Machine Learning-Assisted Operation Monitoring Analytics on a Hydro Power Plant

Globally, amidst the growing energy needs, the energy transition has led to the massive incorporation of renewable energy sources like solar and wind into existing electrical grids, whose variable nature poses serious grid reliability issues. The flexible operation of energy storage solutions like hydro Pumped Storage Plants (PSPs) (in terms of the increased number of start and stop cycles and its operation under varying loading conditions) aimed at mitigating these issues has resulted in the accelerated degradation of its critical components such as the runner, generator etc owing to its prolonged operation under off-design conditions far from its Best Efficiency Point (BEP). It has become expedient to continually monitor and track its operation to ascertain its rate of degradation and inform maintenance and component replacement decisions. This paper facilitates this monitoring objectives by proposing a machine-learning based methodology for detecting/tracking the operating mode of a PSP from a combination of hybrid signals from the on-site Distributed Control System (DCS) using two models - an LSTM-based network and an SVM with overall accuracy, recall and precision of over 99 percent. In a second step, key operation statistics such as the time spent during turbine and pump operations under diverse loading conditions and a count of the start and stop cycles which are extremely useful for remaining useful life estimation and in understanding operational patterns within different time granularities for example are computed.