IGBT service life prediction method, device and system and storage medium

Abstract

The invention provides an IGBT service life prediction method, device and system and a storage medium, and belongs to the technical field of device prediction, and the method comprises the steps: carrying out the preprocessing of original IGBT device aging historical data points, and obtaining the preprocessed IGBT device aging historical data points; constructing an IGBT degradation model through the preprocessed aging historical data points of the IGBT device; and performing prediction analysis on the to-be-predicted IGBT device data points through the IGBT degradation model, the early warning threshold value and the preprocessed IGBT device aging historical data points to obtain an IGBT service life prediction result. According to the method, the defect of particle degradation in particle filtering is effectively overcome, the IGBT service life prediction precision is further improved, and the IGBT service life prediction accuracy is greatly improved.

Description

Technical Field

[0001] The present invention mainly relates to the field of device prediction technology, and in particular to an IGBT life prediction method, device, system and storage medium. Background Art

[0002] In recent years, power electronic equipment has developed rapidly and its application scope is also expanding. However, due to the complex working environment of these devices, their key components, insulated gate bipolar transistors (IGBTs), are prone to failure. Therefore, it is particularly important to evaluate the reliability of IGBT operation. If an effective state monitoring and early warning mechanism can be provided, major accidents can be effectively prevented. Therefore, it is of great significance to study IGBT life prediction technology to avoid catastrophic failures.

[0003] In current technology, IGBT life prediction usually relies on technologies such as analytical models, analytical physical models, and neural network models. Although analytical models and analytical physical models can provide clear prediction mechanisms, their modeling process is complex and model parameters are difficult to obtain. These parameters are also closely related to factors such as the material properties used, the geometry of the product, the load during the life cycle, and the failure model. However, since the working environment of the IGBT module is usually more complex, its internal failure mechanism is often difficult to observe directly. On the other hand, although the neural network model can provide higher prediction accuracy, in order to adapt to different products and working conditions, a large amount of data support is required. These challenges limit the ability of these methods to perform online life prediction in actual industrial environments, and there is still room for improvement in prediction accuracy. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an IGBT life prediction method, device, system and storage medium in view of the deficiencies in the prior art.

[0005] The technical solution of the present invention to solve the above technical problem is as follows: a method for predicting the life of an IGBT, comprising the following steps:

[0006] Importing a plurality of original IGBT device aging history data points at a plurality of time points, preprocessing all the original IGBT device aging history data points, and obtaining a plurality of preprocessed IGBT device aging history data points;

[0007] Constructing an IGBT degradation model through all the IGBT device aging history data points after the preprocessing;

[0008] Import the IGBT device data points to be predicted and the warning threshold, perform prediction analysis on the IGBT device data points to be predicted through the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result.

[0009] Another technical solution of the present invention to solve the above technical problem is as follows: an IGBT life prediction device, comprising:

[0010] Import module, used to import multiple original IGBT device aging history data points at multiple times;

[0011] A preprocessing module, used for preprocessing all the original IGBT device aging history data points to obtain a plurality of preprocessed IGBT device aging history data points;

[0012] A construction module, used to construct an IGBT degradation model through all the pre-processed IGBT device aging history data points;

[0013] The import module is also used to import the IGBT device data points and warning thresholds to be predicted;

[0014] The prediction result acquisition module is used to perform prediction analysis on the IGBT device data points to be predicted through the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result.

[0015] Based on the above-mentioned IGBT life prediction method, the present invention also provides an IGBT life prediction system.

[0016] Another technical solution of the present invention to solve the above technical problem is as follows: an IGBT life prediction system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the IGBT life prediction method as described above is implemented.

[0017] Based on the above-mentioned IGBT life prediction method, the present invention also provides a computer-readable storage medium.

[0018] Another technical solution of the present invention to solve the above technical problem is as follows: a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the IGBT life prediction method as described above is implemented.

[0019] The beneficial effects of the present invention are as follows: preprocessing the original IGBT device aging history data points to obtain preprocessed IGBT device aging history data points, building an IGBT degradation model through the preprocessed IGBT device aging history data points, and obtaining an IGBT life prediction result through the IGBT degradation model, the early warning threshold and the preprocessed IGBT device aging history data points on the IGBT device data points to be predicted, thereby effectively making up for the defect of particle degradation in particle filtering, thereby improving the precision of IGBT life prediction, and greatly improving the accuracy of IGBT life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of a flow chart of an IGBT life prediction method provided by an embodiment of the present invention;

[0021] Figure 2 This is a module block diagram of an IGBT life prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0023] Figure 1 A schematic flow chart of an IGBT life prediction method provided by an embodiment of the present invention.

[0024] like Figure 1 As shown, a method for predicting the life of an IGBT comprises the following steps:

[0025] Importing a plurality of original IGBT device aging history data points at a plurality of time points, preprocessing all the original IGBT device aging history data points, and obtaining a plurality of preprocessed IGBT device aging history data points;

[0026] Constructing an IGBT degradation model through all the IGBT device aging history data points after the preprocessing;

[0027] Import the IGBT device data points to be predicted and the warning threshold, perform prediction analysis on the IGBT device data points to be predicted through the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result.

[0028] It should be understood that the saturated on-state collector-emitter voltage V of the IGBT device disclosed by NASA is obtained ce(on) Aging history data (i.e., original IGBT device aging history data points).

[0029] Specifically, according to the saturated on-state collector-emitter voltage V of the pre-processed IGBT device ce(on) The aging history data (i.e., the aging history data points of the IGBT device after preprocessing) are used to fit the IGBT degradation model.

[0030] In the above embodiment, the original IGBT device aging history data points are preprocessed to obtain the preprocessed IGBT device aging history data points, an IGBT degradation model is constructed through the preprocessed IGBT device aging history data points, and the IGBT life prediction result is obtained through the IGBT degradation model, the early warning threshold and the preprocessed IGBT device aging history data points on the IGBT device data points to be predicted, which effectively compensates for the defect of particle degradation in the particle filter, thereby improving the precision of the IGBT life prediction, and greatly improving the accuracy of the IGBT life prediction.

