Analysis of Gas Turbine Compressor Degradation Using Random Forest-based Machine Learning Model

When a gas turbine for power generation is operated for a long time, foreign particles in the air adhere to the compressor blade and the blade surface becomes contaminated Compressor contamination degrades the performance of the compressor and the gas turbine, and the compressor surge margin is reduced. To prevent this problem, the gas turbine operator periodically performs compressor water washing to remove contaminant particles adhering to the compressor blades to maintain gas turbine performance. However, there is no precise method to evaluate the contamination and deterioration of the compressor in the current performance monitoring system. Consequently, the compressor washing cycle is determined empirically by anticipating the degradation of compressor performance. Therefore, in this study, a machine learning model predicting compressor performance was developed by using real gas turbine plant operating data and compressor performance degradation resulting from compressor contamination was analyzed.