CN114611411A - IGBT junction temperature prediction method based on ISFO-SVM model - Google Patents

IGBT junction temperature prediction method based on ISFO-SVM model Download PDF

Info

Publication number
CN114611411A
CN114611411A CN202210328272.9A CN202210328272A CN114611411A CN 114611411 A CN114611411 A CN 114611411A CN 202210328272 A CN202210328272 A CN 202210328272A CN 114611411 A CN114611411 A CN 114611411A
Authority
CN
China
Prior art keywords
igbt
junction temperature
optimal
formula
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210328272.9A
Other languages
Chinese (zh)
Inventor
李玲玲
武定山
刘佳琪
杨海跃
刘伯颖
李忠涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hebei University of Technology
Original Assignee
Hebei University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hebei University of Technology filed Critical Hebei University of Technology
Priority to CN202210328272.9A priority Critical patent/CN114611411A/en
Publication of CN114611411A publication Critical patent/CN114611411A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a method for predicting IGBT junction temperature. The method comprises the following steps: simulating an IGBT aging process through an IGBT aging acceleration test, acquiring data such as IGBT junction temperature, saturation voltage drop, collector current and aging times, and carrying out normalization processing on the data; setting parameters of an improved flag fish algorithm and a support vector machine model; running an improved flag fish algorithm to obtain an optimal penalty factor and optimal parameters of a kernel function in a support vector machine model; bringing the optimized optimal parameters into a support vector machine model, and training a support vector machine (ISFO-SVM) model optimized by an improved flag fish algorithm; and inputting the prediction data into an ISFO-SVM model to obtain a prediction result, and performing inverse normalization on the prediction result. The result shows that under the performance indexes of RMSE, MAPE and R2, the ISFO-SVM model has better prediction performance and higher fitting degree of the junction temperature predicted value and the actual junction temperature, the defect of low prediction precision of the existing IGBT junction temperature prediction method is made up, and the IGBT junction temperature is effectively predicted.

