Data-driven and knowledge-based digital twin modeling for distributed PV systems

With the growing integration of photovoltaic (PV) systems in modern distribution networks, achieving low-carbon and intelligent grid operation has become increasingly important. However, conventional modelling can't continuously track the performance of PV throughout the normal operation and Low Voltage Ride Through (LVRT) operation. This paper proposes a data-driven and knowledge-based digital twin modeling method for PV systems, which is developed by only requiring grid-side measurements without reliance on internal device details. The proposed method can distinguish between normal operation and LVRT states and continuously simulate the performance of PV throughout the whole duration. In normal operation, PV system is modeled using a bidirectional long short-term memory network, while LVRT behavior is modeled using experimental data by identifying control strategies to capture the current response. Validation through simulations and experimental results confirms that the proposed method can mirror the performance of physical PV and provide basis for PV digital twin application.