Model generation method and apparatus, computer device, and computer-readable storage medium

Abstract

The application discloses a model generation method and device, computer equipment and a computer readable storage medium. The method comprises: acquiring a plurality of panel parameters; the panel parameters comprise a basic value; determining a first training data set based on the plurality of panel parameters and the predetermined importance of each panel parameter; determining a target training data set based on the predetermined linear analysis result of each panel parameter and the first training data set; and training a basic model based on the target training data set to generate a target model. The application can reduce the acquisition time of training data and alleviate the problem of data space sparsity of training data.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a model generation method, apparatus, computer device, and computer-readable storage medium. Background Technology

[0002] In the design process of display panels, the simulation characteristics of the display panel determine its performance, which in turn are determined by the panel layout parameters. To obtain optimal panel performance, it is necessary to explore the panel's parameter space. Artificial intelligence (AI) algorithms are widely used in electronic design automation (EDA) of display panels because they can accelerate the search speed of the parameter space. However, in existing AI-based EDA designs, a large amount of training data is required for each panel product to generate a simulation result prediction model corresponding to each panel product, resulting in long training data acquisition times.

[0003] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0004] This application provides a model generation method, apparatus, computer device, and computer-readable storage medium, aiming to solve the problem that existing designs for each panel product require a large amount of training data and have long training data acquisition times.

[0005] The technical solution adopted by this invention to solve the problem is as follows:

[0006] On the one hand, this application provides a model generation method, including:

[0007] Retrieve multiple panel parameters; panel parameters include basic values;

[0008] The first training dataset is determined based on multiple panel parameters and the pre-determined importance of each panel parameter.

[0009] Based on the linear analysis results of each pre-determined panel parameter and the first training dataset, the target training dataset is determined; wherein, the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results;

[0010] The base model is trained based on the target training dataset to generate the target model.

[0011] In some embodiments of this application, a first training dataset is determined based on multiple panel parameters and a pre-determined importance level for each panel parameter, including:

[0012] Based on multiple panel parameters and the pre-determined importance of each panel parameter, several first target parameters are determined. These first target parameters are the parameters whose importance is ranked from high to low among the multiple panel parameters.

[0013] By taking boundary values ​​for several first target parameters, multiple first boundary values ​​are obtained for each first target parameter.

[0014] The first training dataset is determined based on multiple first boundary values ​​and basic values.

[0015] In some embodiments of this application, before determining the first training dataset based on multiple panel parameters and the predetermined importance of each panel parameter, the method further includes:

[0016] Based on the base value, boundary values ​​are taken for each panel parameter to obtain multiple second boundary values ​​for each panel parameter;

[0017] Multiple sets of first parameters are determined based on multiple second boundary values ​​and basic values;

[0018] Simulations were performed on multiple sets of first parameters to obtain the first simulation results for each set of first parameters.

[0019] Based on multiple sets of first parameters and the first simulation results, the importance of each panel parameter is determined.

[0020] In some embodiments of this application, the second simulation results include multidimensional simulation results;

[0021] Based on multiple sets of first parameters and the first simulation results, the importance of each panel parameter is determined, including:

[0022] For any panel parameter, select multiple first parameter groups from multiple first parameter groups where other panel parameters remain unchanged except for that panel parameter as multiple first target parameter groups;

[0023] Based on multiple first simulation results and multiple first target parameter sets, multiple first target simulation results are determined; each first target simulation result corresponds to a first target parameter set.

[0024] Based on multiple first target parameter groups, determine the maximum value and minimum value of the panel parameter of the panel parameter;

[0025] For any simulation result of the first objective, determine the maximum and minimum values ​​of the simulation result from the multidimensional simulation results of the first objective simulation result;

[0026] Based on the maximum value of the panel parameters, the minimum value of the panel parameters, the maximum value of the simulation results, and the minimum value of the simulation results, the influence coefficient corresponding to the simulation result of the first objective is determined;

[0027] Based on all the influence coefficients of the panel parameter, determine the importance of the panel parameter.

[0028] In some embodiments of this application, before determining the target training dataset based on the linear analysis results of each pre-determined panel parameter and the first training dataset, the method further includes:

[0029] Based on multiple panel parameters and their importance, several second objective parameters are determined. These second objective parameters are the parameters whose importance is ranked from high to low among the multiple panel parameters.

[0030] By taking boundary values ​​for several second objective parameters, multiple third boundary values ​​are obtained for each of the several second objective parameters;

[0031] A second training dataset is determined based on multiple third boundary values ​​and basic values; wherein, the second training dataset includes multiple second parameter groups and the second simulation results of each of the multiple second parameter groups;

[0032] Based on multiple sets of second parameters and the second simulation results, the linear analysis results for each panel parameter are determined.

[0033] In some embodiments of this application, before training the base model based on the target training dataset to generate the target model, the method further includes:

[0034] Based on the linear analysis results, the second training dataset is augmented to obtain the third training dataset; the third training dataset includes multiple third parameter groups and the third simulation results of each of the multiple third parameter groups;

[0035] Input multiple third parameter groups from the third training dataset into a preset network model, and output the first prediction result for each third parameter group through the preset network model;

[0036] The basic model is obtained by training the preset network model based on the third simulation results, the first prediction results, and the loss function of the preset network model.

[0037] In some embodiments of this application, after training the base model based on the target training dataset to generate the target model, the method further includes:

[0038] The pre-constructed panel parameter space is input into the target model, and multiple fourth simulation results of the panel parameter space are output through the target model.

[0039] Based on multiple fourth simulation results and preset second target simulation results, several second target parameter sets are searched from the panel parameter space.

