UDC 621.372.049.77: 004.032.26
ALGORITHM TO CREATE NEURAL NETWORK MODEL BASED ON THE EXAMPLE OF MICROSTRIP MEMS SWITCH
D. H. Nguyen, post-graduate student, RSREU, Ryazan, Russia;
orcid.org/0000-0002-2456-5220, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
E. P. Vasiliev, Doctor in Technical Sciences, department of space technologies, Professor of the Department, RSREU, Ryazan, Russia;
orcid.org/0000-0003-2747-7012, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The algorithm for modeling shunt-type capacitive MEMS switches based on artificial neural networks is proposed. Modern methods to analyze microstrip devices, including electromagnetic modeling, equivalent circuit models, and machine learning-based models are reviewed. The aim of this work is to develop an al gorithm for creating a neural network model to predict S-parameters, providing balance between accuracy and computational time. The algorithm proposed includes stages of design synthesis, training sample gener ation using Latin Hypercube Sampling (N = 1000 computational experiments), electromagnetic modeling in High Frequency Structure Simulator (HFSS) environment via PyAEDT Python library interface, data nor malization, and training of multilayer neural network with 16×16×16 architecture. The developed model achieves high approximation accuracy (R² = 0,9085 – coefficient of determination, MAPE = 1,15 % – mean absolute percentage error) while reducing computation time by more than 104 times compared to full-wave electromagnetic modeling. Independent test set verification confirms prediction accuracy for resonant fre quencies (average error 1,3 – 1,4 %) and isolation (average error 2,2 %). The universality of the algorithm is confirmed by additional study on PIN diode switch, achieving R² = 0,9684 and MAPE = 2,98 % with N = 975 samples. The proposed approach for approximating frequency-dependent output S-parameters is applicable to a wide class of microstrip switches.
Key words: circuit and electromagnetic modeling; microstrip MEMS switch; approximation; S parameters; artificial neural network; neural network model; PIN diode switch.