Silicone rubber material performance prediction method and device based on LSTM neural network

Abstract

The application discloses a kind of based on LSTM neural network's silicon rubber material performance prediction method and device, by obtaining the dielectric property data of silicon rubber material under multiple aging stages, the dielectric property data is fused and handled, and the comprehensive performance index sequence of silicon rubber material is obtained;According to the comprehensive performance index sequence, the network parameters of long short-term memory neural network are optimized by discrete differential evolution method, until long short-term memory neural network reaches preset convergence condition, and the silicon rubber material performance prediction model is obtained, effectively avoids the problem that fitting degree is not enough and precision is low by experience to select key parameters;Using silicon rubber material performance prediction model, according to the initial sequence of the comprehensive performance index of the silicon rubber material to be predicted, the target comprehensive performance index data of the silicon rubber material to be predicted in each aging stage is predicted, the aging performance of silicon rubber material is predicted, and detection basis is provided for the life prediction of silicon rubber material in dry-type transformer.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of performance prediction of insulating materials of power equipment, and particularly relates to a silicon rubber material performance prediction method and device based on an LSTM neural network. BACKGROUND

[0002] The service life of a dry-type transformer is largely dependent on the aging characteristics of the insulating material of the transformer, and the silicon rubber material can improve partial discharge defects, has strong heat dissipation capacity, and has the advantages of high strength, strong overload and short-circuit resistance, and the like, and the silicon rubber material can be recycled and utilized, thereby avoiding pollution of solid waste. Liquid silicon rubber vacuum casting or vacuum impregnation of a winding is adopted, and the silicon rubber penetrating between turns and layers solidifies the winding into an integral structure, and the silicon rubber encapsulation layer formed by casting or impregnation on the inner and outer surfaces of the winding can effectively improve the insulation performance of the dry-type transformer.

[0003] However, at present, the research on aging modeling and life prediction of the silicon rubber material under the solid insulation working condition of the winding of the dry-type transformer is relatively less. Therefore, a silicon rubber material performance prediction method for the solid insulation working condition of the winding of the dry-type transformer is urgently needed. SUMMARY

[0004] The application provides a silicon rubber material performance prediction method and device based on an LSTM neural network, so as to realize prediction of the aging performance of the silicon rubber material and provide a detection basis for life prediction of the silicon rubber material in the dry-type transformer.

[0005] In order to solve the above technical problems, in a first aspect, the application provides a silicon rubber material performance prediction method based on an LSTM neural network, comprising:

[0006] Obtaining dielectric performance data of the silicon rubber material under multiple aging stages;

[0007] Performing fusion processing on the dielectric performance data to obtain a comprehensive performance index sequence of the silicon rubber material, the comprehensive performance index sequence being capable of representing the evolution of the electrical performance of the silicon rubber material in the entire aging process, the entire aging process including the multiple aging stages;

[0008] According to the comprehensive performance index sequence, optimizing network parameters of the long short-term memory neural network by using a discrete differential evolution method until the long short-term memory neural network reaches a preset convergence condition, and obtaining a silicon rubber material performance prediction model;

[0009] Using the silicon rubber material performance prediction model, predicting target comprehensive performance index data of the silicon rubber material to be predicted at each aging stage according to an initial comprehensive performance index sequence of the silicon rubber material to be predicted.

[0010] In some implementations, the dielectric performance data includes volume resistivity, relative dielectric constant, and dielectric loss value, the dielectric performance data is fused to obtain a comprehensive performance index sequence of the silicone rubber material, including:

[0011] The volume resistivity, the relative dielectric constant at the target frequency point, and the dielectric loss value are taken as performance indexes to construct a sample index matrix.

[0012] The dielectric performance data is fused according to the sample index matrix by using the principal component analysis method to generate a comprehensive performance index sequence.

[0013] In some implementations, the dielectric performance data is fused according to the sample index matrix by using the principal component analysis method to generate a comprehensive performance index sequence, including:

[0014] The correlation coefficient matrix between the performance indexes is calculated according to the sample index matrix by using the principal component analysis method, and the eigenvalues of the correlation coefficient matrix are calculated.

[0015] A target eigenvector corresponding to a target eigenvalue meeting a component contribution rate condition is determined.

[0016] The dielectric performance data is fused according to the target eigenvector to obtain a comprehensive performance index sequence.

[0017] In some implementations, the comprehensive performance index sequence is:

[0018]

[0019] a1=[a R ,a ei ,a ti ] T ;

[0020] Wherein, Z represents the comprehensive performance index sequence, a1 is the target eigenvector, R Vol is a normalized sequence of volume resistivity, Eps i is a normalized sequence of the relative dielectric constant at the i-th frequency point, Tanδ i is a normalized sequence of the dielectric loss value at the i-th frequency point.