[0040] Secondly, embodiments of the present invention also provide a model generation apparatus, comprising:

[0041] The parameter acquisition unit is used to acquire multiple panel parameters; panel parameters include basic values.

[0042] The first determining unit is used to determine the first training dataset based on multiple panel parameters and the pre-determined importance of each panel parameter.

[0043] The second determining unit is used to determine the target training dataset based on the linear analysis results of each pre-determined panel parameter and the first training dataset; wherein the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results;

[0044] The model generation unit is used to train the base model based on the target training dataset and generate the target model.

[0045] Thirdly, this application also provides a computer device, which includes:

[0046] One or more processors;

[0047] Memory; and

[0048] One or more applications, wherein the applications are stored in memory and configured to be executed by a processor to implement the model generation method of any of the first aspects.

[0049] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps in the model generation method of any of the first aspects.

[0050] The beneficial effects of this invention are as follows: the target model is generated based on the basic model, and only a small amount of training data is needed to achieve the effect of the model trained on large-scale data, which greatly saves the time cost of acquiring training data. The training dataset is constructed based on the importance of each panel parameter and the results of linear analysis. While ensuring the reliability of the training dataset, the acquisition time of training data can be further reduced, and the problem of data sparsity in the training data space can be alleviated. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating the model generation method provided in an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram showing the order of importance of multiple panel parameters provided in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram showing a linear relationship between the panel parameters and simulation results provided in this embodiment of the invention;

[0055] Figure 4 This is a schematic diagram showing a basic linear relationship between the panel parameters and simulation results provided in the embodiments of the present invention;

[0056] Figure 5 This is a schematic diagram illustrating the nonlinear relationship between panel parameters and simulation results provided in this embodiment of the invention;

[0057] Figure 6 This is a schematic diagram of the structure of the target model provided in the embodiment of the present invention;

[0058] Figure 7 This is a schematic block diagram of the model generation device provided in the embodiments of the present invention;

[0059] Figure 8 This is a schematic diagram of an embodiment of the computer device provided in this invention. Detailed Implementation

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first," "second," "third," "fourth," "fifth," and "sixth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," "third," "fourth," "fifth," and "sixth" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, and "several" means one or more, unless otherwise explicitly specified.

[0062] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0063] It should be noted that since the method in this application embodiment is executed in a computer device, the processing objects of each computer device exist in the form of data or information, such as time, which is essentially time information. It is understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data that exist so that the computer device can process them. Specific details will not be elaborated here.

[0064] Existing methods for exploring display panel performance include traditional manual methods, black-box optimization methods, and artificial intelligence (AI)-based methods. The traditional manual method involves parameterizing the panel layout into partitions, manually initializing a set of TFT parameters based on experience, inputting these parameters into simulation software to obtain simulation results, and then manually adjusting the parameters based on experience, repeating the simulation iteration to find satisfactory TFT parameters. This method suffers from long simulation times, difficulty in finding the global optimum within limited working time, and requires significant expertise in parameter tuning. Black-box optimization methods include Bayesian optimization algorithms, genetic algorithms, and particle swarm optimization. While existing black-box algorithms can maximize display performance with relatively low computational cost, they still require online simulation, making them unsuitable for industrial projects with long simulation data acquisition times. Existing AI-based methods use convolutional neural networks (CNNs) to replace traditional simulation software, reducing simulation result acquisition time to a few seconds. However, current AI-based EDA designs require a large amount of training data to generate a simulation result prediction model for each panel product, resulting in long training data acquisition times.

[0065] Based on this, in this embodiment, multiple panel parameters are obtained; the panel parameters include basic values; a first training dataset is determined based on the multiple panel parameters and the predetermined importance of each panel parameter; a target training dataset is determined based on the linear analysis results of each predetermined panel parameter and the first training dataset; wherein, the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results; the base model is trained based on the target training dataset to generate the target model. This application generates the target model based on the base model, requiring only a small amount of training data to achieve the model effect trained on large-scale data, greatly saving the time cost of acquiring training data. Constructing the training dataset based on the importance of each panel parameter and the linear analysis results ensures the reliability of the training dataset while further reducing the time for acquiring training data and alleviating the problem of data sparsity in the training data space.

[0066] The content of this application will be further explained below with reference to the accompanying drawings and the description of the embodiments.

[0067] This embodiment provides a model generation method, such as Figure 1 As shown, the method includes:

[0068] Step S10: Obtain multiple panel parameters; panel parameters include basic values.

[0069] Specifically, panel parameters are the basic parameters of a panel product. These parameters can be obtained through parameterization using automated software. For example, multiple panel parameters may include 12-dimensional T-series parameters such as T11, T21, T22, T31, T32T33, T41, T42T43, T44, T51T61, T52T62, T53T63, and T54T64, as well as 1-dimensional CBoost parameters. Panel parameters include base values, which are the parameter values ​​of each panel parameter obtained by initializing multiple panel parameters. Adjusting the parameter value of each panel parameter can affect the basic specifications of the panel product, such as border width, charging time, and voltage.

[0070] In order to generate a simulation result prediction model for the panel product, this embodiment first obtains multiple panel parameters of the panel product so that the training dataset for generating the model can be determined based on the multiple panel parameters in subsequent steps.

[0071] Step S20: Determine the first training dataset based on multiple panel parameters and the pre-determined importance of each panel parameter.

