TIDAL CURRENT PREDICTION BY USING GAUSSIAN PROCESS REGRESSION

• Main Author: William Rogers, Ben
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/55504
• Institution: Institut Teknologi Bandung

This Undergraduate Thesis aims to assess the accuracy of tidal current prediction using a machine learning algorithm, the Gaussian Process Regression. The method used in this study is by desktop simulation. The first step is to analyze the current ellipse parameters of each tidal constituent used in this study. Subsequently, the expected tidal current velocity caused by each tidal constituent is calculated and used as predictors in training the tidal current response of the machine learning algorithm. The verification is done through comparisons between the predicted tidal current and the actual tidal current in the same period. The error of the predicted tidal currents using machine learning are compared to the error of the conventional method of predicting tides, the harmonic analysis method. The result of this study is that the harmonic analysis method generally has higher accuracy, correlation to the actual data, and is more reliable compared to the method using machine learning. However, certain unspecified signal frequencies present in the data are also present in the prediction using machine learning, yet the amplitudes of these signals are magnified. These frequencies are absent in the prediction using harmonic analysis. The results confirm that tidal current prediction using machine learning generally has a lower accuracy than the harmonic analysis method. However, the machine learning method may be beneficial in showing present signal frequencies. The Gaussian process regression model’s hyperparameters may be tuned to yield a higher degree of accuracy similar to predictions using the harmonic analysis method.