Abstract:
This research developed a highly efficient deep learning model designed to forecast the thermal behavior of hydropower generators, focusing on the analysis of variables that influence the cooling system in the power generation process. The model aids preventive maintenance by evaluating critical variables and their relationships to system efficiency In this study, a deep learning model based on the Long Short-Term Memory (LSTM) neural network was selected to predict the thermal behavior of a hydropower generation system. The model considers input parameters such as inlet water temperature, water flow rate, power generation capacity, reservoir water level, and variables reflecting seasonal and environmental impacts on the power generation system. The target outputs for thermal prediction include the inlet and outlet air temperatures of the heat exchanger, stator temperature, and outlet water temperature from the heat exchange system. Through training, validation, and testing, the model demonstrated its ability to accurately forecast unseen data for the year 2019, with prediction errors for inlet air temperature, outlet air temperature, outlet water temperature from the heat exchanger, and stator temperature of 0.46%, 0.09%, 0.16%, and 0.93%, respectively. Additionally, the model effectively predicted data from 2021 to 2024, even under conditions differing from the original training dataset. These results highlight the flexibility of the model and the potential for future power generation system management The developed model was utilized to analyze the impact of input parameters using SHAP (SHapley Additive exPlanations) values and Sobol's sensitivity analysis. The results indicated that inlet water temperature and water flow rate were critical factors influencing the thermal behavior of the power generation system. Specifically, inlet water temperature exhibited a first-order and total-order sensitivity index exceeding 75% during critical periods. Conversely, significantly increasing the water flow rate did not substantially enhance cooling performance due to the inherent system heat transfer limitations. Additionally, higher power generation levels contributed to increased heat accumulation in the stator winding generator. Scenario simulations under varying conditions further demonstrated that a 20% increase in inlet water temperature led to a significant rise in the system temperature, whereas a 20% decrease improved cooling efficiency. Moreover, doubling the water flow rate revealed a thermal transfer limit of approximately 37°C. These findings emphasize the importance of controlling inlet water temperature and flow rate to maintain efficient cooling system performance This research supports the application of the developed model in preventive maintenance systems, such as the development of automated alert systems and strategic decision-making support. These applications aim to enhance the efficiency and reliability of the cooling system in the hydropower generator while promoting sustainable energy management in the future.