Leveraging multi-granularity energy data for accurate energy demand forecast in smart grids

Abstract

Accurate energy demand prediction is very important for smart grids to conduct demand response and stabilize the grids. In previous work, many prediction algorithms are proposed to improve the energy consumption prediction accuracy based on the aggregated energy consumption in the whole grid. Recently, with the increasing installations of smart meters in individual homes, high granularity (e.g., per minute) energy consumption data in individual homes becomes available and provides us a great opportunity for better energy consumption prediction. In this paper, we propose M-Pred to utilize the high granularity energy consumption data collected by smart meters in individual homes for better energy consumption prediction in smart grids. In M-Pred, we propose a learning algorithm to learn energy consumption patterns of individual homes from the high granularity energy consumption data. The consumption patterns we learn from homes are then applied for energy consumption prediction in smart grids. Furthermore, since not every home in a smart grid is equipped with a smart meter, we propose a matching and prediction algorithm to leverage the multi-granularity energy data for accurate consumption prediction. We conducted extensive system evaluations with 726 homes' minute-level power consumption data for more than 12 months. The simulation results show that our design can provide accurate energy consumption prediction for the next hour with negligible errors (e.g., Mean Absolute Percentage Error is 2.12%).