A method and system for predicting electronic component performance based on artificial intelligence

Abstract

The present invention relates to the technical field of electrical digital data processing, and discloses an electronic component performance prediction method and system based on artificial intelligence. The present invention first obtains the junction voltage drop and base voltage drop of the electronic component, calculates the contact voltage drop using a contact voltage drop detection method, calculates the forward voltage drop of the electronic component, and records it as the first electronic component performance influencing factor; then establishes a gray model, and then performs residual correction on the gray model to obtain a corrected gray model function, uses a BP neural network to predict the resistance of the electronic component, substitutes the predicted resistance into the corrected gray model function, obtains the life of the electronic component, and records it as the second electronic component performance influencing factor; then, the first electronic component performance influencing factor and the second electronic component performance influencing factor are processed by an interval algorithm and normalization to obtain a final set of electronic component performance influencing factors; finally, an Elman neural network is used to realize the prediction of electronic component performance.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to an electronic component performance prediction method and system based on artificial intelligence. Background Art

[0002] During long-term storage, electronic components will be subject to external influences such as temperature, humidity, and chemical factors, and their internal materials will undergo physical or chemical changes such as corrosion and aging. The linear changes in the performance parameters of electronic components during this process are very complex. Traditional methods cannot accurately predict the performance of electronic components, and traditional methods are time-consuming and labor-intensive. Artificial intelligence has the ability to learn autonomously and can process large amounts of data, providing a solution to complex linear changes. Summary of the invention

[0003] In response to the problems in the related art, the present invention provides an electronic component performance prediction method and system based on artificial intelligence to overcome the above-mentioned technical problems existing in the existing related art.

[0004] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0005] The present invention is an electronic component performance prediction method based on artificial intelligence, comprising the following steps:

[0006] S1. Obtaining the junction voltage drop and base voltage drop of the electronic component, calculating the contact voltage drop using a contact voltage drop detection method, establishing a forward voltage drop model of the electronic component, calculating the forward voltage drop of the electronic component, and recording it as a first electronic component performance influencing factor;

[0007] S2. After modeling the sample data in the electronic component life matrix, a gray model is established, and then the gray model is corrected for residual errors to obtain a corrected gray model function. The electronic component resistance is predicted using a BP neural network, and the predicted resistance is substituted into the corrected gray model function to obtain the electronic component life, which is recorded as the second electronic component performance influencing factor;

[0008] S3, obtaining a final set of electronic component performance influencing factors after applying an interval algorithm and normalization to the first electronic component performance influencing factors and the second electronic component performance influencing factors;

[0009] S4. Finally, the set of factors affecting the performance of electronic components is divided into a sample training set and a sample test set. The sample training set is input into the Elman neural network for training to obtain a trained Elman neural network. The sample test set verifies the trained Elman neural network to achieve the prediction of the performance of electronic components.

[0010] The invention firstly calculates the base voltage drop to establish the forward voltage drop model of the electronic component, uses the periodic voltage drop to replace the sample data in the contact voltage detection set, calculates the contact voltage drop of the electronic component, and records it as the first electronic component performance influencing factor. The method of use is ingenious and simple, and easy to operate. Secondly, the electronic component life matrix is ​​established, and the gray model is established after the whitening differential equation discretization processing, and then the gray model is corrected for the residual to obtain the corrected gray model function, and the BP neural network is trained. The BP neural network is used to predict the resistance of the electronic component, and the predicted resistance is substituted into the corrected gray model function to obtain the electronic component life, which is recorded as the second electronic component performance influencing factor, and the gray system is introduced. The model and BP neural network break through the limitations of traditional methods and make the prediction results more accurate. The first electronic component performance influencing factor and the second electronic component performance influencing factor are processed by interval algorithm and normalization to obtain the final set of electronic component performance influencing factors. The data distribution will not be changed after processing the original data, and the difficulty of subsequent data processing is reduced. The sample training set is used to train the Elman neural network, and the sample test set is used to verify the trained Elman neural network. Finally, the performance of electronic components is predicted. The introduction of variable learning rate algorithm helps to improve the training speed of neural network. The Elman neural network used has fast convergence speed and high approximation accuracy, which is better than other neural networks.

[0011] Preferably, the S1 comprises the following steps:

[0012] S11. A constant forward current passes through the electronic component, and a voltage value is generated at both ends of the electronic component. The forward voltage drop calculation formula is as follows:

[0013] A = A 1+ A 2+ A 3

[0014] in, A represents the forward voltage drop, A 1 means junction voltage drop and A 1=0.6, unit volt, A 2 represents the contact voltage drop, A 3 represents the base voltage drop;

[0015] The specific calculation steps of the base voltage drop and the contact voltage drop are as follows:

[0016] S111, assuming that a represents the base width, a1 represents the minority carrier lifetime, a2 represents the carrier mobility, a3 represents the thermal voltage, and a4 represents the diffusion length, then the base voltage drop calculation formula is as follows:

[0017] ;

