Recent Advances and Future Trends in Neuro-TF for EM Optimization

Artificial neural networks (ANNs) are important tools to perform electromagnetic (EM) parametric modeling and design optimization for microwave structures. Recently, an advanced knowledge-based ANN technique, called neuro-transfer function (short for neuro-TF), has been developed. In the neuro-TF method, transfer function is used as the prior knowledge that expresses the highly nonlinear EM responses versus frequency. Using this transfer function knowledge, the remaining relationships of the transfer function parameters versus geometrical variables are less nonlinear for ANN to learn, resulting neuro-TF to be more accurate and robust. The trained neuro-TF model can be used as surrogate model to perform fast surrogate-based EM design optimization. Future trends for neuro-TF technique can be exploring new ways to incorporate different transfer functions into various intermediate parts of neural structures and training, and combining with various generic optimization methods. Furthermore, incorporation of EM internal formulations into neuro-TF may lead to new solutions to conquer the expense of EM design optimization of microwave structures.