Energy-efficient smart grid operations through dynamic digital twin models and deep learning

Abstract

Adopting the dynamic digital twin (DDT) model in smart grid distribution networks is a revolutionary breakthrough toward advanced dynamic energy management and control. However, even the most advanced systems fail to describe static architectural configuration adequately or they do not offer sufficient automation in this process, they are unable to handle dynamic interactions or topological hierarchy. To overcome such restrictions, this research presents a new framework for building DDT models based on Graph Neural Networks (GNNs). GNNs outperform other deep learning models when it comes to modeling graph-structured data which has application in modeling nodes and edges of smart grids. The adopted model expands the critical technical parameters' achievements and indicates a high Voltage Regulation Efficiency of 92 % and Network Efficiency belonging to 95 %; therefore, the distribution of power and operation reliability is considered optimal. The advantage of these findings is also echoed by the Voltage Profile Deviation of 0.015 p.u. and the Power Loss Reduction of 18.3 % which suggest that the proposed method offers better voltage profile stability and less energy losses than existing static models. The usefulness and applicability of the framework can be shown by performing experiments in MATLAB Simulink and Python-based libraries such as PyTorch Geometric. This study provides a novel approach to address issues in applied research and provides the basis for further advancements in realistic digital twin applications concerning smart grids.