Description

IGBT junction temperature prediction method based on ISFO-SVM model
Technical Field
The technical scheme of the invention belongs to the technical field of IGBT reliability, and particularly relates to an IGBT junction temperature prediction method based on an ISFO-SVM model.
Background
The IGBT is a key part for power conversion in a new energy system, and the reliability of the IGBT greatly restricts the safe and reliable operation of a working system. Due to different thermal expansion coefficients of materials of all layers of the IGBT module, the internal connection part of the IGBT module bears different thermal stress due to temperature cycle fluctuation, so that the IGBT module fails. Especially in a high-frequency and high-power working environment, the IGBT can generate larger switching loss, and larger junction temperature fluctuation of the IGBT module is caused. The higher the working environment temperature is, the higher the failure probability of the IGBT module is. When the power generated by the IGBT chip cannot be dissipated in time, the junction temperature is continuously increased until the device fails. Because the aging degree of the IGBT module has certain influence on the conversion efficiency of the power converter and the safety of the system, the accurate prediction of the IGBT junction temperature has important significance on the evaluation of the reliability of the IGBT junction temperature and the accurate judgment of the safety of the system.
The method for acquiring the junction temperature of the IGBT comprises the following two steps: a measurement method and a calculation model method of relevant experimental equipment; the laboratory measurement method mainly obtains the IGBT junction temperature through a thermal sensor method, an optical fiber detection method and a temperature-sensitive parameter method, but the method breaks the packaging structure of the IGBT and is not suitable for the IGBT junction temperature prediction under the working condition; the model calculation method calculates the junction temperature of the IGBT by establishing a junction temperature calculation model based on electrothermal coupling or a junction temperature prediction model based on an intelligent algorithm, and predicts the junction temperature of the IGBT by establishing a machine learning model, so as to evaluate the reliability of the IGBT.
In the prior art, a simple machine learning model is difficult to realize high-precision prediction, so that an intelligent evolution algorithm is usually adopted to optimize the machine learning model and construct a combined model to realize prediction. However, the algorithms have the problems of poor optimizing capability, premature convergence and the like, and the precision, accuracy and the like of the final junction temperature prediction result can still be improved. Therefore, the method for predicting the IGBT junction temperature with high prediction accuracy and small prediction error has important significance for the replacement and maintenance of the IGBT and the safety and reliability of the system where the IGBT is located.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: providing a method for predicting the junction temperature of the IGBT; the method is based on an improved flag fish algorithm optimization support vector machine model (ISFO-SVM) to predict IGBT junction temperature, self-adaptive nonlinear iteration factors are introduced to improve the optimizing capability of flag fish individuals, a Levy flight strategy is utilized to improve the diversity of a search space, and a DE/current to best/1 strategy in a differential variation strategy is introduced to increase the diversity of a population in order to avoid the algorithm from falling into local convergence in the search process.
The technical scheme adopted by the invention for solving the technical problems is as follows: the method for predicting the junction temperature of the IGBT is provided, and particularly relates to a method for predicting the junction temperature of the IGBT based on an improved flag fish algorithm optimization support vector machine model (ISFO-SVM), which comprises the following steps:
step 1, obtaining data such as IGBT junction temperature, saturation voltage drop, collector current, aging times and the like through an IGBT power cycle aging acceleration test, and carrying out normalization processing on the data;
step 2, setting parameters of an improved flag fish algorithm and a support vector machine model;
step 3, operating an improved sailfish algorithm to obtain an optimal penalty factor and optimal parameters of a kernel function in the support vector machine model;
step 4, bringing the optimized optimal punishment factor and the optimized kernel function parameter into a support vector machine model, and training an improved flag fish algorithm optimized support vector machine (ISFO-SVM) model;
step 5, inputting the prediction data into an ISFO-SVM model to obtain a prediction result, and performing reverse normalization on the prediction result;
and 6, displaying and outputting the IGBT junction temperature prediction result.
Further, the specific implementation method of step 1 includes the following steps:
step 1.1, in order to obtain the degradation data of the IGBT module in the full life cycle, the invention designs an IGBT power cycle aging test and a single pulse test, thereby obtaining a data set S containing the saturation voltage drop, collector current, junction temperature and aging cycle times of the IGBT;
step 1.2, carrying out random confusion and normalization processing on the acquired data set S, dividing the data set S into training data and testing data respectively according to proportion, and taking the power cycle times N, the saturation voltage drop Vce and the collector current Ic as input parts of a model; the junction temperature Tj is used as the output part of the model;
step 1.3, carrying out normalization processing on the data;
Figure BSA0000269956710000021
in the formula, A is the value of the variable to be normalized, such as the power cycle number N, the saturation voltage drop Vce, the collector current Ic and the junction temperature Tj，AminIs the minimum value of a variable, AmaxIs the maximum value of the variable, ANNormalizing the value of the variable;
further, the parameters set in step 2 include: improving the population quantity, the maximum iteration times and the population dimension in the sailfish algorithm, and supporting the search range of a penalty factor C in a vector machine model and the search range of a kernel function parameter g;
further, the specific implementation method of step 3 includes the following steps:
step 3.1, initializing individual positions of the flag fish population and the sardine population of the improved flag fish algorithm, and calculating a target function value of each individual;
step 3.2, sequencing the position and the objective function value of each individual, and recording the current optimal individual position and the optimal objective function value;
3.3, introducing a self-adaptive nonlinear iteration factor to update the position of the flag fish individual, and introducing a Levy flight strategy to update the position of the sardine individual;
step 3.4, introducing a differential variation strategy, continuously searching and updating the positions of individuals in the flag fish population and the sardine population, and judging whether optimal convergence is achieved; if the optimal punishment factor C meets the optimal punishment factor C of the SVM model, obtaining the optimal parameter of the model, and outputting the optimal individual position X (X1, X2) (which respectively corresponds to the optimal punishment factor C of the SVM model and the optimal parameter g of the kernel function); if not, returning and continuing to execute the step 3.2;
