Collaborative Research: Crosslayer Optimization of Energy and Cost Through Unified Modeling of User Behavior and Storage in Multiple Buildings

This proposal addresses the problem of the increasing amount of energy consumed by the buildings sector and the lack of cost-effective solutions to reduce energy consumption and to cut energy costs. The goal is to develop several novel machine learning based algorithms to address the problem of energy optimization at the building and district levels. These algorithms are integrated within a two-level hierarchical simulation framework that combines in a unified model the user behavior with the collaboration between buildings equipped with PV arrays, Energy Storage Systems (ESS), and smart grid meters. Specifically, we introduce models to capture the interaction between the distribution network and buildings as well as to capture social behavioral aspects of user interactions with smart appliances. We introduce novel problem formulations of energy optimization under user well-being constraints and propose solutions that exploit user and group level social habits. In addition, we employ machine learning models based on Recurrent Neural Networks (RNNs) to predict energy loads and PV generation and to optimize for cost reduction using a Deep Reinforcement Learning (DRL) approach that satisfies ESS scheduling constraints identified at the district level.