Converter power device junction temperature real-time estimation method

Abstract

The invention belongs to the technical field of junction temperature measurement of converter power electronic devices, and particularly relates to a converter power device junction temperature real-time estimation method. In order to overcome one or more defects of insufficient measurement precision, repeated parameter adjustment and incapability of adapting to system parameter changes of an existing junction temperature prediction method, the invention adopts the following technical scheme: the converter power device junction temperature real-time estimation method comprises the following steps: obtaining device loss at a current sampling moment according to real-time operation parameters, and obtaining shell temperature information; as the input of an adaptive Kalman filter, the adaptive Kalman filter obtains an estimated junction temperature according to a thermal impedance model prediction method and device loss, and the adaptive Kalman filter also performs adaptive estimation on a measurement noise covariance matrix and a process noise covariance matrix; and the estimated junction temperature of the model is fed back and corrected in real time in combination with the measured junction temperature. The method has the advantages that the Kalman filter is improved in a self-adaptive mode, and the adaptability to system operation parameter changes is enhanced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of junction temperature measurement of power electronic devices of a converter, and particularly relates to a real-time estimation method for junction temperature of power devices of a converter. BACKGROUND

[0002] Power electronic converters are the core equipment of renewable energy and electrified transportation. As the "core" of power electronic converters, the energy efficiency and reliable operation of power semiconductor devices are of great significance to achieving efficient energy conversion. Insulated gate bipolar transistors (IGBTs) and freewheeling diodes (Diodes) are key power electronic devices that determine the reliability of converters, and more than 50% of converter failures are caused by power electronic devices (IGBTs + Diodes).

[0003] Excessive temperature is one of the most important reasons for the failure of power electronic devices (IGBTs + Diodes). The device junction point with the highest temperature is inside the device package, making it difficult to measure the junction temperature (Junction Temperature, the actual operating temperature of a semiconductor in an electronic device).

[0004] The temperature sensitive electrical parameter method (Temperature Sensitive Electrical Parameter, TSEP) is a method that uses the relationship between intrinsic physical parameters and temperature inside the chip to measure the temperature-sensitive external electrical parameters corresponding to the chip junction temperature. The method has the advantages of no need to change the package structure, fast response, easy integration, etc. However, due to the influence of TSEP parameter sensitivity and measurement noise, it is difficult to obtain accurate junction temperature estimation. The thermal impedance model prediction method mainly uses thermal conduction modeling to establish a thermal model from the chip to the external reference temperature point by equivalent heat transfer network and electrical network, and uses the loss of the device as the input of the model to solve the junction temperature of the chip. The method has the advantage of not requiring additional hardware, but the accuracy of the junction temperature prediction depends on the accuracy of the modeling, and changes in related boundary conditions such as water cooling system flow rate will greatly affect the prediction accuracy.

[0005] The latest related research combines the TSEP method with the thermal impedance model and proposes a junction temperature prediction method based on Kalman filter, which can greatly improve the junction temperature monitoring noise in the TSEP measurement method and overcome the problems of single model method being easily affected by boundary conditions or inaccurate modeling. However, the existing junction temperature prediction method based on Kalman filter has the problems of high degree of complexity in estimating performance depending on repeated parameter adjustment of initial values, and the model parameters cannot be adapted under electromagnetic noise changes, resulting in reduced junction temperature prediction accuracy.

[0006] Specifically, the existing Kalman filter needs to statistically obtain the covariance of the measurement noise and the process noise in advance when applied to match the model parameters. However, the noise covariance is difficult to be statistically obtained, and can only be repeatedly adjusted according to experience, and there are problems of multiple adjustments and inaccuracy. Moreover, once the noise changes during operation, the noise covariance cannot be or is difficult to be modified and adapted, which also affects the effect of the filter. SUMMARY

[0007] The present application aims at one or more of the deficiencies of the existing power electronic device junction temperature prediction method of the converter, such as insufficient measurement accuracy, repeated parameter adjustment, and inability to adapt to system parameter changes, and provides a converter junction temperature real-time estimation method based on an adaptive Kalman filter. Through real-time and accurate estimation of the junction temperature, the performance of the power device in actual application is effectively improved, and the stability and reliability of the system are ensured.