further, in the step 3, a root mean square error is selected as an objective function, and training data are adopted to optimize internal parameters of the support vector machine model;
Figure BSA0000269956710000022
the improved swordfish algorithm judges whether the updated position is better than the original position according to the objective function value of the individual position at the moment, and determines whether the updated position is used in the following searching process, wherein the objective function is described as follows:
Figure BSA0000269956710000031
in the formula (I), the compound is shown in the specification,
Figure BSA0000269956710000032
is [0, 1 ]]The random number of (2);
the specific implementation method of the step 3.3 is as follows:
step 3.3.1, introducing a self-adaptive nonlinear iteration factor to update the position of the sailfish individual;
because flag fish individuals are randomly distributed at the initial iteration stage, in order to improve the optimizing capability of the flag fish individuals, the self-adaptive nonlinear iteration factor is introduced into the flag fish position updating formula, and the optimizing capability of the flag fish individuals is accelerated; wherein the flag fish population uses XSFRepresents;
the updating formula of the adaptive nonlinear iteration factor of the ith flag fish individual in the t iteration is described as follows:
Figure BSA0000269956710000033
the flag fish position updating formula is as follows:
Figure BSA0000269956710000034
in the formula (I), the compound is shown in the specification,
Figure BSA0000269956710000035
representing the optimal individual position in the sailfish population during the t-th iteration;
Figure BSA0000269956710000036
representing the optimal individual position in the sardine population at the time of the tth iteration;
Figure BSA0000269956710000037
representing the position of the smelt individual to be updated during the t-th iteration; lambda [ alpha ]iThe coefficients are defined in equation (6):
λi＝2×rand(0，1)×PD-PD (6)
wherein PD represents the density of the prey population, as detailed in equation (7):
Figure BSA0000269956710000038
in the formula, NSFRepresenting the number of flag fishes, NSRepresenting the number of sardines;
step 3.3.2, introducing a Levy flight strategy to update the position of the sardine individual; wherein the sardine population uses XFRepresents;
the updating formula of the position of the sardine in the original sailfish algorithm is shown as the formula (8):
Figure BSA0000269956710000039
in the formula (I), the compound is shown in the specification,
Figure BSA00002699567100000310
represents the self-adaptive nonlinear iteration factor of the ith flag fish individual at the t-th iteration,
Figure BSA00002699567100000311
is shown asthe optimal individual position in the flag fish population during t iterations,
Figure BSA00002699567100000312
representing the position of the sardine individual to be updated during the t-th iteration; AP represents the attack strength of the flag fish, and the detailed expression is shown in formula (9):
AP＝A×(1-2×Itr×e) (9)
in the formula, A and e represent control coefficients of the flag fish attack force, so that the flag fish attack force is linearly converted from A to 0;
when the AP is more than 0.5, namely the attack strength of the flag fish is strong, updating the positions of all sardines by using a formula (8); when the AP is less than 0.5, the attack strength of the flag fish is low, and only part of the positions of the sardines need to be updated;
the range of the partial sardine positions is defined as follows:
α＝NS×AP (10)
β＝di×AP (11)
wherein alpha represents the number of renewed sardines, beta represents the number of renewed dimensions of the sardines, and diThe number of variables at the ith iteration;
in order to improve the randomness of sardine population and the diversity of search space, the invention introduces a Levy flight strategy, and the position updating formula of the sardine at the moment is as follows:
Figure BSA0000269956710000041
where t is the current iteration number, d is the dimension of the position vector,
Figure BSA0000269956710000042
representing the position of the sardine individual to be updated during the t-th iteration;
the equation for Le' vy flight can be described as:
Figure BSA0000269956710000043
in the formula, r1 r2Is two random numbers with the value range of [0, 1%]β ═ 1.5, σ can be calculated as:
Figure BSA0000269956710000044
wherein Γ (x) ═ x-1!
Therefore, the attack strength AP of the flag fish is calculated according to the formula (9), and when the AP is larger than 0.5, namely the attack strength of the flag fish is stronger, the positions of all the sardines are updated by the improved sardine position updating formula (12); when the AP is less than 0.5 and the attack strength of the flag fish is low, calculating the number and the dimensionality of the sardines needing position updating according to the formulas (10) and (11), and updating the sardines by using an improved sardine position updating formula (12);
the specific implementation method of the step 3.4 is as follows:
in order to avoid trapping partial convergence in the searching process, the invention introduces a DE/current to best/1 strategy in a differential variation strategy to perform variation on vectors of population individuals, and adds the differential variation strategy in the later stage of each round of searching to increase the diversity of the population; the formula is expressed as follows:
Figure BSA0000269956710000046
in the formula, p1≠p2≠p3，
Figure BSA0000269956710000047
For the difference vector, F ∈ [0.1, 0.9 ]]As a scaling factor, hi，tObtaining a variation vector for the ith position in the t-th search, and then performing a crossover operation by:
Figure BSA0000269956710000045
in the formula, vi，tFor the cross-variables of the ith search position, j0 is a random value in the dimension, each cross-operation only involves one dimension of the individual, pCR ∈ [0, 1]Is the cross probability;
and carrying out selection operation, reserving the optimal vector of the objective function value as a next generation individual, and expressing the selection operation as follows:
Figure BSA0000269956710000051
continuously searching and updating the positions of the population individuals according to the formula, and judging whether optimal convergence is achieved; if the optimal punishment factor C meets the optimal punishment factor C of the SVM model, obtaining the optimal parameter of the model, and outputting the optimal individual position X (X1, X2) (which respectively corresponds to the optimal punishment factor C of the SVM model and the optimal parameter g of the kernel function); if not, returning and continuing to execute the step 3.2;
further, the specific implementation method of step 4 is as follows: inputting the optimal punishment factor C obtained in the step 3 and the optimal parameter g of the kernel function into a support vector machine model to form an improved flag fish algorithm optimized support vector machine (ISFO-SVM) model, and training the support vector machine model by using the optimized optimal punishment factor C and the optimal parameter g of the kernel function.
Further, the step 5 performs an inverse normalization process as shown in formula (18):
T′i＝T′soale，i×(Tmax-Tmin)+Tmin (18)
in the formula (II), T'iIs a predicted value of junction temperature after reverse normalization, T'soale，iFor the normalized junction temperature prediction value, T, obtained in step 4max、TminThe maximum and minimum values of the junction temperature variation in step 1.3.
Further, the specific implementation method of step 6 is as follows: and (4) outputting the prediction graph of the ISFO-SVM model obtained in the step (5) on the IGBT junction temperature on a display screen of a computer, and displaying an error curve graph and an error histogram of the IGBT junction temperature prediction by using different models.
