Multi-path-based machine learning auxiliary antenna design method

Abstract

The invention discloses a multi-path-based machine learning aided antenna design method, which comprises the following steps: multi-fidelity training set data is learnt by using a deep Gaussian process regression machine learning algorithm to obtain an agent model; and on the basis of a confidence lower bound pre-screening method, a multi-path strategy is provided, and when an evolutionary algorithm is used for performing global optimal value prediction on an agent model, different confidence lower bound constants dynamically balance the convergence and exploratory performance of the algorithm, and search of a global optimal solution is performed through multiple paths. The method has robustness while ensuring the convergence speed of the algorithm.

Description

technical field [0001] The invention belongs to the technical field of antenna design, and relates to a multipath-based machine learning-assisted antenna design method. Background technique [0002] In the past two decades, machine learning methods based on artificial neural networks and Gaussian process regression have been widely used in the design of electromagnetic devices such as antennas and antenna arrays, and have achieved good results. Machine learning establishes cheap proxy models by learning antenna structure parameters and target responses, which reduces the computational burden of thousands of function evaluations using expensive full-wave simulation calculations and meta-heuristic algorithms in traditional optimization. In order to prevent the evolutionary algorithm from falling into a local optimum, the method of Lower Confidence Bound (LCB) is widely used in the optimization process, but in the traditional machine learning-assisted optimization design, the s...

AI Technical Summary

Problems solved by technology

Improper selection of LCB constants will reduce the speed of convergence and the performance of the algorithm

Images