[0021] In some implementations, the network parameters of the long short-term memory neural network are optimized by using the discrete differential evolution method according to the comprehensive performance index sequence until the long short-term memory neural network reaches a preset convergence condition to obtain a silicone rubber material performance prediction model, including:

[0022] The comprehensive performance index sequence is divided into a training set and a test set.

[0023] The long short-term memory neural network is trained by using the training set, and a trained target long short-term memory neural network is obtained.

[0024] The network parameters of the target long short-term memory neural network are optimized by using the discrete differential evolution method according to the test set until the target long short-term memory neural network reaches a preset convergence condition, and a silicone rubber material performance prediction model is obtained, and the network parameters include a time window length, a number of hidden layer neurons, and a number of hidden layer layers.

[0025] In some implementations, the network parameters of the target long short-term memory neural network are optimized by using the discrete differential evolution method according to the test set until the target long short-term memory neural network reaches a preset convergence condition, and a silicone rubber material performance prediction model is obtained, including:

[0026] An initial population is generated based on the time window length, the number of hidden layer neurons, and the number of hidden layer layers.

[0027] The target long short-term memory neural network is trained by using the initial population according to the test set, and an individual fitness is calculated by using a root mean square error.

[0028] The initial population is subjected to mutation operation and crossover operation to obtain a latest population.

[0029] The target long short-term memory neural network is trained by using the latest population according to the test set, and a latest individual fitness is calculated by using a root mean square error.

[0030] If the latest individual fitness is not greater than the individual fitness, the latest population is used as a new initial population for next round of mutation operation and crossover operation until a termination condition is met, and an optimal individual is output.

[0031] The target long short-term memory neural network is updated by using the optimal individual, and a silicone rubber material performance prediction model is obtained.

[0032] In some implementations, the silicone rubber material performance prediction model is used to predict target comprehensive performance index data of a to-be-predicted silicone rubber material at each aging stage according to an initial sequence of comprehensive performance indexes of the to-be-predicted silicone rubber material, including:

[0033] The initial sequence of comprehensive performance indexes is used as model input of the silicone rubber material performance prediction model to predict target comprehensive performance index data of the to-be-predicted silicone rubber material at a next aging stage.

[0034] The target comprehensive performance index data is inserted into the initial sequence of comprehensive performance indexes to obtain a new initial sequence of comprehensive performance indexes.

[0035] With the initial sequence of the new comprehensive performance index, the target comprehensive performance index data of the next aging stage is continuously predicted until the target comprehensive performance index data is lower than the preset threshold, and the remaining life of the silicon rubber material to be predicted is predicted according to the number of iterations experienced when the target comprehensive performance index data is lower than the preset threshold.

[0036] In a second aspect, the present application provides a silicon rubber material performance prediction device based on an LSTM neural network, comprising:

[0037] The acquisition module is configured to acquire dielectric performance data of the silicon rubber material at multiple aging stages.

[0038] The fusion module is configured to perform fusion processing on the dielectric performance data to obtain a comprehensive performance index sequence of the silicon rubber material, the comprehensive performance index sequence being capable of representing the electrical performance evolution of the silicon rubber material in the entire aging process, the entire aging process including multiple aging stages.

[0039] The optimization module is configured to optimize network parameters of the long short-term memory neural network according to the comprehensive performance index sequence by using a discrete differential evolution method until the long short-term memory neural network reaches a preset convergence condition, thereby obtaining a silicon rubber material performance prediction model.

[0040] The prediction module is configured to utilize the silicon rubber material performance prediction model to predict target comprehensive performance index data of the silicon rubber material to be predicted at each aging stage according to an initial sequence of the comprehensive performance index of the silicon rubber material to be predicted.

[0041] In a third aspect, the present application provides a computer device comprising a processor and a memory, the memory being configured to store a computer program, the computer program being configured to implement the silicon rubber material performance prediction method based on the LSTM neural network according to the first aspect when executed by the processor.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, the computer program being configured to implement the silicon rubber material performance prediction method based on the LSTM neural network according to the first aspect when executed by a processor.

[0043] Compared with the prior art, the present application has at least the following beneficial effects:

[0044] By acquiring the dielectric performance data of the silicone rubber material at multiple aging stages, the dielectric performance data is fused to obtain a comprehensive performance index sequence of the silicone rubber material, so that the long short-term memory neural network can be modeled by using the comprehensive performance index sequence representing the evolution of the electrical performance of the silicone rubber material in the entire aging process, so that the electrical performance characteristics of the silicone rubber material at each aging stage are learned by the aging model; according to the comprehensive performance index sequence, the network parameters of the long short-term memory neural network are optimized by the discrete differential evolution method until the long short-term memory neural network reaches the preset convergence condition, and a silicone rubber material performance prediction model (aging model) is obtained, which effectively avoids the problems of insufficient fitting degree and low prediction accuracy caused by selecting key parameters by experience, and improves the model accuracy; finally, the silicone rubber material performance prediction model is used to predict the target comprehensive performance index data of the to-be-predicted silicone rubber material at each aging stage according to the initial sequence of the comprehensive performance index of the to-be-predicted silicone rubber material, so as to realize the prediction of the aging performance of the silicone rubber material, and provide a detection basis for the life prediction of the silicone rubber material in the dry-type transformer. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 A flowchart of the silicone rubber material performance prediction method based on the LSTM neural network shown in the embodiments of the present application is shown.

