Deep Learning Approach for Predicting Thermal Behavior of Hydropower Generator-Stator: A Case Study of a Hydropower Power Plant in Thailand

Hydropower generation is a cost-effective and environmentally friendly energy source that converts the kinetic energy of flowing water into electricity. However, temperature control in power generators, particularly in the conductor windings in the stator, remains a significant challenge for maintaining power generation performance. Several factors influence temperature, and their relationships are quite complex, making it difficult to solve the problem using standard theoretical approaches. This research developed a deep learning model to monitor temperature trends in the conductor windings of a 125 MW hydropower plant in Thailand. Data collected between 2018 and 2021 on electricity generation, reservoir water levels, water and air flow rates, inlet temperatures at the heat exchanger, and conductor winding temperatures were used to train and validate the models. The study implemented three neural network models: a Feedforward Neural Network (FNN), a Multilayer Feedforward Neural Network (MFNN), and a Long Short-Term Memory (LSTM) network. The results showed that the LSTM model provided the most accurate predictions, with a Mean Squared Error (MSE) of 0.00373. Shapley Additive exPlanations (SHAP) values were used to interpret the model predictions, identifying key variables such as electricity generation, water temperature, and water flow rate as the most influential factors affecting system behavior. The findings suggest that deep learning models can effectively predict temperature variations, enabling proactive maintenance and improving operational efficiency in hydropower plants.