Electric power prediction of a two-body hinge-barge wave energy converter using machine learning techniques

Abstract

Investigating the electric power generated by wave energy converters (WECs) has attracted increasing attention along with the developments of wave energy converting technologies in recent years. Electric power estimation currently depends on wave-to-wire model-based numerical analysis and experimental studies of scale-down prototypes. However, traditional numerical simulations are time-consuming and expensive for power orediction of a full-scale WECs operated in real sea. More importantly, they cannot accurately predict WECs’ power performance in real time, thereby cannot be used for prediction-based real-time control. To address those problems, four machine learning techniques are employed as an alternative to traditional, complex physical modelling processes to evaluate and predict the generated electric power, based on historical experimental data of a two-body hinge-barge WEC. Comparative results of a Backpropagation (BP) Neural Network algorithm, a Long Short-Term Memory (LSTM) algorithm, a Support Vector Machine (SVM) algorithm and a Radial Basis Neural Network (RBFF) algorithm indicate that machine-learning-based approaches can accurately predict the electric performance of this complex multi-degree-of-freedom WEC, where the coefficient of determination (R2) of the BP Neural Network on the test dataset is up to 0.9987.