the remaining service life of each component of the SST is output. Finally, it is determined whether the health indicator or remaining service life of any component in the SST reaches the corresponding preset threshold. If it does, an early warning signal is triggered, and targeted predictive maintenance decisions are generated accordingly. Compared with existing SST predictive maintenance schemes based on static or offline calibrated digital twin models, this scheme uses a continuously self-calibrating multi-level parameterized model of the SST, which can adaptively evolve with the aging of the physical entity of the SST, improving the reliability of the state assessment results output by the multi-level parameterized model of the SST, thereby improving the accuracy and effectiveness of predictive maintenance decisions.
[0017] Preferably, in step S3, the multi-level parameterization model of SST includes the device-level parameterization model of SST and the circuit-level parameterization model of SST. SST's device-level parameterization model includes a power MOSFET switching model, a capacitor RC degradation model, and an inductor loss model. The power MOSFET switching model includes a switching transient mathematical model during the turn-on phase and a switching transient mathematical model during the turn-off phase. The specific mathematical expression of the transient switching mathematical model during the conduction phase is as follows: ; in, This represents the instantaneous current between the drain and source stages during the conduction phase; Indicates the DC bus voltage; This represents the gate drive resistance; t represents the time variable. Indicates the conduction time constant; The specific mathematical expression for the transient mathematical model of the switching phase during the turn-off phase is as follows: ; in, This represents the instantaneous voltage between the drain and source during the turn-off phase. This represents the equivalent resistance between the drain and source of the power MOSFET when it is turned on. Indicates the output capacitance of the power MOSFET; The specific mathematical expression for the capacitor RC degradation model is as follows: ; ; in, This indicates the instantaneous capacitance of an electrolytic capacitor. This indicates the nominal capacitance of an electrolytic capacitor in its initial state. Aging rate coefficient representing capacitance; This indicates the cumulative operating time of the electrolytic capacitor; This indicates the amplitude of temperature fluctuations during operation; This represents the equivalent series resistance of an electrolytic capacitor. This represents the nominal equivalent series resistance of an electrolytic capacitor in its initial state. The aging rate coefficient represents the equivalent series resistance. This represents the effective value of the ripple current flowing through the capacitor; The specific mathematical expression for the inductor loss model is as follows: ; ; ; in, This represents the total power loss of the inductor; This represents the copper loss power of the inductor. This represents the iron loss power of the inductor; This represents the effective value of the ripple current flowing through the inductor winding; Indicates the inductor winding at the reference temperature DC resistance below; This indicates the temperature coefficient of resistance of the copper conductor; T represents the current operating temperature of the inductor winding. Indicates the hysteresis loss coefficient; The eddy current loss coefficient is represented by f; the alternating magnetic field frequency of the inductor during operation is represented by f. Indicates the magnitude of magnetic flux density; This represents the material constant of the magnetic core.
[0018] In this embodiment, the power MOSFET switching model accurately characterizes the electrical characteristic drift during the switching process through the current transient equation during the conduction phase and the voltage transient equation during the turn-off phase. The capacitor RC degradation model achieves real-time quantification of capacitance decay and ESR rise by using aging factors such as the cumulative operating time of the coupled electrolytic capacitor, the temperature fluctuation amplitude during operation, and the effective value of the ripple current flowing through the capacitor. The inductor loss model can accurately calculate the accumulated losses of the magnetic core and windings by separating copper losses and iron losses.
[0019] Preferably, the circuit-level parameterization model of SST includes a topology-based time-domain simulation model, wherein the state equations of the topology-based time-domain simulation model are as follows: ; in, Represents the state variables of the SST system; This represents the input vector of the SST system; and Both represent the SST system matrix; This represents a parameter vector.
[0020] In this embodiment, the time-domain simulation model based on topology is based on the actual topology of SST, and uses state variables... Characterize the core electrical quantities within the SST system, and through the system matrix and Accurate characterization of parameter vectors The coupling relationship between the model and the state variables and input vectors enables the model to realistically reproduce the dynamic response process of SST under different operating conditions.
[0021] Preferably, in step S5, the specific formula for calculating the error value is as follows: ; in, This represents the error value at time k; This represents the SST time-series running data collected in real time at time k; This represents the time step k, based on the parameter vector to be corrected. The predicted value of the SST running status output by the effective SST multi-level parameterized model.
