Pietro’s research focuses on enhancing the operational efficiency of complex energy storage assets by integrating exact optimization techniques with machine learning methods.

As the energy transition accelerates, power systems increasingly rely on weather-dependent renewable energy sources like wind turbines and solar panels. The inherent intermittency of these sources introduces uncertainty in power system operations. Therefore, flexible assets are needed to maintain balance between electricity generation and consumption. Energy storage plays a crucial role in providing this flexibility. However, storage systems are challenging to operate due to their complex, nonlinear dynamics. For example, pumped hydro energy storage relies on potential energy in water to store electricity, combining both electrical and hydraulic constraints. Buildings, on the other hand, can serve as distributed thermal storage but require precise modeling of heating, ventilation, and air conditioning (HVAC) systems, which can be highly complex.

Machine learning offers powerful tools to model such nonlinear systems using data-driven approaches. However, traditional machine learning models, like neural networks, often lack the ability to incorporate physics-based knowledge and handle strict operational constraints. Exact optimization methods, in contrast, bring these benefits, offering stronger guarantees on solution quality, but cannot learn from data and often necessitate to simplify the dynamics equations. To bridge this gap, Pietro’s research explores new methods to integrate optimization with machine learning, such as neural-network-constrained optimization and decision-focused learning, to advance the operational decisions for energy storage systems.