PREDIKSI ARUS PASUT RERATA BERDASARKAN KEDALAMAN DAN UJI LAPANGAN MENGGUNAKAN KOMBINASI GAUSSIAN PROCESS REGRESSION DAN ANALISIS HARMONIK

• Main Author: William Rogers, Ben
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75103
• Institution: Institut Teknologi Bandung

The Gaussian Process Regression method – occasionally combined with a least-squared harmonic analysis – of tidal current prediction has been confirmed to be reasonably accurate using numerically simulated tide and tidal current data. However, the capability of this method in handling water level and current data acquired from the field has not been verified. Here we assess the performance of the Combined Gaussian Process Regression – Harmonic Analysis method of tidal current prediction using water level and current data acquired from the field with varying training data length. This is done to show the general performance of the method and an indication of the optimal training data acquisition duration. Assessing the performance of the method based on deviation, correlation, and error pattern aspects, we find that the resulting predictions have acceptable accuracy when using data with less significant non-tidal signals and at least two weeks of training data. However, with data containing strong non-tidal signals, the method is capable of generating predictions without overfitting the model to non-tidal correlated data.