[0072] Importance is used to characterize the magnitude of the influence of each panel parameter on the simulation results; that is, the higher the importance of a panel parameter, the greater its influence on the simulation results, and the lower the importance of a panel parameter, the smaller its influence on the simulation results. Figure 2 The diagram illustrates the importance ranking of the 12-dimensional T-series parameters and the 1-dimensional CBoost parameters. Considering that directly constructing a training dataset using multiple panel parameters results in a large data volume and a long training data acquisition time, this embodiment pre-determines the importance of each panel parameter. After acquiring multiple panel parameters, a first training dataset is determined based on the multiple panel parameters and the importance of each panel parameter. Determining the first training dataset based on the importance of each panel parameter reduces the training data acquisition time while ensuring the reliability of the training data.

[0073] In one specific embodiment, step S20 includes:

[0074] Step S21: Based on multiple panel parameters and the pre-determined importance of each panel parameter, determine several first target parameters. The several first target parameters are the parameters with the highest importance among the multiple panel parameters, sorted from high to low.

[0075] Step S22: Perform boundary value taking on several first target parameters respectively to obtain multiple first boundary values ​​for each first target parameter among several first target parameters;

[0076] Step S23: Determine the first training dataset based on multiple first boundary values ​​and basic values.

[0077] The first target parameters are the parameters ranked highest in importance from lowest to highest among the multiple panel parameters. The number of first target parameters can be set as needed. For example, when there are 6 first target parameters, they are the 6 panel parameters ranked highest in importance from lowest to highest. The multiple first boundary values ​​are the expanded parameter values ​​of each first target parameter obtained by taking boundary values ​​for each first target parameter according to a preset first step length. The preset first step length can be set as needed; for example, the first step length can be 20%.

[0078] In this embodiment, when determining the first training dataset based on multiple panel parameters and the pre-determined importance of each panel parameter, several first target parameters are first determined based on the multiple panel parameters and the pre-determined importance of each panel parameter. Then, boundary values ​​are taken for each of the several first target parameters according to a preset first step length, resulting in multiple first boundary values ​​for each of the several first target parameters. Finally, the first training dataset is determined based on the multiple first boundary values ​​and the base values. Because several first target parameters with higher importance are selected for value taking, instead of taking values ​​for all panel parameters, the time for acquiring training data is reduced, while ensuring the reliability of the training data.

[0079] The first training dataset includes multiple fourth parameter sets and a fifth simulation result for each fourth parameter set. The multiple fourth parameter sets are panel parameter combinations obtained by combining parameters from multiple first boundary values ​​and base values. The dimension of the fourth parameter set is the same as the number of panel parameters; for example, when the number of panel parameters is 13, the fourth parameter set is a 13-dimensional panel parameter combination. The fifth simulation result is the simulation result obtained by simulating the fourth parameter sets using simulation software.

[0080] In this embodiment, when determining the first training sample set based on multiple first boundary values ​​and base values, the parameters of the multiple first boundary values ​​and base values ​​are first combined to obtain multiple fourth parameter groups. Then, simulation software is used to simulate each of the multiple fourth parameter groups to obtain a fifth simulation result for each fourth parameter group. For example, when the multiple panel parameters are 12-dimensional T-series parameters and 1-dimensional CBoost parameters, boundary values ​​are taken for the top 6 panel parameters with higher importance, and the parameters of the boundary values ​​and base values ​​are combined to obtain 162 fourth parameter groups. Then, simulation is performed on each of the 162 fourth parameter groups to obtain 162 fifth simulation results.

[0081] In one specific embodiment, before step S20, the method further includes:

[0082] Step M11: Take values ​​for each panel parameter based on the base values ​​to obtain multiple second boundary values ​​for each panel parameter;

[0083] Step M12: Determine multiple sets of first parameters based on multiple second boundary values ​​and basic values;

[0084] Step M13: Perform simulations on multiple first parameter groups respectively to obtain the first simulation results for each first parameter group in the multiple first parameter groups;

[0085] Step M14: Based on multiple first parameter groups and the first simulation results, determine the importance of each panel parameter.

[0086] Multiple second boundary values ​​are extended parameter values ​​for each panel parameter obtained by taking values ​​for the base values ​​and a preset second step size. The second step size can be set as needed; for example, the second step size can be 5%. Multiple first parameter groups are panel parameter combinations obtained by combining multiple second boundary values ​​and base values. The dimension of the first parameter group is the same as the number of panel parameters. For example, when the number of panel parameters is 13, the first parameter group is a 13-dimensional panel parameter combination. The first simulation result is the simulation result obtained by simulating the first parameter group using simulation software.

[0087] To determine the importance of each panel parameter, this embodiment first assigns values ​​to each panel parameter based on the base value and a preset second step size before determining the first training dataset, obtaining multiple second boundary values ​​for each panel parameter. Then, the multiple second boundary values ​​and the base value are combined to obtain multiple first parameter groups. Next, simulation software is used to simulate the multiple first parameter groups respectively to obtain the first simulation result of each first parameter group. Finally, based on the multiple first parameter groups and the first simulation results, the importance of each panel parameter is determined.

[0088] For example, multiple panel parameters are 12-dimensional T-series parameters and 1-dimensional CBoost parameters. The 13 panel parameters are selected with a step size of 5% within ±20% of the base value. Then, the second boundary value and the base value obtained after selection are combined to obtain 100 different first parameter groups. The 100 first parameter groups are then simulated separately. Based on the 100 first parameter groups and the first simulation results of each first parameter group, the importance of each panel parameter is determined.