The method of inputting data into a computer in the above-described steps is a known method; the computers, displays and MATLAB computer software used were all commercially available.
Compared with the prior art, the invention has the beneficial effects that:
(1) an improved flag fish algorithm comprising a self-adaptive nonlinear iteration factor, a Levy flight strategy and a differential variation strategy is provided; the improved sailfish algorithm not only keeps the solving stability and robustness of the original algorithm, but also obtains the optimal solution with good convergence and optimizing precision;
(2) in order to enable the junction temperature variation to meet the set requirements of the test and accelerate the aging of the IGBT module, the invention designs an IGBT aging acceleration experiment and a single-pulse test platform according to the IEC60068-2-14JEDEC standard of the International Electrotechnical Commission (IEC) on power cycle test, and takes the increase of the saturation voltage drop by 5 percent as the reference quantity of the IGBT failure; obtaining data of saturation voltage drop, collector current, junction temperature and aging times of the IGBT through tests, and constructing an IGBT test data set;
(3) an ISFO optimization SVM model is provided, and an IGBT junction temperature prediction model based on an ISFO-SVM is constructed; compared with an SFO-SVM model and an SVM model, the IGBT junction temperature prediction model based on the ISFO-SVM has better performance of predicting junction temperature and higher fitting degree; in conclusion, the ISFO-SVM model constructed by the invention has a better prediction result on the IGBT junction temperature.
Drawings
The invention is further illustrated with reference to the following figures and examples.
Fig. 1 is a schematic diagram of the junction temperature prediction step of the ISFO-SVM model provided by the present invention;
FIG. 2 is a circuit diagram of an IGBT power cycle aging test provided by the embodiment of the invention;
FIG. 3 is a circuit diagram of a single pulse test provided by an embodiment of the present invention;
fig. 4 is a diagram of predicting the junction temperature of the IGBT by the ISFO-SVM model provided in the embodiment of the present invention;
FIG. 5 is a graph of error curves of different models for IGBT junction temperature prediction provided by embodiments of the present invention;
FIG. 6 is an error histogram of different models for IGBT junction temperature prediction provided by an embodiment of the present invention;
Detailed Description
Fig. 1 shows that the general steps of the IGBT junction temperature prediction method provided by the present invention are, start → IGBT power cycle aging test and single pulse test → obtain the saturation voltage drop, junction temperature, collector current data of the IGBT under the corresponding power cycle aging times → set the IGBT saturation voltage drop, collector current and aging time as input data sets, junction temperature as output data set → normalize the obtained data sets and other preprocessing → initialize the swordfish population, the sardine population and algorithm parameters → choose the root mean square error as objective function, calculate the fitness values of the swordfish and the sardine, sequence the values of the individual objective functions and record the optimal fitness value and position → introduce adaptive nonlinear iteration factor to update the position of the individual swordfish → introduce Levy flight strategy to update the position of the swordfish → introduce DE/current best/1 strategy in the differential variation strategy to increase the diversity of the population → judge whether to reach the optimal convergence, if the optimal parameters are met, the optimal parameters of the model are obtained, the optimal flag fish individual position X (X1, X2) (respectively corresponding to the optimal penalty factor C of the SVM model and the optimal parameter g of the kernel function) is output → an ISFO-SVM-based IGBT junction temperature prediction model is constructed → the test data is predicted, and the output IGBT junction temperature prediction result is displayed → the end is finished.
FIG. 2 shows the circuit principle of the IGBT power cycle aging test provided by the invention, and the research test object is 2 IGBT modules; when the switch is closed, the lower tube is connected with back pressure and is continuously turned off; an aging acceleration test is carried out on the upper tube, a grid signal drives the upper tube IGBT to be conducted, and at the moment, the program-controlled direct-current power supply outputs 75A current to enable the junction temperature and the shell temperature of the IGBT to rise rapidly, so that the purpose of accelerating the aging of the IGBT is achieved; the basic test steps of the IGBT power cycle aging test are as follows:
(1) firstly, a switch S is closed, the output current of a program-controlled constant current source is set to be 50A, a gate pole outputs a driving voltage to be 15V, so that an upper tube of an IGBT power module is conducted, a test sample module generates power loss to cause junction temperature and shell temperature to rise, and an initial shell temperature value is set to be 40 ℃;
(2) monitoring the change condition of temperature through a temperature sensor arranged in the bottom of the module, disconnecting a switch S when the highest shell temperature is 90 ℃, starting an air-cooled radiator to work so that the IGBT power module is quickly cooled to 40 ℃ when the shell temperature is reached, and finishing one-time shell temperature fluctuation power cycle aging;
(3) repeating the steps (1) and (2) until the module approaches the failure standard, stopping the test, finishing 6000 power cycle tests in the test, pausing once every 1000 power cycle tests, taking down the module, and putting the module into a thermostat to perform a short-time single-pulse test; and recording collector current and saturation voltage drop values of the IGBT module at different junction temperatures.
FIG. 3 shows the principle of the single pulse test circuit provided by the invention, wherein a single pulse trigger current is introduced to the IGBT through a DSP development board, an amplifying circuit and a driver to an IGBT power module by a single pulse driving signal; the test is set as follows: the temperature regulation range of the constant temperature box is [0 ℃, 100 ℃), the temperature regulation interval is 10 ℃, the setting range of the collector current is [25A, 70A ], and the regulation interval is 5A; the single pulse test procedure was as follows:
(1) placing the IGBT power module to be tested into a thermostat, adjusting the temperature of the thermostat, and considering that the module reaches thermal balance at the moment after the temperature of the thermostat is stable;
(2) and when the thermal balance condition is met, adjusting the set values of the temperature of the thermostat and the collector current, and sequentially recording the data of the saturation voltage drop, the junction temperature and the collector current under the corresponding power cycle aging times.
In order to illustrate the technical solution of the present invention, the following is described by specific embodiments;
step 1, obtaining data such as IGBT junction temperature, saturation voltage drop, collector current, aging times and the like through an IGBT power cycle aging acceleration test, and carrying out normalization processing on the data;
in order to facilitate the construction of a prediction model without loss of generality, 386 groups of data are randomly selected from the data of the IGBT aging test to establish a data set; the first 70% of the data set is used as a training sample, the data is used for training a prediction model after random confusion and normalization treatment, and the remaining 30% is used as a test sample for testingTesting and verifying the effectiveness of the IGBT junction temperature prediction model; the invention is provided with: the power cycle number N, the saturation voltage drop Vce and the collector current Ic are used as input parts of the model, and the junction temperature TjAs an output part of the model; normalizing the data according to a formula (1);
