AuGrid: Edge-Enabled Distributed Load Management for Smart Grid Service Providers

In this paper, we propose and design AuGrid, an LSTM-based model for geographically aware smart grid service providers, which predicts the hourly load requests from users. We also develop a pricing model which depends on the predictions obtained from AuGrid for deciding per unit cost of energy in contrast to the existing schemes that focused solely on the load requests. The crux of this work is that the suppliers may plan better with forecasts than being in uncertainty. Since smart grids are well connected, logically neighboring smart grids may exchange information and energy on the requirement. We train AuGrid with a lookback set to 2 using real-world datasets and demonstrate its robustness by predicting the load requests for different suppliers. We propose deploying the AuGrid system on geographically aware suppliers for facilitating intelligence on the edge while reducing the user sample space and increasing data security. On extensive implementation and deployment, we observe that AuGrid offers minuscule loss (below 0.1) and the pricing model offers a reduction in per-unit cost by almost 75% in comparison to existing solutions. Additionally, AuGrid requires 30% CPU and 40% RAM of single processor boards on deployment, which illustrates its suitability for resource-constrained devices.