[0018] S112, obtaining the contact voltage of the electronic component, and setting the contact voltage set to A '={b1,b2,b3,...,b i}, where b i Indicates i electronic component contact voltages, the characteristic dimension of the contact voltage in the contact voltage set is j , the contact voltage matrix B is obtained as follows, ;

[0019] in, b ij Represents the contact voltage matrix i The feature dimension is j The contact voltage;

[0020] S113, using the contact voltage drop detection method to calculate the contact voltage drop, taking a row in the contact voltage matrix as a contact voltage detection set ,in Represents the contact voltage matrix The feature dimension is j contact voltage; setting a periodic voltage drop and a critical value of the periodic voltage drop so that the periodic voltage drop reaches the critical value of the periodic voltage drop, adding the periodic voltage drop to a contact voltage detection set to obtain a critical contact voltage detection set; setting a critical threshold value to ω, when the absolute value of the difference between the contact voltage and the periodic voltage drop in the critical contact voltage detection set is greater than or equal to ω, discarding the corresponding contact voltage in the critical contact voltage detection set to obtain a processed critical contact voltage detection set;

[0021] S114. Select a new periodic voltage drop again, set the first threshold value to ω1, and the new periodic voltage drop satisfies that the absolute value of the difference between the periodic voltage drop and the new periodic voltage drop is less than ω1; add the new periodic voltage drop to the processed critical contact voltage detection set to obtain a new critical contact voltage detection set; calculate the average value of the critical contact in the new critical contact voltage detection set, calculate the difference between the new periodic voltage drop and the average value of the critical contact in the new critical contact voltage detection set, and record it as Repeat S113 and S114 for other contact voltage detection sets in the contact voltage matrix to obtain a contact voltage set, recorded as ,in Indicates i The difference between a new periodic voltage drop and an average value of critical contacts in a new critical contact voltage detection set is used to calculate an average value of the contact voltage set, where the average value of the contact voltage set is the contact voltage drop of the electronic component;

[0022] S12. Calculate and obtain a forward voltage drop of the electronic component, and record the forward voltage drop of the electronic component as a first electronic component performance influencing factor.

[0023] The invention establishes a forward voltage drop model of electronic components by calculating the base voltage drop, uses periodic voltage drop to replace sample data in a contact voltage detection set, and obtains the contact voltage drop of the electronic component by calculating the average value of the contact voltage set, which is recorded as the first electronic component performance influencing factor. The method of use is ingenious, simple, and easy to operate.

[0024] Preferably, S2 comprises the following steps:

[0025] S21. Set the electronic component storage failure rate β The basic failure rate of electronic components during working period is β 1 , the environmental factor of electronic components is β 2 , the temperature coefficient of electronic components is β 3 , the quality factor of electronic components is β 4 , the electronic component rating factor is β 5 , the electronic component cycle coefficient is β 6 , then the calculation formula for electronic component storage failure rate is as follows:

[0026] β=β 1 β 2 β 3 β 4 β 5 β 6

[0027] The electronic component storage failure rate is calculated, and the electronic component storage failure rate is the electronic component life. The electronic component life set is obtained, which is recorded as ,in express i' The characteristic dimension of the electronic components in the electronic component life set is set to j' , generate electronic component lifetime matrix B' as follows,

[0028] ;

[0029] in, Represents the first i' The feature dimension is j'The life of electronic components;

[0030] S22, selecting a row in the electronic component life matrix as a grey model test set to be processed ,in Represents the first The feature dimension is j 'Electronic component life; the gray model test set to be processed is accumulated to obtain a gray model test set ,in express Corresponding accumulated value; establish whitening differential equation for the grey model test set, set c 1 represents the development coefficient, c 2 represents the gray parameter. The whitening differential equation is discretized and the calculation formula is as follows:

[0031] ;

[0032] Set the whitening differential matrix to B'' , the grey model test matrix to be processed is B''' ,as follows,

[0033] , ;

[0034] in, express The corresponding accumulated value;

[0035] Calculate using matrix inverse B '''= B ''c3 and satisfies c3=[c 1, c2] T , obtain the response function c3 of the whitened differential equation, the response function of the whitened differential equation is the grey model function;

[0036] S23, performing residual correction on the grey model function, and establishing a residual set for the grey model test set to be processed ,in Indicates the grey model test set to be processed and Calculate the average value of the residuals in the residual set, recorded as the standard residual , set the second threshold value to ω2, when the absolute value of the difference between the residual in the residual set and the standard residual is less than ω2, add the corresponding residual in the residual set to the new residual set, otherwise discard the corresponding residual in the residual set; use the new residual set to restore and correct the grey model function, the specific process is as follows:

[0037] S231, setting the new residual set to D = d 1, d 2, d 3,..., d c},in d c Denote the cth new residual, calculate the relative error of the new residuals in the new residual set and the average relative error of the new residuals in the new residual set; substitute the new residual set into the grey model function to obtain the grey model set, denoted as D ' ={d ' 1,d ' 2,d ' 3,...,d ' c}, where d ' c Represents the data in the cth grey model set, and calculates the average value of the data in the grey model set and the residual average value of the data in the grey model set;

