Towards Statistical Modeling and Machine Learning Based Energy Usage Forecasting in Smart Grid

Developing effective energy resource management strategies in the smart grid is challenging due to the entities on both the demand and supply sides experiencing numerous fluctuations. In this paper, we address the issue of quantifying uncertainties on the energy demand side. Specifically, we first develop approaches using statistical modeling analysis to derive a statistical distribution of energy usage. We then utilize several machine learning based approaches such as the Support Vector Machines (SVM) and neural networks to carry out accurate forecasting on energy usage. We perform extensive experiments of our proposed approaches using a real-world imeter reading data set. Our experimental data shows that the statistical distribution of meter reading data call be largely approximated with a Gaussian dist ribution and the two SVNI-based machine learning approaches to achieve a high accuracy of forecasting energy usage. Extensions to other smart grid applications (e.g., forecasting energy generation, determining optimal demand response, and anomaly detection of malicious energy usage) are discussed as well.(1)