[0008] To achieve the above-mentioned purpose, the present application adopts the following technical solution: a converter power device junction temperature real-time estimation method, the converter power device junction temperature real-time estimation method comprising: During real-time operation of the converter, the device loss at the current sampling time is obtained according to the real-time operation parameters, and the shell temperature information is obtained; The device loss at the current sampling time and the shell temperature information are taken as inputs of the adaptive Kalman filter, the adaptive Kalman filter adopts the state space equation of the discretized electro-thermal network model, the measurement part is the measured junction temperature obtained by real-time table lookup based on the TSEP parameter at each sampling time, the electro-thermal network model obtains the estimated junction temperature according to the thermal impedance model prediction method and the device loss, and the adaptive Kalman filter further estimates the covariance matrix of the measurement noise and the covariance matrix of the process noise in real time, and combines the measured junction temperature to feed back and correct the estimated junction temperature of the model in real time, and outputs the corrected real-time junction temperature information.

[0009] The converter power device junction temperature real-time estimation method of the present application combines the heat-sensitive electrical parameter method and the thermal impedance model prediction method, can greatly reduce the measurement noise of the heat-sensitive electrical parameter in the application process, and reduces the dependence of the thermal model prediction method on the model parameters and the boundary conditions; simultaneously estimates the measurement noise covariance and the process noise covariance in the filter, achieves the parameter adaptation ability under the condition of no experience adjustment and adaptation to noise changes, enhances the adaptation ability to the system operation parameter changes, and is more accurate.

[0010] As an improvement, the real-time operation parameters include one or more of bus voltage, load current, and switching frequency; and / or, The device loss includes IGBT power loss and power diode loss in the converter. The IGBT power loss includes switching loss and conduction loss, the switching loss is obtained by table lookup, and the conduction loss is obtained by the product of the conduction voltage drop and the current.

[0011] As an improvement, real-time operating parameters are input as a table lookup of a pre-prepared loss model to obtain the device loss at the current sampling time; and / or, The TSEP parameters include the device conduction voltage drop.

[0012] As an improvement, the state space equation of the discretized electro-thermal network model is represented as follows: (0-5) Wherein, represents the system matrix, represents the input matrix, represents the output matrix, represents the direct transfer matrix, , represents the state quantity at the current sampling time and the last sampling time respectively, represents the system input at the current sampling time, represents the estimated junction temperature of the model at the current sampling time, k represents the discretized k sampling time, w k is the process noise in the electro-thermal network model prediction, v k is the measurement noise of the junction temperature estimation based on the conduction voltage drop V ce,on .

[0013] As an improvement, the implementation process of the adaptive Kalman filter includes: Variable initialization: initialize the junction temperature prediction parameters of the Kalman filter to 0, the junction temperature prediction parameters include the posterior error covariance matrix and the posterior state quantity; Prediction: estimate the prior state quantity at the current sampling time according to the posterior state quantity at the last sampling time and the system input at the current sampling time, calculate the predicted junction temperature according to the prior state quantity, and update the prior error covariance matrix at the current sampling time; Correction: the innovation value at the current sampling time is obtained by subtracting the measured junction temperature from the predicted junction temperature based on the prior estimation d k , the gain K k of the Kalman filter at the current sampling time is calculated at the same time K k , and the posterior state quantity at the current sampling time is obtained based on the gain Output: Calculate the real-time junction temperature information based on the posterior state variables at the current sampling time.

[0014] As an improvement, variable initialization can be expressed by the formula: (0-8) In equation (0-8), the symbol "^" directly above the character represents the estimated value, and the superscript "+" above the character indicates that this estimate is posterior. Let be the initial covariance matrix of the posterior state estimation error. Represents the posterior initial state quantity, and Δ represents the initial temperature difference between layers in the IGBT electrothermal model. T n , This refers to the chip's operating junction temperature and the temperature of the external water cooling system. T coolant equal; Process noise in electrothermal network model prediction w k and based on conduction voltage drop V ce,on Measurement noise for junction temperature estimation v k The covariance matrix is ​​generally set to an initial value based on the test conditions. Q 0、 R 0, (0-7) in, and They represent w k and v k The expected value of the mathematical value, and They represent w k and v k The covariances are denoted as . Q k and R k .