[0038] S232, calculating the relative error of the new residuals in the new residual set and the variance of the average value of the data in the grey model set, recorded as δ1; calculating the average relative error of the new residuals in the new residual set and the variance of the residual average value of the data in the grey model set, recorded as δ2; setting the third threshold value to ω3, when the ratio of δ1 to δ2 is greater than or ω3, the grey model function does not need to be restored and corrected; otherwise, the grey model function is restored and corrected using the new residual set to obtain a corrected grey model function;

[0039] S24, obtaining electronic component resistance, obtaining an electronic component resistance set, dividing the electronic component resistance set into an electronic component resistance training set and an electronic component resistance test set; setting the number of input layer neurons of the BP (multi-layer forward error back propagation) neural network to 1, the number of output layer neurons to 1, and the number of hidden layer neurons to F , the learning rate is e , the activation function is Sigmoid function; the training error threshold is set to g 1 After initializing the weights, input the electronic component resistance training set to train the BP neural network, and continuously iterate. When the error of the BP neural network is less than g , get the trained BP neural network; otherwise adjust the number of hidden layer nodes and potential factors until the error of the BP neural network is less than g ; Then input the electronic component resistance test set into the trained BP neural network, and set the test error threshold to g2 , when the error of the trained BP neural network is less than g 2 , and obtain the final BP neural network; otherwise, repeat S24 until the error of the trained BP neural network is less than g 2 ;

[0040] The electronic component detection data of previous years is obtained to obtain the electronic component resistance set of previous years, the electronic component resistance set of previous years is input into the final BP neural network, the resistance prediction result is output, the resistance prediction result is input into the modified grey model function, and the function value is output, the function value is the electronic component life, and the electronic component life is recorded as the second electronic component performance influencing factor.

[0041] The invention establishes an electronic component life matrix, establishes a gray model after discretization processing of whitened differential equations, and then performs residual correction on the gray model to obtain a corrected gray model function. After training the BP neural network, the BP neural network is used to predict the resistance of the electronic component, and the predicted resistance is substituted into the corrected gray model function to obtain the electronic component life, which is recorded as the second electronic component performance influencing factor. The gray system model and BP neural network are introduced to make the prediction result highly accurate.

[0042] Preferably, S3 comprises the following steps:

[0043] S31, setting the first electronic component performance influencing factor set to E' ={e ' 1, e ' 2, e ' 3,..., e ' f '}, where e ' f ' Indicates f' The first electronic component performance influencing factors, the second electronic component performance influencing factors set is E'' ={e '' 1, e '' 2, e '' 3,..., e '' f ''},in , e '' f '' Indicates f''a second electronic component performance influencing factor; and a second electronic component performance influencing factor set selected from the first electronic component performance influencing factor set and the second electronic component performance influencing factor set. f data elements constitute the set of factors affecting the performance of the electronic components to be processed E ={(e ' 1, e '' 1),(e ' 2, e '' 2),(e ' 3, e '' 3),...,(e ' f, e '' f )}, where e ' f Indicates f The first factor affecting the performance of electronic components, e '' f Indicates f The second factor affecting the performance of electronic components;

[0044] S32, setting the average value of the sample data in the first electronic component performance influencing factor set to , the standard deviation of the sample data in the first electronic component performance influencing factor set is , the largest sample data in the first electronic component performance influencing factor set is g 1; Set the average value of the sample data in the second electronic component performance influencing factor set to , the standard deviation of the sample data in the second electronic component performance influencing factor set is , the largest sample data in the second electronic component performance influencing factor set is g 2; The calculation formula of interval algorithm is as follows:

[0045] , ;

[0046] in, F 1 represents the set interval of factors affecting the performance of the first electronic component, F 2 represents the set interval of factors affecting the performance of the second electronic component;

[0047] The sample data in the to-be-processed electronic component performance influencing factor set that are not in the first electronic component performance influencing factor set interval and the second electronic component performance influencing factor set interval are discarded to obtain the electronic component performance influencing factor set;

[0048] S33, setting the maximum sample data in the electronic component performance influencing factor set to be e' max and e ' ' max ， The minimum sample data in the set of factors affecting the performance of electronic components are e ' min and e '' min ， The normalized calculation formula is as follows:

[0049] , ;

[0050] Among them, e ''' 1 and e ''' 2 represents the normalized data of the set of factors affecting the performance of electronic components;

[0051] The normalized data of the data samples in the electronic component performance influencing factor set are calculated in sequence to obtain a final electronic component performance influencing factor set.

[0052] The invention uses interval algorithm and normalization processing on the first electronic component performance influencing factors and the second electronic component performance influencing factors, considers the individual differences of electronic components and discards unreasonable data, and obtains the final set of electronic component performance influencing factors. The data distribution will not be changed after processing the original data, and the difficulty of subsequent data processing is reduced.

