Predicting near-real-time total water level with an artificial intelligence model based on Australia's tidal wave energy belt dataset

Predicting near-real-time total water level with an artificial intelligence model based on Australia's tidal wave energy belt dataset

Wave energy resulting from interactions of Earth’s gravitational field with the Sun and Moon is considered a significant resource of distributed variable renewable energy to contribute to the supply of consumer electricity. In this study, we developed an artificial intelligent model based on extreme learning machine (ELM) and hourly seawater level (Ht) data collected between February 02-January 2001 and 31-December 2005 at wave energy sites in Broome, Darwin, Cape Ferguson, and Milner Bay in Northern Australia to predict Ht over near real-time hourly scales. The proposed ELM model is benchmarked against the emotional neural network (EmNN) and extreme gradient boosting (XGBoost) models. The proposed ELM is shown to outperform the EmNN and XGBoost models concerning training, validation, and testing data. For all four study sites, the proposed ELM model achieved a correlation coefficient of 0.998–0.999 vs. 0.975–0.993 (for the EmNN) and 0.975–0.998 (for the XGBoost). Correspondingly, the Legates & McCabe’s Index were 0.936–0.973 vs. 0.775–0.879 and 0.775–0.953 for the ELM and EmNN models and the XGBoost model's testing phase (0.775–0.879), resulted in a significant reduction in root mean square (0.092, 0.069, 0.042 and 0.027 for the sites Broome, Darwin, Cape Ferguson and Milner Bay, respectively) and mean absolute error (0.044, 0.054, 0.033 and 0.021 for the sites Broome, Darwin, Cape Ferguson and Milner Bay, respectively), while Willmott’s Index (1.00 for all sites) and Nash–Sutcliffe’s coefficient (0.998 for Broome and Darwin, 0.996 for Cape Ferguson, and 0.995 for Milner Bay) comparing the predicted and observed Ht registered the highest values compared to all benchmark models. The ELM model also produced the greatest frequency of errors in the smallest error bracket, thus demonstrating its efficacy in predicting hourly seawater levels. In addition, the study also extracted results by developing a multiple linear regression (MLR) model, one of the well-known forecasting technique, and compared it with the machine learning models used in this study. According to the experiment, the study finding that the MLR model has slightly better accuracy in some cases when it was compared with the EmNN and XGBoost models. However, the study objective model (ELM) has relatively better performance in all study sites comparing with the MLR model. We, therefore, conclude that the proposed ELM model may be a useful stratagem for monitoring seawater levels in near-real-time and adopted for forecasting wave energy potentials in tidal energy belt regions.

Files associated with this item cannot be displayed due to copyright restrictions.