NPC converter closed-loop feedback data-driven IGBT saturation voltage drop monitoring method

Abstract

The application discloses an NPC converter closed-loop feedback data-driven IGBT saturation voltage drop monitoring method and relates to the field of electronic power device state monitoring, which comprises the following steps: collecting operation data of an NPC energy storage converter; performing analog-digital conversion on the collected operation data; performing data cleaning on the operation data; deriving a mathematical model of the energy storage converter, substituting the processed data into the mathematical model; performing curve fitting, calculating a target function, finding an optimal solution to minimize the target function, obtaining IGBT identification voltage drop parameters; according to an IGBT output characteristic curve, extracting parameters under a healthy state, comparing the parameters with the identification parameters to perform state evaluation; introducing a parameter self-correction or online correction mechanism; classifying IGBT health states through normal distribution; and realizing device state online evolution evaluation and abnormal early warning. The application adopts the above method, simulates the state and behavior of an actual IGBT by using a digital model, and thus realizes predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of electronic power device condition monitoring technology, and in particular to an IGBT saturation voltage drop monitoring method driven by closed-loop feedback data in an NPC converter. Background Technology

[0002] In medium- and high-voltage power conversion applications, traditional two-level inverters, due to their large output voltage waveform distortion, high voltage stress, and complex filter design, are no longer sufficient to meet the demands for high-performance and high-efficiency power conversion. To overcome these shortcomings, three-level neutral point clamped (NPC) converters have become one of the mainstream choices for medium- and high-voltage converter structures due to their significant advantages, such as multiple output voltage levels, low device voltage stress, and low harmonic content. The NPC structure introduces a neutral point clamping diode in each phase arm, ensuring that each power device only bears half the DC bus voltage, effectively improving the safety margin of the devices and the overall reliability of the system. It is widely used in new energy power generation, high-voltage frequency conversion, smart grids, and large-scale energy storage.

[0003] Insulated-gate bipolar transistors (IGBTs) are key components of energy storage converters and are widely used in renewable energy grid-connected power generation, energy storage systems, electric vehicles, and high-voltage direct current transmission. However, according to industrial survey data, the failure rate of power devices is 20%, and IGBTs, as critical components in power electronic systems, are also among the most prone to failure. Their failures can be mainly divided into two categories: sudden failures and aging failures. Sudden failures are usually caused by overcurrent or overvoltage; these failures occur instantaneously and are difficult to predict. Aging failures accumulate gradually during long-term normal operation. If aging-related IGBT failures are not detected in advance, they may spread from a single component failure to damage the entire power electronic system, leading to serious consequences. Therefore, effective monitoring and maintenance of IGBTs are crucial to prevent potential major losses. However, energy storage converter topologies are relatively complex, operating states are diverse, and integration and power density are increasing year by year. Meanwhile, IGBT condition monitoring remains a technical challenge that needs to be addressed. When effective condition monitoring and fault diagnosis measures are lacking, energy storage systems may experience shutdowns or damage, affecting the reliable operation of the overall system.

[0004] Condition monitoring can be used for predictive maintenance based on estimated degradation levels and the remaining lifespan of components of interest. Component-level health metrics for monitoring the degradation of power semiconductors and capacitors have been proposed, and can be categorized into two types: electrical metrics and thermal metrics. Furthermore, electrical metrics, such as the on-state voltage or resistance of a power semiconductor, can be obtained from the drain-source / collector-emitter terminals and measured using measurement circuitry. Among existing metrics, these exhibit higher sensitivity to power semiconductor degradation. However, they require additional circuitry, increasing implementation complexity.

[0005] From a system-level perspective, various methods have been proposed to monitor power semiconductors and capacitors separately. The frequency response of a converter is sensitive to the on-resistance of the power semiconductor. Furthermore, the output current harmonics of the converter are used to monitor the degradation of the solder layer on the power semiconductor. Both of these methods require additional setup and are intrusive to the target system. Moreover, they cannot distinguish between the degradation of power semiconductors and capacitors. Artificial neural networks are also a potential method for monitoring capacitor degradation; however, this requires offline testing to obtain sufficient training data. Summary of the Invention

[0006] The purpose of this invention is to provide a closed-loop feedback data-driven IGBT saturation voltage drop monitoring method for NPC converters, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides a method for monitoring IGBT saturation voltage drop driven by closed-loop feedback data in an NPC converter, comprising the following steps:

[0008] S1. Collect the operating data of the NPC energy storage converter on the experimental platform, including closed-loop control data and data collected by sensors;

[0009] S2. Use the data acquisition board to perform analog-to-digital conversion on the acquired running data, converting analog signals into digital signals;

[0010] S3. Clean the running data using the Kalman algorithm;

[0011] S4. Derive the mathematical model of the energy storage converter and substitute the cleaned operating data into the mathematical model.

[0012] S5. Build a model library and selection mechanism to reduce dependence on operating conditions;

[0013] S6. The trust region reflection nonlinearity minimum error sum of squares parameter identification method is used to perform curve fitting, calculate the objective function, find the optimal solution that minimizes the objective function, and obtain the IGBT identification voltage drop parameter.

