This letter proposes a new surrogate-based multiphysics optimization technique for microwave devices incorporating artificial neural networks (ANNs) and trust-region algorithm. In the proposed technique, at each optimization iteration, we build an accurate and efficient ANN surrogate model using multiple multiphysics training samples around the optimized solution from the previous iteration. A parallel data generation technique is exploited to accelerate the optimization process. To improve the convergence of the proposed technique, we use a trust-region algorithm to recalculate the ANN surrogate model range at each optimization iteration. By using the proposed technique, the values of design parameters have a large and effective update toward the optimal solution at each iteration, and the optimization can converge in fewer iterations. Therefore, we can achieve the optimal solution faster than existing multiphysics optimization methods. A waveguide filter using piezo actuator is used as an example to demonstrate the validity of our technique.

中文翻译：

---

利用神经网络代理的微波设备并行多物理场优化

这封信为微波设备提出了一种新的基于代理的多物理场优化技术，该技术结合了人工神经网络（ANN）和信任区域算法。在提出的技术中，在每个优化迭代中，我们围绕着来自先前迭代的优化解决方案，使用多个多物理场训练样本来构建准确而高效的ANN替代模型。利用并行数据生成技术来加速优化过程。为了提高所提出技术的收敛性，我们使用信任域算法在每次优化迭代时重新计算ANN代理模型范围。通过使用所提出的技术，设计参数的值在每次迭代时都朝着最佳解决方案进行了较大且有效的更新，并且优化可以收敛于更少的迭代中。所以，我们可以比现有的多物理场优化方法更快地获得最佳解决方案。以使用压电致动器的波导滤波器为例来说明我们技术的有效性。