[0031] Optionally, as an embodiment of the present invention, the process of preprocessing all the original IGBT device aging history data points to obtain a plurality of preprocessed IGBT device aging history data points includes:

[0032] Calculate the median of all the original IGBT device aging history data points to obtain the median of the IGBT device aging history data;

[0033] The abnormality detection threshold is calculated by using the first formula on the median of the IGBT device aging history data and all the original IGBT device aging history data points to obtain the abnormality detection threshold. The first formula is:

[0034] threshold=x m +k*MAD,

[0035] in,

[0036] Among them, threshold is the anomaly detection threshold, x m is the median of the IGBT device aging historical data, k is a constant, MAD is the median absolute deviation, Median is the median function, is the i-th original IGBT device aging history data point at the k-th moment;

[0037] Determine whether the original IGBT device aging history data point is greater than the abnormal detection threshold. If so, fill in the original IGBT device aging history data point to obtain the pre-processed IGBT device aging history data point; if not, use the original IGBT device aging history data point as the pre-processed IGBT device aging history data point, thereby obtaining multiple pre-processed IGBT device aging history data points.

[0038] It should be understood that the saturated on-state collector-emitter voltage V ce(on) The complete aging history data (i.e., the original IGBT device aging history data points) is preprocessed. For the outliers in the aging history data, the median absolute deviation (MAD) outlier recognition method is used to screen out the data that falls within the outlier detection range, and the outliers are filled with other valid values. The processed data is then scaled and mapped to the same range according to a unified standard.

[0039] Specifically, calculate the median of the sample data, denoted by x m (i.e., the median of the IGBT device aging history data); calculate the absolute deviation of each data point from the median, i.e., the absolute value of the difference between each data point and the median, recorded as: |x i -x m |, where x i represents the i-th data point; calculate the median of all absolute deviations (i.e., median of absolute deviations), i.e.: MAD = Median(|x i -x m |), when performing outlier detection, the outlier threshold (i.e., anomaly detection threshold) is set to the median plus a constant k multiplied by MAD. That is, if the absolute difference of a data point exceeds k times the MAD, it can be considered an outlier, where k is an empirical constant, usually 2 or 3. If x i >threshold, it is treated as an outlier. The formula for the anomaly detection threshold is as follows:

[0040] threshold=x m +k*MAD,

[0041] The abnormal values ​​are filled with other valid values, and the mean of the data in a single cycle is used as the valid value of the cycle for effective value filling. The above-mentioned processed aging history data is mapped to the same range according to a unified standard by means of scale transformation.

[0042] In the above embodiment, all original IGBT device aging history data points are preprocessed to obtain preprocessed IGBT device aging history data points, which can filter out data falling within the abnormal value detection range, effectively making up for the defect of particle degradation in particle filtering, thereby improving the accuracy of IGBT life prediction.

[0043] Optionally, as an embodiment of the present invention, the process of performing predictive analysis on the IGBT device data points to be predicted by using the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result includes:

[0044] Establishing an initial state equation and an initial observation equation through the IGBT degradation model;

[0045] Performing a weighted average analysis on the initial state equation, the initial observation equation, and all the pre-processed IGBT device aging history data points to obtain a weighted average value;

[0046] According to the weighted average value, the parameters of the initial state equation and the initial observation equation are respectively updated to obtain an updated state equation corresponding to the initial state equation and an updated observation equation corresponding to the initial observation equation;

[0047] Constructing a prediction model through the updated state equation and the updated observation equation;

[0048] The IGBT device data point to be predicted is predicted using the prediction model and the warning threshold to obtain an IGBT life prediction result.

[0049] It should be understood that the state equation f (ie, the initial state equation) and the observation equation h (ie, the initial observation equation) are established using the IGBT degradation model.

[0050] Specifically, the IGBT degradation model is V = a*exp(b*k)+c*exp(d*k), where a, b, c, d are model parameters, k is the time, i.e., the number of aging cycles, and the state equation f(k) and observation equation h(k) are as follows:

[0051]

[0052] h(k)=a(k)exp(b(k)*k)+c(k)exp(d(k)*k)+v(k),

[0053] Where a, b, c, d represent the model parameters obtained from the degradation model, u(k) represents the process noise; v(k) represents the observation noise. 0 ,u k )=0,Cov(x 0 ,v k )=0, we can see that both process noise and observation noise have mean 0 and variance σ v Gaussian white noise.

[0054] It should be understood that setting a warning threshold and setting Vce(on) The threshold is 1.2 times of the initial value. The warning threshold is substituted into the prediction equation (ie, the prediction model), and then the failure time of the IGBT device can be calculated.

[0055] It should be understood that the weighted average As output, the state of the nonlinear system is updated (i.e., the initial state equation and the initial observation equation), that is, the parameters in the life model are updated to determine the prediction equation.

[0056] In the above embodiment, the IGBT life prediction result is obtained by predicting and analyzing the IGBT device data points to be predicted based on the IGBT degradation model, the early warning threshold and the pre-processed IGBT device aging historical data points, which effectively compensates for the defect of particle degradation in the particle filter, thereby improving the accuracy of the IGBT life prediction and greatly improving the accuracy of the IGBT life prediction.

