Multi-condition antenna optimization method based on large model
By employing a multi-condition antenna optimization method based on a large model, utilizing the Adam algorithm and TensorBoard to monitor performance indicators, and optimizing the hyperparameters of the GPT model, the problem of low efficiency and performance in existing antenna designs is solved, achieving a more efficient antenna design.
Patent Information
- Application Number
- CN202411377562.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-29
- Publication Date
- 2024-11-01
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing antenna design methods are based on experience and trial and error, which makes it difficult to meet the diverse needs of high-speed data transmission, high-capacity communication, and low latency, resulting in low efficiency and performance.
A multi-condition antenna optimization method based on a large model is adopted. The Adam algorithm is used to train a generative pre-trained GPT model. Through data cleaning and standardization, a confusion matrix is constructed. TensorBoard is used to monitor performance indicators and adjust hyperparameters to optimize antenna design.
It improves the accuracy and intelligence of antenna design, enhances multi-target optimization capabilities, and meets the needs of high-speed data transmission, high-capacity communication, and low latency.
Smart Images

Figure CN118886342A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the fields of communication engineering and deep learning technology, and in particular to a multi-condition antenna optimization method based on a large model. Background Art
[0002] At present, antenna design plays a vital role in the field of communication engineering. With the rapid development of 5G and Internet of Things technologies, the demand for antennas has become more and more complex and urgent. However, antenna design faces many challenges. Antenna performance is affected by multiple factors, such as frequency, bandwidth, size, material, etc., and there are complex relationships between these factors. In the design process, it is necessary to weigh and choose between multiple performance indicators to achieve the optimal design solution. Therefore, it is particularly important to seek an antenna design optimization method that can comprehensively consider multiple performance indicators.
[0003] In related technologies, the performance of antennas is often affected by multiple conditions, which may conflict with each other. For example, an increase in the S11 parameter (reflection loss) may lead to a degradation of the S21 parameter (transmission loss); an increase in bandwidth may lead to a decrease in gain; and an increase in radiation directivity may lead to a narrowing of the beam.
[0004] However, the antenna design methods in related technologies are often based on experience and trial and error, which makes it difficult to meet various requirements such as high-speed data transmission, large-capacity communication, and low latency. The efficiency and performance are low and need to be improved urgently. Summary of the invention
[0005] The present application provides a multi-condition antenna optimization method based on a large model to solve the problems that the antenna design method in the related art is often based on experience and trial and error, and it is difficult to meet various requirements such as high-speed data transmission, large-capacity communication, low latency, etc., and the efficiency and performance are low.
[0006] The first aspect of the present application provides a large-model-based multi-condition antenna optimization method, comprising the following steps: collecting design data of at least one target antenna, and dividing the design data of the at least one target antenna into at least one of a training set, a validation set, and a test set; using the Adam algorithm to set a first hyperparameter in the training process of a generative pre-trained GPT model, and using the training set to iteratively adjust the first hyperparameter to obtain a second hyperparameter that meets the target performance condition of the GPT model; using the visualization tool TensorBoard to record the performance indicators in the training process of the GPT model, and when the performance indicators do not meet the preset improvement conditions of the validation set, stopping the iterative training of the GPT model to obtain a GPT final model; based on the GPT final model, constructing a confusion matrix, and using the confusion matrix to analyze the output results of the GPT final model on the test set to obtain an analysis result, evaluating the GPT final model's ability to understand the antenna performance conditions based on the analysis result, and adjusting and optimizing the GPT final model and the second hyperparameter based on the understanding ability to optimize the design of the target antenna.
[0007] Optionally, in one embodiment of the present application, it also includes: sampling the GPT final model to generate at least one output result, determining the final output result among the at least one output result according to a user's selection instruction, and re-adjusting and optimizing the GPT final model according to the final output result to update the GPT final model.
[0008] Optionally, in one embodiment of the present application, the collecting of design data of at least one target antenna and dividing the design data of at least one target antenna into at least one of a training set, a validation set and a test set include: collecting design data of at least one target antenna of different types, frequency bands and performance indicators; performing data cleaning, outlier processing and standardization on the design data of the target antenna to obtain processed data; dividing the processed data according to a preset ratio to obtain at least one of the training set, the validation set and the test set.
[0009] Optionally, in one embodiment of the present application, the use of the Adam algorithm to set the first hyperparameter in the generative pre-trained GPT model training process includes: initializing the first hyperparameter in the GPT model training process to obtain an initialization result; based on the initialization result, when the first hyperparameter does not reach a preset convergence condition, training the GPT model to obtain a training result; based on the training result, when the first hyperparameter reaches the preset convergence condition, stopping the iteration to obtain the first hyperparameter.
[0010] Optionally, in one embodiment of the present application, the iterative adjustment of the first hyperparameter using the training set to obtain the second hyperparameter that meets the target performance conditions of the GPT model includes: forward propagating the input sequence of the antenna according to the GPT model to map the input sequence to a corresponding output sequence to obtain a predicted output result; comparing the predicted output result with the actual antenna design scheme to obtain a comparison result, and calculating the loss function of the GPT model based on the comparison result; obtaining the gradient of the loss function with respect to the first hyperparameter, and back-propagating the gradient to each layer of the network to calculate the error contribution of each layer to obtain the calculated gradient information; based on the calculated gradient information, using the Adam algorithm to update the first hyperparameter to obtain the second hyperparameter that meets the target performance conditions.
[0011] Optionally, in one embodiment of the present application, the visualization tool TensorBoard is used to record the performance indicators of the GPT model during training, and when the performance indicators do not meet the target improvement conditions of the validation set, the iterative training of the GPT model is stopped to obtain the GPT final model, including: recording the performance indicators of the GPT model during training, and using the validation set to evaluate the actual performance of the GPT model; comparing the actual performance of the GPT model with the historical best performance to obtain a comparison result, and judging whether the performance indicators meet the preset improvement conditions of the validation set based on the comparison result; if the validation loss of the GPT model is in a preset decrease condition or the performance indicator is in a preset condition, it is determined that the performance indicator does not meet the preset improvement conditions of the validation set, and the iterative training of the GPT model is stopped to obtain the GPT final model.
