Data-driven modelling of inverter efficiency curves for the digital twin of a large-scale battery storage system

Stationary battery systems, and in particular large-scale battery storage systems (BESS), are increasingly being used for a variety of applications in different markets. A deep understanding of the efficiency behaviour of LBSS is crucial for the development of new operating and bidding strategies for single and combined or multi-use applications. Depending on the application, and thus on the power and state of charge (SOC) ranges in which an LBSS is operated, the efficiency of the inverters in the system will vary significantly. This study proposes two methods for the data-driven characterization of inverter efficiency curves for BESS based on operational data of the 6 MW / 7.5 MWh research focused hybrid LBSS M5BAT (Modular Multi-Megawatt Multi-Technology Medium Voltage - Battery Storage System). These efficiency models were developed for their application within the digital twin of M5BAT and are meant to be a practical approach for BESS operators. The first method uses a quadratic function to model the relationship between AC and DC power of the inverter. Using neural networks, the second method predicts the efficiency of inverters based on the AC power and battery voltage. The results show that both models are sufficient for the simulation of BESS operation in constant power phases.