[0057] Optionally, as an embodiment of the present invention, the process of performing weighted average analysis on the initial state equation, the initial observation equation and all the pre-processed IGBT device aging historical data points to obtain the weighted average includes:

[0058] S321: performing a priori calculation on each of the pre-processed IGBT device aging history data points at each moment by using the second formula to obtain a particle a priori data point corresponding to each of the pre-processed IGBT device aging history data points at each moment, wherein the second formula is:

[0059]

[0060] in, is the particle priori data point corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th time, E is the expected value, is the aging historical data point of the i-th pre-processed IGBT device at the k-th moment;

[0061] S322: Calculate the covariance matrix of each pre-processed IGBT device aging historical data point at each moment and the particle priori data point corresponding to each pre-processed IGBT device aging historical data point at each moment by using the third formula to obtain the particle error covariance matrix corresponding to each pre-processed IGBT device aging historical data point at each moment. The third formula is:

[0062]

[0063] in, is the particle error covariance matrix corresponding to the ith preprocessed IGBT device aging historical data point at the kth time, E is the expected value, is the particle priori data point corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th time, is the aging historical data point of the i-th pre-processed IGBT device at the k-th moment;

[0064] S323: Calculate the Sigma point set of the particle error covariance matrix corresponding to each of the particle priori data points and each of the pre-processed IGBT device aging history data points at each moment by using the fourth formula to obtain the Sigma point set corresponding to each of the pre-processed IGBT device aging history data points at each moment. The fourth formula is:

[0065]

[0066] in, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th time, is the particle priori data point corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th time, is the particle error covariance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th moment;

[0067] S324: Calculate the state vector prediction value of the initial state equation and the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment by using the fifth formula to obtain the state vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The fifth formula is:

[0068]

[0069] in, is the predicted value of the state vector corresponding to the ith pre-processed IGBT device aging historical data point at the k+1th moment, f() is the initial state equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th moment;

[0070] S325: Importing the system noise covariance matrix at multiple moments, respectively performing prediction analysis on the initial state equation, the system noise covariance matrix at each moment, each pre-processed IGBT device aging historical data point at each moment, the particle priori data points corresponding to each pre-processed IGBT device aging historical data point at each moment, and the Sigma point set corresponding to each pre-processed IGBT device aging historical data point at each moment, to obtain the state vector prediction error variance matrix corresponding to each pre-processed IGBT device aging historical data point at each moment and the prediction error value corresponding to each pre-processed IGBT device aging historical data point at each moment;

[0071] S326: Calculate the observation vector prediction value for the initial observation equation and the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the sixth formula, to obtain the observation vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the sixth formula being:

[0072]

[0073] in, is the predicted value of the observation vector corresponding to the ith pre-processed IGBT device aging historical data point at the k+1th moment, h() is the initial observation equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th moment;

[0074] S327: respectively update and analyze the initial observation equation, the observation vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the state vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the prediction error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, to obtain the state vector estimation value corresponding to each of the pre-processed IGBT device aging historical data points at each moment and the estimation error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment;

[0075] S328: updating the data points of the preprocessed IGBT device aging history data points at each moment according to the estimated state vector corresponding to each of the preprocessed IGBT device aging history data points at each moment and the estimated error variance matrix corresponding to each of the preprocessed IGBT device aging history data points at each moment, obtaining updated IGBT device aging history data points corresponding to each of the preprocessed IGBT device aging history data points at each moment, and gathering all of the updated IGBT device aging history data points to obtain an updated IGBT device aging history data set;

[0076] S329: performing weight analysis on the updated IGBT device aging history data set, the updated IGBT device aging history data points corresponding to the pre-processed IGBT device aging history data points at each moment, and the observation vector prediction values ​​corresponding to the pre-processed IGBT device aging history data points at each moment, respectively, to obtain the to-be-processed weights corresponding to the pre-processed IGBT device aging history data points at each moment;

[0077] S3210: Calculate all the weights to be processed by the seventh formula to obtain the total number of target data points. The seventh formula is:

[0078]

[0079] Among them, M eff is the total number of target data points, is the weight to be processed corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, and M is the total number of pre-processed IGBT device aging historical data points;

[0080] S3211: Determine whether the total number of target data points is less than a preset data point threshold. If so, execute S3212; if not, use all the weights to be processed as target weights, so as to obtain the target weights corresponding to the pre-processed IGBT device aging history data points at each moment, and execute 3214;

[0081] S3212: Counting the total number of all the pre-processed IGBT device aging history data points to obtain the total number of pre-processed IGBT device aging history data points;

[0082] S3213: Calculate the weight of the total number of the pre-processed IGBT device aging history data points by the eighth formula to obtain the target weight corresponding to each of the pre-processed IGBT device aging history data points at each moment, and execute S3214. The eighth formula is:

[0083]

[0084] in, is the target weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, and M is the total number of pre-processed IGBT device aging historical data points;

[0085] S3214: Calculate the weighted average of all the target weights and all the updated IGBT device aging history data points using the ninth formula to obtain a weighted average value. The ninth formula is:

[0086]

[0087] in, is the weighted average, is the target weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, is the updated IGBT device aging historical data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, and M is the total number of IGBT device aging historical data points after pre-processing.

[0088] It should be understood that initialization; calculation of Sigma point set; further prediction of Sigma point set (k ≥ 1), time update of nonlinear system equation; further prediction of Sigma point set (k ≥ 1), observation update of nonlinear system equation; in the sampling particle stage, the estimated value of the updated state vector is used (i.e., the state vector estimate) and the estimated error variance matrix (i.e., the estimated error variance matrix) is used to sample and update the particle set.

[0089] It should be understood that in the resampling stage, if (M eff <M thres ), reset the weights to Get a set of samples with equal weights In the state estimation phase, the weighted average of the particle set is calculated

[0090] Specifically, in the initialization stage (k=0), the number of particles is set to N, and the particle set is initialized. The specific calculation is as follows;

[0091]

[0092] The first formula above generates the particle set at the initial stage (k=0); the second formula above calculates the particle prior information at the initial stage (k=0) (i.e., particle prior data points); the third and fifth equations above calculate the particle error covariance matrix in the initial stage (k=0) (i.e., particle error covariance matrix); in the fourth formula above is the error function.

[0093] Specifically, the Sigma point set is calculated as follows:

[0094]

[0095] Assuming that the process noise and observation noise of the nonlinear system are Gaussian white noise, the above formula represents the Sigma point set at time k-1 (i.e., Sigma point set), represents the prior information at time k-1 (i.e., the particle prior data point), Represents the particle error covariance matrix at time k-1.

[0096] It should be understood that the formula for the one-step prediction value of the state vector is obtained as follows:

[0097]

[0098] in, is the state equation function (i.e., the predicted value of the state vector), Indicates the time k Predict k+1 time Estimated value (i.e., predicted value of the state vector), f represents a nonlinear function.

