An Ensemble Learning Approach for Energy Demand Forecasting in Microgrids Using Fog Computing

Abstract

Increased usage of smart meters enables information exchange between customers and utility providers in smart grid systems. Nowadays, the cloud-centric architecture has become a bottleneck for the decentralized and data-driven microgrids evolving from centralized Smart grids. Hence, fog computing is an appropriate paradigm to build distributed, latency-aware, and privacy-preserving energy demand applications in microgrid systems. In this work, we proposed a 3-tier architecture of a microgrid energy demand management system comprising edge, fog, and cloud layers. We set up a simulation environment where Raspberry Pi devices act as fog nodes and resource-efficient Docker applications run on these nodes. As the main contribution of the work, we developed a short-term load forecasting application based on an ensemble model that integrates support vector regression (SVR) and long-short term memory (LSTM) by leveraging the potential of distributed and low-latency fog nodes for complex models. We evaluated the forecasting model deployed in a fog-based simulation environment using the public REFIT Electrical Load dataset. We also tested the deployed fog-based simulation environment based on latency and execution time metrics.