[0015] As an improvement, the formula for calculating the prior error covariance matrix at the current sampling time is as follows: (0-9) in, Indicates the first k State quantity at the next sampling time x Prior estimates, Indicates the first k The system input at the next sampling time Indicates the firstk The sampling time is a prior estimate of the junction temperature at the current time predicted by the model. Indicates the first k The prior error covariance matrix at each sampling time, with the superscript T indicating transpose. Indicates the first k The process noise covariance matrix at each sampling time.

[0016] As an improvement, the correction was made based on the on-state voltage drop. V ce,on The measured junction temperature T vj The innovation value is obtained by subtracting the predicted junction temperature value based on prior estimation. d k Simultaneously calculate the gain of the Kalman filter at the current time. K k And based on this gain K k Obtain state variables x The posterior estimate is calculated using the following formula: (0-10) in, Indicates the first k The sampling time is based on the on-voltage drop. V ce,on Measured junction temperature, Indicates the first k Measurement noise covariance matrix at the next sampling time H Represented as the system's output measurement matrix, Indicates the first k Kalman filter gain at the next sampling time Indicates the first k State quantity at the next sampling time x The posterior estimate, Indicates the first k Posterior estimate of the covariance matrix at the next sampling time f It is a flag variable value, used when in a specific current region. V ce,on Measurement and estimation of junction temperature T vj At that time, variable f Set to 1, otherwise set to 0; Through real-time iteration of the above steps, the real-time junction temperature information predicted by the adaptive Kalman filter can be obtained, as follows: .

[0017] As an improvement, the difference between the current measured value and the current posterior predicted value is... The calculation formula is expressed as: (0-16) wherein, denotes the posterior estimate of the junction temperature at the k-th sampling time; k denotes the posterior estimate of the junction temperature at the k-th sampling time; process noise w k and its posterior estimate is denoted as: (0-20).

[0018] As an improvement, the measurement noise covariance matrix R k and the process noise covariance matrix Q k is estimated as: (0-19) (0-22) wherein, denotes a forgetting factor, 0 A larger forgetting factor gives more weight to the previous parameter estimates, so that R k , Q k the fluctuations of the junction temperature are relatively small over time.

[0019] The beneficial effects of the real-time junction temperature estimation method for power devices of a converter of the present application are: the Kalman filter is used to combine the heat-sensitive electrical parameter method and the thermal impedance model prediction method, which can greatly reduce the measurement noise of the heat-sensitive electrical parameter in the application process, and at the same time reduces the dependence of the thermal model prediction method on the model parameters and boundary conditions; the measurement noise covariance and the process noise covariance in the filter are estimated synchronously, the Kalman filter is improved adaptively, and the adaptability to the changes of system operating parameters is enhanced, and the accuracy is better. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is the principle diagram of the real-time junction temperature prediction method for the converter of the embodiment of the present application.

[0021] Figure 2 is the flow chart of the real-time junction temperature prediction method for the converter of the embodiment of the present application.

[0022] Figure 3 is the self-heating and mutual-heating network model schematic diagram of the power device to be measured.

[0023] Figure 4 is the comparison diagram of the junction temperature prediction test results of the real-time junction temperature prediction method for the converter of the embodiment of the present application and other methods (without interference noise).

[0024] Figure 5 is a comparison chart of junction temperature test results of the converter real-time junction temperature prediction method of the embodiment of the present application and other methods (interference noise is introduced).