[0014] S7. Based on the IGBT output characteristic curve obtained from the datasheet, extract the parameters under the healthy state and compare them with the identified parameters to conduct a state assessment.

[0015] S8. Based on the existing mathematical model, introduce a parameter self-calibration or online correction mechanism;

[0016] S9. Introduce a multi-dimensional state-space model based on the identified voltage drop parameters and historical device data, and classify the IGBT health status through a normal distribution;

[0017] S10, a dynamic threshold judgment module based on temperature and aging evolution trends, is used to realize online evolution evaluation of device status and early warning of anomalies.

[0018] Preferably, step S4 includes:

[0019] S41. When the NPC converter is running in grid-connected mode, collect the three-phase voltage, three-phase current and DC side voltage in real time.

[0020] S42. The measured three-phase voltage needs to be processed by a phase-locked loop to obtain the phase angle θ. The three-phase current is combined with the phase angle θ signal to perform Park transformation to obtain the dq axis current.

[0021] The S43 and dq axis currents are adjusted by proportional-integral control to generate three-phase voltage reference values, which are then controlled by a sinusoidal pulse width modulation strategy to regulate the on / off state of the switching devices in the three-phase inverter.

[0022] Preferably, step S43 includes:

[0023] Based on the circuit topology and Kirchhoff's voltage law, the mathematical model of the three-phase voltage of the NPC energy storage converter is established as follows:

[0024]

[0025] In the formula, u ao′ u represents the voltage at the output point of phase A bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. ao u represents the voltage at the output point of phase A bridge arm of the converter relative to the neutral point O of the power grid. oo′ Represents the common-mode voltage, u bo′ u represents the voltage at the output point of the B-phase bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. bo u represents the voltage at the output point of the B-phase bridge arm of the converter relative to the neutral point O of the power grid. co′ u represents the voltage at the output point of the C-phase bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. co This represents the voltage at the output point of the C-phase bridge arm of the converter relative to the neutral point O of the power grid.

[0026] Based on the three-phase voltage balance of the grid-connected converter:

[0027] u ao +u bo +u co =0;

[0028] Converter common mode voltage u o′o Represented as:

[0029]

[0030] Phase voltage reconstruction is expressed as:

[0031]

[0032] Preferably, the mathematical model for the three-phase output current of the energy storage converter is expressed as follows:

[0033]

[0034] In the formula, i a i b and i c These represent the currents in phases A, B, and C, respectively, where t is the time variable and L... a L b and L c R represents the filter inductance for phases A, B, and C, respectively. a R represents the sum of the stray resistance of the A-phase filter inductor and the line parasitic resistance. b R represents the sum of the stray resistance of the B-phase filter inductor and the line parasitic resistance. c e represents the sum of the stray resistance of the C-phase filter inductor and the line parasitic resistance. a e b and e c L represents the grid voltage of phases A, B, and C, respectively. g Let L be the sum of the filter inductance and the grid-side inductance, and L = L i +L g L i L represents the AC side filter inductor. g Let R be the equivalent inductance of the power grid, and R be the sum of the equivalent series resistance of the filter inductor and the resistance of the power grid, and R = R i +R g R i For inductor L i The stray resistance on, R g The resistance is the equivalent impedance of the power grid.

[0035] Preferably, the mathematical model of the three-phase output current of the PCS is solved using the Runge-Kutta method, and the inductor current in the next sampling interval is expressed as:

[0036]

[0037] In the formula, h is the time step between the nth sampling point and the (n+1)th sampling point, and k is the average rate of change between the nth sampling point and the (n+1)th sampling point. a1 k b1 and k c1 These are the initial values ​​of the current variable i. a i b and i c Rate of change estimation, k a2 k b2 and k c2 They are using k a1 k b1 and k c1 The current variable i obtained after the first intermediate estimation a i b and i c rate of change, k a3 k b3 and k c3 They are respectively at k a2 k b2 and k c2 Based on the intermediate estimation of the current variable i a i b and i c rate of change, k a4 k b4 and k c4 These are the further intermediate estimates of the current variable i. a i b and i c The rate of change;

[0038]

[0039] In the formula, f = [di a / dt,di b / dt,di c / dt],a n b n and c n Let i represent the current variable at step n. a i b and i c Approximate values;

[0040]

[0041] In the formula, i a,n+1 This represents the calculated predicted value of the current in phase A at the (n+1)th discrete time step, used to describe the current state of the system at subsequent times. b,n+1This represents the calculated predicted value of the current in phase B at the (n+1)th discrete time step, used to describe the current state of the system at subsequent times. c,n+1 This represents the calculated predicted value of the current in phase C at the (n+1)th discrete time step, used to describe the current state of the system at subsequent times, V. oi Represents the threshold voltage of a certain IGBT, R oni Represents the on-resistance of a specific IGBT, V di Represents the threshold voltage of a certain anti-parallel diode, R doni Represents the on-resistance of a certain anti-parallel diode, i = 1, 2, 3, 4, 5, 6, i a,n This represents the value of the phase A current at the nth discrete time step. It is the current state quantity at the current moment and serves as the basic input for calculating the current at the next time step (n+1). b,n This represents the value of the B-phase current at the nth discrete time step. It is the current state quantity at the current moment and serves as the basic input for calculating the current at the next (n+1)th time step. c,n This represents the value of the C-phase current at the nth discrete time step. It is the current state quantity at the current moment and serves as the basic input for calculating the current at the next (n+1)th time step.