[0012] Optionally, in one embodiment of the present application, the adjusting and optimizing the GPT final model and the second hyperparameter includes: adjusting and optimizing the GPT final model according to at least one of the adjustment structure of the GPT final model, the second hyperparameter adjustment result of the GPT final model, and the adjustment result of the training data of the GPT final model.
[0013] The second aspect of the present application provides a multi-condition antenna optimization device based on a large model, including: an acquisition module, used to acquire design data of at least one target antenna, and divide the design data of the at least one target antenna into at least one of a training set, a validation set and a test set; a setting module, used to set a first hyperparameter in the training process of a generative pre-trained GPT model using an Adam algorithm, and iteratively adjust the first hyperparameter using the training set to obtain a second hyperparameter that meets the target performance condition of the GPT model; a recording module, used to record the performance indicators in the training process of the GPT model using a visualization tool TensorBoard, and when the performance indicators do not meet the preset improvement conditions of the validation set, stop the iterative training of the GPT model to obtain a GPT final model; an optimization module, used to construct a confusion matrix based on the GPT final model, and use the confusion matrix to analyze the output results of the GPT final model on the test set to obtain an analysis result, evaluate the GPT final model's ability to understand the antenna performance conditions based on the analysis result, and adjust and optimize the GPT final model and the second hyperparameter based on the understanding ability to optimize the design of the target antenna.
[0014] Optionally, in one embodiment of the present application, it also includes: an adjustment module, which is used to sample the GPT final model, generate at least one output result, determine the final output result among the at least one output result according to a user's selection instruction, and adjust and optimize the GPT final model again according to the final output result to update the GPT final model.
[0015] Optionally, in one embodiment of the present application, the acquisition module includes: an acquisition unit, used to collect design data of at least one target antenna of different types, frequency bands and performance indicators; a processing unit, used to perform data cleaning, outlier processing and standardization on the design data of the target antenna to obtain processed data; a division unit, used to divide the processed data according to a preset ratio to obtain at least one of the training set, the verification set and the test set.
[0016] Optionally, in one embodiment of the present application, the setting module includes: an initialization unit, used to initialize the first hyperparameter in the GPT model training process to obtain an initialization result; a training unit, used to train the GPT model based on the initialization result to obtain a training result when the first hyperparameter does not reach a preset convergence condition; and a stopping unit, used to stop iteration based on the training result to obtain the first hyperparameter when the first hyperparameter reaches the preset convergence condition.
[0017] Optionally, in one embodiment of the present application, the setting module also includes: a mapping unit, used to forward propagate the input sequence of the antenna according to the GPT model to map the input sequence to a corresponding output sequence to obtain a predicted output result; a comparison unit, used to compare the predicted output result with the actual antenna design scheme to obtain a comparison result, and calculate the loss function of the GPT model based on the comparison result; a calculation unit, used to obtain the gradient of the loss function with respect to the first hyperparameter, and back-propagate the gradient to each layer of the network to calculate the error contribution of each layer to obtain the calculated gradient information; an update unit, used to update the first hyperparameter based on the calculated gradient information using the Adam algorithm to obtain the second hyperparameter that meets the target performance condition.
[0018] Optionally, in one embodiment of the present application, the recording module includes: a recording unit, used to use the visualization tool TensorBoard to record the performance indicators of the GPT model during training, and use the verification set to evaluate the actual performance of the GPT model; a judgment unit, used to compare the actual performance of the GPT model with the historical best performance to obtain a comparison result, and judge whether the performance indicator meets the preset improvement condition of the verification set based on the comparison result; a judgment unit, used to judge that the performance indicator does not meet the preset improvement condition of the verification set when the verification loss of the GPT model is in a preset decrease condition or the performance indicator is in a preset condition, stop the iterative training of the GPT model, and obtain the GPT final model.
[0019] Optionally, in one embodiment of the present application, the optimization module includes: an optimization unit, used to adjust and optimize the GPT final model according to at least one of the adjustment structure of the GPT final model, the second hyperparameter adjustment result of the GPT final model, and the adjustment result of the training data of the GPT final model.
[0020] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a large model-based multi-condition antenna optimization method as described in the above embodiment.
[0021] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned large model-based multi-condition antenna optimization method.
[0022] The embodiment of the present application can collect and divide antenna design scheme data, train and optimize the GPT model, monitor and adjust the performance indicators during the training process, evaluate and adjust the GPT model, and adjust and optimize the GPT model again according to the selection, thereby improving the accuracy and stability of the GPT model, enhancing the multi-objective optimization capability of the antenna design, improving the accuracy of the antenna design, and improving the intelligent level of the antenna design. Thus, the problem that the antenna design method in the related art is often based on experience and trial and error, and it is difficult to meet the various requirements of high-speed data transmission, large-capacity communication, low latency, etc., and the efficiency and performance are low.
