Estimating electric energy from wavepower with machine learning

In this thesis, a machine learning-based hybrid model is developed to predict the electrical energy generated from wave power using meteorological data in the Black Sea. In the study, a one-year dataset for the Samsun and Ordu buoys, obtained from the General Directorate of Meteorology for the year 2023, was used. This dataset contains parameters such as wave height, wave period, wind speed, water temperature and so on. To predict wave power, basic regression models including LSTM, ELM, ResNet, SVM, Bayesian, and RNN algorithms were developed, and their strengths and weaknesses were analyzed. A hybrid model was created by applying a stacking approach using the XGB algorithm to improve the performance of the base models. The performance of the hybrid model was evaluated using MAE, RMSE, MSE, and R² metrics, and it was observed that the XGB algorithm produced better results compared to other models. For example, in the forecasts at the Samsun buoy, the MAE value of 0.0491 for the ResNet algorithm was reduced to 0.0079 in the hybrid model. These results prove that the hybrid model provides high accuracy in wave power prediction and serves as an effective tool for regulating the irregular characteristics of wave energy. As a result of the thesis, it is concluded that the proposed model can contribute to energy regulation by predicting the irregularities in the electrical energy system generated from wave power in the short term. It is also expected that the model can be used to calculate the monthly and annual electricity generation potential in regions where wave energy conversion systems are planned for long-term deployment. This study demonstrates the effectiveness of machine learning algorithms in wave energy forecasting and offers an innovative approach to energy conversion system planning.