[0022] In this embodiment, the error value is obtained by directly comparing it with the actual output of the SST physical entity. Output of an effective SST multilevel parameterized model This enables precise quantification of the deviation between the actual operating state of SST and the output of the effective SST multi-level parameterized model.
[0023] Preferably, step S6 specifically includes the following sub-steps: Step S61: Update the parameter vectors in the effective SST multi-level parameterized model using the RLS algorithm. ,in, The specific update formula is as follows: ; ; ; in, This represents the parameter vector at time k; This represents the Kalman gain at time k; This represents the error value at time k; Let the covariance matrix at time k be represented. Indicates the forgetting factor; This represents the sensitivity vector at time k; Step S62: Use the APF algorithm to update the... Perform global sampling on the parameter space to find the global optimum. Correction results.
[0024] In this embodiment, in step S61, in During the update process, through Control parameter update step size, while utilizing The uncertainty in quantifying parameter estimation enables the RLS algorithm to quickly converge the parameter estimates to the true values while maintaining stability. Further explanation involves the forgetting factor. Used to track time-varying parameters. In step S62, the APF algorithm updates the parameters using RLS. The importance suggestion distribution is globally sampled across the entire parameter space, which can effectively prevent the RLS algorithm from getting trapped in local optima, thereby ensuring the global convergence of parameter correction.
[0025] Preferably, in step S7, the specific calculation formulas for the health indicators of each component of SST are as follows: ; in, This represents the health index of the i-th component in the SST at time k; n represents the total number of corrected model parameters for the i-th component in the SST. This represents the weight corresponding to the j-th corrected model parameter of the i-th component in SST; This represents the normalization function corresponding to the j-th corrected model parameter of the i-th component in SST; This represents the j-th corrected model parameter of the i-th component in the SST at time k; This represents the nominal value of the j-th corrected model parameter of the i-th component in SST.
[0026] In this embodiment, by adopting a weighted normalization method, the multi-dimensional corrected model parameters of each component of the SST are mapped into a unified quantitative health indicator, thereby achieving a comprehensive assessment of the health status of each component of the SST.
[0027] Preferably, in step S8, the specific calculation formula for the remaining service life of each component of the SST is as follows: ; in, This represents the remaining service life of the i-th component in the SST at time k. Denotes the infimum function; Indicates the time offset; Indicates the failure threshold of health indicators; This represents the historical health index of the i-th component in SST from the initial time to the k-th time.
[0028] In this embodiment, the infimum function is used. In historical health indicators Under the premise of finding the first time a health indicator falls below its failure threshold The time offset is used to obtain the remaining service life. This design enables accurate prediction of remaining service life.
[0029] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.
[0030] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A predictive maintenance method for solid-state transformers based on model parameter self-calibration, characterized in that: Includes the following steps: Step S1: Collect timing operation data of the solid-state transformer (SST); Step S2: Based on the time-series running data of SST, extract health feature vectors related to the degradation process of SST components; Step S3: Construct a multi-level parameterized model of SST, and input the health feature vector into the multi-level parameterized model of SST for training to obtain an effective multi-level parameterized model of SST. Step S4: Input the real-time collected SST time series running data into the effective SST multi-level parameterized model for simulation calculation, and output the predicted value of SST running status; Step S5: Calculate the error value based on the predicted value of SST running status and the real-time collected SST time series running data; Step S6: Based on the error value, the algorithm combining Recursive Least Squares (RLS) and Adaptive Particle Filter (APF) is used to self-calibrate the parameters in the effective SST multi-level parameterized model to obtain the corrected model parameters. Step S7: Obtain the nominal values of the corrected model parameters, and calculate the health indicators of each component of SST based on the corrected model parameters and their nominal values. Step S8: Input the health index sequence of each component of SST into the preset Long Short-Term Memory (LSTM) network model for calculation, and output the remaining service life of each component of SST. Step S9: Determine whether the health index of any component in the SST is less than the first preset threshold, or whether the remaining service life of any component in the SST is less than the second preset threshold. If yes, generate an early warning signal and generate a corresponding predictive maintenance decision based on the early warning signal; otherwise, do not perform the decision generation operation.