[0099] It should be understood that the one-step prediction value of the observation vector is obtained, and the observation equation formula is as follows:

[0100]

[0101] in, represents the observation equation function (i.e., the observation vector prediction value), and h represents the nonlinear function.

[0102] Specifically, in the sampling particle stage, the estimated value of the updated state vector is used (i.e., the state vector estimate) and the estimated error variance matrix (i.e., particle error covariance matrix) to sample and update the particle set. The specific calculation is as follows:

[0103]

[0104] Among them, the above formula is to update the particle set

[0105] Specifically, in the resampling phase, if (M eff <M thres ), reset the weights to Get a set of samples with equal weights Among them, M eff (i.e. the total number of target data points) The specific calculation formula is as follows:

[0106]

[0107] It should be understood that in the state estimation phase, the weighted average of the particle set is calculated The specific calculation is as follows:

[0108]

[0109] In the above embodiment, a weighted average value is obtained by performing a weighted average analysis on the initial state equation, the initial observation equation and the historical data points of the IGBT device aging after preprocessing, which effectively compensates for the defect of particle degradation in the particle filter, thereby improving the accuracy of the IGBT life prediction and greatly improving the accuracy of the IGBT life prediction.

[0110] Optionally, as an embodiment of the present invention, in S325, the initial state equation, the system noise covariance matrix at each moment, each pre-processed IGBT device aging history data point at each moment, the particle priori data points corresponding to each of the pre-processed IGBT device aging history data points at each moment, and the Sigma point set corresponding to each of the pre-processed IGBT device aging history data points at each moment are respectively predicted and analyzed to obtain the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging history data points at each moment and the prediction error value corresponding to each of the pre-processed IGBT device aging history data points at each moment, and the process includes:

[0111] The prediction error value is calculated for the initial state equation, each pre-processed IGBT device aging history data point at each moment, the particle priori data point corresponding to each pre-processed IGBT device aging history data point at each moment, and the Sigma point set corresponding to each pre-processed IGBT device aging history data point at each moment through the tenth formula, to obtain the prediction error value corresponding to each pre-processed IGBT device aging history data point at each moment. The tenth formula is:

[0112]

[0113] in,

[0114] in, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, f() is the initial state equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th time, is the initial error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, is the aging historical data point of the i-th pre-processed IGBT device at the k-1th moment, is the particle priori data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment;

[0115] The state vector prediction error variance matrix is ​​calculated by respectively calculating the system noise covariance matrix at each moment and the prediction error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the eleventh formula, so as to obtain the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The eleventh formula is:

[0116]

[0117] in, is the state vector prediction error variance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th time, Q k is the system noise covariance matrix at the kth moment.

[0118] Specifically, the error function (i.e., prediction error value) in the prediction stage (k ≥ 1) is as follows:

[0119]

[0120] The one-step prediction error variance matrix (i.e., the state vector prediction error variance matrix) is obtained, and the calculation formula is as follows:

[0121]

[0122] in, is the error function (i.e., prediction error value) in the prediction stage (k ≥ 1), Represents the error function at time k-1 (i.e., the initial error value); the first and third equations above are used to calculate the Sigma point set in the prediction stage (k≥1) to obtain the one-step prediction error variance matrix (i.e., the state vector prediction error variance matrix), Indicates the time k Predict k+1 time Q k represents the system noise covariance matrix at time k;

[0123] In the above embodiment, the initial state equation, the system noise covariance matrix, the preprocessed IGBT device aging historical data points, the particle prior data points and the Sigma point set are predicted and analyzed to obtain the state vector prediction error variance matrix and the prediction error value, which effectively compensates for the defect of particle degradation in the particle filter, thereby improving the accuracy of IGBT life prediction and greatly improving the accuracy of IGBT life prediction.

[0124] Optionally, as an embodiment of the present invention, the process of S327 includes:

[0125] The updated error value is calculated by respectively using the initial observation equation, the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the predicted error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the twelfth formula, so as to obtain the updated error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The twelfth formula is:

[0126]

[0127] in, is the updated error value corresponding to the ith pre-processed IGBT device aging historical data point at the k+1th moment, h() is the initial observation equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k+1th time, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment;

[0128] The updated error values ​​corresponding to the aging historical data points of the pre-processed IGBT devices at each moment are respectively calculated by the thirteenth formula to obtain the observation vector prediction error variance matrix corresponding to the aging historical data points of the pre-processed IGBT devices at each moment. The thirteenth formula is:

[0129]

[0130] in, is the prediction error variance matrix of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the updated error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time;

[0131] The Kalman gain matrix corresponding to each of the preprocessed IGBT device aging historical data points at each moment is calculated by the fourteenth formula, respectively, for the observation vector prediction error variance matrix corresponding to each of the preprocessed IGBT device aging historical data points at each moment, the update error value corresponding to each of the preprocessed IGBT device aging historical data points at each moment, and the prediction error value corresponding to each of the preprocessed IGBT device aging historical data points at each moment, to obtain the Kalman gain matrix corresponding to each of the preprocessed IGBT device aging historical data points at each moment. The fourteenth formula is:

[0132]

[0133] in,

[0134] in, is the Kalman gain matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the cross-covariance matrix corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time, is the prediction error variance matrix of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time, is the updated error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time;

[0135] The state vector estimation value is calculated by respectively calculating the initial observation equation, the state vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the Kalman gain matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the observation vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the fifteenth formula, so as to obtain the state vector estimation value corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The fifteenth formula is:

[0136]

[0137] in, is the estimated value of the state vector corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time, h() is the initial observation equation function, is the predicted value of the state vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the Kalman gain matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the predicted value of the observation vector corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment;

[0138] The Kalman gain matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the observation vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment are calculated by the sixteenth formula to obtain the estimated error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The sixteenth formula is:

[0139]

[0140] in, is the estimated error variance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the state vector prediction error variance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the Kalman gain matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, It is the prediction error variance matrix of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time.

