Neural network current prediction for shipping guidance

Accurate current prediction is important for the safe and efficient guidance of deep draft ships using the port of Amsterdam. Because the present numerical models cannot yet generate operational current forecasts, in 1993 a feasibility study was carried out on using a neural network on a personal computer to make predictions of currents. This study resulted in a neural network that has been operational since December 1993. The error backpropagation network was trained on nine months of current measurements, with wind and water level data as input. A check against off-line current measurements carried out from January to April 1994 shows a hindcast root mean square error of 13.4 cm/s. Under astronomical conditions the root mean square error falls to under 8 cm/s