[0089] In one specific implementation, the first simulation result includes multi-dimensional simulation results. Step M14 includes:

[0090] Step M141: For any panel parameter, select multiple first parameter groups from multiple first parameter groups where other panel parameters besides that panel parameter remain unchanged as multiple first target parameter groups;

[0091] Step M142: Based on multiple first simulation results and multiple first target parameter sets, determine multiple first target simulation results; each first target simulation result corresponds to a first target parameter set;

[0092] Step M143: Based on multiple first target parameter groups, determine the maximum value and minimum value of the panel parameter of the panel parameter;

[0093] Step M144: For any simulation result of the first target, determine the maximum value and minimum value of the simulation result from the multidimensional simulation results of the first target simulation result;

[0094] Step M145: Based on the maximum value of the panel parameters, the minimum value of the panel parameters, the maximum value of the simulation results, and the minimum value of the simulation results, determine the influence coefficient corresponding to the simulation result of the first target.

[0095] Step M146: Determine the importance of the panel parameter based on all its influence coefficients.

[0096] The maximum value of each panel parameter is the maximum value among multiple combinations of first parameters, and the minimum value of each panel parameter is the minimum value among multiple combinations of first parameters. The first simulation result includes multi-dimensional simulation results; for example, when the first parameter group is a 13-dimensional panel parameter combination, the first simulation result includes 31-dimensional simulation results. Each dimension of the simulation result contains multiple simulation values, and the maximum value of each dimension is the maximum value among the multiple simulation values, and the minimum value is the minimum value among the multiple simulation values. The influence coefficient is used to characterize the magnitude of the influence of each panel parameter on each dimension of the simulation result.

[0097] In this embodiment, when determining the importance of each panel parameter based on multiple first parameter groups and first simulation results, firstly, for any panel parameter, multiple first parameter groups are selected from the multiple first parameter groups where other panel parameters remain unchanged, as multiple first target parameter groups. Then, based on the multiple first simulation results and multiple first target parameter groups, multiple first target simulation results are determined. Next, based on the multiple first target parameter groups, the maximum and minimum values ​​of the panel parameter are determined. Simultaneously, for any first target simulation result, the maximum and minimum values ​​of the simulation result are determined from the multidimensional simulation results of the first target simulation result. Then, based on the maximum and minimum values ​​of the panel parameter, the maximum and minimum values ​​of the simulation result, the influence coefficient corresponding to the first target simulation result is determined. The formula for calculating the influence coefficient is as follows: Among them, Importance(x,y) i ) represents the simulation result y of panel parameter x in the i-th dimension.i The influence coefficient in The i-th dimension simulation result y represents i The maximum value in, The i-th dimension simulation result y represents i The minimum value of x max This represents the maximum value of the panel parameter x. min This represents the minimum value of the panel parameter x.

[0098] The importance of each panel parameter is the weighted sum of its influence coefficients in each dimension of the simulation result. After determining the influence coefficients corresponding to the first target simulation result, the importance of each panel parameter can be determined based on these coefficients. The formula for calculating the importance of each panel parameter is as follows: Among them, I x The importance of panel parameter x is indicated by n, where n represents the number of dimensions in the second simulation result. i ) represents the simulation result y of panel parameter x in the i-th dimension. i The influence coefficient in.

[0099] Step S30: Determine the target training dataset based on the linear analysis results of each pre-determined panel parameter and the first training dataset; wherein, the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results.

[0100] The linear analysis results for each panel parameter are used to characterize the linear correlation between each panel parameter and the simulation results. The linear analysis results include linear relationships, basic linear relationships, and nonlinear relationships. A linear relationship means that the parameter values ​​of the panel parameters and the simulation results are linearly related; a basic linear relationship means that the parameter values ​​of the panel parameters and the simulation results are basically linearly related; and a nonlinear relationship means that the parameter values ​​of the panel parameters and the simulation results are nonlinearly related. Figures 3-5 As shown, the CBoost parameters have a linear relationship with the simulation results, the T11 parameters have a basically linear relationship with the simulation results, and the T32 and T33 parameters have a nonlinear relationship with the simulation results.

[0101] The second training dataset is an augmented dataset obtained by augmenting the first training dataset based on the linear analysis results of each panel parameter. To address the data sparsity problem of the first training dataset, this embodiment, after determining the first training dataset, augments it based on the pre-determined linear analysis results of each panel parameter to obtain the target training dataset. The target training dataset effectively alleviates the data spatial sparsity problem of the first training dataset, and augmenting based on the linear analysis results reduces the time required to acquire training data. For example, 162 sets of training data can be augmented to 6250 sets of training data based on the linear analysis results.

[0102] Considering that only panel parameters whose linear analysis results show a linear or near-linear relationship can be augmented based on linear relationships, this embodiment determines the target training dataset based on the linear analysis results of each pre-determined panel parameter and the first training dataset. First, based on the pre-determined linear analysis results of each panel parameter, several third target parameters are determined. These third target parameters are parameters among the multiple panel parameters whose linear analysis results show a linear or near-linear relationship. Then, data augmentation is performed on the first training dataset based on these third target parameters to obtain the target training dataset. When augmenting the first training dataset based on these third target parameters, interpolation can be performed on the data corresponding to these third target parameters in the first training dataset.

[0103] In one specific embodiment, before step S30, the method further includes:

[0104] Step M21: Based on multiple panel parameters and their importance, determine several second target parameters. These second target parameters are the parameters whose importance is ranked from highest to lowest among the multiple panel parameters.

[0105] Step M22: Take boundary values ​​for several second target parameters respectively to obtain multiple third boundary values ​​for each of the several second target parameters;

[0106] Step M23: Determine the second training dataset based on multiple third boundary values ​​and base values; wherein, the second training dataset includes multiple second parameter groups and the second simulation results of each of the multiple second parameter groups;

[0107] Step M24: Based on multiple second parameter groups and second simulation results, determine the linear analysis results for each panel parameter.