[0023] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a multi-condition antenna optimization method based on a large model according to an embodiment of the present application; Figure 2 A flowchart of a multi-condition antenna optimization method based on a large model according to an embodiment of the present application; Figure 3 A schematic diagram of the structure of a multi-condition antenna optimization device based on a large model according to an embodiment of the present application; Figure 4 It is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION
[0025] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0026] The following describes a multi-condition antenna optimization method based on a large model in an embodiment of the present application with reference to the accompanying drawings. In view of the fact that the antenna design method in the related art mentioned in the above background technology is often based on experience and trial and error, it is difficult to meet the various requirements of high-speed data transmission, large-capacity communication, low latency, etc., and the efficiency and performance are low, the present application provides a multi-condition antenna optimization method based on a large model, in which the antenna design scheme data can be collected and divided, the GPT model can be trained and optimized, the performance indicators in the training process can be monitored and adjusted, the GPT model can be evaluated and adjusted, and the GPT model can be adjusted and optimized again according to the selection, thereby improving the accuracy and stability of the GPT model, enhancing the multi-objective optimization capability of the antenna design, improving the accuracy of the antenna design, and improving the intelligent level of the antenna design. As a result, the problem that the antenna design method in the related art is often based on experience and trial and error, it is difficult to meet the various requirements of high-speed data transmission, large-capacity communication, low latency, etc., and the efficiency and performance are low is solved.
[0027] Specifically, Figure 1 A flowchart of a multi-condition antenna optimization method based on a large model provided in an embodiment of the present application.
[0028] like Figure 1 As shown, the multi-condition antenna optimization method based on a large model includes the following steps: In step S101, design data of at least one target antenna is collected, and the design data of at least one target antenna is divided into at least one of a training set, a validation set, and a test set.
[0029] It can be understood that the design data of at least one target antenna in the embodiment of the present application may be design data from a large number of representative antenna design solutions.
[0030] During the actual implementation process, the embodiments of the present application can collect a large number of representative antenna design schemes, which cover the design data of at least one target antenna, and divide the design data of at least one target antenna into a training set, a validation set, and a test set, etc., so as to provide support for subsequent training and optimization of the GPT model (Generative Pre-Trained Transformer, generative pre-training model), thereby ensuring the fairness and accuracy of model training and evaluation.
[0031] Optionally, in one embodiment of the present application, design data of at least one target antenna is collected, and the design data of at least one target antenna is divided into at least one of a training set, a validation set, and a test set, including: collecting design data of at least one target antenna of different types, frequency bands, and performance indicators; performing data cleaning, outlier processing, and standardization on the design data of the target antenna to obtain processed data; dividing the processed data according to a preset ratio to obtain at least one of a training set, a validation set, and a test set.
[0032] It can be understood that the design data of the target antenna in the embodiment of the present application is both the design scheme data in the antenna design scheme and the original data; the preset ratio division method in the embodiment of the present application can be based on , and The proportion of the division is divided in this way, where , and Represent the proportion of data in the training set, validation set, and test set in the dataset respectively.
[0033] As a possible implementation method, the embodiment of the present application can collect design solution data covering different types, frequency bands and performance indicators, and perform data cleaning on the original data to remove erroneous, repeated or incomplete parts. At the same time, possible outliers are identified and processed to avoid adverse effects on model training and evaluation. Since the original data may have different dimensions and distribution characteristics, in order to enable the model to better learn and understand the characteristics of the data, the data needs to be standardized so that it has a similar mean and variance to be closer to the characteristics of the standard normal distribution.
[0034] Furthermore, the embodiment of the present application can process the standardized data according to , and The ratio is divided into training set, validation set and test set, satisfying: .
[0035] When dividing the data set, it is necessary to ensure the representativeness and randomness of the data to avoid deviations in the model during training and evaluation. In addition, it is also necessary to consider the balance of the data and try to ensure that each category or each performance indicator is evenly distributed in the training, validation, and test sets to ensure the fairness and accuracy of model training and evaluation.
[0036] It should be noted that the preset ratio can be set by those skilled in the art according to actual conditions and is not specifically limited here.
[0037] In step S102, the Adam algorithm is used to set the first hyperparameter in the generative pre-trained GPT model training process, and the first hyperparameter is iteratively adjusted using the training set to obtain a second hyperparameter that meets the target performance conditions of the GPT model.
[0038] It can be understood that the first hyperparameter in the embodiment of the present application can be a GPT model parameter during the training process, and the second hyperparameter can be a GPT model parameter after iterative training; the Adam algorithm in the embodiment of the present application is a first-order gradient descent optimization algorithm for optimizing neural networks and machine learning models. The Adam algorithm combines the momentum method and the adaptive learning rate technology to adjust the step size of the parameter update by calculating the adaptive learning rate of each parameter.
[0039] In the actual implementation process, the embodiment of the present application can use the Adam optimization algorithm to optimize the GPT model, use the Adam algorithm to set the first hyperparameter in the GPT model training process, and initialize, train, optimize and iterate the first hyperparameter, so as to minimize the objective function by continuously iteratively adjusting the model parameters. The embodiment of the present application can use the training set to iteratively train the GPT model while using the Adam algorithm, that is, iteratively adjust the first hyperparameter using the training set to obtain a second hyperparameter that meets the target performance conditions of the GPT model, thereby optimizing the model performance.
[0040] The embodiment of the present application can use the Adam algorithm to set appropriate hyperparameters and use iterative training to optimize the GPT model, thereby improving the speed and effect of optimization. The combination of the Adam algorithm and the GPT model can effectively accelerate the training process of the model and better control the learning rate and momentum during the training process, thereby improving the accuracy and stability of the model.
[0041] Optionally, in one embodiment of the present application, the Adam algorithm is used to set the first hyperparameter in the training process of the generative pre-trained GPT model, including: initializing the first hyperparameter in the training process of the GPT model to obtain an initialization result; based on the initialization result, when the first hyperparameter does not reach a preset convergence condition, training the GPT model to obtain a training result; based on the training result, when the first hyperparameter reaches a preset convergence condition, stopping the iteration to obtain the first hyperparameter.
[0042] It can be understood that the preset convergence condition in the embodiment of the present application can be a condition that the GPT model parameters in the training process meet the gradual stabilization condition.
[0043] Specifically, the embodiment of the present application can initialize the first hyperparameter in the GPT model training process to obtain an initialization result as shown below: (1) Parameter initialization: Initialize the parameter vector: .
[0044] Initialize the first-order moment estimate vector and the second-order moment estimate vector , are all zero vectors.
[0045] Initialize time step is zero.
[0046] Initialize learning rate , the exponential decay rate of the first-order moment estimate , the exponential decay rate of the second-order moment estimate , and a constant to prevent the denominator from being zero .
[0047] Define the parameters as The random objective function is: .
[0048] Furthermore, the embodiment of the present application can train the GPT model based on the initialization result when the first hyperparameter does not reach the preset convergence condition to obtain a training result as shown below: (2) Model training and optimization: When the parameter If convergence has not yet occurred, perform the following steps: Increase the time step : .
[0049] The calculation time step is The gradient of the stochastic objective : , in, As parameters, The time step is The gradient of the stochastic objective is expressed in parameter space, where the objective function The direction of fastest change.
[0050] Update the first-order moment estimate vector and the second-order moment estimate vector : , , in, The time step is The first-order moment estimation vector of is the momentum term of the gradient, which represents the exponentially weighted moving average of the gradient; is the exponential decay rate of the first-order moment estimate; The time step is The gradient of the stochastic objective; The time step is The second-order moment estimation vector of is the momentum term of the squared gradient, which represents the exponentially weighted moving average of the squared gradient; is the exponential decay rate of the second-order moment estimate.
[0051] Calculate the bias-corrected first-order moment estimate vector and the second-order moment estimate vector : , , in, The time step is The bias-corrected first-order moment estimate vector, The time step is The bias-corrected second-order moment estimate vector.
[0052] Update Parameters : , in, To initialize the learning rate, A constant to prevent the denominator from being zero.
[0053] Furthermore, the embodiment of the present application can stop iteration when the first hyperparameter reaches a preset convergence condition based on the training result, and obtain the first hyperparameter as shown below: (3) Stop conditions: When the parameter When convergence occurs, stop iteration and return the result parameter .
[0054] The embodiments of the present application can further utilize the Adam algorithm to set appropriate hyperparameters and use iterative training to optimize the GPT model, thereby further improving the accuracy and stability of the model, thereby improving the speed and effect of optimization, and enhancing the multi-objective optimization capability of antenna design.
[0055] It should be noted that the preset convergence condition can be set by those skilled in the art according to actual conditions and is not specifically limited here.
[0056] Optionally, in one embodiment of the present application, a first hyperparameter is iteratively adjusted using a training set to obtain a second hyperparameter that meets the target performance conditions of the GPT model, including: forward propagating the input sequence of the antenna according to the GPT model to map the input sequence to a corresponding output sequence to obtain a predicted output result; comparing the predicted output result with the actual antenna design scheme to obtain a comparison result, and calculating the loss function of the GPT model based on the comparison result; obtaining the gradient of the loss function with respect to the first hyperparameter, and back-propagating the gradient to each layer of the network to calculate the error contribution of each layer to obtain the calculated gradient information; based on the calculated gradient information, using the Adam algorithm to update the first hyperparameter to obtain a second hyperparameter that meets the target performance conditions.
[0057] Specifically, the embodiment of the present application can use the training set to iteratively train the GPT model while using the Adam algorithm to optimize the model performance. The training process is as follows: The embodiment of the present application can forward propagate the input sequence of the antenna according to the GPT model to map the input sequence to the corresponding output sequence to obtain a predicted output result, as shown below: (1) Forward propagation: For a given input sequence (antenna design conditions), forward propagation is performed through the GPT model. During the forward propagation process, the model maps the input sequence to the corresponding output sequence, which is the predicted antenna design solution.
[0058] Furthermore, the embodiment of the present application can compare the predicted output result with the actual antenna design solution to obtain a comparison result, and calculate the loss function of the GPT model according to the comparison result, as shown below: (2) Loss calculation: Compare the model output with the actual antenna design and calculate the loss function. The loss function measures the difference between the model prediction and the actual situation. It is the objective function in the optimization process and can be in the form of mean square error or cross entropy. The formula for calculating the mean square error is: , The calculation formula for cross entropy is: , in, n is the assumed sample size, For the i The true labels of samples, is the predicted value of the model.
[0059] Furthermore, the embodiment of the present application can obtain the gradient of the loss function with respect to the first hyperparameter, and back-propagate the gradient to each layer of the network to calculate the error contribution of each layer, and obtain the calculated gradient information, as shown below: (3) Back propagation: Based on the chain rule, the output layer calculates the gradient of the loss function for all learnable parameters in the model layer by layer. After obtaining the gradient of the loss function for the model parameters, these gradients can be back propagated to each layer of the network to calculate the error contribution of each layer.
[0060] Furthermore, the embodiment of the present application can update the first hyperparameter using the Adam algorithm based on the calculated gradient information to obtain the second hyperparameter that meets the target performance condition, as shown below: (4) Parameter update: Based on the calculated gradient information, the parameter update rule in the Adam optimization algorithm is used to update the parameters. .
[0061] The embodiment of the present application can use the training set to iteratively train the GPT model to optimize the model performance, thereby providing support for the subsequent improvement of various aspects of the antenna performance and improving the accuracy of the antenna design.
[0062] In step S103, the performance indicators of the GPT model training process are recorded using the visualization tool TensorBoard, and when the performance indicators do not meet the preset improvement conditions of the validation set, the iterative training of the GPT model is stopped to obtain the final GPT model.
[0063] It is understandable that TensorBoard (a tool that provides the measurements and presentations required during the machine learning workflow) in the embodiment of the present application is a visualization tool provided by TensorFlow (an end-to-end open source machine learning platform) for monitoring various indicators during the training process.
[0064] During the actual execution process, the embodiment of the present application can use the visualization tool TensorBoard to record the performance indicators during the GPT model training process, and when the performance indicators do not meet certain improvement conditions of the validation set, stop the iterative training of the GPT model to obtain the GPT final model, thereby ensuring that training is stopped when the performance of the validation set no longer improves to prevent overfitting.
[0065] Optionally, in one embodiment of the present application, the performance indicators of the GPT model training process are recorded using the visualization tool TensorBoard, and when the performance indicators do not meet the target improvement conditions of the validation set, the iterative training of the GPT model is stopped to obtain the GPT final model, including: recording the performance indicators of the GPT model training process, and using the validation set to evaluate the actual performance of the GPT model; comparing the actual performance of the GPT model with the historical best performance to obtain a comparison result, and judging whether the performance indicators meet the preset improvement conditions of the validation set based on the comparison result; if the validation loss of the GPT model is in a preset decrease condition or the performance indicator is in a preset condition, it is determined that the performance indicator does not meet the preset improvement conditions of the validation set, and the iterative training of the GPT model is stopped to obtain the GPT final model.
[0066] It can be understood that the preset improvement condition in the embodiment of the present application can be the condition that the performance on the verification set meets the requirement of no longer being improved, the preset decrease condition can be the condition that the verification loss meets the requirement of no longer being decreased, and the preset condition can be the condition that other performance indicators meet the requirement of no longer being improved.
[0067] As a possible implementation method, the embodiment of the present application can record key performance indicators, such as training loss, validation loss and other performance indicators, into the TensorBoard log file during the training process, and use the validation set to evaluate the performance of the GPT model after each training iteration. By comparing the performance of the current model on the validation set with the previous best performance, it is determined whether the model has the potential for further improvement. When the performance on the validation set no longer improves, that is, when the validation loss no longer decreases or other performance indicators no longer improve, the training process is stopped to obtain the final GPT model, thereby avoiding overfitting of the model to the training data, ensuring that the model strikes a balance between generalization ability and prediction ability, and thus improving the level of intelligence in antenna design.
[0068] Assuming that the validation set contains n samples, the true output of the model is , the predicted output is , the validation loss can be expressed as: , here is the loss function, which can be in the form of mean square error or cross entropy.
[0069] It should be noted that the preset boosting conditions can be set by those skilled in the art according to actual conditions and are not specifically limited here.
[0070] In step S104, a confusion matrix is constructed based on the GPT final model, and the confusion matrix is used to analyze the output results of the GPT final model on the test set to obtain analysis results. The ability of the GPT final model to understand the antenna performance conditions is evaluated based on the analysis results, and the GPT final model and the second hyperparameter are adjusted and optimized based on the understanding ability to optimize the design of the target antenna.
[0071] During the actual implementation process, the embodiment of the present application can construct a confusion matrix and use the confusion matrix comparison model to classify different antenna performance conditions, thereby analyzing the output results of the GPT final model on the test set to obtain the analysis results.
[0072] Assume that different performance conditions, the size of the confusion matrix is , define the confusion matrix C as follows: , in, represents the number of samples where the true performance condition is 1 and the model predicts 1. represents the number of samples where the true performance condition is 1 and the model predicts 2, and so on. It represents the number of samples where the true performance condition is k and the model predicts k.
[0073] In the embodiment of the present application, based on the number of true positives (True Positives, TP), true negatives (True Negatives, TN), false positives (False Positives, FP), and false negatives (False Negatives, FN) in the confusion matrix, the precision (Precision) and recall (Recall) can be obtained. ) and accuracy: , , .
[0074] Through precision and recall, we can also calculate value: .
[0075] The commonly used performance evaluation indicators in the embodiments of the present application can assist in evaluating the classification performance of the model on the test set, thereby judging the quality of the model. In addition, by observing the value of each cell in the confusion matrix, the classification of the model under different performance conditions can be understood, thereby judging whether the model has a good understanding of the relationship between antenna performance and conditions, and adjusting and optimizing the GPT final model and the second hyperparameter according to the understanding ability to optimize the design of the target antenna.
[0076] The embodiment of the present application can use the confusion matrix to analyze the output results of the model on the test set, evaluate the GPT model's ability to understand the relationship between antenna performance and conditions, and make improvements such as model adjustment, hyperparameter adjustment, and adding data, thereby enhancing the multi-objective optimization capability of antenna design and improving the accuracy of antenna design.
[0077] Optionally, in one embodiment of the present application, the GPT final model and the second hyperparameter are adjusted and optimized, including: adjusting and optimizing the GPT final model according to at least one of the adjustment structure of the GPT final model, the adjustment result of the second hyperparameter of the GPT final model, and the adjustment result of the training data of the GPT final model.
[0078] In the actual implementation process, the embodiment of the present application can adjust and improve the performance of the performance indicators of the GPT final model according to the test set results. Specific improvement methods include adjusting the structure of the GPT final model (such as modifying the number of layers, number of nodes, activation function and other structural parameters of the neural network, and adding or deleting specific levels and functional units), adjusting the second hyperparameter of the GPT final model, and increasing the training data of the GPT final model.
[0079] The embodiments of the present application improve the classification performance and generalization ability of the model by continuously optimizing the model, thereby providing more accurate guidance for antenna design and having stronger adaptability.
[0080] Optionally, in one embodiment of the present application, it also includes: sampling the GPT final model, generating at least one output result, determining the final output result among the at least one output result according to a user's selection instruction, and re-adjusting and optimizing the GPT final model according to the final output result to update the GPT final model.
[0081] It can be understood that at least one output result in the embodiment of the present application can be 10 different output results, and the final output result can be but is not limited to one or more most suitable results; the users in the embodiment of the present application can be but are not limited to human experts or field-related personnel.
[0082] During the actual implementation process, the embodiment of the present application can sample the GPT final model multiple times to generate, for example, 10 different output results, and human experts or field-related personnel can select one or more most appropriate results from the multiple output results generated, and adjust and optimize the GPT final model again based on the one or more most appropriate results to update the GPT final model, so as to ensure that the commonalities, advantages and disadvantages can be analyzed according to the manually selected results, and local fine-tuning can be performed accordingly.
[0083] The embodiment of the present application can complete multiple outputs through the GPT model, and then manually select one or more appropriate results. The model is adjusted and optimized again based on the selection, thereby increasing the judgment probability to a level close to that of humans, thereby improving the intelligence level of antenna design.
[0084] Specifically, it can be combined Figure 2 As shown, the working principle of a multi-condition antenna optimization method based on a large model in an embodiment of the present application is described in detail with a specific embodiment.
[0085] like Figure 2 As shown, the embodiment of the present application may include the following steps: Step S201: A large amount of antenna design data.
[0086] Among them, the embodiments of the present application can collect a large number of representative antenna design schemes, which should cover design scheme data of different types, frequency bands and performance indicators.
[0087] Step S202: Data cleaning, outlier processing and standardization.
[0088] Among them, the embodiments of the present application can clean the original data to remove erroneous, duplicate or incomplete parts; identify and process possible outliers to avoid adverse effects on model training and evaluation; and standardize the data to have similar means and variances, as well as characteristics that are closer to standard normal distribution.
[0089] Step S203: Divide the data into a training set, a validation set, and a test set.
[0090] Among them, the embodiment of the present application can convert the standardized data into , and The set is divided into training set, validation set and test set in the same proportion.
[0091] Step S204: Initialize GPT model parameters.
[0092] Step S205: Loading the data set.
[0093] Step S206: setting hyperparameters in the model training process.
[0094] Step S207: Determine whether the number of iterations or convergence conditions are reached, if so, execute step S208; if not, execute step S209.
[0095] Among them, the embodiment of the present application can determine whether the model training reaches the number of iterations or convergence conditions. If so, step S208 is executed to record the performance indicators using TensorBoard; if not, step S209 is executed to forward propagate the calculation prediction output. The embodiment of the present application can continuously adjust the model parameters and continuously optimize the model through multiple iterations so that its performance gradually converges to the ideal value.
[0096] Step S208: Use TensorBoard to record performance indicators.
[0097] Among them, the embodiment of the present application can record key performance indicators, such as training loss, validation loss and other performance indicators, into the TensorBoard log file during the training process.
[0098] Step S209: forward propagation to calculate the predicted output.
[0099] Among them, the embodiment of the present application can perform forward propagation through the GPT model for a given input sequence (antenna design condition).
[0100] Step S210: Calculate the loss function.
[0101] Among them, the embodiment of the present application can compare the output of the model with the actual antenna design solution and calculate the loss function.
[0102] Step S211: Back propagate the gradient of the model parameters.
[0103] Among them, the embodiment of the present application can back-propagate the gradient of the loss function to the model parameters to each layer of the network, thereby calculating the error contribution of each layer.
[0104] Step S212: According to the Adam algorithm, the model parameters are updated using parameter gradients.
[0105] Among them, the embodiment of the present application can update the parameters using the parameter update rules in the Adam optimization algorithm based on the calculated gradient information.
[0106] Step S213: Determine whether the model performance can be further improved, if so, execute step S207; if not, execute step S214.
[0107] Among them, the embodiment of the present application can determine whether the model performance can continue to improve. If so, step S207 is re-executed to determine whether the number of iterations or convergence conditions are reached; if not, step S214 is executed to stop iterative training. The embodiment of the present application can determine whether the model has the possibility of further improvement by comparing the performance of the current model on the validation set with the previous best performance.
[0108] Step S214: stop iterative training.
[0109] Among them, the embodiment of the present application can stop the training process when the performance on the verification set no longer improves, that is, when the verification loss no longer decreases or other performance indicators no longer improve.
[0110] Step S215: construct a confusion matrix.
[0111] Among them, the embodiment of the present application can construct a confusion matrix and use the confusion matrix comparison model to classify different antenna performance conditions.
[0112] Step S216: Calculate performance evaluation indicators.
[0113] Among them, the embodiment of the present application can use performance evaluation indicators to assist in evaluating the classification performance of the model on the test set, thereby judging the quality of the model.
[0114] Step S217: Analyze the output results of the model on the test set.
[0115] Step S218: Optimize and improve the model.
[0116] Among them, the embodiments of the present application can improve the classification performance and generalization ability of the model by continuously optimizing the model, thereby providing more accurate guidance for antenna design and having stronger adaptability.
[0117] Step S219: perform multiple sampling on the GPT model.
[0118] Among them, the embodiment of the present application can sample the GPT model multiple times to generate, for example, 10 different output results.
[0119] Step S220: Manually select a suitable result.
[0120] Among them, the embodiment of the present application can select one or more most suitable results from the multiple output results generated, and analyze the commonalities and advantages and disadvantages therein based on the manually selected results.
[0121] Step S221: fine-tune the model locally based on manual selection.
[0122] According to a multi-condition antenna optimization method based on a large model proposed in an embodiment of the present application, it is possible to collect and divide antenna design solution data, train and optimize the GPT model, monitor and adjust the performance indicators during the training process, evaluate and adjust the GPT model, and adjust and optimize the GPT model again according to the selection, thereby improving the accuracy and stability of the GPT model, enhancing the multi-objective optimization capability of the antenna design, improving the accuracy of the antenna design, and improving the intelligent level of the antenna design. Thus, the problem that the antenna design method in the related art is often based on experience and trial and error, and it is difficult to meet multiple requirements such as high-speed data transmission, large-capacity communication, and low latency, and the efficiency and performance are low is solved.
[0123] Next, a multi-condition antenna optimization device based on a large model proposed according to an embodiment of the present application is described with reference to the accompanying drawings.
[0124] Figure 3 It is a structural schematic diagram of a large-model-based multi-condition antenna optimization device in an embodiment of the present application.
[0125] like Figure 3 As shown, the large model-based multi-condition antenna optimization device 10 includes: a collection module 100, a setting module 200, a recording module 300 and an optimization module 400.
[0126] The acquisition module 100 is used to acquire design data of at least one target antenna, and divide the design data of at least one target antenna into at least one of a training set, a validation set and a test set.
[0127] The setting module 200 is used to set the first hyperparameter in the training process of the generative pre-trained GPT model using the Adam algorithm, and iteratively adjust the first hyperparameter using the training set to obtain a second hyperparameter that meets the target performance conditions of the GPT model.
[0128] The recording module 300 is used to use the visualization tool TensorBoard to record the performance indicators during the GPT model training process, and when the performance indicators do not meet the preset improvement conditions of the validation set, stop the iterative training of the GPT model to obtain the GPT final model.
[0129] The optimization module 400 is used to construct a confusion matrix based on the GPT final model, and use the confusion matrix to analyze the output results of the GPT final model on the test set to obtain analysis results, evaluate the GPT final model's understanding ability of the antenna performance conditions based on the analysis results, and adjust and optimize the GPT final model and the second hyperparameter based on the understanding ability to optimize the design of the target antenna.
[0130] Optionally, in one embodiment of the present application, a large model-based multi-condition antenna optimization device 10 further includes: an adjustment module.
[0131] Among them, the adjustment module is used to sample the GPT final model, generate at least one output result, determine the final output result among the at least one output result according to the user's selection instruction, and adjust and optimize the GPT final model again according to the final output result to update the GPT final model.
[0132] Optionally, in one embodiment of the present application, the acquisition module 100 includes: a acquisition unit, a processing unit and a division unit.
[0133] The acquisition unit is used to acquire design data of at least one target antenna of different types, frequency bands and performance indicators.
[0134] The processing unit is used to perform data cleaning, outlier processing and standardization on the design data of the target antenna to obtain processed data.
[0135] The partitioning unit is used to partition the processed data according to a preset ratio to obtain at least one of a training set, a validation set and a test set.
[0136] Optionally, in one embodiment of the present application, the setting module 200 includes: an initialization unit, a training unit and a stop unit.
[0137] Among them, the initialization unit is used to initialize the first hyperparameter in the GPT model training process to obtain an initialization result.
[0138] The training unit is used to train the GPT model based on the initialization result and obtain the training result when the first hyperparameter does not reach the preset convergence condition.
[0139] The stopping unit is used to stop iteration based on the training result and obtain the first hyperparameter when the first hyperparameter reaches a preset convergence condition.
[0140] Optionally, in one embodiment of the present application, the setting module 200 further includes: a mapping unit, a comparison unit, a calculation unit and an update unit.
[0141] Among them, the mapping unit is used to forward propagate the input sequence of the antenna according to the GPT model to map the input sequence to the corresponding output sequence to obtain a predicted output result.
[0142] The comparison unit is used to compare the predicted output result with the actual antenna design scheme to obtain a comparison result, and calculate the loss function of the GPT model according to the comparison result.
[0143] The computing unit is used to obtain the gradient of the loss function with respect to the first hyperparameter, and back-propagate the gradient to each layer of the network to calculate the error contribution of each layer and obtain the calculated gradient information.
[0144] The updating unit is used to update the first hyperparameter by using the Adam algorithm based on the calculated gradient information to obtain a second hyperparameter that meets the target performance condition.
[0145] Optionally, in one embodiment of the present application, the recording module 300 includes: a recording unit, a judging unit and a determining unit.
[0146] Among them, the recording unit is used to use the visualization tool TensorBoard to record the performance indicators during the GPT model training process, and use the verification set to evaluate the actual performance of the GPT model.
[0147] The judgment unit is used to compare the actual performance of the GPT model with the historical best performance to obtain a comparison result, and judge whether the performance index meets the preset improvement conditions of the validation set based on the comparison result.
[0148] The determination unit is used to determine that the performance index does not meet the preset improvement condition of the validation set when the verification loss of the GPT model is in a preset decrease condition or the performance index is in a preset condition, stop the iterative training of the GPT model, and obtain the GPT final model.
[0149] Optionally, in one embodiment of the present application, the optimization module 400 includes: an optimization unit.
[0150] Among them, the optimization unit is used to adjust and optimize the GPT final model according to at least one of the adjustment structure of the GPT final model, the second hyperparameter adjustment result of the GPT final model and the adjustment result of the training data of the GPT final model.
[0151] It should be noted that the above explanation of an embodiment of a multi-condition antenna optimization method based on a large model is also applicable to a multi-condition antenna optimization device based on a large model in this embodiment, and will not be repeated here.
[0152] According to the multi-condition antenna optimization device based on a large model proposed in the embodiment of the present application, it is possible to collect and divide antenna design scheme data, train and optimize the GPT model, monitor and adjust the performance indicators during the training process, evaluate and adjust the GPT model, and adjust and optimize the GPT model again according to the selection, thereby improving the accuracy and stability of the GPT model, enhancing the multi-objective optimization capability of the antenna design, improving the accuracy of the antenna design, and improving the intelligent level of the antenna design. Thus, the problem that the antenna design method in the related art is often based on experience and trial and error, and it is difficult to meet the various requirements of high-speed data transmission, large-capacity communication, low latency, etc., and the efficiency and performance are low is solved.
[0153] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include: Memory 401 , processor 402 , and a computer program stored in the memory 401 and executable on the processor 402 .
[0154] When the processor 402 executes the program, a multi-condition antenna optimization method based on a large model provided in the above embodiment is implemented.
[0155] Furthermore, the electronic device further comprises: The communication interface 403 is used for communication between the memory 401 and the processor 402 .
[0156] The memory 401 is used to store computer programs that can be executed on the processor 402 .
[0157] The memory 401 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
[0158] If the memory 401, the processor 402 and the communication interface 403 are implemented independently, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0159] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can communicate with each other through an internal interface.
[0160] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.
[0161] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned multi-condition antenna optimization method based on a large model is implemented.
[0162] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0163] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0164] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.
[0165] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection with one or N wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.
[0166] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0167] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0168] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0169] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A multi-condition antenna optimization method based on a large model, characterized in that: The following steps are involved: Collecting design data of at least one target antenna, and dividing the design data of the at least one target antenna into at least one of a training set, a validation set, and a test set; Using the Adam algorithm to set a first hyperparameter in the training process of the generative pre-trained GPT model, and iteratively adjusting the first hyperparameter using the training set to obtain a second hyperparameter that meets the target performance condition of the GPT model; Using the visualization tool TensorBoard to record the performance indicators during the GPT model training process, and when the performance indicators do not meet the preset improvement conditions of the validation set, stopping the iterative training of the GPT model to obtain the GPT final model; Based on the GPT final model, a confusion matrix is constructed, and the confusion matrix is used to analyze the output results of the GPT final model on the test set to obtain analysis results, and the ability of the GPT final model to understand antenna performance conditions is evaluated according to the analysis results, and the GPT final model and the second hyperparameters are adjusted and optimized according to the understanding ability to optimize the design of the target antenna.
2. The multi-condition antenna optimization method based on a large model according to claim 1, characterized in that: Also includes: The GPT final model is sampled to generate at least one output result, a final output result among the at least one output result is determined according to a user's selection instruction, and the GPT final model is adjusted and optimized again according to the final output result to update the GPT final model.
3. The multi-condition antenna optimization method based on a large model according to claim 1, characterized in that: The collecting design data of at least one target antenna and dividing the design data of the at least one target antenna into at least one of a training set, a validation set, and a test set comprises: Collect design data of at least one target antenna of different types, frequency bands and performance indicators; Performing data cleaning, outlier processing and standardization processing on the design data of the target antenna to obtain processed data; The processed data is divided according to a preset ratio to obtain at least one of the training set, the validation set and the test set.
4. The multi-condition antenna optimization method based on a large model according to claim 1, characterized in that: The method of using the Adam algorithm to set the first hyperparameter in the generative pre-training GPT model training process includes: Initializing a first hyperparameter in the GPT model training process to obtain an initialization result; Based on the initialization result, when the first hyperparameter does not reach a preset convergence condition, training the GPT model to obtain a training result; Based on the training result, when the first hyperparameter reaches the preset convergence condition, the iteration is stopped to obtain the first hyperparameter.
5. The method for optimizing antennas based on a large model according to claim 1, characterized in that: The iteratively adjusting the first hyperparameter using the training set to obtain a second hyperparameter that satisfies a target performance condition of the GPT model includes: Performing forward propagation on the antenna input sequence according to the GPT model to map the input sequence to a corresponding output sequence to obtain a predicted output result; Comparing the predicted output result with the actual antenna design solution to obtain a comparison result, and calculating the loss function of the GPT model according to the comparison result; Obtaining the gradient of the loss function with respect to the first hyperparameter, and back-propagating the gradient to each layer of the network to calculate the error contribution of each layer, and obtaining calculated gradient information; Based on the calculated gradient information, the first hyperparameter is updated using the Adam algorithm to obtain the second hyperparameter that meets the target performance condition.
6. The multi-condition antenna optimization method based on a large model according to claim 1, characterized in that: The use of the visualization tool TensorBoard to record the performance indicators of the GPT model training process, and when the performance indicators do not meet the target improvement conditions of the validation set, the iterative training of the GPT model is stopped to obtain the GPT final model, including: Recording the performance indicators during the training of the GPT model, and using the validation set to evaluate the actual performance of the GPT model; Compare the actual performance of the GPT model with the historical best performance to obtain a comparison result, and determine whether the performance indicator meets the preset improvement condition of the validation set according to the comparison result; If the verification loss of the GPT model is in a preset decreasing condition or the performance index is in a preset condition, it is determined that the performance index does not meet the preset improving condition of the verification set, the iterative training of the GPT model is stopped, and the GPT final model is obtained.
7. The multi-condition antenna optimization method based on a large model according to claim 1, characterized in that: The adjusting and optimizing the GPT final model and the second hyperparameter includes: The GPT final model is adjusted and optimized according to at least one of the adjustment structure of the GPT final model, the second hyperparameter adjustment result of the GPT final model, and the adjustment result of the training data of the GPT final model.
8. A multi-condition antenna optimization device based on a large model, characterized in that: include: A collection module, used to collect design data of at least one target antenna, and divide the design data of the at least one target antenna into at least one of a training set, a validation set and a test set; A setting module, configured to set a first hyperparameter in a training process of a generative pre-trained GPT model using an Adam algorithm, and iteratively adjust the first hyperparameter using the training set to obtain a second hyperparameter that meets a target performance condition of the GPT model; A recording module is used to record the performance indicators of the GPT model during training using the visualization tool TensorBoard, and when the performance indicators do not meet the preset improvement conditions of the validation set, stop the iterative training of the GPT model to obtain the GPT final model; An optimization module is used to construct a confusion matrix based on the GPT final model, and use the confusion matrix to analyze the output results of the GPT final model on the test set to obtain analysis results, evaluate the GPT final model's understanding ability of antenna performance conditions according to the analysis results, and adjust and optimize the GPT final model and the second hyperparameter according to the understanding ability to optimize the design of the target antenna.
9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the large model-based multi-condition antenna optimization method as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the large model-based multi-condition antenna optimization method as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Multi-target antenna design method based on sequence model
CN111967184A
Power distribution network dynamic topology identification method, system and device and storage medium
CN116316637A
Drug recommendation method based on GPT-4 and LangChain
CN118116578A
Resume analysis model training method, resume analysis method, electronic equipment and medium
CN118133008A
Systems and Methods for Generative Models for Design
US20210390396A1
