Forecasting Electricity Generation of a Geothermal Power Plant Using LSTM and GRU Networks

Abstract

The growing concerns over environmental variations and the depletion of fossil fuel reserves have encouraged many countries to prioritise on renewable energy sources. Geothermal power provides a clean and consistent energy source which enhances the diversification of energy supply profiles. However, accurate forecasting of geothermal electricity generation remains a challenging task due to the complex and dynamic characteristics of underground heat reservoirs. This study utilises machine learning (ML)-based methods to forecast an hour-ahead energy generation of the Kizildere 3 Geothermal Power Plant (GPP) which holds the largest installed capacity in Türkiye. ML-based algorithms present a robust alternative to conventional numerical modelling approaches by capturing non-linear relationships and ascending accuracy of forecast. By employing historical operational data, long short-term memory (LSTM) and gated recurrent unit (GRU) networks are evaluated and compared in terms of prediction performance metrics such as coefficient of determination (R2) and normalised root mean squared error (nRMSE). The results highlight the potential of data-driven models in improving short-term geothermal energy forecasting which contributes to more efficient grid integration and operational planning. © 2025 IEEE.