step 2, setting parameters of an improved flag fish algorithm and a support vector machine model;
setting the number of populations, the maximum iteration times and the population dimension in an improved sailfish algorithm, the search range of a penalty factor C in a support vector machine model and the search range of a kernel function parameter g;
in this embodiment, the number of populations in the improved flagfish algorithm is recorded as 30, the maximum iteration number is 500, the population dimension is 2, the search range of the penalty factor C in the support vector machine model is [0.1, 1200], the range of the kernel function parameter g of the support vector machine model is [0.01, 100], and the rest parameters are default values;
step 3, operating an improved sailfish algorithm to obtain an optimal penalty factor and optimal parameters of a kernel function in the support vector machine model;
step 3.1, initializing individual positions of the flag fish population and the sardine population of the improved flag fish algorithm, and calculating a target function value of each individual;
selecting a root mean square error as a target function according to the formula (2), and optimizing internal parameters of the support vector machine model by adopting training data;
step 3.2, sequencing the position and the objective function value of each individual, and recording the current optimal individual position and the optimal objective function value;
3.3, introducing a self-adaptive nonlinear iteration factor to update the position of the flag fish individual, and introducing a Levy flight strategy to update the position of the sardine individual;
because flag fish individuals are randomly distributed at the initial iteration stage, in order to improve the optimizing capability of the flag fish individuals, a self-adaptive nonlinear iteration factor is introduced into a flag fish position updating formula, and the optimizing capability of the flag fish individuals is accelerated; wherein, the position updating formula of the flag fish is shown in formula (5);
in order to improve the randomness of sardine population and the diversity of search space, the invention introduces a Levy flight strategy, and the position updating formula of the sardine is shown in formula (12); firstly, calculating attack strength AP of the flag fish according to a formula (9), and updating the positions of all the sardines by using an improved sardine position updating formula (12) when the AP is more than 0.5, namely the attack strength of the flag fish is stronger; when the AP is less than 0.5 and the attack strength of the flag fish is low, calculating the number and the dimensionality of the sardines needing position updating according to the formulas (10) and (11), and updating the sardines by using an improved sardine position updating formula (12);
3.4, in order to avoid trapping partial convergence in the searching process, the invention introduces a DE/current to best/1 strategy in a differential variation strategy to perform variation on the vectors of population individuals, and adds the differential variation strategy in the later stage of each round of searching to increase the diversity of the population; specifically describing the formula (15); after obtaining the variation vector, performing cross operation according to a formula (16); then, selecting operation is carried out according to a formula (17), the optimal vector of the objective function value is reserved as a next generation individual, the positions of population individuals are continuously searched and updated, and whether optimal convergence is achieved is judged; if so, acquiring an optimal penalty factor C and an optimal parameter g of the kernel function in the support vector machine model; if not, returning and continuing to execute the step 3.2;
and 4, inputting the optimal punishment factor C obtained in the step 3 and the optimal parameter g of the kernel function into a support vector machine model to form an improved sailfish algorithm optimized support vector machine (ISFO-SVM) model, and training the support vector machine model by using the optimized optimal punishment factor C and the optimal parameter g of the kernel function.
Step 5, inputting the prediction data into an ISFO-SVM model to obtain a prediction result, and performing reverse normalization on the prediction result;
step 6, displaying and outputting the IGBT junction temperature prediction result;
and (4) outputting the prediction graph of the ISFO-SVM model obtained in the step (5) on the IGBT junction temperature on a display screen of a computer, and displaying an error curve graph and an error histogram of the IGBT junction temperature prediction by using different models.
In addition, in the embodiment of the invention, in order to better show the good performance of the proposed prediction model, an SVM model, an SFO-SVM model and an ISFO-SVM model are selected for comparison, and RMSE, MAPE and R2 are selected as evaluation indexes of the models; the evaluation index values of IGBT junction temperature prediction under different models are shown in table 1;
Figure BSA0000269956710000081
Figure BSA0000269956710000082
Figure BSA0000269956710000083
wherein N is the number of test samples; t'iIs a predicted value; t isiIs the actual value.
TABLE 1 analysis table of junction temperature prediction and evaluation indexes
Figure BSA0000269956710000084
Combining table 1 and fig. 5, it was found that RMSE, MAPE and R2 of the ISFO-SVM model are superior to the other two models. The difference between the maximum value, the minimum value and the average value of the RMSE and MAPE indexes of the SVM model is relatively small, which shows that the robustness of the SVM model is good; RMSE, MAPE and R2 of the ISFO-SVM model are superior to other two models, good prediction performance is shown, and high prediction precision is achieved for predicting the IGBT junction temperature; compared with the SFO-SVM model, the average value of RMSE is reduced by 67.16%, the average value of MAPE is reduced by 56.52%, and R2 is increased by 0.71%, so that the ISFO-SVM model has better performance of predicting junction temperature and higher fitting degree.
As can be seen from fig. 6, the sample individual prediction errors in the SVM model account for 64.65% in the range of [ -3 ℃, 3 ℃), and account for 0.86% in the range of [9 ℃, 12 ℃), and the SFO-SVM model has slightly improved junction temperature prediction results compared with the SVM model, and is more suitable for the true value of junction temperature; by improving the SFO algorithm, the error of the constructed ISFO-SVM model for predicting the junction temperature is relatively small and concentrated, sample individuals mainly distributed in the range of-3 ℃ and 3 ℃ account for 91.38% of the total number of the samples, and only 8.82% of the absolute value of the error of the sample individuals is at the temperature of 3 ℃ and 6 ℃; in conclusion, the IGBT junction temperature prediction method based on the ISFO-SVM model provided by the invention effectively improves the prediction accuracy and robustness of the IGBT junction temperature prediction model.
In the above method for predicting the IGBT junction temperature based on the ISFO-SVM model, the method for inputting the data into the computer is a known method; the computers, displays and MATLAB computer software used were all commercially available.
The above description is only for the specific embodiments of the method of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can substitute or change the technical solution of the present invention and its inventive concept within the technical scope of the present invention.