[0053] Preferably, S4 comprises the following steps:

[0054] S41, setting the final electronic component performance influencing factor set to ,in Indicates f''' The final electronic component performance influencing factors are divided into the final electronic component performance influencing factors set into the electronic component performance sample training set and electronic component performance sample test sets , where e ''' g represents the gth factor affecting the performance of the final electronic component; the number of input layer nodes is set to g, the number of output layer nodes is 2, and the number of hidden layer nodes is , initialize the input layer, input layer and hidden layer, and set the connection weights to c and The initial learning rate is or and or =0.1, the initial learning rate is constantly corrected during the training process, the correction formula is as follows,

[0055] ;

[0056] in, or ( h +1) indicates the number of iterations is h +1 learning rate, or ( h ) indicates the number of iterations is h The learning rate, k ( h ) indicates the number of iterations is h The network error and k ( h -1) indicates the number of iterations is h -1 network error sum;

[0057] S42, inputting the electronic component performance sample training set into the Elman neural network, setting the expected error to P , the current number of iterations is l , the maximum number of iterations is L , train the Elman neural network, iterate continuously, and the current number of iterations is greater than L Or the training error is less than P , the trained Elman neural network is obtained; otherwise, the connection weights are adjusted until the current number of iterations is greater than L Or the training error is less than P ; Input the electronic component performance sample test set into the trained Elman neural network, and set the accuracy error to g , when the error of the trained Elman neural network is less than g , the final Elman neural network is obtained; otherwise, repeat S41 and S42 until the error of the trained Elman neural network is less than g ;

[0058] The forward voltage drop of electronic components in previous years and the life of electronic components in previous years are input into the final Elman neural network, and the predicted forward voltage drop of electronic components and the predicted life of electronic components are output to realize the prediction of electronic component performance.

[0059] The invention trains the Elman neural network by using a sample training set, verifies the trained Elman neural network by using a sample test set, obtains the final Elman neural network after the training conditions are met, and finally outputs the predicted forward voltage drop of the electronic component and the predicted life of the electronic component to achieve performance prediction of the electronic component. The introduction of a variable learning rate algorithm helps to improve the neural network training speed and prediction accuracy compared to the traditional Elman neural network, and has good dynamic performance.

[0060] This embodiment also discloses an electronic component performance prediction system based on artificial intelligence, which specifically includes: an electronic component forward voltage drop module, an electronic component life module, an electronic component performance influencing factor processing module and an electronic component performance prediction module;

[0061] The electronic component forward voltage drop module is used to calculate the electronic component forward voltage drop from the junction voltage drop, the base voltage drop and the contact voltage drop;

[0062] The electronic component life module is used to predict the resistance of the electronic component using a neural network and obtain the electronic component life through a grey model function;

[0063] The electronic component performance influencing factor processing module is used to process the electronic component performance influencing factors through interval algorithm and normalization;

[0064] The electronic component performance prediction module is used to train the Elman neural network and then use the final Elman neural network to predict the performance of the electronic component.

[0065] The present invention has the following beneficial effects:

[0066] 1. The invention first calculates the base voltage drop to establish a forward voltage drop model for electronic components, uses the periodic voltage drop to replace the sample data in the contact voltage detection set, calculates the contact voltage drop of the electronic component, and records it as the first electronic component performance influencing factor. The method of use is ingenious and simple and easy to operate.

[0067] 2. The invention establishes an electronic component life matrix, establishes a gray model after discretization processing of whitened differential equations, and then performs residual correction on the gray model to obtain a corrected gray model function, trains a BP neural network, and uses the BP neural network to predict the resistance of the electronic component. The predicted resistance is substituted into the corrected gray model function to obtain the electronic component life, which is recorded as the second electronic component performance influencing factor. The gray system model and BP neural network are introduced to break through the limitations of traditional methods and make the prediction results highly accurate.

[0068] 3. The invention obtains a final set of electronic component performance influencing factors after using interval algorithm and normalization processing on the first electronic component performance influencing factors and the second electronic component performance influencing factors. The data distribution will not be changed after processing the original data, and the difficulty of subsequent data processing is reduced. The sample training set is then used to train the Elman neural network, and the sample test set is used to verify the trained Elman neural network. Finally, the performance of the electronic component is predicted. The introduction of a variable learning rate algorithm helps to improve the training speed of the neural network. The Elman neural network used has a fast convergence speed and high approximation accuracy.

[0069] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0071] Figure 1 The present invention provides a flow chart of an electronic component performance prediction system based on artificial intelligence for predicting electronic component performance. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0074] This embodiment discloses an electronic component performance prediction method based on artificial intelligence, which specifically includes the following contents:

[0075] S1. Obtaining the junction voltage drop and base voltage drop of the electronic component, calculating the contact voltage drop using a contact voltage drop detection method, establishing a forward voltage drop model of the electronic component, calculating the forward voltage drop of the electronic component, and recording it as a first electronic component performance influencing factor;

[0076] The S1 comprises the following steps:

[0077] S11. A constant forward current passes through the electronic component, and a voltage value is generated at both ends of the electronic component. The forward voltage drop calculation formula is as follows:

[0078] A = A 1+ A 2+ A 3

[0079] in, A represents the forward voltage drop, A 1 means junction voltage drop and A 1=0.6, unit volt,A 2 represents the contact voltage drop, A 3 represents the base voltage drop;

[0080] The specific calculation steps of the base voltage drop and the contact voltage drop are as follows:

[0081] S111, assuming that a represents the base width, a1 represents the minority carrier lifetime, a2 represents the carrier mobility, a3 represents the thermal voltage, and a4 represents the diffusion length, then the base voltage drop calculation formula is as follows:

[0082] ;

[0083] S112, obtaining the contact voltage of the electronic component, and setting the contact voltage set to A '={b1,b2,b3,...,b i}, where b i Indicates i electronic component contact voltages, the characteristic dimension of the contact voltage in the contact voltage set is j , the contact voltage matrix B is obtained as follows,

[0084] ;

[0085] in, b ij Represents the contact voltage matrix i The feature dimension is j The contact voltage;

[0086] S113, using the contact voltage drop detection method to calculate the contact voltage drop, taking a row in the contact voltage matrix as a contact voltage detection set ,in Represents the contact voltage matrix The feature dimension is j contact voltage; setting a periodic voltage drop and a critical value of the periodic voltage drop so that the periodic voltage drop reaches the critical value of the periodic voltage drop, adding the periodic voltage drop to a contact voltage detection set to obtain a critical contact voltage detection set; setting a critical threshold value to ω, when the absolute value of the difference between the contact voltage and the periodic voltage drop in the critical contact voltage detection set is greater than or equal to ω, discarding the corresponding contact voltage in the critical contact voltage detection set to obtain a processed critical contact voltage detection set;

[0087] S114. Select a new periodic voltage drop again, set the first threshold value to ω1, and the new periodic voltage drop satisfies that the absolute value of the difference between the periodic voltage drop and the new periodic voltage drop is less than ω1; add the new periodic voltage drop to the processed critical contact voltage detection set to obtain a new critical contact voltage detection set; calculate the average value of the critical contact in the new critical contact voltage detection set, calculate the difference between the new periodic voltage drop and the average value of the critical contact in the new critical contact voltage detection set, and record it as Repeat S113 and S114 for other contact voltage detection sets in the contact voltage matrix to obtain a contact voltage set, recorded as ,in Indicates i The difference between a new periodic voltage drop and an average value of critical contacts in a new critical contact voltage detection set is used to calculate an average value of the contact voltage set, where the average value of the contact voltage set is the contact voltage drop of the electronic component;

[0088] S12, calculating and obtaining a forward voltage drop of the electronic component, and recording the forward voltage drop of the electronic component as a first electronic component performance influencing factor;

[0089] S2. After modeling the sample data in the electronic component life matrix, a gray model is established, and then the gray model is corrected for residual errors to obtain a corrected gray model function. The electronic component resistance is predicted using a BP neural network, and the predicted resistance is substituted into the corrected gray model function to obtain the electronic component life, which is recorded as the second electronic component performance influencing factor;

[0090] The S2 comprises the following steps:

[0091] S21. Set the electronic component storage failure rate β The basic failure rate of electronic components during working period is β 1 , the environmental factor of electronic components is β 2 , the temperature coefficient of electronic components is β 3 , the quality factor of electronic components is β 4 , the electronic component rating factor is β 5 , the electronic component cycle coefficient is β 6 , then the calculation formula for electronic component storage failure rate is as follows:

[0092] β=β 1 β 2 β 3 β4 β 5 β 6

[0093] The electronic component storage failure rate is calculated, and the electronic component storage failure rate is the electronic component life. The electronic component life set is obtained, which is recorded as ,in express i' The characteristic dimension of the electronic components in the electronic component life set is set to j' , generate electronic component lifetime matrix B' as follows,

[0094] ;

[0095] in, Represents the first i' The feature dimension is j' The life of electronic components;

[0096] S22, selecting a row in the electronic component life matrix as a grey model test set to be processed ,in Represents the first The feature dimension is j' The electronic component life; the gray model test set to be processed is accumulated to obtain a gray model test set ,in express Corresponding accumulated value; establish whitening differential equation for the grey model test set, set c 1 represents the development coefficient, c 2 represents the gray parameter. The whitening differential equation is discretized and the calculation formula is as follows:

[0097] ;

[0098] Set the whitening differential matrix to B'' , the grey model test matrix to be processed is B''' ,as follows,

[0099] , ;

[0100] in, express The corresponding accumulated value;

[0101] Calculate using matrix inverse B '''= B''c3 and satisfies c3=[c 1, c2] T , obtain the response function c3 of the whitened differential equation, the response function of the whitened differential equation is the grey model function;

[0102] S23, performing residual correction on the grey model function, and establishing a residual set for the grey model test set to be processed ,in Indicates the grey model test set to be processed and Calculate the average value of the residuals in the residual set, recorded as the standard residual , set the second threshold value to ω2, when the absolute value of the difference between the residual in the residual set and the standard residual is less than ω2, add the corresponding residual in the residual set to the new residual set, otherwise discard the corresponding residual in the residual set; use the new residual set to restore and correct the grey model function, the specific process is as follows:

