Near Term Ship Motion Forecasting From Prior Motion

Abstract

Ship-motion forecasting can be useful for naval operations such as aircraft landing, cargo transfer, off-loading of small boats, and ship “mating” between a large transport ship and smaller ships. The forecasted ship motion is particularly useful in sea states above SS3 when unanticipated large motions can suddenly occur. A 5- to 10-second forecast of future ship motion provides the operator time to compensate for the motion to avoid serious collisions. Ship motion forecasting can enable autonomous landing during higher sea states and can also provide an alternative ship location estimate during emergency procedures such as loss of communication link. This paper summarizes the evaluation of four categories of forecasting methodologies: subspace algorithms, autoregressive algorithms, nonlinear autoregressive using a wavelet network, and perturbation error methods. Simulated Model 5415 ship motion was evaluated in 24 conditions including sea states 4–6, ship speeds of 5, 10, 20, and 30 kts, and wave headings 150 and 180 deg (bow and head seas). For each condition, the simulation motion data was divided into twenty 7.5-minute segments consisting of 0.5 to 5 minutes of training data and 2.5 minutes of testing data. Two types of forecasting accuracy metrics were developed. One metric was based on forecasting simulated ship motion and the other metric involved forecasting periods when the motion exceeds threshold limits within a 4-second window, representing a quiescent period. The results indicate that for simulation forecast accuracy, the correlation coefficient between forecasted and actual motion was greater than 80% for 5-sec forecast horizons, and greater than 60% for 10-sec forecast horizons. For motion threshold forecasting, the forecasting accuracy was greater than 90% for 5-sec horizons and greater than 60% for 10-sec horizons. A qualitative assessment of both simulation and threshold metrics indicated that 80% accuracy produces a good forecast and 60% accuracy produces an acceptable forecast. Threshold forecasting can forecast the presence and duration of near-future quiescent periods, enabling safer, more efficient operations and reduced cost of ship-based aviation operations such as launch, recovery, and movement of aircraft.