2. The predictive maintenance method for solid-state transformers based on model parameter self-calibration according to claim 1, characterized in that: In step S3, the multi-level parameterization model of SST includes the device-level parameterization model of SST and the circuit-level parameterization model of SST. SST's device-level parameterization model includes a power MOSFET switching model, a capacitor RC degradation model, and an inductor loss model. The power MOSFET switching model includes a switching transient mathematical model during the turn-on phase and a switching transient mathematical model during the turn-off phase. The specific mathematical expression of the transient switching mathematical model during the conduction phase is as follows: ； in, This represents the instantaneous current between the drain and source stages during the conduction phase; Indicates the DC bus voltage; This represents the gate drive resistance; t represents the time variable. Indicates the conduction time constant; The specific mathematical expression for the transient mathematical model of the switching phase during the turn-off phase is as follows: ； in, This represents the instantaneous voltage between the drain and source during the turn-off phase. This represents the equivalent resistance between the drain and source of the power MOSFET when it is turned on. Indicates the output capacitance of the power MOSFET; The specific mathematical expression for the capacitor RC degradation model is as follows: ； ； in, This indicates the instantaneous capacitance of an electrolytic capacitor. This indicates the nominal capacitance of an electrolytic capacitor in its initial state. Aging rate coefficient representing capacitance; This indicates the cumulative operating time of the electrolytic capacitor; This indicates the amplitude of temperature fluctuations during operation; This represents the equivalent series resistance of an electrolytic capacitor. This represents the nominal equivalent series resistance of an electrolytic capacitor in its initial state. The aging rate coefficient represents the equivalent series resistance. This represents the effective value of the ripple current flowing through the capacitor; The specific mathematical expression for the inductor loss model is as follows: ； ； ； in, This represents the total power loss of the inductor; This represents the copper loss power of the inductor. This represents the iron loss power of the inductor; This represents the effective value of the ripple current flowing through the inductor winding; Indicates the inductor winding at the reference temperature DC resistance below; This indicates the temperature coefficient of resistance of the copper conductor; T represents the current operating temperature of the inductor winding. Indicates the hysteresis loss coefficient; The eddy current loss coefficient is represented by f; the alternating magnetic field frequency of the inductor during operation is represented by f. Indicates the magnitude of magnetic flux density; This represents the material constant of the magnetic core.
3. The predictive maintenance method for solid-state transformers based on model parameter self-calibration according to claim 2, characterized in that: SST's circuit-level parameterization model includes a topology-based time-domain simulation model, where the state equations of the topology-based time-domain simulation model are as follows: ； in, Represents the state variables of the SST system; This represents the input vector of the SST system; and Both represent the SST system matrix; This represents a parameter vector.
4. The predictive maintenance method for solid-state transformers based on model parameter self-calibration according to claim 1, characterized in that: In step S5, the specific formula for calculating the error value is as follows: ； in, This represents the error value at time k; This represents the SST time-series running data collected in real time at time k; This represents the time step k, based on the parameter vector to be corrected. The predicted value of the SST running status output by the effective SST multi-level parameterized model.
5. The predictive maintenance method for solid-state transformers based on model parameter self-calibration according to claim 4, characterized in that: Step S6 specifically includes the following sub-steps: Step S61: Update the parameter vectors in the effective SST multi-level parameterized model using the RLS algorithm. ,in, The specific update formula is as follows: ； ； ； in, This represents the parameter vector at time k; This represents the Kalman gain at time k; This represents the error value at time k; Let the covariance matrix at time k be represented. Indicates the forgetting factor; This represents the sensitivity vector at time k. Step S62: Use the APF algorithm to update the... Perform global sampling on the parameter space to find the global optimum. Correction results.
6. The predictive maintenance method for solid-state transformers based on model parameter self-calibration according to claim 5, characterized in that: In step S7, the specific calculation formulas for the health indicators of each component of SST are as follows: ； in, This represents the health index of the i-th component in the SST at time k; n represents the total number of corrected model parameters for the i-th component in the SST. This represents the weight corresponding to the j-th corrected model parameter of the i-th component in SST; This represents the normalization function corresponding to the j-th corrected model parameter of the i-th component in SST; This represents the j-th corrected model parameter of the i-th component in the SST at time k; This represents the nominal value of the j-th corrected model parameter of the i-th component in SST.
7. The predictive maintenance method for solid-state transformers based on model parameter self-calibration according to claim 6, characterized in that: In step S8, the specific calculation formula for the remaining service life of each component of SST is as follows: ； in, This represents the remaining service life of the i-th component in the SST at time k. Denotes the infimum function; Indicates the time offset; Indicates the failure threshold of health indicators; This represents the historical health index of the i-th component in SST from the initial time to the k-th time.