[0025] Figure 6 is a result chart of the converter real-time junction temperature prediction method of the embodiment of the present application when the system boundary condition changes. DETAILED DESCRIPTION

[0026] The technical solutions of the embodiments of the present application are explained and described below, but the following embodiments are only preferred embodiments of the present application, not all. Based on the embodiments in the embodiments, other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0027] Referring to Figure 1 , the principle or system architecture of the junction temperature prediction method of the first embodiment of the present application is that, in the process of real-time running of the converter, the running parameters such as bus voltage V dc , load current I ph , switching frequency f s , etc. are taken as the table lookup inputs of the loss model to obtain the current time device loss P loss , the device loss P loss and the shell temperature T c information are taken as the inputs of the model in the Kalman filter, wherein the model in the Kalman filter adopts the state space equation of the thermal network model (hereinafter referred to as IGBT electro-thermal model), the measurement part is the junction temperature obtained by real-time table lookup based on TSEP parameters at each sampling time T j_mea , and the estimated junction temperature of the model T j_mea T j ^ is fed back and corrected in real time in combination with the measured junction temperature information . The TSEP parameters include the on-state voltage drop of the device to be measured K . In the figure, Kd represents the gain of the Kalman filter multiplied by the innovation d .

[0028] The IGBT electro-thermal model is shown in Figure 2 , wherein the green area is the network part of IGBT self-heating, and the pink area is the thermal network part coupled by the same bridge arm power diode, and the corresponding model inputs are the power loss of IGBT P ​IGBT and power diode losses P Diode , the external reference temperature is the temperature of the water cooling liquid T coolant .

[0029] In order to obtain real-time predicted junction temperature information in the controller of the converter, a corresponding state space model is established. Figure 2 The temperature difference ΔT of each RC network (RC network is a circuit containing resistors, capacitors operating with voltage source, current source drive) of the IGBT electro-thermal model in the middle T n is the state quantity x ( t ), the input of the junction temperature estimation system (hereinafter referred to as the system) u ( t ) is [ P IGBT P Diode T coolant ] T The output of the system is the predicted junction temperature of the device under test T j ^ The obtained state space equation is as follows: (0-1) Wherein, represents the system matrix in the continuous domain, represents the input matrix in the continuous domain, , represents a constant vector, represents the nth order thermal resistance parameter of the RC network, represents the nth order thermal capacity parameter of the RC network, and the subscript n represents the number of layers of the RC network.

[0030] According to the sampling frequency of the system, the above state space equation is discretized, and the discretization of the state of the continuous system is used to improve the solving precision. The state space equation of the system after discretization can be expressed as: (0-2) Wherein, k represents the kth sampling time of the discrete system, k The corresponding system matrix after system discretization can be calculated according to the following formula: (0-3) Wherein, T s represents the sampling (switching) period of the system, System matrix representing the continuous domain, Input matrix representing the continuous domain.

[0031] The loss of the input part of the system mainly contains the loss of the device under test P IGBT With the loss of the anti-parallel power diode P Diode , P IGBT Can be divided into two parts, the switching loss P swi And the conduction loss P cond Two parts, represented as: (0-4) Where, Represents the on-state voltage drop of the device under test, Represents the current of the device, Represents the duty cycle of the device at the current time, Represents the turn-on energy in a switching process, Represents the turn-off energy in a switching process.

[0032] To improve accuracy, the formula (0-2) is improved, and the improved formula is: (0-5) Where, H Represents the output measurement matrix of the system, which satisfies: (0-6) Where, w k Process noise in the prediction of the electro-thermal network model, v k The measurement noise of the junction temperature estimation based on the on-state voltage drop V ce，on First, it is considered that the two noise sequences are independent of each other, and are approximately equivalent to white noise, and are independent of the state variables x k Therefore, it has the following statistical characteristics: (0-7) Where, And Respectively represent w k And v k The mathematical expectation of And Respectively represent w k And vk The covariances are denoted as . Q k and R k .

[0033] The implementation process of the junction temperature prediction method based on Kalman filter can be summarized as follows:

[0034] Step 1 - Variable Initialization: The initialization steps for the relevant parameters of the Kalman filter-based junction temperature prediction method are as follows: (0-8) In equation (0-8), the superscript "+" on the character indicates that this estimate is posterior. Let be the initial covariance matrix of the posterior state estimation error. This represents the posterior initial state quantity.