[0046] Figure 2 A structure diagram of the LSTM neural network shown in the embodiments of the present application is shown.

[0047] Figure 3 A flowchart of the DDE algorithm shown in the embodiments of the present application is shown.

[0048] Figure 4 A division result diagram of dividing the training set and the test set by the comprehensive performance index sequence shown in the embodiments of the present application is shown.

[0049] Figure 5 A prediction result diagram of the to-be-predicted silicone rubber material shown in the embodiments of the present application is shown.

[0050] Figure 6 A structure diagram of the silicone rubber material performance prediction device based on the LSTM neural network shown in the embodiments of the present application is shown.

[0051] Figure 7 A structure diagram of the computer device shown in the embodiments of the present application is shown. DETAILED DESCRIPTION

[0052] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0053] Please refer to Figure 1 , Figure 1 A flowchart of a silicon rubber material performance prediction method based on an LSTM neural network provided by an embodiment of the present application is shown. The silicon rubber material performance prediction method based on an LSTM neural network of the present embodiment can be applied to a computer device, including but not limited to a smart phone, a notebook computer, a tablet computer, a desktop computer, a physical server, and a cloud server, etc. As shown in Figure 1 , the silicon rubber material performance prediction method based on an LSTM neural network of the present embodiment includes steps S101 to S105, which are described in detail as follows:

[0054] Step S101, obtaining dielectric performance data of the silicon rubber material at multiple aging stages.

[0055] In this step, the entire aging process includes multiple aging stages, and different aging stages of the present embodiment can be regarded as different time points of the aging process. The dielectric performance data includes but is not limited to volume resistivity, relative dielectric constant, and dielectric loss value.

[0056] Optionally, the silicon rubber material is processed into a sheet sample, the surface of the sample is cleaned, the cleaned sample is loaded into a glass container, and the glass container is placed in a vacuum drying oven for drying. Then, the sample is subjected to a heat aging test, and part of the sample is taken out at a certain interval and vacuum stored. The volume resistivity, relative dielectric constant, and dielectric loss value of the taken-out sample are measured.

[0057] Exemplarily, the silicon rubber material is processed into a square sheet sample with a thickness of 1 mm, a length and width of 40x40 mm 2 . The surface of the sample is cleaned with anhydrous ethanol, the cleaned sample is loaded into a glass container, and the glass container is placed in a vacuum drying oven at 120℃ and 50Pa for drying for 2h. Then, the dried sample is subjected to a heat aging test: the environmental temperature is set to 250℃, the longest aging time is set to 125 days, part of the sample is taken out at an interval of 12h and vacuum stored, and a total of n=250 groups of test data under different aging states are obtained. The test data includes volume resistivity, relative dielectric constant, and dielectric loss value at target frequencies within a frequency band of 10 －1 Hz to 10 6 Hz.

[0058] In step S102, the dielectric performance data is fused to obtain a comprehensive performance index sequence of the silicone rubber material, which can represent the evolution of the electrical properties of the silicone rubber material in the entire aging process including multiple aging stages.

[0059] In this step, the comprehensive performance index sequence is obtained by combining the dielectric performance data of the multiple aging dielectrics in sequence of aging time. Optionally, the fusion processing includes data dimension reduction and data fusion. The comprehensive performance index sequence is generated by reducing the dimension of the dielectric performance data of the multiple aging dielectrics and then fusing the reduced dielectric performance data.

[0060] In step S103, the network parameters of the long short-term memory neural network are optimized by the discrete differential evolution method according to the comprehensive performance index sequence until the long short-term memory neural network reaches the preset convergence condition, and a silicone rubber material performance prediction model is obtained.

[0061] In this step, the discrete differential evolution method (DDE) is an algorithm constructed after the rounding processing in the optimization process of the standard differential evolution method (DE). It has the characteristics of less controlled parameters, strong robustness and fast convergence speed.

[0062] Optionally, the comprehensive performance index sequence is decomposed into a training set and a test set. The long short-term memory (LSTM) neural network is trained. The prediction error on the test set is used as the target. The network parameters of the LSTM neural network are optimized by the DDE algorithm. The network parameters include the time window length, the number of hidden layer neurons and the number of hidden layer layers. The finally optimized LSTM neural network is used as the silicone rubber material performance prediction model, i.e. the silicone rubber material aging model. It should be noted that compared with the current setting of network parameters according to user experience, the network parameters are determined by the DDE algorithm in this embodiment, which can improve the model fitting degree and the model prediction accuracy.

