Deep neural network based digital twin framework for per-day energy extraction in partially illuminated photovoltaic systems

Digital Twins (DTs) in digitalization have become a potent device in real-time optimization of the photovoltaic (PV) system. Partial shading in PV systems is a serious issue resulting in significant energy wastage. The methodology discussed in this paper is to maximize the per-day energy extraction (PDEE) in the case of shading through the utilization of a supplementary PV source. The methodology combines two important features: (1) Digital Twin Framework design for PV Systems Using Calibrated Voltage, Current, Temperature, and Solar insolation Data to facilitate reliable online power assessment with reduced dependence on hardware sensors, thereby improving scalability and maintainability. (2) Incorporating external biasing units in series to mitigate partial shading effects in PV arrays to maintain V m ref across all PV arrays to optimize the overall power retrieval. A PV system’s digital twin is a fusion of analytical Formulation of the Photovoltaic Model and deep neural networks (DNNs) optimized using the improved Harris Hawks Optimization (IHHO) algorithm. Moreover, a machine learning model can be applied to the Digital Twin features to predict the Global maximum reference voltage. Simulations and experimental findings indicate an increase in power extraction by 15.4%, which points to a viable approach to reduce the effect of shading in a realistic PV system.