[0035] Before system parameter initialization, the IGBT power module under test is not working, so the chip's junction temperature is... Temperature of external water coolant system T coolant Equal to (and equal to the shell temperature), and the initial temperature difference Δ between each layer in the IGBT electrothermal model is... T n state variables x 0 is 0, process noise w k and measuring noise v k The covariance matrix is ​​generally set to an initial value based on the test conditions. Q 0、 R 0.

[0036] Step 2 - Prediction: This step mainly involves state variables. x Prior estimates and current junction temperature obtained based on model predictions Prior error covariance matrix P k - The update and calculation are performed, where “-” indicates that the estimate is prior.

[0037] (0-9) in, Indicates the first k State quantity at the next sampling time x Prior estimates, Indicates the first k The system input at the next sampling time Indicates the first kThe sampling time is a prior estimate of the junction temperature at the current time predicted by the model. Indicates the first k The prior error covariance matrix at each sampling time, with the superscript T indicating transpose; Indicates the first k The process noise covariance matrix at each sampling time; the superscript "-" on the character indicates that this estimate is prior.

[0038] Step 3 - Calibration: Based on on-voltage drop V ce，on Measured junction temperature T vj The innovation value is obtained by subtracting the predicted junction temperature value based on prior estimation. d k It can also calculate the gain of the Kalman filter at the current moment. K k And based on this gain K k Obtain state variables x The posterior estimate is shown in the following equation: (0-10) in, Indicates the first k The sampling time is based on the on-voltage drop. V ce，on Measured junction temperature, Indicates the first k Measurement noise covariance matrix at the next sampling time H Represented as the system's output measurement matrix, Indicates the first k Kalman filter gain at the next sampling time Indicates the first k State quantity at the next sampling time x The posterior estimate, Indicates the first k Posterior estimate of the covariance matrix at each sampling time. f It is a flag variable value, used when in a specific current region. V ce，on Measurement and estimation of junction temperature T vj At that time, variable f Set it to 1 if the condition is not met, otherwise set it to 0. Through real-time iteration of the above steps, real-time junction temperature information based on Kalman filter prediction can be obtained. .

[0039] In step 1 above, the value of the noise covariance is measured. R kIt can be calculated from the measurement results, but the corresponding statistical steps are quite cumbersome. Meanwhile, the covariance matrix of the process noise... Q k This data typically requires continuous debugging and is easily affected by changes in system conditions. Therefore, an initial value for the measurement noise covariance is needed before starting the junction temperature prediction algorithm in the controller. R Initial values ​​of the covariance matrix of 0 and process noise Q Repeated parameter tuning tests were conducted to achieve better real-time filtering results. However, inevitably, changes in the observed noise level or sudden interference during system operation can easily introduce large errors or interference into the estimation results.

[0040] To address this issue, based on existing Kalman filter-based junction temperature prediction methods, this invention proposes a real-time junction temperature prediction method for power devices based on an adaptive Kalman filter. This invention's real-time junction temperature prediction method for power devices based on an adaptive Kalman filter, while simultaneously performing real-time estimation of the system state, also calculates the measurement noise covariance matrix. R k Covariance matrix of process noise Q k Adaptive estimation is performed to match the system parameters, thereby improving the system's robustness and adaptability.

[0041] In the aforementioned existing junction temperature prediction algorithms based on Kalman filters, the innovation... d k Based on on-voltage drop V ce，on Measured junction temperature value T vj The difference from the current predicted value, on the other hand, can also be used with the measured junction temperature value. T vj With the k The optimal estimate at the next time step Perform the subtraction to obtain the residual information at the current time. Specifically, it is represented as follows: (0-11) in, Indicates the first k Posterior estimate of the junction temperature at the next sampling time.