[0063] In step S104, the silicone rubber material performance prediction model is used to predict the target comprehensive performance index data of the to-be-predicted silicone rubber material in each aging stage according to the initial comprehensive performance index sequence of the to-be-predicted silicone rubber material.

[0064] In the step, the dielectric performance data of a preset number of samples of the silicon rubber material to be predicted at the current stage is measured, the dielectric performance data is fused into an initial sequence of comprehensive performance indexes, the initial sequence of comprehensive performance indexes is taken as model input of a performance prediction model of the silicon rubber material, target comprehensive performance index data of the silicon rubber material to be predicted at a next aging stage is predicted, the initial sequence of comprehensive performance indexes is updated by using the target comprehensive performance index data, target comprehensive performance index data of a next aging stage is continuously predicted, and the target comprehensive performance index data of the silicon rubber material to be predicted at each aging stage is predicted in turn.

[0065] In some embodiments, the step S102 comprises:

[0066] The volume resistivity, the relative dielectric constant and the dielectric loss value at the target frequency point are taken as performance indexes to construct a sample index matrix.

[0067] The dielectric performance data is fused according to the sample index matrix by using a principal component analysis method to generate the comprehensive performance index sequence.

[0068] In the embodiment, for the silicon rubber material sampled for the ith time, the volume resistivity R Vol , the relative dielectric constant Eps and the dielectric loss value Tanδ at the target frequency point are measured, and a sample index matrix x={x ij} is constructed, i=1, 2, …, p, and j=1, 2, …, p. Then, the dielectric performance data is reduced in dimension and fused according to the sample index matrix by using a principal component analysis method to generate a comprehensive performance index sequence.

[0069] Optionally, the step of fusing the dielectric performance data according to the sample index matrix by using the principal component analysis method to generate the comprehensive performance index sequence comprises:

[0070] The correlation coefficient matrix between the performance indexes is calculated according to the sample index matrix by using the principal component analysis method, and the eigenvalues of the correlation coefficient matrix are calculated.

[0071] A target eigenvector corresponding to a target eigenvalue meeting a component contribution rate condition is determined.

[0072] The dielectric performance data is fused according to the target eigenvector to obtain the comprehensive performance index sequence.

[0073] In the optional embodiment, the sample index matrix is normalized as an example:

[0074]

[0075]

[0076]

[0077] wherein, is the mean of column data in the sample index matrix S j is the standard deviation of column data in the sample index matrix S

[0078] According to the normalized sample index matrix X, the correlation coefficient matrix R={r ij} between each performance index X ij is calculated, and the eigenvalues (λ1, λ2, …, λ p ) and corresponding eigenvectors a1, …, ap of the correlation coefficient matrix R are calculated. p

[0079] The mth component sequence F m =Xa m , m=1, 2, …, p, and the contribution rate b m of the component sequence are calculated.

[0080]

[0081] The component contribution rate condition is that if the maximum value b1>0.85, it is determined that the first principal component sequence F1 basically retains the main information of the original performance index, and the sequence Z=F1 is taken as the comprehensive performance index sequence.

[0082] Optionally, the comprehensive performance index sequence is:

[0083]

[0084] a1=[a R ,a ei ,a ti ] T ;

[0085] wherein, Z represents the comprehensive performance index sequence, a1 is the target eigenvector corresponding to the maximum eigenvalue of the correlation coefficient matrix R, R Vol is the normalized sequence of volume resistivity, Eps i is the normalized sequence of relative permittivity at the ith frequency point, Tanδ i is the normalized sequence of dielectric loss value at the ith frequency point.

[0086] In some embodiments, the step S103 comprises:

[0087] dividing the comprehensive performance index sequence into a training set and a test set;

[0088] ​The long short-term memory neural network is trained by using the training set to obtain a trained target long short-term memory neural network.

[0089] The network parameters of the target long short-term memory neural network are optimized by using the discrete differential evolution method according to the test set until the target long short-term memory neural network reaches the preset convergence condition, so as to obtain the silicone material performance prediction model, and the network parameters include a time window length, a number of hidden layer neurons and a number of hidden layer layers.

[0090] Exemplarily, a network structure of the LSTM neural network is as shown in Figure 2 The LSTM network includes four gate structures, each of which is a neural network including num_layer layers of hidden layers, and each layer of hidden layers includes num_hidden neurons and is fully connected with an input vector [h(t-1), x(t)], h(t-1) is an output at a previous moment, which is a vector with a dimension of num_hidden, and x(t) is an input at a current moment, which is composed of L previous outputs, L being a time window length.

