Assuming that a generator is monitored by the system operator via a PMU device positioned at the generator's terminal bus, we pose and resolve the question of the real-time, data-driven and automatic monitoring of the generator's performance. We establish regimes of optimal performance for four complementary techniques ranging from the computationally light (a) Vector Auto-Regressive Model, suitable for normal, linear or almost linear regime, via (b) Long-Short-Term-Memory and (c) Neural ODE Deep Learning models, appropriate to monitor mildly nonlinear regimes, and finally to the (d) physics-informed model. For example, the physics-informed model is capable of fast identification of nonlinear transients and providing interpretable results, suitable, in particular, for corrective actions. The conclusions are reached in the result of validating the models on synthetic data generated in a realistic setting from an open-source, state-of-the-art modeling software. Advanced analysis is followed by a summary and conclusion suitable for the next step - validation of the hierarchy of the suggested data-driven schemes in the industry setting.
- Machine learning
- Neural networks
- Physics informed machine learning
- Power generator model
ASJC Scopus subject areas
- Energy Engineering and Power Technology
- Electrical and Electronic Engineering