[0042] Preferably, step S5 specifically involves: model library + selection mechanism, which specifically involves: constructing multiple circuit mathematical sub-models based on the operating conditions of the energy storage converter PCS, and selecting the most suitable model for voltage drop identification through a real-time operating condition identification mechanism.

[0043] Preferably, step S6 includes:

[0044] S61. Using the parameter identification method of minimum sum of squared errors for trust region reflection nonlinearity, the physical model and the mathematical analytical model are curve-fitted to solve for the coefficients x, resulting in:

[0045]

[0046] In the formula, F(·) represents the model function, x represents the model coefficients to be solved, and y represents the actual observed data;

[0047] S62 and TRRLS are essentially solutions to optimization problems, calculating s to minimize q(s). The trust region subproblem can be expressed as:

[0048]

[0049] In the formula, s is the search step size, and H is the search step size. k D is a Hessian matrix. k It is a diagonal scale matrix, Δ k For the trust region scale, g k Let q be the gradient of f(x).k (s) represents a quadratic approximation model of the objective function in the trust region subproblem, s T Represents the search direction vector;

[0050] S63, if f(x) k+1 ) <f(x k If x k+1 =x k +α k s k α k Constrained by the boundary, α k The step size factor is represented by the reflective trust region method, which "reflects" the solution back into the trust region; otherwise, the iteration point remains unchanged, and the trust region is further reduced.

[0051] S64, the identification voltage drop parameter of IGBT is contained in u ao′ In the example of phase A, the phase voltage is expressed as:

[0052]

[0053] In the formula, U dc V represents the DC voltage source voltage. ce1 This represents the on-state voltage of the upper arm IGBT, V. F1 This represents the forward voltage of the anti-parallel diode in the upper bridge arm, V. ce2 This indicates the on-state voltage of the lower IGBT arm, V. F2 Indicates the forward voltage of the anti-parallel diode in the lower bridge arm, sign(i a ) represents the direction of the current. When the current flows from the energy storage converter PCS to the grid, sign(i a If the value is 1, then the value is 0; otherwise, the value is 0. Indicates the direction of current flow; when current flows from the grid to the energy storage converter PCS, Q is 1 if the input is positive and 0 otherwise. Q represents the IGBT's on / off state, as shown below:

[0054]

[0055] S65. Based on the operating and switching states of the energy storage converter, an ideal model is established as follows:

[0056]

[0057] Preferably, step S8 specifically involves: based on the derivation of the mathematical model of the energy storage converter, dynamically correcting the model parameters by combining actual sampling data, and adopting a sliding window-based error minimization correction strategy to improve the online accuracy and adaptability of the model.

[0058] Preferably, step S9 includes:

[0059] S91. Examine the histogram and kernel density curve of the measurement data to determine whether they are approximately a unimodal symmetrical normal distribution;

[0060] S92. Calculate the mean and standard deviation of the data to describe the central tendency and dispersion of the data;

[0061] S93. Apply the “3σ criterion”: Define a valid data range around the mean and remove data outside that range;

[0062] S94. Recalculate statistical characteristics and verify distribution patterns through iterative optimization.

[0063] Preferably, step S10 includes:

[0064] The dynamic threshold judgment module for temperature and aging evolution trend utilizes the temperature characteristics of the device and the aging evolution trend of the voltage drop parameter during long-term operation to construct a multi-dimensional threshold model for temperature compensation.

[0065] By fitting historical data, the growth trend of saturated pressure drop at different temperatures is obtained, and dynamic adjustments are made in conjunction with real-time monitoring values.

[0066] When real-time data deviates significantly from the predicted aging curve, the dynamic threshold judgment module for temperature and aging evolution trend automatically triggers an abnormal warning.

[0067] Therefore, the IGBT saturation voltage drop monitoring method driven by closed-loop feedback data of NPC converter described above has the following beneficial effects:

[0068] (1) It can monitor the working status and health status of IGBT in real time without disassembling or damaging the equipment; it is accomplished by correlating external electrical signals (such as voltage, current, temperature) with the health status of IGBT; it infers the loss and life of IGBT by collecting specific external signals and analyzing their characteristics, thus avoiding the actual measurement of the physical parameters inside the IGBT and achieving non-invasiveness.

[0069] (2) It can extract effective health status indicators from the external electrical signals of IGBT and perform status assessment based on them to ensure high accuracy and reliability; it can use digital models to simulate the status and behavior of actual IGBTs to achieve predictive maintenance.

[0070] (3) Based on the scenario of grid connection of energy storage converter power generation, this invention uses the nonlinear minimum error sum of squares algorithm to monitor the health parameters of IGBT. In order to minimize the impact of interference signals in the collected data on the accuracy of parameter identification, the feature data sampled by the physical converter is cleaned using the Kalman algorithm to filter out noise and interference from other signals, and then the optimal parameter set is output; otherwise, if the global optimal parameter set is close to the true value of the parameters in the physical circuit, it proves that the parameter identification method is effective.