[0103] S231, setting the new residual set to D = d 1, d 2, d 3,..., d c},in d c Denotes the cth new residual, calculates the relative error of the new residuals in the new residual set and the average relative error of the new residuals in the new residual set; substitutes the new residual set into the grey model function to obtain the grey model set, recorded as D' = d' 1, d' 2, d' 3,..., d' c},in d' c Represents the data in the cth grey model set, and calculates the average value of the data in the grey model set and the residual average value of the data in the grey model set;

[0104] S232, calculating the relative error of the new residuals in the new residual set and the variance of the average value of the data in the grey model set, recorded as δ1; calculating the average relative error of the new residuals in the new residual set and the variance of the residual average value of the data in the grey model set, recorded as δ2; setting the third threshold value to ω3, when the ratio of δ1 to δ2 is greater than or equal to ω3, there is no need to restore and correct the grey model function; otherwise, the grey model function is restored and corrected using the new residual set to obtain a corrected grey model function;

[0105] S24, obtaining electronic component resistance, obtaining an electronic component resistance set, and dividing the electronic component resistance set into an electronic component resistance training set and an electronic component resistance test set; setting the number of input layer neurons of the BP neural network to 1, the number of output layer neurons to 1, and the number of hidden layer neurons to F , the learning rate is e , the activation function is Sigmoid function; the training error threshold is set to g 1 After initializing the weights, input the electronic component resistance training set to train the BP neural network, and continuously iterate. When the error of the BP neural network is less than g , get the trained BP neural network; otherwise adjust the number of hidden layer nodes and potential factors until the error of the BP neural network is less than g ; Then input the electronic component resistance test set into the trained BP neural network, and set the test error threshold to g 2 , when the error of the trained BP neural network is less than g 2 , and obtain the final BP neural network; otherwise, repeat S24 until the error of the trained BP neural network is less than g 2 ;

[0106] Acquire the electronic component detection data of previous years to obtain the electronic component resistance set of previous years, input the electronic component resistance set of previous years into the final BP neural network, output the resistance prediction result, input the resistance prediction result into the modified grey model function, output the function value, the function value is the electronic component life, and record the electronic component life as the second electronic component performance influencing factor;

[0107] S3, obtaining a final set of electronic component performance influencing factors after applying an interval algorithm and normalization to the first electronic component performance influencing factors and the second electronic component performance influencing factors;

[0108] The S3 comprises the following steps:

[0109] S31, setting the first electronic component performance influencing factor set to E' ={e ' 1, e ' 2, e ' 3,..., e ' f '}, where e ' f ' Indicatesf' The first electronic component performance influencing factors, the second electronic component performance influencing factors set is E'' ={e '' 1, e '' 2, e '' 3,..., e '' f ''},in , e '' f '' Indicates f'' a second electronic component performance influencing factor; and a second electronic component performance influencing factor set selected from the first electronic component performance influencing factor set and the second electronic component performance influencing factor set. f data elements constitute the set of factors affecting the performance of the electronic components to be processed E ={(e ' 1, e '' 1),(e ' 2, e '' 2),(e ' 3, e '' 3),...,(e ' f, e '' f )}, where e ' f Indicates f The first factor affecting the performance of electronic components, e '' f Indicates f Second factor affecting the performance of electronic components

[0110] S32, setting the average value of the sample data in the first electronic component performance influencing factor set to , the standard deviation of the sample data in the first electronic component performance influencing factor set is , the largest sample data in the first electronic component performance influencing factor set is g 1; Set the average value of the sample data in the second electronic component performance influencing factor set to , the standard deviation of the sample data in the second electronic component performance influencing factor set is , the largest sample data in the second electronic component performance influencing factor set is g 2; The calculation formula of interval algorithm is as follows:

[0111] , ;

[0112] in, F 1 represents the set interval of factors affecting the performance of the first electronic component, F 2 represents the set interval of factors affecting the performance of the second electronic component;

[0113] The sample data in the to-be-processed electronic component performance influencing factor set that are not in the first electronic component performance influencing factor set interval and the second electronic component performance influencing factor set interval are discarded to obtain the electronic component performance influencing factor set;

[0114] S33, setting the maximum sample data in the electronic component performance influencing factor set to be e ' max and e ' ' max ， The minimum sample data in the set of factors affecting the performance of electronic components are e ' min and e '' min ， The normalized calculation formula is as follows:

[0115] , ;

[0116] Among them, e ''' 1 and e ''' 2 represents the normalized data of the set of factors affecting the performance of electronic components;

[0117] Calculating the normalized data of the data samples in the electronic component performance influencing factor set in sequence to obtain a final electronic component performance influencing factor set;

[0118] S4. The final set of factors affecting the performance of electronic components is divided into a sample training set and a sample test set. The sample training set is input into the Elman neural network for training to obtain a trained Elman neural network. The sample test set verifies the trained Elman neural network to achieve performance prediction of electronic components.

