Real Time Estimation of Local Wave Characteristics from Ship Motions Using Artificial Neural Networks

Wave characteristics such as significant wave height, peak period and main direction have an important effect on all aspects of ship operations. Yet, estimation of these characteristics in real time in an accurate, robust and efficient manner poses a significant challenge. Current approaches include using wave buoys, wave radars, weather forecasts and regression based on ship motion statistics. External sources are not always available or reliable, and wave radars are expensive. Statistics based methods work reasonably well for wave height, but fail to estimate peak period and relative wave direction. In this study the attention is focused on estimating the wave characteristics from ship motions using artificial neural networks. The inputs to the neural networks were the time histories of 6-DOF ship motions, and the outputs were the wave characteristics. Using time series instead of statistics preserves phase differences between signals. Especially when estimating the relative wave direction, these phase differences encode essential features. Therefore, the problem was treated as multivariate time series regression. Two different data sources were used with neural networks: in-service measurement data and numerical simulation data. In-service measurement data was collected from a frigate type vessel for a period of two years. The numerical simulation data was produced by using a time-domain seakeeping and manoeuvring software. In this paper, we present methods to successfully implement neural networks for real time estimation of local wave characteristics. The paper will include details about the measurement data, numerical simulations, the networks architectures and their performances on the measurement and simulation data.