Uncertainty Estimation of PV and Load Using Deep Learning for Networked Microgrid

Microgrid systems, particularly networked micro-grids, play a pivotal role in modern energy infrastructure by offering localized generation and distribution of electricity. However, integrating renewable energy sources, scheduling generation and performing energy management in these systems present challenges due to dynamic energy generation and demand patterns. To quantify this uncertain nature of photovoltaic and load patterns, this study focuses on leveraging deep learning models to obtain a preparation range for welfare maximization in networked microgrid energy management. First, the suggested framework utilizes light deep-learning techniques, enabling fog deployment, to predict PV production and load demand. Second, the preparation range is calculated using prediction error mean and standard deviation to facilitate better scheduling for the networked microgrid. Subsequently, model prediction uncertainty is quantified using prediction interval convergence probability (PICP), and prediction interval normalized average width (PINAW), helping to analyze model prediction capability. The simulation results demonstrate the Gated Recurrent Unit (GRU) model's superior performance in accurately predicting load demand and PV generation, realizing an R² score of 0.94 and 0.99. In addition, the GRU model achieves a perfect PICP score, proving its reliability for uncertainty predictions.