A Novel Probabilistic Method to Model the Uncertainty of Tidal Prediction

This paper develops a probabilistic model to predict the tidal current for modeling the prediction uncertainty, and thereby the forecast error. This requires the extension of the deterministic models from a point-by-point forecast to the probabilistic models with prediction intervals (PIs). The proposed model uses PIs to construct the bandwidth, which models the uncertainty of tidal current prediction properly. It uses the lower upper bound estimation method to train the neural network (NN) without making any assumption about the distribution of the forecast error. In order to adjust the weighting and biasing factors of NN, firefly algorithm with a new two-phase modification method is developed to search the problem space globally. Two benchmarks are used to show the search ability of the algorithm. The high accuracy of the proposed model is examined on the practical tidal data collected from the Bay of Fundy, NS, Canada.