CN-122021353-A - Artificial intelligence auxiliary design and performance prediction method for radio frequency coaxial connector

CN122021353ACN 122021353 ACN122021353 ACN 122021353ACN-122021353-A

Abstract

The invention discloses an artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector, which relates to the technical field of radio frequency and microwave device simulation, and comprises the steps of constructing a mapping relation between an ANN model representation key size parameter and corresponding electrical performance simulation data, acquiring input target electrical performance, selecting an initial value of the key size parameter, inputting the initial value into a trained ANN model, calling a plurality of optimization algorithms, calling a prediction loss function to calculate deviation between an electrical performance predicted value and the target electrical performance, carrying out iterative search in a scanning range of the key size parameter with the deviation minimized as an optimization target, and outputting the optimal key size parameter for enabling the deviation to meet preset requirements. The mapping relation between the key size parameters and the electrical performance is established through the ANN model, and the electrical performance index is used as a target to directly optimize by adopting a reverse optimization method, so that manual iteration is avoided, re-simulation is not needed when the design requirement is changed, good design reusability is achieved, and the overall design period is remarkably shortened.

Inventors

XIAO FEI
WANG JINJIE
LI JIANKAI

Assignees

成都华兴汇明科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The artificial intelligence auxiliary design and performance prediction method of the radio frequency coaxial connector is characterized by comprising the following steps of: S1, establishing a parameterized model of a radio frequency coaxial connector, setting a scanning range of key size parameters in the parameterized model, and performing sweep, so as to generate corresponding electrical performance simulation data; S2, constructing an ANN model according to the key size parameters and the corresponding electrical performance simulation data thereof, and training to obtain a trained ANN model which is used for representing the mapping relation between the key size parameters and the corresponding electrical performance simulation data thereof; s3, acquiring input target electrical performance, selecting an initial value of a critical dimension parameter, inputting the initial value of the critical dimension parameter into a trained ANN model to obtain an initial electrical performance predicted value, and executing S4 if the deviation between the initial electrical performance predicted value and the target electrical performance does not meet the preset requirement; S4, invoking a plurality of optimization algorithms, executing a reverse iterative optimization method with target electrical performance as a target, inputting current critical dimension parameters into a trained ANN model in each iteration to obtain an electrical performance predicted value, invoking a predicted loss function to calculate deviation between the electrical performance predicted value and the target electrical performance, performing iterative search in a scanning range of the critical dimension parameters with the deviation minimized as the optimization target, and outputting optimal critical dimension parameters with the deviation meeting preset requirements.
2. The artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector of claim 1, further comprising the step of, prior to step S2: Interpolation processing is carried out on the electrical performance simulation data, the electrical performance simulation data are converted into a serial archive file format to be stored, a training data file is obtained, and the training data file is called to train the ANN model.
3. The artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector of claim 1, wherein the ANN model comprises an input layer, a multi-stage neural network layer and an output layer, wherein the number of neurons of the input layer corresponds to the number of critical dimension parameters of the radio frequency coaxial connector, and the number of neurons of the output layer corresponds to the number of frequency points of an electrical performance curve output by the ANN model.
4. The artificial intelligence aided design and performance prediction method of the radio frequency coaxial connector according to claim 3, wherein the specific process of the step S2 is as follows: Performing linear normalization processing on the critical dimension parameters, and mapping the critical dimension parameters into a preset interval to obtain normalized critical dimension parameters; Performing nonlinear processing on each layer of neurons of the ANN model by applying an activation function, so that the ANN model learns nonlinear correlation between normalized critical dimension parameters and electrical performance simulation data; and training the ANN model by utilizing the electrical performance simulation data to finally obtain a trained ANN model.
5. The artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector of claim 4, wherein applying an activation function to perform nonlinear processing on each layer of neurons of an ANN model comprises: each neuron of the ANN model input layer is processed using a Sigmoid activation function, And processing the neurons of each level of neural network layer of the ANN model by using a Leaky ReLU activation function.
6. The artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector of claim 1, wherein the training process of the ANN model is: Defining a training loss function, setting iteration times, taking the key size parameter and the corresponding electrical performance simulation data as a training data file, and dividing the training data file into a plurality of training batches; In a plurality of training batches, performing iterative training on the ANN model by adopting an optimizer to obtain a training result corresponding to each training batch; comparing the training results corresponding to each training batch with the training data files corresponding to the training batches to verify the training accuracy of the ANN model; judging whether the training precision meets a preset threshold, if so, stopping training and outputting the ANN model which is trained currently, otherwise, adjusting training parameters or continuing iteration until the set iteration times are reached, and outputting the ANN model which is trained finally.
7. The artificial intelligence aided design and performance prediction method of radio frequency coaxial connector of claim 1, wherein the predicting loss function calculates the deviation between the predicted value of the electrical performance and the target electrical performance further comprises the step of using a single-side penalty mechanism to only hold the part of the predicted value of the electrical performance inferior to the target electrical performance for weighting operation, and the rest value is set to 0.
8. The artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector of claim 1, wherein each iteration of the reverse iterative optimization method performs the following operations: s41, calling a target preprocessing function, and generating a frequency point index according to the target electrical property; S42, inputting the current critical dimension parameters into the trained ANN model to obtain corresponding electrical property predicted values, wherein the electrical property predicted values comprise an S parameter array; s43, calling an array interception function, transferring the S parameter array to the array interception function, intercepting a sub-array in the range from zero frequency to the frequency index by utilizing the frequency index generated by the target preprocessing function, and obtaining an intercepted electrical property predicted value; S44, calling a predictive loss function, and calculating deviation between the intercepted electrical property predictive value and the target electrical property; s45, sequentially calling a plurality of optimization algorithms according to the deviation, and starting optimization iteration respectively with the same initial value and boundary condition; S46, repeatedly calling the trained ANN model and the predicted loss function in the optimization iteration process, repeating the steps S42-S45, and continuously updating the critical dimension parameter until the maximum iteration number is reached or the deviation of adjacent iteration results is smaller than a preset error, and outputting the optimal critical dimension parameter.
9. The artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector of claim 8, wherein the target electrical performance includes a target operating frequency band and a target standing wave VSWR, and the target preprocessing function performs the following operations: determining the position of the target working frequency band in the discrete frequency spectrum through an argmin operator according to the target working frequency band, and generating a frequency point index; Converting the target standing wave VSWR to a target return loss S 11 : ; The frequency point index is used for intercepting the S parameter array as the limit of the array intercepting function to obtain an electric performance predicted value after interception in the target working frequency band; the target return loss is used for being input into the predicted loss function and calculating deviation from the intercepted electrical property predicted value.
10. The artificial intelligence aided design and performance prediction method of radio frequency coaxial connector of claim 1, wherein the plurality of optimization algorithms includes at least two of Nelder-Mead algorithm, powell algorithm, DIFFERENTIAL EVOLUTION algorithm and COBYLA algorithm, and the optimal critical dimension parameters are screened out by running in parallel and comparing the predicted loss function values of the optimal solutions returned by the optimization algorithms.

Description

Artificial intelligence auxiliary design and performance prediction method for radio frequency coaxial connector Technical Field The invention relates to the technical field of simulation of radio frequency and microwave devices, in particular to an artificial intelligent auxiliary design and performance prediction method of a radio frequency coaxial connector. Background The radio frequency coaxial connector is used as a core basic element for realizing high-efficiency and low-loss transmission of high-frequency signals, and is widely applied to scenes such as communication base stations, phased array radars, high-precision test equipment and the like. With the development of radio frequency systems, design indexes of radio frequency coaxial connectors are increasingly demanding, and demands thereof are increasingly increasing. Therefore, the rapid and reliable design of the rf coaxial connector structure has become a key element in the rf system engineering. The existing design of the rf coaxial connector mainly depends on the conventional means such as theoretical estimation, electromagnetic simulation or manual tuning. The designer firstly determines the basic radial dimension based on the characteristic impedance formula of the coaxial transmission line, then adopts three-dimensional electromagnetic simulation software to perform parameterization scanning (hereinafter referred to as' sweeping) on structures such as an impedance matching section, an insulating medium and the like, and continuously and iteratively searches for a dimension parameter combination meeting the electrical performance index. The electrical performance index of the rf coaxial connector generally refers to scattering parameters, such as voltage standing wave ratio (hereinafter referred to as "standing wave") and insertion loss, which measure the matching and transmission efficiency of the rf coaxial connector. However, the above conventional design method has limitations. Firstly, the nonlinear relation exists between the critical dimension parameters and the electrical performance of the radio frequency coaxial connector, a designer is required to repeatedly sweep parameters of the critical dimension parameters and iterate manually, so that the design period is longer, the increasing requirement of the radio frequency coaxial connector at the present stage cannot be met, and secondly, when the working frequency band or the interface standard is changed, repeated sweeping parameters and electromagnetic simulation are required to be carried out again, so that the reusability is poor. Therefore, a design method capable of automatically predicting the structure size, reducing electromagnetic simulation dependency and having good design reusability is important. Disclosure of Invention The invention aims to provide an artificial intelligence auxiliary design and performance prediction method of a radio frequency coaxial connector. An artificial neural network (ARTIFICIAL NEURAL NETWORK, ANN for short) is an important technical means for realizing artificial intelligence. The invention establishes the mapping relation between the critical dimension parameter and the electrical property through the ANN model, adopts the reverse optimization method to directly optimize the electrical property index as a target, avoids manual iteration, does not need to simulate again when the design requirement is changed, has good design reusability and obviously shortens the whole design period. In order to achieve the above object, the present application provides the following solutions: In one aspect, the invention provides an artificial intelligence aided design and performance prediction method of a radio frequency coaxial connector, which specifically comprises the following steps: S1, establishing a parameterized model of a radio frequency coaxial connector, setting a scanning range of key size parameters in the parameterized model, and performing sweep, so as to generate corresponding electrical performance simulation data; S2, constructing an ANN model according to the key size parameters and the corresponding electrical performance simulation data thereof, and training to obtain a trained ANN model which is used for representing the mapping relation between the key size parameters and the corresponding electrical performance simulation data thereof; s3, acquiring input target electrical performance, selecting an initial value of a critical dimension parameter, inputting the initial value of the critical dimension parameter into a trained ANN model to obtain an initial electrical performance predicted value, and executing S4 if the deviation between the initial electrical performance predicted value and the target electrical performance does not meet the preset requirement; S4, invoking a plurality of optimization algorithms, executing a reverse iterative optimization method with target electrical performance as a target, inputting current cr