[0119] The S4 comprises the following steps:

[0120] S41, setting the final electronic component performance influencing factor set to ,in Indicates f''' The final electronic component performance influencing factors are divided into the final electronic component performance influencing factors set into the electronic component performance sample training set and electronic component performance sample test sets , where e '''g represents the gth factor affecting the performance of the final electronic component; the number of input layer nodes is set to g, the number of output layer nodes is 2, and the number of hidden layer nodes is , initialize the input layer, input layer and hidden layer, and set the connection weights to c and The initial learning rate is or and or =0.1, the initial learning rate is constantly corrected during the training process, the correction formula is as follows,

[0121] ;

[0122] in, or ( h +1) indicates the number of iterations is h +1 learning rate, or ( h ) indicates the number of iterations is h The learning rate, k ( h ) indicates the number of iterations is h The network error and k ( h -1) indicates the number of iterations is h -1 network error sum;

[0123] S42, input the electronic component performance sample training set into the Elman neural network, and set the expected error to P , the current number of iterations is l , the maximum number of iterations is L , train the Elman neural network, iterate continuously, and the current number of iterations is greater than L Or the training error is less than P , the trained Elman neural network is obtained; otherwise, the connection weights are adjusted until the current number of iterations is greater than L Or the training error is less than P ; Input the electronic component performance sample test set into the trained Elman neural network, and set the accuracy error to g , when the error of the trained Elman neural network is less than g , the final Elman neural network is obtained; otherwise, repeat S41 and S42 until the error of the trained Elman neural network is less than g ;

[0124] The forward voltage drop of electronic components in previous years and the life of electronic components in previous years are input into the final Elman neural network, and the predicted forward voltage drop of electronic components and the predicted life of electronic components are output to realize the prediction of electronic component performance.

[0125] This embodiment also discloses an electronic component performance prediction system based on artificial intelligence, which specifically includes: an electronic component forward voltage drop module, an electronic component life module, an electronic component performance influencing factor processing module and an electronic component performance prediction module;

[0126] The electronic component forward voltage drop module is used to calculate the electronic component forward voltage drop from the junction voltage drop, the base voltage drop and the contact voltage drop;

[0127] The electronic component life module is used to predict the resistance of the electronic component using a neural network and obtain the electronic component life through a grey model function;

[0128] The electronic component performance influencing factor processing module is used to process the electronic component performance influencing factors through interval algorithm and normalization;

[0129] The electronic component performance prediction module is used to train the Elman neural network and then use the final Elman neural network to predict the performance of the electronic component.

[0130] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0131] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. An electronic component performance prediction method based on artificial intelligence, characterized in that: The steps include: S1. Obtaining the junction voltage drop and base voltage drop of the electronic component, calculating the contact voltage drop using a contact voltage drop detection method, establishing a forward voltage drop model of the electronic component, calculating the forward voltage drop of the electronic component, and recording it as a first electronic component performance influencing factor; S2. After modeling the sample data in the electronic component life matrix, a gray model is established, and then the gray model is corrected for residual errors to obtain a corrected gray model function. The electronic component resistance is predicted using a BP neural network, and the predicted resistance is substituted into the corrected gray model function to obtain the electronic component life, which is recorded as the second electronic component performance influencing factor; S3, obtaining a final set of electronic component performance influencing factors after applying an interval algorithm and normalization to the first electronic component performance influencing factors and the second electronic component performance influencing factors; S4. Finally, the set of factors affecting the performance of electronic components is divided into a sample training set and a sample test set. The sample training set is input into the Elman neural network for training to obtain a trained Elman neural network. The sample test set verifies the trained Elman neural network to achieve the prediction of the performance of electronic components.

2. The method for predicting electronic component performance based on artificial intelligence according to claim 1, characterized in that: The S1 comprises the following steps: S11. A constant forward current passes through the electronic component, a voltage value is generated at both ends of the electronic component, and the forward voltage drop of the electronic component is calculated. A = A 1+ A 2+ A 3. A 1 means junction voltage drop and A 1=0.6, unit volt, A 2 represents the contact voltage drop, A 3 represents the base voltage drop; the forward voltage drop of the electronic component is recorded as the first electronic component performance influencing factor.

3. The method for predicting electronic component performance based on artificial intelligence according to claim 2, characterized in that: The S11 comprises the following steps: S111, setting the base width, minority carrier lifetime, carrier mobility, thermal voltage and diffusion length, and calculating the base voltage drop; S112, obtaining the contact voltage of the electronic component, setting the contact voltage set to generate a contact voltage matrix, and using a contact voltage drop detection method to calculate the contact voltage drop.

