Methodology of Neural Design: Applications in Microwave Engineering

• Main Authors: Z. Raida, P. Pomenka
• Format: Article
• Language: English
• Published: Spolecnost pro radioelektronicke inzenyrstvi 2006-06-01
• Series: Radioengineering
• Subjects: Artificial neural networks; genetic algorithms; metho-dology of developing neural models Bayesian regula-rization; Levenberg-Marquardt algorithm
• Online Access: http://www.radioeng.cz/fulltexts/2006/06_02_12_17.pdf

In the paper, an original methodology for the automatic creation of neural models of microwave structures is proposed and verified. Following the methodology, neural models of the prescribed accuracy are built within the minimum CPU time. Validity of the proposed methodology is verified by developing neural models of selected microwave structures. Functionality of neural models is verified in a design - a neural model is joined with a genetic algorithm to find a global minimum of a formulated objective function. The objective function is minimized using different versions of genetic algorithms, and their mutual combinations. The verified methodology of the automated creation of accurate neural models of microwave structures, and their association with global optimization routines are the most important original features of the paper.