Study on the Performance of Machine Learning Forecasting Models for Coal Mill Equipment Failure Prediction in Steam Power Plant

Digitalization at Indramayu Coal-fired Power Plant has been implemented to improve equipment efficiency and reliability, but challenges remain in identifying various operational failure modes that require further development. This study proposes the application of machine learning algorithms, specifically Long Short-Term Memory (LSTM) and Gradient Boosting Classifier, to predict and detect potential plugging in Coal Mill, one of the main causes of output decline in the plant. This method utilizes real-time Coal Mill sensor data, including motor current, coal flow, inlet and outlet temperatures, and air-to- fuel ratio. This historical data is processed to produce accurate predictions of operating parameters, allowing preventive actions against potential failures. This model is validated with Root Mean Square Error (RMSE) and F1 Score, producing the highest RMSE value of 0.88 and F1 Score of 0.97, indicating high plugging detection accuracy. The implementation of this method allows operators to receive early warnings regarding potential plugging, so that quick actions can be taken to maintain operational continuity and avoid damage. The results show that this model contributes to significantly improving the efficiency and reliability of Coal Mill operations, supporting optimal achievement in power plant management. The development of this machine learning-based system is expected to continue to support digitalization efforts in the energy sector, maintaining the efficiency, safety, and stability of coal-fired power plant operations.