[0091] In the embodiment, the comprehensive performance index sequence is divided into (N-L) groups of samples, 80% of which are used as a training set for network training, and 20% of which are used as a test set for testing, and a division result is as shown in Figure 4 In the test, the output at the previous moment is added to the comprehensive performance index sequence, a sliding time window forms new L inputs, and a root mean squared error (RMSE) of the neural network is defined as:

[0092]

[0093] where T is a number of test set samples, Z i is an i-th comprehensive performance evaluation index in the test set, f(x i ) is a predicted value of the LSTM neural network for an i-th sample input x i .

[0094] The target LSTM neural network trained based on the training set is optimized based on the discrete differential evolution method combined with the root mean squared error.

[0095] Optionally, the network parameters of the target long short-term memory neural network are optimized by using the discrete differential evolution method according to the test set until the target long short-term memory neural network reaches the preset convergence condition, so as to obtain the silicone material performance prediction model, and the network parameters include a time window length, a number of hidden layer neurons and a number of hidden layer layers.

[0096] generate an initial population based on the time window length, the number of hidden layer neurons, and the number of hidden layer layers;

[0097] train the target LSTM neural network with the initial population according to the test set, and calculate individual fitness with the root mean square error;

[0098] perform mutation operation and crossover operation on the initial population to obtain a latest population;

[0099] train the target LSTM neural network with the latest population according to the test set, and calculate the latest individual fitness with the root mean square error;

[0100] if the latest individual fitness is not greater than the individual fitness, perform the next round of mutation operation and crossover operation with the latest population as a new initial population until a termination condition is met, and output an optimal individual;

[0101] update the target LSTM neural network with the optimal individual to obtain the silicone rubber material performance prediction model.

[0102] In the optional embodiment, the network parameters that determine the prediction effect of the LSTM neural network include the number of hidden layer layers num_layer, the number of hidden layer neurons num_hidden, and the time window length L. In order to ensure the prediction effect of the model, the DDE algorithm is used to optimize the three network parameters of the LSTM neural network with the RMSE as the optimization target.

[0103] As shown in Figure 3 , the ranges of num_hidden, num_layer, and L are set, and the maximum number of iterations t max , the population size N p , the mutation operator F0, and the crossover operator CR are set.

[0104] generate an initial population, train the LSTM neural network with the initial population, and calculate the individual fitness RSME(X(i)) based on the above root mean square error formula;

[0105] The mutation operation is:

[0106] V = X(1) + floor{F0(X(2) - X(3))};

[0107] wherein X(1), X(2), and X(3) are three different individuals in the population X, and floor represents rounding down;

[0108] The crossover operation is: replace the individuals in the population X with the individuals V in the mutation population with the crossover operator CR, handle the out-of-bound individuals in V in a boundary absorption manner to obtain the latest population V(i).

[0109] The LSTM neural network is trained using the latest population V(i), and the latest individual fitness RSME(V(i)) is calculated based on the above root mean square error formula;

[0110] If RSME(V(i))≤RSME(X(i)), the latest population V(i) is taken as the new initial population X(i) for the next mutation operation, crossover operation and selection operation, until the termination condition is met, and the optimal individual is output.

[0111] The termination condition is RSME(V(i))≤preset value, or the iteration number reaches t max The optimal individual determined after DDE optimization is the parameter of the LSTM model, num_layer=2, num_hidden=8, and L=3.

[0112] In some embodiments, the step S104 comprises:

[0113] The initial sequence of the comprehensive performance indicators is taken as the model input of the silicone rubber material performance prediction model, and the target comprehensive performance indicator data of the to-be-predicted silicone rubber material at the next aging stage is predicted.

[0114] The target comprehensive performance indicator data is inserted into the initial sequence of the comprehensive performance indicators to obtain a new initial sequence of the comprehensive performance indicators.

[0115] The new initial sequence of the comprehensive performance indicators is used to continue to predict the target comprehensive performance indicator data of the next aging stage until the target comprehensive performance indicator data is lower than the preset threshold, and the remaining life of the to-be-predicted silicone rubber material is predicted according to the iteration number experienced when the target comprehensive performance indicator data is lower than the preset threshold.

[0116] In this embodiment, for example, three consecutive volume resistivities, dielectric constants and dielectric losses of the to-be-predicted silicone rubber material are obtained, and the three dielectric performance data are taken as the initial values of the initial sequence Z of the comprehensive performance indicators, and the threshold Zend of the comprehensive performance indicators is calculated according to the dielectric performance data when the material fails. The LSTM neural network takes the three initial values of the initial sequence Z of the comprehensive performance indicators as the input, predicts the target comprehensive performance indicator data at the next time, inserts the target comprehensive performance indicator data into the last one of the original sequence Z, deletes the first one, and keeps the sample amount of the original sequence unchanged, realizes the dynamic updating of the sequence of the comprehensive performance indicators, and estimates the remaining life of the to-be-predicted silicone rubber material according to the iteration number experienced when the comprehensive performance indicators are lower than the threshold Zend.