Claims (3)

1. The IGBT junction temperature prediction method based on the ISFO-SVM model is characterized in that the IGBT junction temperature is predicted by constructing an optimized support vector machine model (ISFO-SVM) based on an improved flag fish algorithm, and the method comprises the following specific steps:
step 1, obtaining data such as IGBT junction temperature, saturation voltage drop, collector current, aging times and the like through an IGBT power cycle aging acceleration test, and carrying out normalization processing on the data;
step 2, setting parameters of an improved flag fish algorithm and a support vector machine model;
step 3, operating an improved sailfish algorithm to obtain an optimal penalty factor and optimal parameters of a kernel function in the support vector machine model;
step 4, bringing the optimized optimal punishment factor and the optimized kernel function parameter into a support vector machine model, and training an improved flag fish algorithm optimized support vector machine (ISFO-SVM) model;
step 5, inputting the prediction data into an ISFO-SVM model to obtain a prediction result, and performing reverse normalization on the prediction result;
step 6, displaying and outputting the IGBT junction temperature prediction result;
further, the specific implementation method of step 1 includes the following steps:
step 1.1, in order to obtain the degradation data of the IGBT module in the full life cycle, the invention designs an IGBT power cycle aging test and a single pulse test, thereby obtaining a data set S containing the saturation voltage drop, collector current, junction temperature and aging cycle times of the IGBT;
step 1.2, carrying out random confusion and normalization processing on the acquired data set S, dividing the data set S into training data and testing data respectively according to proportion, and taking the power cycle times N, the saturation voltage drop Vce and the collector current Ic as input parts of a model; junction temperature TjAs an output part of the model;
step 1.3, carrying out normalization processing on the data;
Figure FSA0000269956700000011
in the formula, A is the value of the variable to be normalized, such as the power cycle number N, the saturation voltage drop Vce, the collector current Ic and the junction temperature Tj，AminIs the minimum value of a variable, AmaxIs the maximum value of the variable, ANNormalizing the value of the variable;
further, the parameters set in step 2 include: improving the population quantity, the maximum iteration times and the population dimension in the sailfish algorithm, and supporting the search range of a penalty factor C in a vector machine model and the search range of a kernel function parameter g;
further, the specific implementation method of step 3 includes the following steps:
step 3.1, initializing individual positions of the flag fish population and the sardine population of the improved flag fish algorithm, and calculating a target function value of each individual;
step 3.2, sequencing the position and the objective function value of each individual, and recording the current optimal individual position and the optimal objective function value;
3.3, introducing a self-adaptive nonlinear iteration factor to update the position of the flag fish individual, and introducing a Levy flight strategy to update the position of the sardine individual;
step 3.4, introducing a differential variation strategy, continuously searching and updating the positions of individuals in the flag fish population and the sardine population, and judging whether optimal convergence is achieved; if the optimal punishment factor C meets the optimal punishment factor C of the SVM model, obtaining the optimal parameter of the model, and outputting the optimal individual position X (X1, X2) (which respectively corresponds to the optimal punishment factor C of the SVM model and the optimal parameter g of the kernel function); if not, returning and continuing to execute the step 3.2;
further, in the step 3, a root mean square error is selected as an objective function, and training data are adopted to optimize internal parameters of the support vector machine model;
Figure FSA0000269956700000021
the improved swordfish algorithm judges whether the updated position is better than the original position according to the objective function value of the individual position at the moment, and determines whether the updated position is used in the following searching process, wherein the objective function is described as follows:
Figure FSA0000269956700000022
in the formula (I), the compound is shown in the specification,
Figure FSA0000269956700000023
is [0, 1 ]]The random number of (2);
the specific implementation method of the step 3.3 is as follows:
step 3.3.1, introducing a self-adaptive nonlinear iteration factor to update the position of the flagfish individual;
because flag fish individuals are randomly distributed at the initial iteration stage, in order to improve the optimizing capability of the flag fish individuals, the method is used in the flag fish position updating formulaIntroducing a self-adaptive nonlinear iteration factor to accelerate the optimizing capability of the smelt individual; wherein the flag fish population uses XSFRepresenting;
the updating formula of the adaptive nonlinear iteration factor of the ith flag fish individual in the t iteration is described as follows:
Figure FSA0000269956700000024
the flag fish position updating formula is as follows:
Figure FSA0000269956700000025
in the formula (I), the compound is shown in the specification,
Figure FSA0000269956700000026
representing the optimal individual position in the sailfish population during the t-th iteration;
Figure FSA0000269956700000027
representing the optimal individual position in the sardine population at the time of the tth iteration;
Figure FSA0000269956700000028
representing the position of the smelt individual to be updated during the t-th iteration; lambda [ alpha ]iThe coefficients are defined in equation (6):
λi＝2×rand(0，1)×PD-PD (6)
wherein PD represents the density of the prey population, as detailed in equation (7):
Figure FSA0000269956700000029
in the formula, NSFRepresenting the number of flag fishes, NSRepresenting the number of sardines;
step 3.3.2, introducing Levy flight strategy to update sardine individualsThe position of (a); wherein the sardine population uses XFRepresents;
the updating formula of the position of the sardine in the original flagfish algorithm is shown as the formula (8):
Figure FSA00002699567000000210
in the formula (I), the compound is shown in the specification,
Figure FSA00002699567000000211
represents the self-adaptive nonlinear iteration factor of the ith flag fish individual at the t-th iteration,
Figure FSA00002699567000000212
represents the best individual position in the flag fish population at the time of the t-th iteration,
Figure FSA00002699567000000213
representing the position of the sardine individual to be updated during the t-th iteration; AP represents the attack strength of the flag fish, and the detailed expression is shown in formula (9):
AP＝A×(1-2×Itr×e) (9)
in the formula, A and e represent control coefficients of the flag fish attack force, so that the flag fish attack force is linearly converted from A to 0;
when the AP is more than 0.5, namely the attack strength of the flag fish is stronger, updating the positions of all the sardines by using a formula (8); when the AP is less than 0.5, the attack strength of the flag fish is low, and only part of the positions of the sardines need to be updated;
the range of partial sardine positions is defined as follows:
α＝NS×AP (10)
β＝di×AP (11)
wherein alpha represents the number of renewed sardines, beta represents the number of renewed dimensions of the sardines, and diThe number of variables at the ith iteration;
