A Data-Driven Sensitivity Analysis Approach for Dynamically Positioned Vessels

Abstract

For safety-critical marine operations, the dynamically positioned (DP) vessel should maintain a predetermined heading and position for varying environmental conditions using the thrusters. Studying the effect of each thruster to the capability of DP vessels is significance but challenging. This paper presents a data-driven and variance-based sensitivity analysis (SA) approach that can dig into the ship sensor data to estimate the influence of each thruster for DP operations. Considering high-computational cost of variance-based SA, an Extreme Learning Machine (ELM) -based SA is proposed. To apply the SA to sensor data, an ANN is built and trained on the basis of ship sensor data and then employed as a surrogate model to generate Monte Carlo (MC) samples. A benchmark test shows the correctness of the proposed approach. A case study of SA in DP operation is conducted and the experimental results show that the proposed approach can rank and identify the most sensitive factors. The proposed approach highlights the application of variance-based SA in data-driven modeling for ship intelligence.