[0141] Specifically, the error function (i.e., updated error value) in the update phase (k ≥ 1) is as follows:

[0142]

[0143] The one-step prediction error variance matrix (i.e., observation vector prediction error variance matrix) is obtained as follows:

[0144]

[0145] The cross covariance matrix (i.e., cross covariance matrix) is given by the following formula:

[0146]

[0147] The corresponding Kalman gain matrix is ​​as follows:

[0148]

[0149] The estimated value of the state vector (i.e., the estimated value of the state vector) is as follows:

[0150]

[0151] The corresponding estimation error variance matrix (i.e., estimation error variance matrix) is given by the following formula:

[0152]

[0153] in, is the error function (i.e., updated error value) in the update phase (k ≥ 1); the second formula above is to calculate the Sigma point set in the update phase (k ≥ 1) to obtain the one-step prediction error variance matrix (i.e., the observation vector prediction error variance matrix); the third formula above is used to calculate the Sigma point set cross covariance matrix in the update phase (k ≥ 1) (i.e., the cross covariance matrix); the above K k+1 is the Kalman gain matrix, and the fifth formula above represents the estimated value of the updated sample (i.e., the estimated value of the state vector); the sixth formula above represents the updated error covariance matrix (i.e., the estimated error variance matrix).

[0154] In the above embodiment, the initial observation equation, observation vector prediction value, state vector prediction value, Sigma point set, state vector prediction error variance matrix and prediction error value are updated and analyzed to obtain the state vector estimation value and the estimation error variance matrix, which effectively compensates for the defect of particle degradation in particle filtering and thereby improves the accuracy of IGBT life prediction.

[0155] Optionally, as an embodiment of the present invention, the process of S329 includes:

[0156] The updated IGBT device aging history data set, the updated IGBT device aging history data points corresponding to the pre-processed IGBT device aging history data points at each moment, and the observation vector prediction values ​​corresponding to the pre-processed IGBT device aging history data points at each moment are respectively calculated by the seventeenth formula to obtain the normalized weights corresponding to the pre-processed IGBT device aging history data points at each moment. The seventeenth formula is:

[0157]

[0158] in,

[0159] in, is the normalized weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, p(|) is the prior probability, is the predicted value of the observation vector corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th time, is the updated IGBT device aging historical data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, is the initial weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, is the initial weight corresponding to the jth pre-processed IGBT device aging historical data point at the k-1th moment, M is the total number of pre-processed IGBT device aging historical data points, α is the improvement factor, is the predicted value of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k-1th moment, is the updated IGBT device aging historical data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, and A is the updated IGBT device aging historical data set;

[0160] Normalization processing is performed on each of the weights to be normalized, respectively, to obtain the weight to be processed corresponding to each of the pre-processed IGBT device aging historical data points at each moment.

[0161] It should be understood that by calculating the weights, introducing the improvement factor α, and improving the calculation of the weights, the phenomenon of a small number of particles occupying large weights can be avoided, thereby effectively compensating for the defect of particle degradation in the particle filter, and further improving the accuracy of IGBT life prediction based on the central difference Kalman particle filter algorithm.

[0162] Specifically, the calculation is as follows:

[0163]

[0164] Among them, the above formula is to calculate the particle weight (i.e. weights to be normalized), represents the prior probability.

[0165] It should be understood that the calculated weight value (i.e., the weight to be normalized) is normalized, and the specific calculation is as follows:

[0166]

[0167] The above formula represents the normalization processing of the weight value (i.e., the weight to be normalized).

[0168] In the above embodiment, weight analysis is performed on the updated IGBT device aging history data set, the updated IGBT device aging history data points and the observation vector prediction value to obtain the weight to be processed, thereby avoiding the phenomenon that a small number of particles occupy large weights, thereby effectively compensating for the defect of particle degradation in particle filtering, and further improving the accuracy of IGBT life prediction.

[0169] Optionally, as another embodiment of the present invention, in an actual application scenario, the present invention connects a saturated on-state collector-emitter voltage measuring device to the collector (C) and emitter (E) of the IGBT to detect the saturated on-state collector-emitter voltage at the moment the IGBT is turned on again when the minimum case temperature (Tcmin) setting value is reached under the same current (IC) conditions, and the saturated on-state collector-emitter voltage and the number of turn-ons are recorded in real time.

[0170] Optionally, as another embodiment of the present invention, the present invention defines the state vector of the particle filter as x k =[a(k),b(k),c(k),d(k)], at time k=0 x 0 It can be obtained by fitting the IGBT degradation model. It should be understood that the state vector is x k =[a(k),b(k),c(k),d(k)], where a, b, c, d are the same coefficients as those in the state equation f(k) and the observation equation h(k), representing the model parameters of the degradation model.

[0171] Optionally, as another embodiment of the present invention, the present invention includes the following steps:

[0172] Step 1: Obtain the saturated on-state collector-emitter voltage V of the IGBT device published by NASA ce(on) Aging history data;

[0173] Step 2: Preprocess the data obtained in step 1;

[0174] Step 3: In the actual application scenario, connect the saturated on-state collector-emitter voltage measurement device to the collector (C) and emitter (E) of the IGBT to detect the saturated on-state collector-emitter voltage at the moment the IGBT is turned on again when the minimum case temperature (Tcmin) setting value is reached under the same current (IC) conditions, and record the saturated on-state collector-emitter voltage and the number of times the IGBT is turned on in real time.

