Increasing power plant efficiency with clustering methods and Variable Importance Index assessment

Power plant performance can decrease along with its life span, and move away from the design commissioning targets. Maintenance issues, operational practices, market restrictions, financial objectives may lead to that behavior, knowledge of appropriate actions could support system retake original performance. This paper applies unsupervised machine learning techniques identify operating patterns based on power plant’s historical data which leads identification steam generator efficiency conditions. The selected variables are evaluated in respect their impact performance, quantified by Variable Importance Index. That metric is proposed among a much wide set monitored whose variation impacts overall operation, should be controlled more attention. Principal Component Analysis (PCA) k-means++ clustering used suitable conditions one-year-long 27 recorded 360MW thermal plant. adequate number clusters identified average Silhouette coefficient Index sorts nine as most relevant ones, finally group recommended settings achieve target Results show gains values 73.5% lowest condition records 68%, 76%.