[0071] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0072] Figure 1 These are the on-state voltage drop aging failure curves of six IGBT devices according to embodiments of the present invention;

[0073] Figure 2 This is a control block diagram of the current inner loop controller according to an embodiment of the present invention;

[0074] Figure 3 This is a simplified structural diagram of the energy storage converter according to an embodiment of the present invention;

[0075] Figure 4 The output characteristics of the IGBT in this embodiment of the invention;

[0076] Figure 5 This is a flowchart illustrating the nonlinear minimum error sum of squares in an embodiment of the present invention;

[0077] Figure 6 The grid-connected current i of the energy storage converter is a set of physical experiments and mathematical models for identification in this invention. L Data comparison chart;

[0078] Figure 7 The following are the identification results of the IGBT in the embodiment of the present invention. Figures (a) and (b) show the threshold voltage and on-resistance of transistor T1, and Figures (c) and (d) show the threshold voltage and on-resistance of transistor T2.

[0079] Figure 8 This is a flowchart of the statistical optimization process for saturation voltage drop error according to an embodiment of the present invention;

[0080] Figure 9 This is a normal distribution fitting diagram of the identification conduction voltage drop according to an embodiment of the present invention;

[0081] Figure 10 This is a flowchart illustrating the model selection mechanism in an embodiment of the present invention. Detailed Implementation

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0083] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0084] Example

[0085] like Figure 1 As shown, during the power cycle, the on-state voltage drop V of the IGBT... ce The voltage drop will gradually increase with aging. Therefore, this invention provides a method for monitoring IGBT saturation voltage drop driven by closed-loop feedback data in NPC converters, including the following steps:

[0086] S1. Collect operating data from the NPC energy storage converter on the experimental platform, including closed-loop control data and data collected by sensors.

[0087] The current inner loop controller in this embodiment is as follows: Figure 2 As shown, the sinusoidal fundamental voltage u on the AC side of the inner-loop output voltage source converter (VSC) is... tdref u tqref The output u after the inverse Park transform abc By adding grid-side voltage feedforward u gd and u gq This can increase the controller's response speed by adding current feedback. gd and i gq It can improve the tracking response characteristics of the current controller, and adding a PI controller can eliminate steady-state error and adjust the response speed.

[0088] S2. The acquired operational data is converted from analog to digital (A / D) signals using a data acquisition board. dSPACE is a hardware platform for the development and testing of real-time control systems, widely used in power electronics, automotive electronics, and other fields. It supports rapid model deployment and closed-loop simulation, enabling real-time execution and verification of control algorithms on hardware.

[0089] S3. In order to improve the accuracy of parameter identification, it is necessary to filter out sampling noise and clean the running data using the Kalman algorithm.

[0090] S4. Derive the mathematical model of the energy storage converter and substitute the cleaned operating data into the mathematical model.

[0091] In this embodiment, step S4 includes:

[0092] S41. When the NPC converter is running in grid-connected mode, collect the three-phase voltage, three-phase current and DC side voltage in real time.

[0093] S42. The measured three-phase voltage needs to be processed by a phase-locked loop to obtain the phase angle θ. The three-phase current is combined with the phase angle θ signal and Park transform is performed to obtain the dq axis current.

[0094] The S43 and dq axis currents are adjusted by proportional-integral control to generate three-phase voltage reference values, and then the on / off state of the switching devices in the three-phase inverter is controlled by a sinusoidal pulse width modulation (SPWM) strategy.

[0095] Specifically, step S43 includes:

[0096] For ease of analysis, the DC bus capacitor is drawn as two capacitors connected in series, and the imaginary midpoint O′ is marked, referring to... Figure 3 U dc The voltage source voltage is the DC side voltage. The voltage inverter adopts a three-level neutral point clamped (NPC) topology. Each phase consists of 4 IGBTs and 2 clamping diodes, for a total of 12 IGBTs forming a three-phase bridge inverter. i For AC side filter inductance, R i For inductor L i stray resistance, L g R g These are the inductance and resistance of the equivalent impedance of the power grid, respectively. i (i = a / b / c) is the equivalent ideal voltage source of the power grid. Based on the circuit topology and Kirchhoff's voltage law, the mathematical model of the three-phase voltage of the NPC energy storage converter can be expressed as:

[0097]

[0098] In the formula, u ao′ u represents the voltage at the output point of phase A bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. ao u represents the voltage at the output point of phase A bridge arm of the converter relative to the neutral point O of the power grid. oo′ Represents the common-mode voltage, u bo′ u represents the voltage at the output point of the B-phase bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. bo u represents the voltage at the output point of the B-phase bridge arm of the converter relative to the neutral point O of the power grid. co′ u represents the voltage at the output point of the C-phase bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. co This represents the voltage at the output point of the C-phase bridge arm of the converter relative to the neutral point O of the power grid.