4. The method for predicting electronic component performance based on artificial intelligence according to claim 3, characterized in that: The S2 comprises the following steps: S21, calculating the electronic component storage failure rate, where the electronic component storage failure rate is the electronic component life, obtaining an electronic component life set, and generating an electronic component life matrix; S22, selecting a row in the electronic component life matrix as a grey model test set to be processed, performing accumulation processing on the grey model test set to obtain a grey model test set, and establishing a whitened differential equation for the grey model test set; using matrix inverse calculation to obtain a response function of the whitened differential equation, wherein the response function of the whitened differential equation is the grey model function; S23, performing residual correction on the grey model function to obtain a new residual set, and using the new residual set to restore and correct the grey model function to obtain a corrected grey model function; S24, acquiring electronic component resistances to obtain an electronic component resistance set, and dividing the electronic component resistance set into an electronic component resistance training set and an electronic component resistance test set; Set the number of neurons in the input layer of the BP neural network to 1, the number of neurons in the output layer to 1, and the number of neurons in the hidden layer to Φ , the learning rate is ε , the activation function is Sigmoid function; the training error threshold is set to ζ 1 After initializing the weights, input the electronic component resistance training set to train the BP neural network, and continuously iterate. When the error of the BP neural network is less than ζ 1 , get the trained BP neural network; Otherwise, adjust the number of hidden layer nodes and potential factors until the error of the BP neural network is ζ 1 ; Then input the electronic component resistance test set into the trained BP neural network, and set the test error threshold to ζ 2 , when the error of the trained BP neural network is less than ζ 2 , get the final BP neural network; Otherwise, repeat S24 until the error of the trained BP neural network is less than ζ 2 .

5. The method for predicting electronic component performance based on artificial intelligence according to claim 4, characterized in that: The electronic component detection data of previous years is obtained to obtain the electronic component resistance set of previous years, the electronic component resistance set of previous years is input into the final BP neural network, the resistance prediction result is output, the resistance prediction result is input into the modified grey model function, and the function value is output, the function value is the electronic component life, and the electronic component life is recorded as the second electronic component performance influencing factor.

6. The method for predicting electronic component performance based on artificial intelligence according to claim 5, characterized in that: The S23 comprises the following steps: S231, setting a new residual set and substituting the new residual set into the grey model function to obtain a grey model set, and calculating the average value of the data in the grey model set and the residual average value of the data in the grey model set; S232. Calculate the relative error of the new residuals in the new residual set and the variance of the average value of the data in the grey model set, recorded as δ1; calculate the average relative error of the new residuals in the new residual set and the variance of the residual average value of the data in the grey model set, recorded as δ2; set the third threshold value to ω3, when the ratio of δ1 and δ2 is greater than or equal to ω3, there is no need to restore and correct the grey model function; otherwise, use the new residual set to restore and correct the grey model function to obtain a corrected grey model function.

7. The method for predicting electronic component performance based on artificial intelligence according to claim 6, characterized in that: The S3 comprises the following steps: S31, setting a first electronic component performance influencing factor set and a second electronic component performance influencing factor set, selecting the first electronic component performance influencing factor set and the second electronic component performance influencing factor set before f The data elements constitute a set of influencing factors of the electronic component performance to be processed; S32, using an interval algorithm to filter the sample data in the set of electronic component performance influencing factors to be processed to obtain the set of electronic component performance influencing factors; S33, calculating the normalized data of the data samples in the electronic component performance influencing factor set in sequence to obtain a final electronic component performance influencing factor set.

8. The method for predicting electronic component performance based on artificial intelligence according to claim 7, characterized in that: The S4 comprises the following steps: S41, setting the final electronic component performance influencing factor set, dividing the final electronic component performance influencing factor set into an electronic component performance sample training set and an electronic component performance sample test set, and continuously correcting the initial learning rate during the training process; S42, input the electronic component performance sample training set into the Elman neural network, and set the expected error to Ψ , the current number of iterations is l , the maximum number of iterations is L , train the Elman neural network, iterate continuously, and the current number of iterations is greater than L Or the training error is less than Ψ When , the trained Elman neural network is obtained; Otherwise, adjust the connection weight until the current number of iterations is greater than L Or the training error is less than Ψ ; The electronic component performance sample test set is input into the trained Elman neural network, and the accuracy error is set to ζ , when the error of the trained Elman neural network is less than ζ When , the final Elman neural network is obtained; Otherwise, repeat S41 and S42 until the error of the trained Elman neural network is less than ζ .

9. The method for predicting electronic component performance based on artificial intelligence according to claim 8, characterized in that: The forward voltage drop of electronic components in previous years and the life of electronic components in previous years are input into the final Elman neural network, and the predicted forward voltage drop of electronic components and the predicted life of electronic components are output to realize the prediction of electronic component performance.

10. An electronic component performance prediction system based on artificial intelligence according to any one of claims 1 to 9, characterized in that: Specifically include: Electronic component forward voltage drop module, electronic component life module, electronic component performance influencing factor processing module and electronic component performance prediction module; The electronic component forward voltage drop module is used to calculate the electronic component forward voltage drop from the junction voltage drop, the base voltage drop and the contact voltage drop; The electronic component life module is used to predict the resistance of the electronic component using a neural network and obtain the electronic component life through a grey model function; The electronic component performance influencing factor processing module is used to process the electronic component performance influencing factors through interval algorithm and normalization; The electronic component performance prediction module is used to train the Elman neural network and then use the final Elman neural network to predict the performance of the electronic component.

Images