[0175] Step 4: Based on the saturated on-state collector-emitter voltage V of the IGBT device pre-processed in step 2 ce(on) Fitting IGBT degradation model with aging history data;

[0176] Step 5: Use the IGBT degradation model to establish the state equation f and observation equation h;

[0177] Step 6: Define the state vector of the particle filter as x k =[a(k),b(k),c(k),d(k)], at time k=0 x 0 It can be obtained by fitting the data obtained in step 4;

[0178] Step 7: Initialization;

[0179] Step 8: Calculate the Sigma point set;

[0180] Step 9: Make further predictions on the Sigma point set (k ≥ 1) and perform time updates on the nonlinear system equations;

[0181] Step 10: Make further predictions on the Sigma point set (k ≥ 1) and perform observation updates on the nonlinear system equations;

[0182] Step 11: In the sampling particle phase, use the estimated value of the updated state vector and the estimated error variance matrix To sample and update the particle set;

[0183] Step 12: Calculate the weights, introduce the improvement factor α, and improve the calculation of the weights, so that the phenomenon of a small number of particles occupying large weights can be avoided, thereby effectively compensating for the defect of particle degradation in the particle filter, and further improving the accuracy of IGBT life prediction based on the central difference Kalman particle filter algorithm;

[0184] Step 13: Normalize the weight value;

[0185] Step 14: In the resampling phase, if (M eff <M thres ), reset the weights to Get a set of samples with equal weights

[0186] Step 15: In the state estimation phase, calculate the weighted average of the particle set

[0187] Step 16: Take the weighted average As output, the state of the nonlinear system is updated, that is, the parameters in the life model are updated to determine the prediction equation;

[0188] Step 17: Set an early warning threshold and set V ce(on) The threshold is 1.2 times of the initial value. Substituting the warning threshold into the prediction equation, the failure time of the IGBT device can be calculated.

[0189] Step 18: Replace the prediction data of the last moment according to the data detected and recorded in step 3, repeat the process from step 6 to step 18, dynamically update the data after the prediction, and realize the real-time life prediction of the IGBT module.

[0190] Optionally, as another embodiment of the present invention, the specific steps of the present invention are as follows:

[0191] Firstly, the idea of ​​difference is introduced into particle filter, so that the central difference Kalman filter and particle filter are integrated, and the algorithm is applied to IGBT life prediction for the first time. Then, the absolute median difference MAD is used to preprocess the data, and the data falling within the outlier detection range is screened out, and the outliers are filled with other valid values. Then the processed data is scaled and mapped to the same range according to a unified standard. Then, when calculating the weight, the improvement factor α is introduced to improve the calculation of the weight, so that the phenomenon of a small number of particles occupying a large weight is avoided, thereby effectively making up for the defect of particle degradation in particle filtering, and then improving the accuracy of IGBT life prediction based on the central difference Kalman particle filter algorithm. Finally, the prediction data of the previous moment is replaced according to the data measured in the actual application scenario, and the process of steps 6 to 17 is repeated to dynamically update the data after the prediction, so as to realize the real-time life prediction of the IGBT module. To a certain extent, the algorithm has simple steps, is easy to implement, and has high accuracy, so as to further improve the prediction accuracy.

[0192] Optionally, as another embodiment of the present invention, the present invention makes four different improvements to the particle filtering algorithm:

[0193] 1. The idea of ​​difference is introduced into particle filtering, so that the central difference Kalman filter and particle filter are integrated and applied to the prediction of the remaining life of IGBT for the first time.

[0194] 2. In data preprocessing, the median absolute deviation (MAD) outlier identification method is used to screen out the data that falls within the outlier detection range, and the outliers are filled with other valid values. The processed data is then scaled and mapped to the same range according to a unified standard.

[0195] 3. In the weight calculation, an improvement factor α (0≤α≤1) is introduced to improve the calculation of weights, so that the phenomenon of a small number of particles occupying large weights can be avoided, thereby effectively compensating for the defect of particle degradation in particle filtering and further improving the accuracy of IGBT life prediction based on the central difference Kalman particle filter algorithm.

[0196] 4. In actual application scenarios, the saturated on-state collector-emitter voltage measurement module is connected to the collector (C) and emitter (E) of the IGBT. When the case temperature reaches the minimum (Tcmin) setting value under the same current (IC) condition, the saturated on-state collector-emitter voltage of the IGBT at the moment of first turn-on is measured and recorded. The predicted data of the previous moment is replaced according to the measured data, and the algorithm process is repeated to update the data after the prediction in real time to realize the real-time life prediction of the IGBT module.

[0197] The above four improvement measures make the IGBT life prediction to a certain extent. The steps of the present invention are relatively simple, easy to implement, and have high accuracy. This method greatly improves the accuracy of the remaining service life prediction of the IGBT.

[0198] Figure 2 A module block diagram of an IGBT life prediction device provided by an embodiment of the present invention.

[0199] Optionally, as another embodiment of the present invention, Figure 2 As shown, an IGBT life prediction device comprises:

[0200] Import module, used to import multiple original IGBT device aging history data points at multiple times;

[0201] A preprocessing module, used for preprocessing all the original IGBT device aging history data points to obtain a plurality of preprocessed IGBT device aging history data points;

[0202] A construction module, used to construct an IGBT degradation model through all the pre-processed IGBT device aging history data points;

[0203] The import module is also used to import the IGBT device data points and warning thresholds to be predicted;

[0204] The prediction result acquisition module is used to perform prediction analysis on the IGBT device data points to be predicted through the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result.

[0205] Optionally, another embodiment of the present invention provides an IGBT life prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the IGBT life prediction method described above is implemented. The system may be a computer or other system.

[0206] Optionally, another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the IGBT life prediction method as described above is implemented.

[0207] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0208] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0210] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0211] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0212] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0213] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting IGBT life, characterized in that: The steps include: Importing a plurality of original IGBT device aging history data points at a plurality of time points, preprocessing all the original IGBT device aging history data points, and obtaining a plurality of preprocessed IGBT device aging history data points; Constructing an IGBT degradation model through all the IGBT device aging history data points after the preprocessing; Import the IGBT device data points to be predicted and the warning threshold, perform prediction analysis on the IGBT device data points to be predicted through the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result.