[0117] For example, the prediction result is as shown in Figure 5 Figure 5 ​It can be known that after 36 iterations, the index performance of the to-be-predicted silicone rubber material is lower than the threshold Zend, so the remaining life of the to-be-predicted silicone rubber material is 36*12 hours = 432 hours.

[0118] In order to perform the silicone rubber material performance prediction method based on the LSTM neural network corresponding to the above-mentioned method embodiment, the corresponding functions and technical effects are realized. Referring to Figure 7 , Figure 7 A structural block diagram of a silicone rubber material performance prediction device based on an LSTM neural network provided by an embodiment of the present application is shown. For ease of illustration, only the parts related to the present embodiment are shown. The silicone rubber material performance prediction device based on the LSTM neural network provided by the present embodiment comprises:

[0119] The acquisition module 601 is configured to acquire dielectric performance data of a silicone rubber material at multiple aging stages.

[0120] The fusion module 602 is configured to perform fusion processing on the dielectric performance data to obtain a comprehensive performance index sequence of the silicone rubber material, the comprehensive performance index sequence being capable of representing the evolution of the electrical performance of the silicone rubber material throughout the entire aging process, the entire aging process including multiple aging stages.

[0121] The optimization module 603 is configured to optimize network parameters of a long short-term memory neural network according to the comprehensive performance index sequence by using a discrete differential evolution method until the long short-term memory neural network reaches a preset convergence condition, thereby obtaining a silicone rubber material performance prediction model.

[0122] The prediction module 604 is configured to use the silicone rubber material performance prediction model to predict target comprehensive performance index data of a to-be-predicted silicone rubber material at each aging stage according to an initial comprehensive performance index sequence of the to-be-predicted silicone rubber material.

[0123] In some embodiments, the dielectric performance data includes volume resistivity, relative dielectric constant, and dielectric loss value, and the fusion module 602 comprises:

[0124] The construction unit is configured to use the volume resistivity, the relative dielectric constant at the target frequency point, and the dielectric loss value as performance indexes to construct a sample index matrix.

[0125] The processing unit is configured to use a principal component analysis method to perform fusion processing on the dielectric performance data according to the sample index matrix, thereby generating the comprehensive performance index sequence.

[0126] In some embodiments, the processing unit is specifically configured to:

[0127] The principal component analysis method is used to calculate a correlation coefficient matrix among the performance indexes according to the sample index matrix, and to calculate eigenvalues of the correlation coefficient matrix;

[0128] A target eigenvector corresponding to a target eigenvalue meeting a component contribution rate condition is determined.

[0129] The dielectric performance data are fused according to the target eigenvector to obtain the comprehensive performance index sequence.

[0130] In some embodiments, the comprehensive performance index sequence is:

[0131]

[0132] a1=[a R ,a ei ,a ti ] T ;

[0133] wherein Z represents the comprehensive performance index sequence, a1 is the target eigenvector, R Vol is a normalized sequence of volume resistivity, Eps i is a normalized sequence of relative permittivity at the i-th frequency point, Tanδ i is a normalized sequence of dielectric loss value at the i-th frequency point.

[0134] In some embodiments, the optimization module 603 comprises:

[0135] A division unit is configured to divide the comprehensive performance index sequence into a training set and a test set.

[0136] A training unit is configured to train the long short-term memory neural network by using the training set to obtain a trained target long short-term memory neural network.

[0137] An optimization unit is configured to optimize network parameters of the target long short-term memory neural network by using the discrete differential evolution method according to the test set until the target long short-term memory neural network reaches a preset convergence condition, to obtain the silicon rubber material performance prediction model, wherein the network parameters include a time window length, a number of hidden layer neurons, and a number of hidden layer layers.

[0138] In some embodiments, the optimization unit is specifically configured to:

[0139] generate an initial population based on the time window length, the number of hidden layer neurons, and the number of hidden layer layers;

[0140] According to the test set, the target long short-term memory neural network is trained by using the initial population, and individual fitness is calculated by using a root mean square error;

[0141] The initial population is subjected to mutation operation and crossover operation to obtain a latest population;

[0142] According to the test set, the target long short-term memory neural network is trained by using the latest population, and latest individual fitness is calculated by using a root mean square error;

[0143] If the latest individual fitness is not greater than the individual fitness, the next round of mutation operation and crossover operation are performed by using the latest population as a new initial population, until a termination condition is met, and an optimal individual is output;

[0144] The target long short-term memory neural network is updated by using the optimal individual, and the silicone rubber material performance prediction model is obtained.