[0042] Based on the above definition, in the real-time prediction method for power device junction temperature based on an adaptive Kalman filter in this embodiment of the invention, the noise covariance matrix is ​​measured. R k Covariance matrix of process noise Q kThe adaptive estimation process of is shown as follows: 1) Estimate the measurement noise covariance matrix based on the residual R k First, the innovation d k can be further expressed as: (0-12) where, represents the state quantity (true value) at the k time of the th sampling, k represents the prior estimation (measurement value) of the state quantity x at the time of the v th sampling, k is the measurement noise at the current time; let represent the prior error between the state measurement value and the true value, ; according to the definition of the error covariance matrix described above, it can be further expressed as (0-13) Accordingly, the innovation d k can also be obtained. The covariance matrix of is: (0-14) Since and the measurement noise v k at the current time are uncorrelated, the above formula can be further expressed as: (0-15).

[0043] Therefore, the measurement noise covariance matrix R k can be calculated by the currently obtained innovation value and related parameters, but this operation involves subtraction and cannot guarantee that R k is positive definite. Once R k becomes negative, problems such as filter divergence will occur. In order to guarantee that R k is positive definite, the residual information is used to estimate the measurement noise covariance.

[0044] The residual information can be further expressed as: (0-16) When the flag valuef =0 means no input is introduced based on the on-voltage drop. V ce，on Measured junction temperature T vj When correcting the model, residual information With new information d k They are equivalent, which is consistent with theoretical analysis; when the value of the flag... f When =1, when introducing measurement information to correct the model observations, k Time residual Numerically smaller than the new information d k This indicates that the measurement component is effectively incorporated into the system estimate, achieving the state fusion effect of the Kalman filter.

[0045] Furthermore, according to the error propagation law, residual information The covariance can be expressed as: (0-17).

[0046] Based on this, residual information can be used... The measurement noise covariance matrix was calculated. R k for: (0-18).

[0047] Since the above formula includes residual information The expected computation would consume significant storage space and computational resources if performed within the controller. Therefore, a sliding window-like approach is used to further simplify the above equation, introducing a forgetting factor. (0 < ≤1) is used to replace the expectation operation of equation (0-18), which will affect the residual information. The relevant expectation calculation is transformed into the previous time step. R k-1 and residual information Weighting of related product operations, while taking into account flag values. f The role of this is to measure the noise covariance matrix in the aforementioned junction temperature prediction system. R k It can be estimated as follows: (0-19).

[0048] It is worth noting that a larger forgetting factor assigns greater weight to previous parameter estimates, thus R k The fluctuations over time are relatively small, which can reduce the forgetting factor. Set in 0.7~0.9 around, can reach effectively to R k Estimate while ensuring convergence speed also has good effect.

[0049] 2) Based on innovation estimation process noise covariance matrix Q k In order to adaptively estimate process noise covariance matrix Q k , process noise w k And its posteriori estimation can be expressed as: (0-20).

[0050] Therefore, Q k-1 Can be estimated as: (0-21).

[0051] Similarly, introduce a forgetting factor (0< ≤1) instead of the expectation operation in formula (0-21), the real-time estimation for Q k Can be obtained: (0-22) Therefore, by introducing formula (0-19) (0-19) and formula (0-22) (0-22) to the existing Kalman filter, the IGBT junction temperature real-time prediction method based on adaptive Kalman filter of the embodiment of the application can be obtained.

[0052] Referring to Figure 2 , the junction temperature prediction method based on adaptive Kalman filter of the embodiment of the application comprises: Filter input; State estimation; Prior error covariance matrix calculation; Prior junction temperature estimation; Innovation calculation; Posterior state estimation; Junction temperature estimation at k time; Residual calculation; Measurement error covariance calculation; Kalman filter gain calculation; Model error covariance calculation; Posterior error covariance matrix calculation; Output and iteration.

[0053] The effect of the junction temperature prediction method based on the adaptive Kalman filter of the embodiment of the application is verified below.

[0054] As shown in Figure 4 , the junction temperature based on the real-time measurement of TSEP is T vj , the junction temperature predicted by the single thermal model is T j-Model , and the junction temperature estimated based on the adaptive Kalman filter of the embodiment of the application is T j_AKF , T j_OTG1 , and the temperature recorded by the optical fiber temperature meter is T vj Due to the influence of measurement noise, there is a certain random distribution, and due to the problems of boundary cooling liquid temperature and model parameters, the single thermal model significantly underestimates the junction temperature of the device under test. T j_AKF The noise problem in the measurement is optimized T vj , and to some extent, the problem of model predicted junction temperature caused by boundary parameters and modeling errors is overcome.

