Higher Granular Sectoral Demand Forecast Under Data Scarcity: An Integrated Physics-Based Top-Down and Bottom-Up Approach

The proliferation of renewable distributed energy resources and transportation electrification has created new challenges for the utilities in integrating these new technologies. Consequently, the utilities and electric distribution companies need to adapt their current business models to maintain service reliability, while higher granular data scarcity hinders accurate decision-making processes. In this context, this article proposes an integrated top-down bottom-up approach that aims at estimating the essential higher granular demand data through a two-layer physics-based framework. In the top-down model, various economic-demographic regression models are developed to unlock higher granular demand in geographic areas and technology/policy adoptions, while in the bottom-up layer, our in-house-developed physics-based building simulator is employed to simulate sectoral disaggregated demand patterns. The simulation error is modeled by the least-squares estimation of a Fourier series, extreme gradient boosting (XGBoost) algorithm, and long short-term memory (LSTM) network, where the numerical results show the superiority of the LSTM in capturing the simulation errors. To reveal the merits of the proposed hierarchical framework, the state of New Jersey in the US is chosen as the test case directed by the Energy Master Plan to pursue a set of targets by 2021, which is the target year in this article, while data scarcity is in place.