[0099] Based on the three-phase voltage balance of the grid-connected converter, we can obtain:

[0100] u ao +u bo +u co =0;

[0101] Converter common mode voltage u o′o It can be represented as:

[0102]

[0103] Therefore, phase voltage reconstruction is expressed as:

[0104]

[0105] The mathematical model of the three-phase output current of a PCS can be expressed as:

[0106]

[0107] In the formula, i a i b and i c These represent the currents in phases A, B, and C, respectively, where t is the time variable and L... a L b and L c R represents the filter inductance for phases A, B, and C, respectively. a R represents the sum of the stray resistance of the A-phase filter inductor and the line parasitic resistance. b R represents the sum of the stray resistance of the B-phase filter inductor and the line parasitic resistance. c e represents the sum of the stray resistance of the C-phase filter inductor and the line parasitic resistance. a e b and e c Let A, B, and C represent the grid voltages respectively, and L be the sum of the filter inductance and the grid-side inductance, where L = L i+L g L g Let R be the equivalent inductance of the power grid, R be the sum of the equivalent series resistance of the filter inductor and the power grid resistance, and R be the sum of the stray resistance of the filter inductor and the power grid resistance, and R = R i +R g R i For inductor L i The stray resistance on, R g The resistance is the equivalent impedance of the power grid.

[0108] The Runge-Kutta method is used to solve the mathematical model of the three-phase output current of the PCS. The Runge-Kutta method is a typical method for solving differential equations, and it can use recursion to find numerical solutions to the differential equations, offering advantages such as high accuracy and simple programming. Therefore, the inductor current in the next sampling interval can be expressed as:

[0109]

[0110] In the formula, h is the time step between the nth sampling point and the (n+1)th sampling point, and k is the average rate of change between the nth sampling point and the (n+1)th sampling point. a1 k b1 and k c1 These are the initial values ​​of the current variable i. a i b and i c Rate of change estimation, k a2 k b2 and k c2 They are using k a1 k b1 and k c1 The current variable i obtained after the first intermediate estimation a i b and i c rate of change, k a3 k b3 and k c3 They are respectively at k a2 k b2 and k c2 Based on the intermediate estimation of the current variable i a i b and i c rate of change, k a4 k b4 and k c4 These are the further intermediate estimates of the current variable i. a i b and i c The rate of change of k at these different stages. aThe values ​​will eventually be combined to obtain a more accurate current variable i from the nth sampling point to the (n+1)th sampling point. a The changes in k b1 ,k b2 ,k b3 ,k b4 Similarly, it is related to the variable current i. b The relevant calculations of the "average rate of change" at different stages are used to accurately calculate the variable current i. b The variation of k between sampling points c1 ,k c2 ,k c3 ,k c4 : is related to variable current variable i c The relevant calculations of the "average rate of change" at different stages are used to accurately calculate the variable current i. c Changes between sampling points.

[0111]

[0112] In the formula, f = [di a / dt,di b / dt,di c / dt],a n b n and c n Let i represent the current variable at step n. a i b and i c Approximate values ​​for .

[0113]

[0114] In the formula, i a,n+1 This represents the calculated predicted value of the current in phase A at the (n+1)th discrete time step, used to describe the current state of the system at subsequent times. b,n+1 This represents the calculated predicted value of the current in phase B at the (n+1)th discrete time step, used to describe the current state of the system at subsequent times. c,n+1 This represents the calculated predicted value of the current in phase C at the (n+1)th discrete time step, used to describe the current state of the system at subsequent times, V. oi Represents the threshold voltage of a certain IGBT, R oni Represents the on-resistance of a specific IGBT, V di Represents the threshold voltage of a certain anti-parallel diode, R doni Represents the on-resistance of a certain anti-parallel diode, i = 1, 2, 3, 4, 5, 6, i a,nThis represents the value of the phase A current at the nth discrete time step. It is the current state quantity at the current moment and serves as the basic input for calculating the current at the next time step (n+1). b,n This represents the value of the B-phase current at the nth discrete time step. It is the current state quantity at the current moment and serves as the basic input for calculating the current at the next (n+1)th time step. c,n This represents the value of the C-phase current at the nth discrete time step. It is the current state quantity at the current moment and serves as the basic input for calculating the current at the next (n+1)th time step.

[0115] S5. Construct a "model library + selection mechanism" to reduce dependence on operating conditions, such as... Figure 10 As shown. The model library and selection mechanism works as follows: based on the operating conditions of the energy storage converter PCS (such as load, grid voltage, etc.), multiple circuit mathematical sub-models are constructed, and the most suitable model is selected for voltage drop identification through a real-time operating condition identification mechanism. Previously, the model had a unique structure; the addition of the strategy of "automatically selecting the model according to the converter's operating conditions" increases its practical applicability.

[0116] S6. The trust-region reflective nonlinear least squares (TRRLS) parameter identification method is used for curve fitting, the objective function is calculated, and the optimal solution that minimizes the objective function is found to obtain the IGBT identification voltage drop parameters. (Refer to...) Figure 5 The “PSC mathematical analytical model” in the figure is a discrete dynamic mathematical model between output voltage and current based on the circuit structure and control strategy of the three-phase two-level energy storage converter.

[0117] In this embodiment, step S6 includes:

[0118] S61. The method of curve fitting between the physical model and the mathematical analytical model using the trust region reflection nonlinearity minimum sum of squared errors is a powerful tool for solving constraint-bounded nonlinearity minimization problems. Typically, TRRLS is used to solve the coefficients x for the following problem:

[0119]

[0120] In the formula, F(·) represents the model function, x represents the model coefficients to be solved, and y represents the actual observation data.

