| Terahertz(THz)waves have a frequency range of 0.1 terahertz to 10 terahertz,which is between the long-wave millimeter wave and the short-wave infrared light.electromagnetic waves.The terahertz perfect absorbing material MPA is an important photonic component for detecting,modulating and manipulating terahertz radiation.Therefore,a novel terahertz all-dielectric wave absorber is designed in this paper.Compared with the conventional metal absorber,the absorber proposed in this paper has the advantages of high pressure resistance,high temperature resistance and high resistivity,and the non-metallic materials used in the absorber in this paper are rich in variety,wide selection,and the properties of different materials.Each has its own characteristics and is cheap.Compared with the all-dielectric absorber,the absorber proposed in this paper has a wider absorption bandwidth and is superior to most non-metallic absorbers in electromagnetic wave absorption performance.Then,in view of the necessary links in the design process of the absorber,that is,the shortcoming of long time and low efficiency in the simulation stage of the absorber,based on the basic principles of deep neural networks,this paper builds a learning all-dielectric terahertz absorber.A six-layer fully-connected neural network based on the mapping relationship between the structural parameters and the reflection spectrum.The time required for simulation is greatly reduced,and the frequency is extracted as the input data of the model,the input and output structure of the model is optimized,the prediction accuracy of the model is greatly improved,the generalization ability of the model is enhanced,and the model’s performance is improved.utilization.Finally,this paper explores the optimization method of the fully connected neural network in the terahertz absorber scenario.The data set is vectorized in the data preprocessing part,which greatly improves the speed of the network operation,and at the same time normalizes the input data.processing,so that the input data has the same degree of influence on the neural network training results.After 200 iterations,the cost function of the neural network is stable at 1.03.Finally,the trained neural network model is tested with the test set data.The results show that,compared with the actual reflectance spectrum,the reflectance spectrum predicted by the neural network is much better than the actual reflectance spectrum.good fitting effect.Therefore,for the reflection spectrum of the absorber,the prediction method using the neural network can improve the calculation efficiency on the premise of ensuring the accuracy of the results.Tests have shown that,under the same equipment and conditions,compared with the traditional data simulation method,the introduction of the neural network reduces the average time consumption for calculating the reflectance spectrum by 300 times each time. |
