Applying Deep Learning for Long-Term Prediction of Water Column Height in the Chamber of Fixed Oscillating Water Column Wave Energy Converters

초록

This study presents a method for predicting the changes in water column height inside the chamber of a Fixed Oscillating Water Column(OWC) Wave Energy Converters(WEC) from the time series of wave heights obtained from a Waverider Buoy. Predicting the changes of the water column height in the chamber solely based on the time series data of incident wave heights had limitations in terms of accuracy. To address this issue, this study compared the spectral correlation between the height of the incident waves and the water column height in the chamber in terms of similarity in the shape of the spectra. The method for calculating similarity involved comparing the shapes of the wave spectra, and Cosine Similarity was used for this purpose. Based on this correlation, it was aimed to predict the behavior of the water column by forecasting the spectrum of the water column height from the incident wave height spectra. Firstly, the tidal characteristics included in the wave height time series were removed, leaving only the changes in wave and water column heights to be converted into spectra. The analysis of the similarity between the spectrum of the incident waves and the spectrum of the water column height in the chamber shows that the magnitude of the similarity is closely related to the characteristics of the incident waves. The similarity of the two spectra changes depending on the size of the wave significant height and energy, and it was possible to identify the range of incident waves that causes actual generate power. It was possible to extract incident waves that are directly related to power generation by setting Cut-in and Cut-out criteria based on the significant height and energy of the incident wave. These waves were used as input data for the deep learning model, and the spectral prediction model was designed to estimate the intensity of the Power Spectral Density (PSD) for each frequency contained in the wave spectra. The input and output of the prediction mode