Deep learning model for predicting power spectral density of water column height in a fixed oscillating water column using remotely measured wave height

This study proposes a method to predict the water column height inside the chamber of a fixed oscillating water column (OWC) wave energy converter (WEC). In OWC WECs, variations in the water column height and the corresponding air pressure changes within the chamber drive a turbine that generates electricity. The objective of this research is to forecast the response of the water column to incident waves by using the data collected from a Waverider buoy located 1.29 km away from the OWC. Given this distance, real-time prediction of the water column height by using only the wave height data is challenging. To address this challenge, we establish a relationship between the power spectral density (PSD) of incident waves and that of the water column height inside the OWC chamber. A fully connected network (FCN) deep learning model was used to capture the continuous frequency-based relationship between the wave spectrum (input) and water column height spectrum (output). Cosine similarity was used to measure the correlation between the two spectra. The results of this analysis demonstrated that power was generated when wave energy and significant wave height (SWH) surpassed a threshold, which is consistent with the principles of energy transfer efficiency in OWC systems. Based on this observation, cut-in criteria were defined to select relevant training data for the model. The model's performance was validated by predicting the water column height spectrum at the current time, 15 min later, and 30 min later. A comparison between the predictions and the actual measured spectra indicated that the accuracy of the predictions exceeded 90 %, especially when the SWH exceeded 1.5 m and the average wave period was 6.5–7 s. These results confirmed that the response characteristics of the water column can be predicted accurately by using the wave data obtained from a remote location, offering valuable insights for optimizing energy generation in OWC WEC systems. • Wave data from a buoy 1.29 km away were used to predict water column variations inside the OWC chamber. • A deep learning model was developed to forecast water column height spectrum up to 30-min ahead. • Tidal height, wave direction, SWH, and mean period were evaluated for their effects on prediction accuracy. • Applying cut-in thresholds for SWH and mean period improved accuracy to over 90% in high-energy sea states. • Prediction accuracy increased with the magnitude of significant wave height and spectral mean wave period.