Transfer Learning-based Adaptive Diagnosis for Power Plants under Varying Operating Conditions

Abstract

Transfer learning is a method that transfers knowledge learned from a source domain to a similar target domain to improve learning. In power plants, obtaining sufficient anomaly data is difficult due to the characteristics of the systems. Transfer learning enables learning with only a small amount of data from the target domain by using a model trained in a similar domain. By applying transfer learning, models developed for one power plant can be expanded and used in other power plants where available data are limited.

Using actual data from an operating combined-cycle power plant, an anomaly diagnosis model was developed and tested. Its applicability to different operating conditions and anomaly cases was evaluated through transfer learning. The fine-tuned pre-trained model was effectively adapted with limited target domain data. Transfer learning was applied despite the limitations of data and distribution differences. The expandability of anomaly diagnosis models to different power plant systems was demonstrated by applying transfer learning.