Modeling of nonlinear dynamic system based on Deep Operator Network
作者
Yanfeng Wang,Longfei Zhang,Zhi Zhou Li
出处
期刊: [Institute of Electrical and Electronics Engineers] 日期:2021-10-22卷期号:: 5914-5918被引量:4
标识
DOI:10.1109/cac53003.2021.9728625
摘要
In this paper, we focus on solving the problem of learning operators to approximate nonliner system. In the neural architecture called DeepONets, we add the CNN(Convolutional Neural Networks) network to speed up the traning set and we use the lakey-relu to awake more information to gain better convergence performance. We build two subnetworks, one for encoding the input functions, another for the location variables, and then we do a mergement to get the output. When the system gains various input functions, and the response which we can get are some signal points,we can apply this architecture to modeling the system. We test DeepONets on a nonliner system which is a gravity pendulum with an external force, and show that DeepONets can achieve better convergence performance. The Test MSE(Mean Square Error) can achieve 5.518e-08, while the Train MSE is 5.712e-08.