[0145] In some embodiments, the prediction module 604 is specifically configured to:

[0146] The initial sequence of the comprehensive performance indicators is taken as model input of the silicone rubber material performance prediction model, and target comprehensive performance indicator data of the to-be-predicted silicone rubber material at a next aging stage is predicted;

[0147] The target comprehensive performance indicator data is inserted into the initial sequence of the comprehensive performance indicators to obtain a new initial sequence of the comprehensive performance indicators;

[0148] The new initial sequence of the comprehensive performance indicators is taken to continue to predict the target comprehensive performance indicator data of the next aging stage, until the target comprehensive performance indicator data is lower than a preset threshold, and a remaining life of the to-be-predicted silicone rubber material is predicted according to an iteration number experienced when the target comprehensive performance indicator data is lower than the preset threshold.

[0149] The above-described silicone rubber material performance prediction device based on an LSTM neural network can implement the above-described method embodiment of the silicone rubber material performance prediction method based on an LSTM neural network. The optional items in the above-described method embodiment are also applicable to the present embodiment, and will not be described in detail herein. The remaining content of the present embodiment can be referred to the content of the above-described method embodiment, and will not be described in detail herein.

[0150] Figure 7 A structural schematic diagram of a computer device is provided in an embodiment of the present application. As shown in the figure, the computer device 7 of this embodiment includes at least one processor 70 (for example, a central processing unit (CPU), a microprocessor, or the like), at least one memory 71 (for example, a random access memory (RAM), a read-only memory (ROM), or the like), and a bus 72 (for example, a motherboard, a bus line, or the like) electrically connecting the processor 70 and the memory 71. Figure 7 The processor 70 is configured to execute various functions of the computer device 7 by running or executing various programs or codes stored in the memory 71. Figure 7Only one is shown in the figure) a processor, a memory 71, and a computer program 72 stored in the memory 71 and executable on the at least one processor 70, wherein the processor 70 implements the steps of any of the above method embodiments when executing the computer program 72.

[0151] The computer device 7 may be a computing device such as a smart phone, a tablet computer, a desktop computer, a cloud server, etc. The computer device may include but is not limited to a processor 70 and a memory 71. It will be understood by those skilled in the art that Figure 7 This is merely an example of the computer device 7 and does not constitute a limitation on the computer device 7 . The computer device 7 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0152] The processor 70 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0153] In some embodiments, the memory 71 may be an internal storage unit of the computer device 7, such as a hard disk or memory of the computer device 7. In other embodiments, the memory 71 may also be an external storage device of the computer device 7, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 7. Furthermore, the memory 71 may include both an internal storage unit of the computer device 7 and an external storage device. The memory 71 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 71 may also be used to temporarily store data that has been output or is about to be output.

[0154] In addition, the embodiment of the present application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps in any method embodiment described above.

[0155] The embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device is caused to execute the steps in each method embodiment described above.

[0156] In several embodiments provided in the present application, it can be understood that each block in the flowchart or block diagram can represent a module, a segment or a portion of code which includes one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in a different order than that noted in the figure. For example, two blocks noted in succession can actually be executed substantially concurrently or in the opposite order, depending on the functionality involved.

[0157] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0158] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are merely for the purpose of the present application and do not limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting the performance of a silicone rubber material based on an LSTM neural network, characterized by, The method comprises the following steps: acquiring dielectric property data of a silicone rubber material at multiple aging stages; performing fusion processing on the dielectric property data to obtain a comprehensive performance index sequence of the silicone rubber material, the comprehensive performance index sequence being capable of representing the evolution of the electrical properties of the silicone rubber material throughout the entire aging process, the entire aging process comprising multiple aging stages; wherein the dielectric property data comprises volume resistivity, relative dielectric constant, and dielectric loss value; the fusion processing on the dielectric property data to obtain the comprehensive performance index sequence comprises: taking the volume resistivity, the relative dielectric constant at a target frequency point, and the dielectric loss value as performance indexes, and constructing a sample index matrix; using principal component analysis to perform fusion processing on the dielectric property data according to the sample index matrix to generate the comprehensive performance index sequence; optimizing network parameters of a long short-term memory neural network according to the comprehensive performance index sequence by using a discrete differential evolution method until the long short-term memory neural network reaches a preset convergence condition, thereby obtaining a silicone rubber material performance prediction model; using the silicone rubber material performance prediction model to predict target comprehensive performance index data of a to-be-predicted silicone rubber material at each aging stage according to an initial comprehensive performance index sequence of the to-be-predicted silicone rubber material.

2. The LSTM neural network based silicone material property prediction method of claim 1, wherein, The fusion processing on the dielectric property data to obtain the comprehensive performance index sequence by using principal component analysis according to the sample index matrix comprises: using principal component analysis to calculate a correlation coefficient matrix among the performance indexes and eigenvalues of the correlation coefficient matrix according to the sample index matrix; determining a target eigenvector corresponding to a target eigenvalue that meets a component contribution rate condition; performing fusion processing on the dielectric property data according to the target eigenvector to obtain the comprehensive performance index sequence.