2. The IGBT life prediction method according to claim 1, characterized in that: The process of preprocessing all the original IGBT device aging history data points to obtain a plurality of preprocessed IGBT device aging history data points comprises: Calculate the median of all the original IGBT device aging history data points to obtain the median of the IGBT device aging history data; The abnormality detection threshold is calculated by using the first formula on the median of the IGBT device aging history data and all the original IGBT device aging history data points to obtain the abnormality detection threshold. The first formula is: threshold=x m +k*MAD, in, Among them, threshold is the abnormal detection threshold, xm is the median of the IGBT device aging history data, k is a constant, MAD is the median absolute deviation, Median is the median function, is the i-th original IGBT device aging history data point at the k-th moment; Determine whether the original IGBT device aging history data point is greater than the abnormal detection threshold. If so, fill in the original IGBT device aging history data point to obtain the pre-processed IGBT device aging history data point; if not, use the original IGBT device aging history data point as the pre-processed IGBT device aging history data point, thereby obtaining multiple pre-processed IGBT device aging history data points.

3. The IGBT life prediction method according to claim 1, characterized in that: The process of performing prediction analysis on the IGBT device data points to be predicted by using the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result includes: Establishing an initial state equation and an initial observation equation through the IGBT degradation model; Performing a weighted average analysis on the initial state equation, the initial observation equation, and all the pre-processed IGBT device aging history data points to obtain a weighted average value; According to the weighted average value, the parameters of the initial state equation and the initial observation equation are respectively updated to obtain an updated state equation corresponding to the initial state equation and an updated observation equation corresponding to the initial observation equation; Constructing a prediction model through the updated state equation and the updated observation equation; The IGBT device data point to be predicted is predicted using the prediction model and the warning threshold to obtain an IGBT life prediction result.

4. The IGBT life prediction method according to claim 3, characterized in that: The process of performing weighted average analysis on the initial state equation, the initial observation equation and all the pre-processed IGBT device aging historical data points to obtain the weighted average includes: S321: performing a priori calculation on each of the pre-processed IGBT device aging history data points at each moment by using the second formula to obtain a particle a priori data point corresponding to each of the pre-processed IGBT device aging history data points at each moment, wherein the second formula is: in, is the particle priori data point corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th time, E is the expected value, is the aging historical data point of the i-th pre-processed IGBT device at the k-th moment; S322: Calculate the covariance matrix of each pre-processed IGBT device aging historical data point at each moment and the particle priori data point corresponding to each pre-processed IGBT device aging historical data point at each moment by using the third formula to obtain the particle error covariance matrix corresponding to each pre-processed IGBT device aging historical data point at each moment. The third formula is: in, is the particle error covariance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th time, E is the expected value, is the particle priori data point corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th time, is the aging historical data point of the i-th pre-processed IGBT device at the k-th moment; S323: Calculate the Sigma point set of the particle error covariance matrix corresponding to each of the particle priori data points and each of the pre-processed IGBT device aging history data points at each moment by using the fourth formula to obtain the Sigma point set corresponding to each of the pre-processed IGBT device aging history data points at each moment. The fourth formula is: in, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th time, is the particle priori data point corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th time, is the particle error covariance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k-th moment; S324: Calculate the state vector prediction value of the initial state equation and the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment by using the fifth formula to obtain the state vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The fifth formula is: in, is the predicted value of the state vector corresponding to the ith pre-processed IGBT device aging historical data point at the k+1th moment, f() is the initial state equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th moment; S325: Importing the system noise covariance matrix at multiple moments, respectively performing prediction analysis on the initial state equation, the system noise covariance matrix at each moment, each pre-processed IGBT device aging historical data point at each moment, the particle priori data points corresponding to each pre-processed IGBT device aging historical data point at each moment, and the Sigma point set corresponding to each pre-processed IGBT device aging historical data point at each moment, to obtain the state vector prediction error variance matrix corresponding to each pre-processed IGBT device aging historical data point at each moment and the prediction error value corresponding to each pre-processed IGBT device aging historical data point at each moment; S326: Calculate the observation vector prediction value for the initial observation equation and the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the sixth formula, to obtain the observation vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the sixth formula being: in, is the predicted value of the observation vector corresponding to the ith pre-processed IGBT device aging historical data point at the k+1th moment, h() is the initial observation equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th moment; S327: respectively update and analyze the initial observation equation, the observation vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the state vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the prediction error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, to obtain the state vector estimation value corresponding to each of the pre-processed IGBT device aging historical data points at each moment and the estimation error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment; S328: updating the data points of the preprocessed IGBT device aging history data points at each moment according to the estimated state vector corresponding to each of the preprocessed IGBT device aging history data points at each moment and the estimated error variance matrix corresponding to each of the preprocessed IGBT device aging history data points at each moment, obtaining updated IGBT device aging history data points corresponding to each of the preprocessed IGBT device aging history data points at each moment, and gathering all of the updated IGBT device aging history data points to obtain an updated IGBT device aging history data set; S329: performing weight analysis on the updated IGBT device aging history data set, the updated IGBT device aging history data points corresponding to the pre-processed IGBT device aging history data points at each moment, and the observation vector prediction values ​​corresponding to the pre-processed IGBT device aging history data points at each moment, respectively, to obtain the to-be-processed weights corresponding to the pre-processed IGBT device aging history data points at each moment; S3210: Calculate all the weights to be processed by the seventh formula to obtain the total number of target data points. The seventh formula is: Among them, M eff is the total number of target data points, is the weight to be processed corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, and M is the total number of pre-processed IGBT device aging historical data points; S3211: Determine whether the total number of target data points is less than a preset data point threshold. If so, execute S3212; if not, use all the weights to be processed as target weights, so as to obtain the target weights corresponding to the pre-processed IGBT device aging history data points at each moment, and execute 3214; S3212: Counting the total number of all the pre-processed IGBT device aging history data points to obtain the total number of pre-processed IGBT device aging history data points; S3213: Calculate the weight of the total number of the pre-processed IGBT device aging history data points by the eighth formula to obtain the target weight corresponding to each of the pre-processed IGBT device aging history data points at each moment, and execute S3214. The eighth formula is: in, is the target weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, and M is the total number of pre-processed IGBT device aging historical data points; S3214: Calculate the weighted average of all the target weights and all the updated IGBT device aging history data points using the ninth formula to obtain a weighted average value. The ninth formula is: in, is the weighted average, is the target weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, is the updated IGBT device aging historical data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, and M is the total number of IGBT device aging historical data points after pre-processing.

