Analysis and modeling of sensor data for ship motion prediction

This paper presents an analyzing and modeling scheme that can dig into the ship sensor data to generate models for ship motion prediction. As the raw sensor data contains information that is noisy and discontinuous, three data cleaning methods used for noise reduction, data continuing and resampling are developed. According to specific ship applications, various ship motion constraints are designed to filter and establish meaningful data sets which will be used in the modeling step. Correlation analysis within the data sets is performed to figure out how significant the sensor data contributes to the predictive target. An flexible neural network based modeling mechanism is implemented, in which the user can design the model based on the correlation analysis. By training and testing the neural network using the generated data sets, the ship motion predictive model is obtained. Through a case study of fine maneuvering, the scheme is verified efficient in analyzing and modeling sensor data for ship motion prediction.