3. The method for predicting silicone rubber material properties based on LSTM neural network according to claim 2, characterized in that: The comprehensive performance index sequence is: a1 = [a R ,a ei ,a ti ] T ; Wherein, Z represents a comprehensive performance index sequence, a1 is a target feature vector, R Vol is a normalized sequence of volume resistivity, Eps i is a normalized sequence of relative permittivity at the i-th frequency point, Tanδ i is a normalized sequence of dielectric loss value at the i-th frequency point.

4. The LSTM neural network based prediction method of silicone rubber material properties as claimed in claim 1, wherein, The optimization of the network parameters of the long short-term memory neural network according to the comprehensive performance index sequence by using the discrete differential evolution method until the long short-term memory neural network reaches the preset convergence condition, thereby obtaining the silicone rubber material performance prediction model, comprises: dividing the comprehensive performance index sequence into a training set and a test set; training the long short-term memory neural network by using the training set to obtain a target long short-term memory neural network after training; optimizing the network parameters of the target long short-term memory neural network according to the test set by using the discrete differential evolution method until the target long short-term memory neural network reaches the preset convergence condition, thereby obtaining the silicone rubber material performance prediction model, the network parameters comprising a time window length, a number of hidden layer neurons, and a number of hidden layers.

5. The LSTM neural network-based prediction method of silicone rubber material properties according to claim 4, wherein, The optimization of the network parameters of the target long short-term memory neural network according to the test set by using the discrete differential evolution method until the target long short-term memory neural network reaches the preset convergence condition, thereby obtaining the silicone rubber material performance prediction model, comprises: generating an initial population based on the time window length, the number of hidden layer neurons, and the number of hidden layer layers; training the target long short-term memory neural network with the initial population according to the test set, and calculating individual fitness with root mean square error; performing mutation operation and crossover operation on the initial population to obtain a latest population; training the target long short-term memory neural network with the latest population according to the test set, and calculating latest individual fitness with root mean square error; if the latest individual fitness is not greater than the individual fitness, performing next round of mutation operation and crossover operation with the latest population as a new initial population until a termination condition is met, and outputting an optimal individual; updating the target long short-term memory neural network with the optimal individual to obtain the silicone rubber material performance prediction model.

6. The LSTM neural network based prediction of silicone rubber material properties method of claim 1, wherein, The silicone rubber material performance prediction model is used to predict target comprehensive performance index data of a to-be-predicted silicone rubber material at each aging stage according to an initial sequence of comprehensive performance indexes of the to-be-predicted silicone rubber material, including: using the initial sequence of comprehensive performance indexes as model input of the silicone rubber material performance prediction model to predict target comprehensive performance index data of the to-be-predicted silicone rubber material at a next aging stage; inserting the target comprehensive performance index data into the initial sequence of comprehensive performance indexes to obtain a new initial sequence of comprehensive performance indexes; continuing to predict target comprehensive performance index data of a next aging stage with the new initial sequence of comprehensive performance indexes until the target comprehensive performance index data is lower than a preset threshold, and predicting a remaining life of the to-be-predicted silicone rubber material according to an iteration number experienced when the target comprehensive performance index data is lower than the preset threshold.

7. A device for predicting the performance of a silicone rubber material based on an LSTM neural network, characterized by including: an acquisition module configured to acquire dielectric performance data of a silicone rubber material at multiple aging stages; a fusion module configured to perform fusion processing on the dielectric performance data to obtain a comprehensive performance index sequence of the silicone rubber material, the comprehensive performance index sequence being capable of representing electrical performance evolution of the silicone rubber material in an entire aging process including the multiple aging stages; wherein the dielectric performance data includes volume resistivity, relative dielectric constant, and dielectric loss value, and the fusion processing on the dielectric performance data to obtain the comprehensive performance index sequence includes: taking the volume resistivity, the relative dielectric constant at a target frequency point, and the dielectric loss value as performance indexes to construct a sample index matrix; and using principal component analysis to perform fusion processing on the dielectric performance data according to the sample index matrix to generate the comprehensive performance index sequence; an optimization module configured to optimize network parameters of a long short-term memory neural network with discrete differential evolution method according to the comprehensive performance index sequence until the long short-term memory neural network reaches a preset convergence condition to obtain a silicone rubber material performance prediction model. A prediction module is configured to utilize the silicone rubber material performance prediction model to predict target comprehensive performance index data of the to-be-predicted silicone rubber material at each aging stage according to an initial sequence of comprehensive performance indexes of the to-be-predicted silicone rubber material.

8. A computer device, comprising: The method comprises a processor and a memory, and the memory is configured to store a computer program, and the computer program is executed by the processor to implement the LSTM neural network-based silicone rubber material performance prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed by the processor to implement the LSTM neural network-based silicone rubber material performance prediction method according to any one of claims 1 to 6.

Images