Memory Efficient DGTD Method Based on Radial Basis Function Neural Network Prediction Scheme

A novel and efficient electromagnetic analysis scheme for periodic structure is proposed, which introduces the radial basis function neural network (RBFNN) based data regression prediction method into the memory efficient discontinuous Galerkin time domain (DGTD) method to achieve low-memory and short-time simulation. Initially, this approach establishes a reasonable number of time step, utilizing the results from this stage as training values for the prediction method. Then, the training data undergoes training via the prediction method, serving as a replacement for subsequent time-stepping processes, thereby achieving a significant optimization in computation speed. The accuracy and advantages of the algorithm were validated through simulation of a periodic antenna array.