Prediction of Powerplant Performance by Machine Learning Technique

The IEEE Standard 762-1987 introduced a comprehensive set of 25 indices for reporting power plant reliability, availability, and productivity. The sheer number of indices can be impractical, and an univariate indicator is preferable for real-world applications. Developing a single indicator derived from the 25 indices is the work's main theme. This study addresses this issue using machine learning to consolidate these indices, ultimately classifying power plant performance into "performed" and "unperformed." Monthly performance data from 2020 were collected from 570 power plants in the Java-Sumatra interconnected electric system. These data were obtained from the Generation Availability Information System, a proprietary application owned by the Indonesia National Electric Power Company. Using the $k$-means clustering method, we classified the performance of these power plants. The clustering process was supplemented with the opinions of seven experts with extensive experience in the system. The comparison between machine learning classifications and expert opinions is presented.