[0121] S62 and TRRLS are essentially solutions to optimization problems, calculating s to minimize q(s). The trust region subproblem can be expressed as:

[0122]

[0123] In the formula, s is the search step size, and H is the search step size.k D is a Hessian matrix. k It is a diagonal scale matrix, Δ k For the trust region scale, g k Let q be the gradient of f(x). k (s) represents a quadratic approximation model of the objective function in the trust region subproblem, s T This represents the search direction vector.

[0124] S63, if f(x) k+1 ) <f(x k If x k+1 =x k +α k s k α k Constrained by the boundary, α k The step size factor is represented by the reflection trust region method, which "reflects" the solution back into the trust region, thereby avoiding invalid solutions and continuing optimization; otherwise, the iteration point remains unchanged, and the trust region is further reduced.

[0125] S64, the identification voltage drop parameter of IGBT is contained in u ao′ In the example of phase A, the phase voltage can be expressed as:

[0126]

[0127] In the formula, U dc V represents the DC voltage source voltage. ce1 This represents the on-state voltage of the upper arm IGBT, V. F1 This represents the forward voltage of the anti-parallel diode in the upper bridge arm, V. ce2 This indicates the on-state voltage of the lower IGBT arm, V. F2 Indicates the forward voltage of the anti-parallel diode in the lower bridge arm, sign(i a ) represents the direction of the current. When the current flows from the energy storage converter PCS to the grid, sign(i a If the value is 1, then the value is 0; otherwise, the value is 0. Indicates the direction of current flow; when current flows from the grid to the energy storage converter PCS, Q is 1 if the input is positive and 0 otherwise. Q represents the IGBT's on / off state, as shown below:

[0128]

[0129] S65. Based on the operating and switching states of the energy storage converter PCS, establish an ideal model:

[0130]

[0131] S6. Based on the IGBT output characteristic curve obtained from the datasheet, extract the parameters under healthy conditions and compare them with the identified voltage drop parameters to perform a condition assessment.

[0132] In this embodiment, step S7 includes: according to the IGBT output characteristic curve in the device datasheet provided by the manufacturer, such as... Figure 4 As shown, the on-state voltage drop parameter of the IGBT under healthy conditions is obtained by performing two-dimensional linear interpolation on the output characteristic curve of the IGBT, and then compared with the identified voltage drop parameter to evaluate the aging state of the IGBT.

[0133] S8. Based on the existing mathematical model, introduce a parameter self-calibration or online correction mechanism, such as... Figure 8 As shown, based on the derivation of the mathematical model of the energy storage converter, the model parameters are further dynamically corrected by combining actual sampling data. A sliding window-based error minimization correction strategy is adopted to improve the online accuracy and adaptability of the model. This mathematical model has dynamic adaptive correction capability, and can correct model parameters based on real-time collected feedback data to improve identification accuracy.

[0134] S9. Enhanced "feature extraction + state classification" function, introducing a multi-dimensional state-space model based on identified voltage drop parameters and historical device data, intelligently classifying IGBT health states using statistical methods such as normal distribution. To improve the accuracy of IGBT on-state voltage identification, data filtering is performed based on normal distribution, such as... Figure 9 As shown.

[0135] In this embodiment, step S9 includes:

[0136] S91. Examine the histogram and kernel density curve of the measurement data to determine whether it approximates a unimodal symmetrical normal distribution.

[0137] S92. Calculate the mean and standard deviation of the data to describe the central tendency and dispersion of the data. These two parameters are key characteristics of the normal distribution.

[0138] S93. Apply the “3σ criterion”: Define a valid data range around the mean (e.g., [0.960, 1.290]V) and remove data outside this range. These outliers are usually caused by noise, interference or abnormal operating conditions.

[0139] S94. Recalculate statistical characteristics and verify distribution patterns through an iterative optimization process to ensure the accuracy and reliability of the filtered data.

[0140] S10, A dynamic threshold judgment module based on temperature and aging evolution trends, used to realize online evolution assessment and anomaly early warning of device status, specifically:

[0141] The dynamic threshold judgment module for temperature and aging evolution trend utilizes the temperature characteristics of the device and the aging evolution trend of the voltage drop parameter during long-term operation to construct a temperature-compensated multidimensional threshold model. By fitting historical data, the growth trend of saturation voltage drop at different temperatures is obtained, and dynamic adjustments are made in conjunction with real-time monitoring values. When the real-time data deviates significantly from the predicted aging curve (such as the saturation voltage drop exceeding the reasonable range after temperature correction or the degradation rate increasing abnormally), the module automatically triggers an anomaly warning, realizing real-time monitoring of the device's health status and early fault identification.

[0142] Depend on Figure 6 It can be seen that the output of the mathematical model matches the output of the physical entity well, proving that the established digital model is accurate, and the parameters identified by the method proposed in this invention are very close to the actual parameters in the physical response. Figure 7 As shown, the IGBT identification results match the voltage drop measurement results.

