基于人工神经网络的光学膜系设计方法

An Optical Coating Designing Method Based on an Artificial Neural Network

提出一种基于人工神经网络算法,应用于膜系的设计。在膜系自动优化过程中,根据确定的边界条件范围内,计算出膜系设计的最优解。采用了新型的评价函数,使得设计结果更符合设计要求。优化过程中能够快速学习网络权重,并且能够摆脱局部极点的问题。结果表明在同样的条件下,人工神经网络算法可以得到更合理的膜系。

A network algorithm based on artificial neural network is proposed for coating design. The optimal solution is calculated during Auto-optimization process of coating according to the boundary conditions, which is more suitable when a new evaluation function is utilized. The network weights can be quickly learned during the optimization process, and can get rid of local pole. The results witness a more reasonable coating under the same conditions. The quantity of training samples determines the quality of neural network. As the sample reaches a certain number, a better training neural networks is achieved. However, an excessive number of samples can slow down the training and computing and thus affect the training effect, whereas insufficient number of training samples cannot result in suitable training, affecting the membrane system optimization.