[0108] The secondary objective parameters are the parameters ranked highest in importance from the panel parameters. The number of secondary objective parameters can be set as needed. For example, when there are 9 secondary objective parameters, they are the 9 panel parameters ranked highest in importance from the panel parameters. Figure 2 The panel parameters are excluding T51T61, T52T62, T53T63, and T54T64. Multiple third boundary values ​​are extended parameter values ​​obtained by taking values ​​for each second objective parameter according to a preset third step size. The preset third step size can be set as needed; for example, the third step size can be 20%. The second training dataset is a dataset determined based on multiple third boundary values ​​and base values. The second training dataset includes multiple second parameter groups and the second simulation results for each second parameter group.

[0109] Before determining the target training dataset in this embodiment, several second target parameters are determined based on multiple panel parameters and the importance of each panel parameter. Then, the values ​​of the several second target parameters are taken according to a preset third step size to obtain multiple third boundary values ​​for each of the several second target parameters. Next, based on the multiple third boundary values ​​and the base values, the second training dataset is determined. Finally, based on the multiple second parameter groups in the second training dataset and the second simulation results of each second parameter group, the linear analysis results of each panel parameter are determined.

[0110] The steps for determining the second training dataset based on multiple third boundary values ​​and base values ​​are the same as the steps for determining the first training dataset based on multiple first boundary values ​​and base values. For details, please refer to the aforementioned steps for determining the first training dataset based on multiple first boundary values ​​and base values. This application will not repeat the details here.

[0111] Step S40: Train the base model based on the target training dataset to generate the target model.

[0112] The base model is a pre-trained model, and the target model is the trained base model obtained by training the base model on the target training dataset. The model structure of the target model is the same as that of the base model. The difference lies in that the model parameters of the base model are pre-trained model parameters, while the model parameters of the target model are trained on the target training dataset. The target model can employ a Deep Neural Network (DNN) structure, including but not limited to convolutional layers, fully connected layers, pooling layers, and linear unit activation functions. The optimizer used in the target model can be gradient descent, momentum methods, etc., including but not limited to stochastic gradient descent (SGD), momentum gradient descent, adaptive subgradient method (AdaGrad), adaptive learning rate algorithm (Root Mean Square Propagation (RMSProp), and adaptive moment estimation (Adam). The input and output dimensions of each layer of the target model's neural network are as follows: Figure 6 As shown, for example, the input to the target model is a 13-dimensional combination of panel parameters, and the output is a 31-dimensional simulation result.

[0113] Considering the similarities between panel products within the same framework, in one specific implementation, the base model can be a simulation result prediction model corresponding to a certain panel product, and the target model is a simulation result prediction model corresponding to another panel product obtained by transferring learning from the base model.

[0114] After determining the target training dataset in this embodiment, the basic model is trained based on the target training dataset to generate a target model for predicting the simulation results of panel parameters in panel products. By generating the target model based on the basic model, only a small amount of training data is needed to achieve the effect of a model trained on large-scale data.

[0115] The inventors discovered through research that by generating a target model based on a base model, only 162 sets of parameters need to be simulated while ensuring comparable predictive performance. In contrast, existing methods require 13,122 sets of parameters to be simulated, and the amount of data to be simulated is about 1.2% of the original amount. This greatly saves the time cost of obtaining simulation data, thereby reducing the time required to obtain training data.

[0116] In one specific implementation, the target training dataset includes multiple fifth parameter groups and a sixth simulation result for each of the multiple fifth parameter groups. When training a pre-trained base model based on the target training dataset, the multiple fifth parameter groups from the target training dataset are first input into the base model. The base model outputs a second prediction result for each fifth parameter group. Then, the base model is trained based on the sixth simulation result, the second prediction result, and the loss function of the base model to generate the target model. The loss function of the base model includes, but is not limited to, Mean Absolute Error Loss (MAE Loss), Mean Square Error Loss (MSE Loss), Wing Loss, and Softmax Loss.

[0117] In one implementation of this embodiment, when training the base model based on the sixth simulation result, the second prediction result, and the loss function of the base model, a first loss value is first determined based on the sixth simulation result, the second prediction result, and the loss function of the base model. If the first loss value does not meet a preset first condition, the process continues by inputting multiple fifth parameter groups from the target training dataset into the base model and outputting the second prediction result for each fifth parameter group through the base model, until the first loss value meets the preset first condition. The first condition for the first loss value to meet the preset first condition can be that the first loss value is less than a preset first threshold, or that the difference between two consecutive first loss values ​​is less than a preset second threshold.

[0118] In one specific embodiment, before step S40, the method further includes:

[0119] M31. Based on the linear analysis results, the second training dataset is augmented to obtain the third training dataset; wherein, the third training dataset includes multiple third parameter groups and the third simulation results of each of the multiple third parameter groups;

[0120] M32. Input multiple third parameter groups from the third training dataset into the preset network model, and output the first prediction result of each third parameter group through the preset network model.

[0121] M33. Based on the third simulation results, the first prediction results, and the loss function of the preset network model, the preset network model is trained to obtain the basic model.

[0122] The third training dataset is an augmented dataset obtained by augmenting the second training dataset based on the linear analysis results. The third training dataset includes multiple third parameter groups and the third simulation results for each of these third parameter groups. The first prediction result is the simulation result for each third parameter group predicted by the preset network model. The loss function of the preset network model includes, but is not limited to, Mean Absolute Error Loss (MAE Loss), Mean Square Error Loss (MSE Loss), Wing Loss, and Softmax Loss.

