Parameter Optimization for CNN-LSTM by Using Uniform Experimental Design

Abstract

This paper focuses on parameter optimization for CNN-LSTM neural networks. Many parameters influence the entire performance in CNN-LSTM. However, there are few studies to consider optimizing the combination of parameters. To obtain the best combination of parameters effectively, this paper used a uniform experimental design (UED) to execute the experiment design. This paper collected current data from a machine tool feed system, including lubricated and unlubricated feed shafts. This paper used a uniform layout of U10(1010) to search the parameter combination of CNN-LSTM. From the experimental results, the combination obtained by UED can get the best performance for CNN-LSTM.