in order to improve the randomness of sardine population and the diversity of search space, the invention introduces a Levy flight strategy, and the position updating formula of the sardine at the moment is as follows:
Figure FSA0000269956700000031
where t is the current iteration number, d is the dimension of the position vector,
Figure FSA0000269956700000032
representing the position of the sardine individual to be updated during the t-th iteration;
the equation for Le' vy flight can be described as:
Figure FSA0000269956700000033
in the formula, r1 r2Is two random numbers with the value range of [0, 1%]β ═ 1.5, σ can be calculated as:
Figure FSA0000269956700000034
wherein Γ (x) ═ x-1!
Therefore, the attack strength AP of the flag fish is calculated according to the formula (9), and when the AP is larger than 0.5, namely the attack strength of the flag fish is stronger, the positions of all the sardines are updated by the improved sardine position updating formula (12); when the AP is less than 0.5 and the attack strength of the flag fish is low, calculating the number and the dimensionality of the sardines needing position updating according to the formulas (10) and (11), and updating the sardines by using an improved sardine position updating formula (12);
the specific implementation method of the step 3.4 is as follows:
in order to avoid trapping partial convergence in the searching process, the invention introduces a DE/current to best/1 strategy in a differential variation strategy to perform variation on vectors of population individuals, and adds the differential variation strategy in the later stage of each round of searching to increase the diversity of the population; the formula is expressed as follows:
Figure FSA0000269956700000035
in the formula, p1≠p2≠p3，
Figure FSA0000269956700000036
For the difference vector, F ∈ [0.1, 0.9 ]]As a scaling factor, hi，tObtaining a variation vector for the ith position in the t-th search, and then performing a crossover operation by:
Figure FSA0000269956700000041
in the formula, vi，tFor the cross-variant of the ith search position, j0 is a random value in the dimension, each cross operation only involves one dimension of the individual, pCR epsilon [0, 1]Is the cross probability;
and carrying out selection operation, reserving the optimal vector of the objective function value as a next generation individual, and expressing the selection operation as follows:
Figure FSA0000269956700000042
continuously searching and updating the positions of population individuals according to the formula, judging whether optimal convergence is achieved, if so, obtaining the optimal parameters of the model, and outputting the optimal smelt individual position X (X1, X2) (which respectively corresponds to the optimal punishment factor C of the SVM model and the optimal parameter g of the kernel function); if not, returning and continuing to execute the step 3.2;
further, the specific implementation method of step 4 is as follows: inputting the optimal punishment factor C obtained in the step (3) and the optimal parameter g of the kernel function into a support vector machine model to form an improved flag fish algorithm optimized support vector machine (ISFO-SVM) model, and training the support vector machine model by using the optimized optimal punishment factor C and the optimal parameter g of the kernel function;
further, the step 5 performs an inverse normalization process as shown in formula (18):
T′i＝T′scale，i×(Tmax-Tmin)+Tmin (18)
in formula (II) T'iIs a predicted value of junction temperature after reverse normalization, T'scale，iFor the normalized junction temperature prediction value, T, obtained in step 4max、TminMaximum and minimum values of the junction temperature variable in step 1.3;
further, the specific implementation method of step 6 is as follows: and (4) outputting the prediction graph of the ISFO-SVM model obtained in the step (5) on the IGBT junction temperature on a display screen of a computer, and displaying an error curve graph and an error histogram of the IGBT junction temperature prediction by using different models.
2. The method for predicting the IGBT junction temperature based on the ISFO-SVM model according to claim 1, wherein: the IGBT aging test data set is characterized in that data of saturation voltage drop, collector current, junction temperature and aging cycle times of the IGBT are obtained by designing an IGBT power cycle aging acceleration test and a single pulse test; in the test case, the model of the IGBT is MMG75S-120B, and the rated value is 1200V/75A; setting the junction temperature fluctuation delta Tj as 100 ℃, and taking the increase of the saturation voltage drop Vce by 5 percent as the judgment standard of the IGBT failure.
3. The method for predicting the IGBT junction temperature based on the ISFO-SVM model according to claim 2, wherein: the IGBT aging test data set is subjected to IGBT power cycle aging test and single pulse test;
the invention builds an IGBT power cycle aging test circuit, as shown in FIG. 2; referring to the IEC60068-2-14JEDEC standard, as defined by the International Electrotechnical Commission (IEC) on Power cycle tests, the basic test procedures for this test are as follows:
(1) firstly, a switch S is closed, a program-controlled constant current source is set to output current 50A, a gate pole outputs driving voltage 15V, so that an upper tube of an IGBT power module is conducted, a test sample module generates power loss to cause the temperature of a junction and the temperature of a shell to rise, and the initial value of the temperature of the shell is set to be 40 ℃;
(2) monitoring the change condition of temperature through a temperature sensor arranged in the bottom of the module, disconnecting a switch S when the highest shell temperature is 90 ℃, starting an air-cooled radiator to work so that the IGBT power module is quickly cooled to 40 ℃ when the shell temperature is reached, and finishing one-time shell temperature fluctuation power cycle aging;
(3) repeating the steps (1) and (2) until the module approaches the failure standard, stopping the test, finishing 6000 power cycle tests in the test, pausing once every 1000 power cycle tests, taking down the module, and putting the module into a thermostat to perform a short-time single-pulse test; recording collector current and saturation voltage drop values of the IGBT module at different junction temperatures;
considering that after one stage of accelerated aging is completed every 1000 times, the saturation voltage drop, the junction temperature and the collector current of the IGBT power module in the current aging stage need to be obtained, a single-pulse test platform is set up as shown in fig. 3, the temperature regulation range of a thermostat is set to be 0 ℃, 100 ℃, and the temperature regulation interval is set to be 10 ℃; the set range of the collector current is [25A, 70A ], and the adjusting interval is 5A; the method comprises the following specific steps:
(1) placing the IGBT power module to be tested into a thermostat, adjusting the temperature of the thermostat, and considering that the module reaches thermal balance at the moment after the temperature of the thermostat is stable;
(2) and when the thermal balance condition is met, adjusting the set values of the temperature of the constant temperature box and the current of the collector, and sequentially recording the data of the saturation voltage drop, the junction temperature and the current of the collector under the corresponding power cycle aging times.
CN202210328272.9A 2022-03-31 2022-03-31 IGBT junction temperature prediction method based on ISFO-SVM model Pending CN114611411A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210328272.9A CN114611411A (en) 2022-03-31 2022-03-31 IGBT junction temperature prediction method based on ISFO-SVM model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210328272.9A CN114611411A (en) 2022-03-31 2022-03-31 IGBT junction temperature prediction method based on ISFO-SVM model