[0123] The base model is trained using a pre-defined network model. The model structure of the base model is the same as that of the pre-defined network model. The difference between the base model and the pre-defined network model is that the model parameters of the pre-defined network model are the preset initial model parameters, while the model parameters of the base model are the model parameters trained based on a third training dataset.

[0124] Before generating the target model, this embodiment further augments the second training dataset based on the linear analysis results to obtain a third training dataset. Then, multiple third parameter groups from the third training dataset are input into a preset network model. The preset network model outputs the first prediction result for each third parameter group. Finally, the preset network model is trained based on the third simulation results, the first prediction results, and the loss function of the preset network model to obtain the base model. The step of augmenting the second training dataset based on the linear analysis results is the same as the step of augmenting the first training dataset based on the linear analysis results; specifically, refer to the aforementioned step of augmenting the first training dataset based on the linear analysis results, which will not be repeated here.

[0125] In one implementation of this embodiment, when training the preset network model based on the third simulation result, the first prediction result, and the loss function of the preset network model, a second loss value is first determined based on the third simulation result, the first prediction result, and the loss function of the preset network model. If the second loss value does not meet the preset second condition, the process continues by inputting multiple third parameter groups from the third training dataset into the preset network model and outputting the first prediction result of each third parameter group through the preset network model, until the second loss value meets the preset second condition. The second loss value meeting the preset second condition can be that the second loss value is less than a preset third threshold, or that the difference between two consecutive second loss values ​​is less than a preset fourth threshold.

[0126] In one specific embodiment, after step S40, the method further includes:

[0127] S51. Input the pre-constructed panel parameter space into the target model, and output multiple fourth simulation results of the panel parameter space through the target model;

[0128] S52. Based on multiple fourth simulation results and preset second target simulation results, search for several second target parameter sets from the panel parameter space.

[0129] The panel parameter space is a pre-constructed parameter space for the panel product. The method for constructing the panel parameter space can be as follows: Several first target parameters are assigned values ​​according to a preset fourth step length, resulting in multiple fourth boundary values ​​for each first target parameter. Then, these multiple first boundary values ​​and the base values ​​are combined to obtain a parameter space composed of multiple sixth parameter groups. The fourth step length is less than the first step length; for example, the fourth step length is 5%, and the first step length is 20%.

[0130] The fourth simulation result is the simulation result of multiple sixth parameter groups in the panel parameter space predicted by the target model. The second target simulation result is the simulation result expected by the user in advance. The second target simulation result can be used to measure whether the multiple fourth simulation results meet the user's expectations. In this embodiment, after generating the target model, the pre-constructed panel parameter space can be input into the target model, and the target model outputs multiple fourth simulation results in the panel parameter space. Then, based on the multiple fourth simulation results and the preset second target simulation results, several second target parameter groups are searched from the panel parameter space. For example, after obtaining multiple fourth simulation results, the error values ​​between the multiple fourth simulation results and the second target simulation results are calculated, and the top 10 parameter groups with smaller error values ​​are output. When using the target model generated in this embodiment to search the panel parameter space, the search speed of the panel parameter space is 2.5ms / time, which is significantly improved compared to the search speed of 10min / time of the traditional method.

[0131] In one specific implementation, after searching for several sets of second target parameters, these sets of second target parameters can be simulated using simulation software. The simulation results of the second set of second target parameters output by the target model are then verified based on the simulation results output by the simulation software. Finally, the verified second set of target parameters and its corresponding simulation results are returned to the user.

[0132] To better implement the model generation method in the embodiments of this application, a model generation apparatus is also provided in the embodiments of this application, such as... Figure 7 As shown, the model generation device includes:

[0133] The parameter acquisition unit 710 is used to acquire multiple panel parameters; the panel parameters include basic values.

[0134] The first determining unit 720 is used to determine the first training dataset based on multiple panel parameters and the pre-determined importance of each panel parameter.

[0135] The second determining unit 730 is used to determine the target training dataset based on the linear analysis results of each pre-determined panel parameter and the first training dataset; wherein the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results.

[0136] The model generation unit 740 is used to train the base model based on the target training dataset to generate the target model.

[0137] In this embodiment, the target model is generated based on the base model. Only a small amount of training data is needed to achieve the effect of a model trained on large-scale data, which greatly saves the time cost of acquiring training data. The training dataset is constructed based on the importance of each panel parameter and the results of linear analysis. While ensuring the reliability of the training dataset, the acquisition time of training data can be further reduced, and the problem of data sparsity in the training data space can be alleviated.

[0138] In some embodiments of this application, the first determining unit 720 is specifically used for:

[0139] Based on multiple panel parameters and the pre-determined importance of each panel parameter, several first target parameters are determined. These first target parameters are the parameters whose importance is ranked from high to low among the multiple panel parameters.

[0140] By taking boundary values ​​for several first target parameters, multiple first boundary values ​​are obtained for each first target parameter.

[0141] The first training dataset is determined based on multiple first boundary values ​​and basic values.

[0142] In some embodiments of this application, the model generation apparatus further includes:

[0143] The first value-taking unit is used to take boundary values ​​for each panel parameter based on the base value, so as to obtain multiple second boundary values ​​for each panel parameter;

[0144] The third determining unit is used to determine multiple sets of first parameters based on multiple second boundary values ​​and basic values;

[0145] The first simulation unit is used to simulate multiple first parameter groups respectively, and obtain the first simulation result of each first parameter group in the multiple first parameter groups;

[0146] The fourth determining unit is used to determine the importance of each panel parameter based on multiple first parameter groups and first simulation results.