[0055] In the case where the cooling liquid temperature, fundamental frequency, and pure absorption reactive working condition of the bridge arm under test remain unchanged, the load current I L amplitude command is switched from 100A to about 175A near 22.8s, and the effect of the real-time predicted junction temperature is as shown in Figure 5 . From Figure 5 , it can be seen that the system is disturbed by an electromagnetic interference near 23.8s, causing the junction temperature calculated based on TSEP to T vj vibrate and deviate, and the deviation error of the junction temperature measurement has reached about 20℃; T j _ KF The junction temperature predicted based on the general Kalman filter deviates from the actual junction temperature, and due to its inability to estimate and adjust the measurement noise covariance R k , it incorrectly predicts the junction temperature information following the change of the system, which will cause false alarms of the system warning information.

[0056] To verify the performance of the junction temperature prediction method of the adaptive Kalman filter of the embodiment of the application when the boundary conditions of the system change, the following test is performed.

[0057] The initial working condition of the system is as follows: the external cooling liquid temperature T coolant = 40℃, the fundamental frequency f 0= 5Hz, and the load currentI L The amplitude is 175A, and the pure active (cos =1) of the bridge arm to be measured is absorbed. The water cooling is suddenly turned off near 40s, and the junction temperature of the device continuously rises due to the sharp decline in heat dissipation capacity. As shown in the figure, Figure 6 The junction temperature real-time estimation method of the embodiment of the present application can resist the influence of such boundary parameter conditions on the model estimated junction temperature, and also shows good dynamic performance in such working condition changes. At the same time, the fourth element Q k of the process noise covariance matrix Q 4 estimated by the system synchronously shows that the algorithm estimates the effect of its parameters in this process. For more intuitive data display, the corresponding statistics of Q 4 are performed with a 1s mean filtering window. As shown in the second column, Figure 6 the uncertainty of the model parameters can be measured, and after the water cooling is turned off near 40s, the innovation calculated by the adaptive Kalman filter k continuously increases, d showing a gradually rising trend, and near 54s, it is close to twice the normal working condition, indicating that the system model parameters are not matched, so it can be used as a criterion for the mismatch of the device thermal model parameters (such as solder layer degradation or heat dissipation system failure).

[0058] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Those skilled in the art should understand that the present application includes but is not limited to the content described in the above specific embodiment. Any modification that does not deviate from the functional and structural principles of the present application will be included in the scope of the claims.​

Claims

1. A method for real-time estimation of junction temperature of power devices of a power converter, characterized in that: The real-time converter power device junction temperature estimation method comprises: During real-time operation of the converter, device loss at a current sampling time is obtained according to real-time operation parameters, and shell temperature information is obtained; The device loss at the current sampling time and the shell temperature information are taken as inputs of an adaptive Kalman filter, the adaptive Kalman filter adopts a state space equation of a discretized electro-thermal network model, the measurement part of the adaptive Kalman filter is a measured junction temperature obtained by real-time look-up table based on TSEP parameters at each sampling time, the electro-thermal network model obtains an estimated junction temperature according to a thermal impedance model prediction method and the device loss, the adaptive Kalman filter further performs real-time feedback and correction on the estimated junction temperature of the model by adaptive estimation of a measurement noise covariance matrix and a process noise covariance matrix, and outputs corrected real-time junction temperature information.

2. The method of claim 1, wherein: The real-time operation parameters comprise one or more of bus voltage, load current and switching frequency; and / or The device loss comprises IGBT power loss and power diode loss in the converter; and / or The real-time operation parameters are taken as look-up table inputs of a pre-prepared loss model to obtain the device loss at the current sampling time.