[0143] Based on the three-level NPC topology energy storage converter, the operating state and voltage drop characteristics of its IGBT devices differ significantly from those of the two-level converter, especially in terms of the more complex conduction path, voltage stress, and current flow control strategies. Therefore, an online IGBT voltage drop monitoring scheme suitable for NPC topologies is proposed. The system modeling and parameter identification process proposed in this invention has topological universality, and the established filter state-space model has good adaptability to different types of converters. Taking the NPC three-level converter as an example, its closed-loop control data can be used as modeling input, and the filter model can still be directly applied without adjusting the state equation form, thus ensuring the universality and structural independence of the method. In addition, a parameter evolution trend modeling and dynamic threshold judgment mechanism based on long-term monitoring data are further introduced to achieve early warning of IGBT aging status and tracking of degradation process.

[0144] Therefore, the present invention adopts the above-mentioned IGBT saturation voltage drop monitoring method driven by closed-loop feedback data of NPC converter to realize real-time monitoring of the working status and health status of IGBT, while ensuring high accuracy and reliability.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring IGBT saturation voltage drop driven by closed-loop feedback data in an NPC converter, characterized in that, Includes the following steps: S1. Collect the operating data of the NPC energy storage converter on the experimental platform, including closed-loop control data and data collected by sensors; S2. Use the data acquisition board to perform analog-to-digital conversion on the acquired running data, converting analog signals into digital signals; S3. Clean the running data using the Kalman algorithm; S4. Derive the mathematical model of the energy storage converter and substitute the cleaned operating data into the mathematical model. Step S4 includes: S41. When the NPC converter is running in grid-connected mode, collect the three-phase voltage, three-phase current and DC side voltage in real time. S42. The measured three-phase voltage needs to be processed by a phase-locked loop to obtain the phase angle θ. The three-phase current is combined with the phase angle θ signal to perform Park transformation to obtain the dq axis current. The S43 and dq axis currents are adjusted by proportional-integral control to generate three-phase voltage reference values, and then the on / off state of the switching devices in the three-phase inverter is controlled by a sinusoidal pulse width modulation strategy. S5. Build a model library and selection mechanism to reduce dependence on operating conditions; S6. The trust region reflection nonlinearity minimum error sum of squares parameter identification method is used to perform curve fitting, calculate the objective function, find the optimal solution that minimizes the objective function, and obtain the IGBT identification voltage drop parameter. Step S6 includes: S61. The coefficients are solved by curve fitting of the physical model and the mathematical analytical model using the trust region reflection nonlinear minimum error sum of squares parameter identification. x ,get: ； In the formula, Represents the model function, These represent the model coefficients to be solved. This represents actual observation data; S62 and TRRLS are essentially solutions to optimization problems, and calculations... s To minimize q ( s The trust domain subproblem is represented as: ； In the formula, s For the search step size, H k It is a Hessian matrix. D k It is a diagonal scale matrix. Δ k For trust region metric, g k for f ( x The gradient of ) This represents a quadratic approximation model of the objective function in the trust region subproblem. Represents the search direction vector; S63, if f ( x k+1 )< f ( x k ),but x k+1 = x k + α k s k , α k Limited by boundaries, α k The step size factor is represented by the reflective trust region method, which "reflects" the solution back into the trust region; otherwise, the iteration point remains unchanged, and the trust region is further reduced. The identification voltage drop parameters of S64 and IGBT are contained in u ao′ In the middle, referring to phase A, the phase voltage is expressed as: ； In the formula, Indicates the DC voltage source voltage. This indicates the on-state voltage of the upper arm IGBT. This indicates the forward voltage of the anti-parallel diode in the upper bridge arm. This indicates the on-state voltage of the lower IGBT arm. This indicates the forward voltage of the anti-parallel diode in the lower bridge arm. This represents the common-mode voltage of the converter. sign ( i a () represents the direction of current. When current flows from the energy storage converter PCS to the power grid, sign ( i a If the value is 1, then the value is 0; otherwise, the value is 0. Indicates the direction of current flow; when current flows from the grid to the energy storage converter PCS, It is 1 if it is 1, otherwise it is 0. Q The on / off states of the IGBT are represented as follows: ； S65. Based on the operating and switching states of the energy storage converter, an ideal model is established as follows: ； S7. Based on the IGBT output characteristic curve obtained from the datasheet, extract the parameters under the healthy state and compare them with the identified parameters to conduct a state assessment. S8. Based on the existing mathematical model, introduce a parameter self-calibration or online correction mechanism; S9. Introduce a multi-dimensional state-space model based on the identified voltage drop parameters and historical device data, and classify the IGBT health status through a normal distribution; S10, a dynamic threshold judgment module based on temperature and aging evolution trends, is used to realize online evolution evaluation of device status and early warning of anomalies.

2. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data in an NPC converter according to claim 1, characterized in that, Step S43 includes: Based on the circuit topology and Kirchhoff's voltage law, the mathematical model of the three-phase voltage of the energy storage NPC converter is established as follows: ； In the formula, This represents the voltage at the output point of phase A bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. This represents the voltage at the output point of phase A bridge arm of the converter relative to the neutral point O of the power grid. Indicates common-mode voltage. This represents the voltage at the output point of the B-phase bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. This represents the voltage at the output point of the B-phase bridge arm of the converter relative to the neutral point O of the power grid. This represents the voltage at the output point of the C-phase bridge arm of the converter relative to the imaginary midpoint of the DC capacitor. This represents the voltage at the output point of the C-phase bridge arm of the converter relative to the neutral point O of the power grid. Based on the three-phase voltage balance of the grid-connected converter: ； Converter common mode voltage u o′o Represented as: ； Phase voltage reconstruction is expressed as: 。 3. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data of an NPC converter according to claim 2, characterized in that: The mathematical model for the three-phase output current of the energy storage converter is expressed as follows: ； In the formula, , and These represent the currents in phases A, B, and C, respectively. It is a time variable. , and These represent the filter inductors for phases A, B, and C, respectively. This represents the sum of the stray resistance of the filter inductor in phase A and the parasitic resistance of the line. This represents the sum of the stray resistance of the B-phase filter inductor and the line parasitic resistance. This represents the sum of the stray resistance of the C-phase filter inductor and the line parasitic resistance. , and These represent the grid voltages for phases A, B, and C, respectively. L This is the sum of the filter inductance and the grid-side inductance, and L = L i + L g , L i Indicates the AC side filter inductor. L g Let R be the equivalent inductance of the power grid, and R be the sum of the equivalent series resistance of the filter inductor and the resistance of the power grid. R = R i + R g , R i For inductance L i Stray resistance on R g The resistance is the equivalent impedance of the power grid.

4. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data of an NPC converter according to claim 3, characterized in that: The mathematical model of the three-phase output current of the energy storage converter PCS is solved using the Runge-Kutta method. The inductor current in the next sampling interval is expressed as: ； In the formula, h For the first n The sampling point and the first n +1 time step between sampling points k For the first n The sampling point and the first n The average rate of change between +1 sampling points k a1 , k b1 and k c1 These are the initial values ​​of the current variable. i a , i b and i c The rate of change estimate, k a2 , k b2 and k c2 They are using k a1 , k b1 and k c1 The current variable obtained after the first intermediate estimation i a , i b and i c rate of change, k a3 , k b3 and k c3 They are in k a2 , k b2 and k c2 Based on the intermediate estimation of current variables i a , i b and i c rate of change k a4 , k b4 and k c4 These are the current variables after further intermediate estimates. i a , i b and i c The rate of change; ； In the formula, f =[ di a / dt , di b / dt , di c / dt ], , and They represent the first n Step-time current variable i a , i b and i c Approximate values; ； In the formula, Indicates the first n +1 discrete time steps, the calculated and predicted value of the current in phase A, used to describe the current state of the system at subsequent time steps. Indicates the first n +1 discrete time steps, the calculated and predicted value of the current in phase B, used to describe the current state of the system at subsequent time steps. Indicates the first n +1 discrete time steps, the calculated and predicted value of the current in phase C, used to describe the current state of the system at subsequent time steps. Represents the threshold voltage of a specific IGBT. Represents the on-resistance of a specific IGBT. The threshold voltage of a certain anti-parallel diode, Represents the on-resistance of a certain anti-parallel diode. i =1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, Indicates the first n At each discrete time step, the value of the phase A current is the current state quantity at the current moment, used to calculate the next moment. n +1 step current base input, Indicates the first n At each discrete time step, the value of the B-phase current is the current state quantity at the current moment, used to calculate the next moment. n +1 step current base input, Indicates the first n The value of the C-phase current at each discrete time step is the current state quantity at the current moment, used to calculate the next moment. n +1 step current base input.

5. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data in an NPC converter according to claim 1, characterized in that, Step S5 specifically involves: Model library + selection mechanism. Specifically, based on the operating conditions of the energy storage converter PCS, multiple circuit mathematical sub-models are constructed, and the most suitable model is selected for voltage drop identification through a real-time operating condition identification mechanism.

6. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data in an NPC converter according to claim 1, characterized in that, Step S8 specifically involves: based on the derivation of the mathematical model of the energy storage converter, dynamically correcting the model parameters by combining actual sampling data, and adopting a sliding window-based error minimization correction strategy to improve the online accuracy and adaptability of the model.

7. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data in an NPC converter according to claim 1, characterized in that, Step S9 includes: S91. Examine the histogram and kernel density curve of the measurement data to determine whether they are approximately a unimodal symmetrical normal distribution; S92. Calculate the mean and standard deviation of the data to describe the central tendency and dispersion of the data; S93. Apply the "3σ criterion": Define a valid data range around the mean and remove data outside that range; S94. Recalculate statistical characteristics and verify distribution patterns through iterative optimization.

8. The IGBT saturation voltage drop monitoring method driven by closed-loop feedback data in an NPC converter according to claim 1, characterized in that, Step S10 includes: The dynamic threshold judgment module for temperature and aging evolution trend utilizes the temperature characteristics of the device and the aging evolution trend of the voltage drop parameter during long-term operation to construct a multi-dimensional threshold model for temperature compensation. By fitting historical data, the growth trend of saturated pressure drop at different temperatures is obtained, and dynamic adjustments are made in conjunction with real-time monitoring values. When real-time data deviates significantly from the predicted aging curve, the dynamic threshold judgment module for temperature and aging evolution trend automatically triggers an abnormal warning.

Images