[0147] In some embodiments of this application, the first simulation result includes multi-dimensional simulation results, and the fourth determining unit is specifically used for:

[0148] For any panel parameter, select multiple first parameter groups from multiple first parameter groups where other panel parameters besides that panel parameter remain unchanged as multiple first target parameter groups;

[0149] Based on multiple first simulation results and multiple first target parameter sets, multiple first target simulation results are determined; each first target simulation result corresponds to a first target parameter set.

[0150] Based on multiple first target parameter groups, determine the maximum value and minimum value of the panel parameter of the panel parameter;

[0151] For any simulation result of the first objective, determine the maximum and minimum values ​​of the simulation result from the multidimensional simulation results of the first objective simulation result;

[0152] Based on the maximum value of the panel parameters, the minimum value of the panel parameters, the maximum value of the simulation results, and the minimum value of the simulation results, the influence coefficient corresponding to the simulation result of the first objective is determined;

[0153] Based on all the influence coefficients of the panel parameter, determine the importance of the panel parameter.

[0154] In some embodiments of this application, the model generation apparatus further includes:

[0155] The fifth determining unit is used to determine several second target parameters based on multiple panel parameters and their importance. The several second target parameters are the parameters with the highest importance among the multiple panel parameters, ranked from highest to lowest.

[0156] The second value-taking unit is used to take boundary values ​​for several second target parameters respectively, so as to obtain multiple third boundary values ​​for each of the several second target parameters;

[0157] The sixth determining unit is used to determine the second training dataset based on multiple third boundary values ​​and basic values; wherein the second training dataset includes multiple second parameter groups and the second simulation results of each of the multiple second parameter groups;

[0158] The seventh determining unit is used to determine the linear analysis result of each panel parameter based on multiple second parameter groups and second simulation results.

[0159] In some embodiments of this application, the model generation apparatus further includes:

[0160] The data augmentation unit is used to augment the second training dataset based on the linear analysis results to obtain the third training dataset; wherein, the third training dataset includes multiple third parameter groups and the third simulation results of each of the multiple third parameter groups;

[0161] The result prediction unit is used to input multiple third parameter groups from the third training dataset into a preset network model, and output the first prediction result of each third parameter group through the preset network model.

[0162] The model training unit is used to train the preset network model based on the third simulation results, the first prediction results, and the loss function of the preset network model to obtain the basic model.

[0163] In some embodiments of this application, the model generation apparatus further includes:

[0164] The result output unit is used to input the pre-built panel parameter space into the target model and output multiple fourth simulation results of the panel parameter space through the target model.

[0165] The parameter search unit is used to search for several sets of second target parameters from the panel parameter space based on multiple fourth simulation results and preset second target simulation results.

[0166] This application also provides a computer device that integrates any of the model generation apparatuses provided in this application. The computer device includes:

[0167] One or more processors;

[0168] Memory; and

[0169] One or more applications, wherein the applications are stored in memory and configured to be executed by a processor from the steps of the model generation method in any of the embodiments described above.

[0170] This application also provides a computer device that integrates any of the model generation apparatuses provided in this application. For example... Figure 8 As shown, it illustrates a structural schematic diagram of the computer device involved in the embodiments of this application, specifically:

[0171] The computer device may include components such as a processor 801 with one or more processing cores, a memory 802 with one or more computer-readable storage media, a power supply 803, and an input unit 804. Those skilled in the art will understand that... Figure 8 The computer device structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0172] The processor 801 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 802, and by calling data stored in the memory 802, it performs various functions of the computer device and processes data, thereby providing overall monitoring of the computer device. Optionally, the processor 801 may include one or more processing cores; preferably, the processor 801 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 801.

[0173] The memory 802 can be used to store software programs and modules. The processor 801 executes various functional applications and data processing by running the software programs and modules stored in the memory 802. The memory 802 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 802 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 802 may also include a memory controller to provide the processor 801 with access to the memory 802.

[0174] The computer device also includes a power supply 803 that supplies power to the various components. Preferably, the power supply 803 can be logically connected to the processor 801 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 803 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0175] The computer device may also include an input unit 804, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0176] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 801 in the computer device loads the executable files corresponding to the processes of one or more application programs into the memory 802 according to the following instructions, and the processor 801 runs the application programs stored in the memory 802 to realize various functions, as follows:

[0177] Retrieve multiple panel parameters; panel parameters include basic values;

[0178] The first training dataset is determined based on multiple panel parameters and the pre-determined importance of each panel parameter.

[0179] Based on the linear analysis results of each pre-determined panel parameter and the first training dataset, the target training dataset is determined; wherein, the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results;

[0180] The base model is trained based on the target training dataset to generate the target model.

[0181] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0182] Therefore, embodiments of this application provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a disk, or an optical disk, etc. A computer program is stored thereon, which is loaded by a processor to execute the steps in any of the model generation methods provided in embodiments of this application. For example, the computer program loaded by the processor can execute the following steps:

[0183] Retrieve multiple panel parameters; panel parameters include basic values;

[0184] The first training dataset is determined based on multiple panel parameters and the pre-determined importance of each panel parameter.

[0185] Based on the linear analysis results of each pre-determined panel parameter and the first training dataset, the target training dataset is determined; wherein, the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results;

[0186] The base model is trained based on the target training dataset to generate the target model.

[0187] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.

[0188] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.