5. The IGBT life prediction method according to claim 4, characterized in that: In S325, the process of performing prediction analysis on the initial state equation, the system noise covariance matrix at each moment, each pre-processed IGBT device aging history data point at each moment, the particle priori data point corresponding to each pre-processed IGBT device aging history data point at each moment, and the Sigma point set corresponding to each pre-processed IGBT device aging history data point at each moment to obtain the state vector prediction error variance matrix corresponding to each pre-processed IGBT device aging history data point at each moment and the prediction error value corresponding to each pre-processed IGBT device aging history data point at each moment includes: The prediction error value is calculated for the initial state equation, each pre-processed IGBT device aging history data point at each moment, the particle priori data point corresponding to each pre-processed IGBT device aging history data point at each moment, and the Sigma point set corresponding to each pre-processed IGBT device aging history data point at each moment through the tenth formula, to obtain the prediction error value corresponding to each pre-processed IGBT device aging history data point at each moment. The tenth formula is: in, in, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, f() is the initial state equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k-th time, is the initial error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, is the aging historical data point of the i-th pre-processed IGBT device at the k-1th moment, is the particle priori data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment; The state vector prediction error variance matrix is ​​calculated by respectively calculating the system noise covariance matrix at each moment and the prediction error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the eleventh formula, so as to obtain the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The eleventh formula is: in, is the state vector prediction error variance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th time, Q k is the system noise covariance matrix at the kth moment.

6. The IGBT life prediction method according to claim 4, characterized in that: The process of S327 includes: The updated error value is calculated by respectively using the initial observation equation, the Sigma point set corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the predicted error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the twelfth formula, so as to obtain the updated error value corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The twelfth formula is: in, is the updated error value corresponding to the ith pre-processed IGBT device aging historical data point at the k+1th moment, h() is the initial observation equation function, is the Sigma point set corresponding to the aging historical data point of the i-th pre-processed IGBT device at the k+1th time, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment; The updated error values ​​corresponding to the aging historical data points of the pre-processed IGBT devices at each moment are respectively calculated by the thirteenth formula to obtain the observation vector prediction error variance matrix corresponding to the aging historical data points of the pre-processed IGBT devices at each moment. The thirteenth formula is: in, is the prediction error variance matrix of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the updated error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time; The Kalman gain matrix corresponding to each of the preprocessed IGBT device aging historical data points at each moment is calculated by the fourteenth formula, respectively, for the observation vector prediction error variance matrix corresponding to each of the preprocessed IGBT device aging historical data points at each moment, the update error value corresponding to each of the preprocessed IGBT device aging historical data points at each moment, and the prediction error value corresponding to each of the preprocessed IGBT device aging historical data points at each moment, to obtain the Kalman gain matrix corresponding to each of the preprocessed IGBT device aging historical data points at each moment. The fourteenth formula is: in, in, is the Kalman gain matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the cross covariance matrix corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time, is the prediction error variance matrix of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the prediction error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time, is the updated error value corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time; The state vector estimation value is calculated by respectively calculating the initial observation equation, the state vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the Kalman gain matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the observation vector prediction value corresponding to each of the pre-processed IGBT device aging historical data points at each moment through the fifteenth formula, so as to obtain the state vector estimation value corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The fifteenth formula is: in, is the estimated value of the state vector corresponding to the i-th pre-processed IGBT device aging historical data point at the k+1th time, h() is the initial observation equation function, is the predicted value of the state vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the Kalman gain matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the predicted value of the observation vector corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment; The Kalman gain matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, the observation vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment, and the state vector prediction error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment are calculated by the sixteenth formula to obtain the estimated error variance matrix corresponding to each of the pre-processed IGBT device aging historical data points at each moment. The sixteenth formula is: in, is the estimated error variance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the state vector prediction error variance matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, is the Kalman gain matrix corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time, It is the prediction error variance matrix of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k+1th time.

7. The IGBT life prediction method according to claim 4, characterized in that: The process of S329 includes: The updated IGBT device aging history data set, the updated IGBT device aging history data points corresponding to the pre-processed IGBT device aging history data points at each moment, and the observation vector prediction values ​​corresponding to the pre-processed IGBT device aging history data points at each moment are respectively calculated by the seventeenth formula to obtain the normalized weights corresponding to the pre-processed IGBT device aging history data points at each moment. The seventeenth formula is: in, in, is the normalized weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, p(|) is the prior probability, is the predicted value of the observation vector corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th time, is the updated IGBT device aging historical data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-th moment, is the initial weight corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, is the initial weight corresponding to the jth pre-processed IGBT device aging historical data point at the k-1th moment, M is the total number of pre-processed IGBT device aging historical data points, α is the improvement factor, is the predicted value of the observation vector corresponding to the i-th preprocessed IGBT device aging historical data point at the k-1th moment, is the updated IGBT device aging historical data point corresponding to the i-th pre-processed IGBT device aging historical data point at the k-1th moment, and A is the updated IGBT device aging historical data set; Normalization processing is performed on each of the weights to be normalized, respectively, to obtain the weight to be processed corresponding to each of the pre-processed IGBT device aging historical data points at each moment.

8. An IGBT life prediction device, characterized in that: include: Import module, used to import multiple original IGBT device aging history data points at multiple times; A preprocessing module, used for preprocessing all the original IGBT device aging history data points to obtain a plurality of preprocessed IGBT device aging history data points; A construction module, used to construct an IGBT degradation model through all the pre-processed IGBT device aging history data points; The import module is also used to import the IGBT device data points and warning thresholds to be predicted; The prediction result acquisition module is used to perform prediction analysis on the IGBT device data points to be predicted through the IGBT degradation model, the warning threshold and all the pre-processed IGBT device aging history data points to obtain the IGBT life prediction result.

9. An IGBT life prediction method system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the IGBT lifetime prediction method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the IGBT lifetime prediction method according to any one of claims 1 to 7 is implemented.

Images