Regression Prediction of Performance Parameters in Ship Propulsion Equipment Simulation Model Based on One-Dimensional Convolutional Neural Network

Deep learning methods such as the one using Convolutional Neural Network (CNN) have made remarkable achievements in computer vision and natural language processing. Compared with the conventional neural network structures, CNN features low complexity, fewer parameters, and higher degree of nonlinearity. As the sizes of sensor signal input are often different from those of image input, using CNN to monitor the equipment status is a new issue compared with image recognition. To examine the impacts of various one-dimensional CNN structures on the regression of performance parameters, this paper conducts a preliminary study on the application of CNN in equipment status recognition, and utilizes published simulation datasets of ship propulsion equipment to train and test one-dimensional CNN models with different structures. The results show that the size of convolution kernels hinges on the attributes of input features when one-dimensional CNN is used for data regression prediction. In the case of independent and direct feature input, the training effect can be effectively improved by using 1 x 1 convolution kernels and the Network In Network (NIN) structure.

Automated summary

This paper investigates the application of CNN in equipment status recognition and explores the impact of different one-dimensional CNN structures on performance parameter regression prediction. The results show that the size of the convolution kernels depends on the attributes of the input features. In the case of independent and direct feature input, using 1x1 convolution kernels and the Network In Network (NIN) structure can effectively improve training performance.