ARABIAN JOURNAL OF GEOSCIENCES, cilt.15, sa.1625, ss.1-19, 2022 (Scopus)
In coal-fired or nuclear power plants installed at the seaside, turbine efficiency is directly dependent on sea water temperature
(SWT). Given the long-term average climatic conditions, the cooling medium temperature plays an important role in
the design of any power plant. Therefore, the efficiency in electricity generation is significantly affected as instantaneous
changes in seawater temperature will cause the cooling environment design temperature of the plant to deviate. In this
respect, accurate SWT estimation plays an important role for electrical output from power plant applications. In this study,
various machine learning approaches namely fuzzy C-means clustering (FCM), grid partition (GP), subtractive clustering
(SC)-based adaptive neuro-fuzzy inference system (ANFIS), and long short-term memory (LSTM) neural network were used
to make one-day ahead SWT predictions. Analyses were made using 5-year daily mean SWTs measured by Turkish State
Meteorological Service between 2014 and 2018 at 3 different stations (Mersin, Izmir, and Samsun provinces) on the coasts
of the country. Mean absolute error (MAE), root-mean-square error (RMSE), and correlation coefficient (R) were used as
evaluation criteria. According to the daily SWT prediction, the best results for MAE, RMSE, and R values were obtained
for Mersin as 0.1003 °C, 0.1654 °C, and 0.999594, respectively, with ANFIS-GP model; for Izmir as 0.1754 °C, 0.2638 °C,
and 0.998962, respectively, with ANFIS-SC model; and for Samsun as 0.2716 °C, 0.3629 °C, and 0.998285, respectively,
with LSTM model.