3. The method of claim 1, wherein: The TSEP parameters comprise device on-state voltage drop.

4. The method of claim 1, wherein: The state space equation of the discretized electro-thermal network model is represented as follows: (0-5) wherein, represents a system matrix, represents an input matrix, represents an output matrix, represents a direct transmission matrix, , respectively represent the state quantity at the current sampling time and the previous sampling time, represents the system input at the current sampling time, represents the estimated junction temperature at the current sampling time of the model, k represents the discretized k sampling time, w k is the process noise in the electrically heated network model prediction, v k is the measurement noise based on the on-state voltage drop V ce,on of the junction temperature estimation.

5. The method of claim 4, wherein: The implementation process of the adaptive Kalman filter comprises: Variable initialization: the junction temperature prediction parameters of the Kalman filter are initialized to 0, the junction temperature prediction parameters comprise posterior error covariance matrix and posterior state quantity; Prediction: the prior state quantity at the current sampling time is estimated according to the posterior state quantity at the previous sampling time and the system input at the current sampling time, the predicted junction temperature is calculated according to the prior state quantity, and the prior error covariance matrix at the current sampling time is updated; Amend: the measured junction temperature is subtracted from the predicted junction temperature based on the prior estimate to obtain an innovation value at the current sampling time d k At the same time, the gain of the Kalman filter at the current sampling time is calculated K k And based on the gain K k The posterior state quantity at the current sampling time is obtained Output: real-time junction temperature information is calculated according to the posterior state quantity at the current sampling time.

6. The method of claim 5, wherein: The variable initialization is represented by a formula as follows: (0-8) In formula (0-8), the upper right corner symbol "+" on the character represents that this estimate is posterior, the initial covariance matrix of the error for the posterior state estimate, represents the posterior initial state quantity, and represents the initial temperature difference Δ T n , is equal to the chip operating junction temperature and the external water cooling system temperature T coolant . Process noise in an electric heat network model prediction w k And based on the on voltage drop V ce,on Measurement noise of junction temperature estimation v k The covariance matrix of the test working conditions is generally set as the initial value Q 0, R 0, (0-7) wherein denote denote w k denote v k the mathematical expectation of denote denote w k denote v k the covariance of Q k and R k .

7. The method of claim 5, wherein: The calculation formula of the prior error covariance matrix at the current sampling time is as follows: (0-9) wherein, represents the prior estimate of the state quantity at the k sampling time instant, x represents the system input at the k sampling time instant, represents the prior estimate of the junction temperature at the current time instant predicted by the model at the k sampling time instant, represents the prior error covariance matrix at the k sampling time instant, the upper index T denotes the transpose, represents the process noise covariance matrix at the k sampling time instant.​ 8. The method of claim 5, wherein: When the correction is made, the predicted junction temperature value based on the on-state voltage drop V ce,on The measured junction temperature T vj The innovation value is obtained by subtracting the predicted junction temperature value based on the prior estimation from the measured junction temperature d k The gain of the Kalman filter at the current time is calculated at the same time K k And the gain is used to obtain the posterior estimation of the state quantity K k x The posterior estimation of the state quantity​ (0-10) wherein, represents the k sampling time instant based on the on-state voltage drop V ce,on measured junction temperature, represents the k sampling time instant measurement noise covariance matrix, H represents the output measurement matrix for the system, represents the k sampling time instant Kalman filter gain, represents the k sampling time instant posterior estimate of the state quantity x represents the sampling time instant posterior estimate of the covariance matrix, k represents the f is a flag variable value, when using the V ce,on measured estimated junction temperature T vj variable f is set to 1, otherwise, it is set to 0; Through real-time iteration of the above steps, real-time junction temperature information predicted based on the adaptive Kalman filter can be obtained, represented as follows: 。 9. The method of claim 8, wherein: Difference between current time measurement and current a posteriori prediction The formula for the calculation is given by (0-16) wherein, represents the posterior estimate of the junction temperature at the k sampling time instant. Process noise w k and its a posteriori estimate is represented as: (0-20)。 10. The method of claim 9, wherein: Measuring a noise covariance matrix R k and a process noise covariance matrix Q k is estimated as: (0-19) (0-22) wherein denotes a forgetting factor, 0 < f < 1 A larger forgetting factor gives more weight to previous parameter estimates, so that R k , Q k The fluctuations of the wavelet are relatively small over time.

Images