Bi-Level Optimization Framework for Networked Microgrids With Forecasting and Uncertainty Analysis in an Edge–Fog–Cloud Architecture

The growing deployment of renewable energy sources and the need for robust low-carbon power systems have accelerated the use of decentralized energy management approaches. Networked Microgrids (NMGs) are an interesting solution to enhance distributed energy coordination and reliability. Nevertheless, the efficient management of energy under uncertainty, with the least possible operating cost and emissions, continues to be a significant challenge. This paper presents a Bi-Level Optimization Framework for NMGs in an Edge–Fog–Cloud infrastructure for supporting real-time, uncertainty-aware energy management. Deep learning models are trained at the cloud level for photovoltaic (PV) and load forecasting and deployed at edge. Forecast uncertainty is measured in terms of Prediction Interval Coverage Probability (PICP) and Prediction Interval Normalized Average Width (PINAW) over a number of different confidence levels. Experimental outcomes prove that Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM) models exhibit better forecasting performance, with GRU achieving lower PINAW and higher PICP under most conditions. The edge layer utilizes a local Mixed-Integer Nonlinear Programming (MINLP) optimizer for energy dispatch within each microgrid, while the fog layer manages inter-microgrid energy exchange and utility interaction via a global MINLP optimizer. Simulation results show that the robust optimization strategy—based on worst-case uncertainty—yields near-optimal results compared to the base case using actual data. On the validation dataset, the base case achieved a maximum profit of 43.75%, and the robust optimization achieved up to 37.40%. On the test dataset, they achieved 19.96% and 13.25%, respectively, at specific time stamps. A scalability study has also been conducted to demonstrate the scalable nature of the proposed framework, ensuring timely emission-conscious and uncertainty-resilient energy management, thereby supporting the requirements of future smart grid environments.