[0189] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0190] The foregoing has provided a detailed description of a model generation method, apparatus, computer device, and computer-readable storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A model generation method, characterized in that, include: Get multiple panel parameters; The panel parameters include basic values; A first training dataset is determined based on multiple panel parameters and the pre-determined importance of each panel parameter. The first training dataset is augmented based on the linear analysis results of each pre-determined panel parameter to obtain the target training dataset; wherein the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results; The base model is trained based on the target training dataset to generate the target model; The step of determining the first training dataset based on multiple panel parameters and a pre-determined importance level of each panel parameter includes: Based on the multiple panel parameters and the pre-determined importance of each panel parameter, a number of first target parameters are determined, wherein the number of first target parameters are the parameters whose importance is ranked from high to low among the multiple panel parameters. By taking boundary values ​​for several first target parameters, multiple first boundary values ​​are obtained for each of the several first target parameters. The first training dataset is determined based on multiple first boundary values ​​and the base values; Before determining the first training dataset based on multiple panel parameters and the predetermined importance of each panel parameter, the method further includes: Based on the base value, boundary values ​​are taken for each panel parameter to obtain multiple second boundary values ​​for each panel parameter; Based on the multiple second boundary values ​​and the basic values, multiple first parameter groups are determined; Simulations are performed on multiple sets of first parameters respectively to obtain the first simulation result for each set of first parameters. Based on multiple sets of the first parameter and the first simulation results, the importance of each panel parameter is determined.

2. The method according to claim 1, characterized in that, The first simulation result includes multi-dimensional simulation results; The determination of the importance of each panel parameter based on multiple first parameter groups and the first simulation results includes: For any panel parameter, select from a plurality of first parameter groups that are unchanged except for the panel parameter itself as a plurality of first target parameter groups; Based on multiple first simulation results and multiple first target parameter sets, multiple first target simulation results are determined; each first target simulation result corresponds to one first target parameter set. Based on multiple first target parameter groups, determine the maximum value of the panel parameter and the minimum value of the panel parameter. For any of the first target simulation results, determine the maximum and minimum values ​​of the simulation results from the multidimensional simulation results of the first target simulation result; Based on the maximum value of the panel parameters, the minimum value of the panel parameters, the maximum value of the simulation results, and the minimum value of the simulation results, the influence coefficient corresponding to the simulation result of the first target is determined; The importance of the panel parameter is determined based on all the influence coefficients of the panel parameter.

3. The method according to claim 1, characterized in that, Before determining the target training dataset based on the linear analysis results of each pre-determined panel parameter and the first training dataset, the method further includes: Based on the multiple panel parameters and their importance, several second target parameters are determined, and the several second target parameters are the parameters whose importance is ranked from high to low among the multiple panel parameters. By taking boundary values ​​for several second target parameters, multiple third boundary values ​​are obtained for each of the several second target parameters; A second training dataset is determined based on multiple third boundary values ​​and the base values; wherein, the second training dataset includes multiple second parameter groups and a second simulation result for each of the multiple second parameter groups; Based on multiple sets of the second parameter and the second simulation results, the linear analysis result for each panel parameter is determined.

4. The method according to claim 3, characterized in that, Before training the base model based on the target training dataset to generate the target model, the method further includes: Based on the linear analysis results, the second training dataset is augmented to obtain a third training dataset; wherein, the third training dataset includes multiple third parameter groups and the third simulation results of each of the multiple third parameter groups; The third parameter group in the third training dataset is input into a preset network model, and the preset network model outputs the first prediction result of each third parameter group. The preset network model is trained based on the third simulation result, the first prediction result, and the loss function of the preset network model to obtain the basic model.

5. The method according to claim 1, characterized in that, After training the base model based on the target training dataset to generate the target model, the method further includes: The pre-constructed panel parameter space is input into the target model, and multiple fourth simulation results of the panel parameter space are output through the target model. Based on multiple fourth simulation results and preset second target simulation results, several second target parameter sets are searched from the panel parameter space.

6. A model generation apparatus, characterized in that, include: The parameter acquisition unit is used to acquire multiple panel parameters; The panel parameters include basic values; The first determining unit is configured to determine a first training dataset based on a plurality of panel parameters and a pre-determined importance level of each of the panel parameters; The second determining unit is used to perform data augmentation on the first training dataset based on the linear analysis results of each pre-determined panel parameter to obtain a target training dataset; wherein the linear analysis results are used to characterize the linear correlation between each panel parameter and the simulation results; The model generation unit is used to train the base model based on the target training dataset to generate the target model; The first determining unit is specifically used for: Based on the multiple panel parameters and the pre-determined importance of each panel parameter, a number of first target parameters are determined, wherein the number of first target parameters are the parameters whose importance is ranked from high to low among the multiple panel parameters. By taking boundary values ​​for several first target parameters, multiple first boundary values ​​are obtained for each of the several first target parameters. The first training dataset is determined based on multiple first boundary values ​​and the base values; The model generation device further includes: The first value-taking unit is used to take boundary values ​​for each panel parameter based on the base value to obtain multiple second boundary values ​​for each panel parameter; The third determining unit is used to determine multiple first parameter groups based on multiple second boundary values ​​and the basic values; The first simulation unit is used to simulate multiple first parameter groups respectively, and obtain the first simulation result of each first parameter group in the multiple first parameter groups; The fourth determining unit is used to determine the importance of each panel parameter based on multiple first parameter groups and the first simulation results.

7. A computer device, characterized in that, The computer device includes: One or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the model generation method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to perform the steps of the model generation method according to any one of claims 1 to 5.

Images