Sea state identification based on vessel motion response learning via multi-layer classifiers

Abstract

In order to extend the operational weather window for marine vessels under Dynamic Positioning (DP) control, a novel sea state identification method with multi-layer classifiers is proposed in this paper. Due to the distinction of system responses for various sea states, four motion signals including surge, sway, roll and yaw are adopted for classification purpose. Firstly, preprocessing techniques, like filtration and k-means clustering are performed to the raw data to filter out the “corrupted” low frequency (LF) information and generate the band-pass filter bank. Then, the processed data is decomposed into 20 categories via Hilbert-Huang transform (HHT), filter bank method and wavelet transform and 11 statistical features are extracted for each category. Subsequently, Max-relevance Min-redundancy (mRMR) method helps to select salient features with best trade-off between relevance and redundancy. With these selected features, a newly developed three-layer classification structure with Adaptive Neuro-Fuzzy Inference System (ANFIS), Random Forest (RF) and Particle Swarm Optimization (PSO) based combination classifiers is proposed to derive the current sea state. The simulation results demonstrate that the proposed identification system can achieve satisfactory classification accuracy. Moreover, the multi-layer classifier outperforms single layer classifier and can rapidly classify the sea state in real-time implementation.