Publications (1)

Publication Number Publication Date
CN114611411A true CN114611411A (en) 2022-06-10

Family

ID=81866097

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210328272.9A Pending CN114611411A (en) 2022-03-31 2022-03-31 IGBT junction temperature prediction method based on ISFO-SVM model

Country Status (1)

Country Link
CN (1) CN114611411A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024031938A1 (en) * 2022-08-11 2024-02-15 贵州电网有限责任公司 Method for inverting concentration of sf6 decomposition component co2 on basis of isfo-vmd-kelm
WO2024036950A1 (en) * 2022-08-15 2024-02-22 国网河北省电力有限公司电力科学研究院 Configuration optimization method for microgrid distributed photovoltaic power generation system and terminal device

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024031938A1 (en) * 2022-08-11 2024-02-15 贵州电网有限责任公司 Method for inverting concentration of sf6 decomposition component co2 on basis of isfo-vmd-kelm
WO2024036950A1 (en) * 2022-08-15 2024-02-22 国网河北省电力有限公司电力科学研究院 Configuration optimization method for microgrid distributed photovoltaic power generation system and terminal device

Similar Documents

Publication Publication Date Title
CN114611411A (en) IGBT junction temperature prediction method based on ISFO-SVM model
CN109101738B (en) IGBT module aging degree evaluation method
CN110398697B (en) Lithium ion health state estimation method based on charging process
Ciappa et al. Lifetime prediction and design of reliability tests for high-power devices in automotive applications
JP7299274B2 (en) Battery model construction method and battery deterioration prediction device
CN112001127B (en) IGBT junction temperature prediction method
CN109284543B (en) IGBT residual life prediction method based on optimal scale Gaussian process model
CN111832221A (en) Lithium battery life prediction method based on feature screening
CN111829688A (en) Temperature field distribution prediction method in working process of lithium ion battery
CN112214862B (en) Battery parameter calibration method, system and equipment based on genetic algorithm
CN112288147A (en) Method for predicting insulation state of generator stator by BP-Adaboost strong predictor
CN116861957A (en) Operator automatic tuning method and related device based on reinforcement learning
CN117406100A (en) Lithium ion battery remaining life prediction method and system
CN112329370B (en) Online fault diagnosis method for SiC power tube based on extreme learning machine
CN113536662A (en) Electronic transformer error state prediction method based on firefly optimized LightGBM algorithm
CN117250490A (en) LSTM-based residual current operated circuit breaker service life prediction method
CN114995545B (en) Control method, device, equipment and medium of vehicle thermal management system
CN116187071A (en) Method for rapidly predicting service life of battery cell
CN113191075B (en) Photovoltaic array fault diagnosis method based on improved goblet sea squirt group algorithm
CN108760813A (en) A kind of turbine blade of gas turbine health monitoring systems and method based on temperature signal
CN115270983A (en) Switch cabinet fault prediction method based on AdaBoost-RBF algorithm
CN113312862A (en) LFSR-based random circuit hardware overhead minimization design method
CN111695291A (en) Method and system for determining breakdown voltage of long air gap
CN117312806B (en) Electric energy meter metering error prediction method and system in complex environment
EP4242676A1 (en) Hybrid parametric - machine learning battery aging model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination