Antenna Intelligent Design Model Based On Machine Learning

With the continuous progress of wireless communication technology,antenna design has attracted widespread attention in wireless equipments such as radar systems and MIMO technology.As the device for transmitting and receiving electromagnetic signals,antenna is very important for affecting electromagnetic signals strength and quality.In the past few years,various new antennas and emerging technologies have emerged.As we all know,antenna synthesis and modeling is always done in full-wave electromagnetic simulation software such as HFSS,CST and so on.Full-wave electromagnetic simulation software is suitable for simulating various antennas to obtain the antenna's scattering parameter,gain,pattern,propagation constant and so on.However,this electromagnetic simulation analysis requires a large amount of computing resources,a large amount of time,a large amount of storage space,and advanced electromagnetic knowledge.In addition,traditional electromagnetic field analysis methods such as the method of moments(MOM)and finite element method(FEM),due to the existence of a large number of formula calculations and consume a lot of time,it is difficult to combine with modern intelligent optimization design procedures to improve the design efficiency of the antenna.This paper combines machine learning and intelligent optimization algorithm to propose a new hybrid machine learning model based on antenna performance requirements for intelligent design.This model requires fewer training dataset and has higher accuracy than a single machine learning model.Firstly,calculate the complex correlation coefficients of the structural size parameters and performance requirements to select the structural size that has the greatest impact on performance requirements.This can reduce the number of samples through choosing of the structure size.After selecting the structural parameters,machine learning is used to map performance characteristics to geometric parameters.The model combines 10 different machine learning models,such as support vector machines,neural networks,random forests and so on.Using cross-validation during training reduces the risk of overfitting Improve the generalization performance of the entire model by learning the advantages of the respective basic learners.In order to make the results more accurate,according to the requirements of multi-objective optimization,the improved structural size results were further optimized by using the improved beetle antennae search algorithm,and the gap between the actual results and the actual results was continuously reduced.A faster and more accurate design of the antenna is achieved,which has great significance for antenna design and analysis.