The Smart Grid group is conducting research in the development of next-generation smart grid technology through cutting-edge techniques in data-driven modeling, highly accurate and rapid predictive analytics and optimizations. This work titled DEEP SOLAR is a part of the ENERGISE program with funding for research from Office of Energy Efficiency and Renewable Energy SunShot Initiative of Department of Energy (DoE).
Previously, the group successfully conducted research into informatics-driven scalable software architectures to address real-time power management in the domain of Smart Power Grids. This work was funded by the US Department of Energy as part of the five-year Los Angeles Smart Grid Demonstration project, to forecast and curtail power consumption by thousands of electricity consumers on-demand through large-scale information processing and consumer pattern detection. This cyber-physical domain provides unique challenges to many existing computer science algorithms, approaches, and frameworks due to the data complexity, dynamism, scale and need for the real-time response.
Areas of Interest: Data-driven modeling, predictive analytics, optimizations, semantic information integration, graph analytics, stream processing, cyber-physical security, and cloud computingResearch
DEEP SOLAR
The DEEP SOLAR project aims to develop the Internet of Energy grid infrastructure of the future through accurate data-driven models of the Smart grid and its components with a focus on integrating renewable energy sources. We use the latest applied machine learning and data analytics techniques to develop accurate predictive models of various distributed energy resources (DERS) such as solar energy generators (distributed Solar PVs) and distributed storage components (EVs, batteries) and their interactions using Smart grid IoT data from our utility partners. The expected outcome of the project is a scalable Dynamic Scenario Analysis Software Toolkit for operational planning under deep solar penetration. More information can be found at the DEEP SOLAR website.
Internet-of-Energy (IoE) Enabled Distributed Energy System
In order to achieve high penetration of renewables, we focus on developing a coordinated controlled ecosystem based on AMI and intelligent devices that we label a “Live Energy Map” (LEM) of the status of all energy things in the network of prosumers. Our LEM abstraction associates a rich set of attributes that can be used to model both static as well as dynamic variables of the grid. The LEM will update in real time with changing dynamic variables.
Data Analytics
In a Smart Microgrid, both the utility and the consumer can benefit from analytics based on data for electricity consumption, generation and related variables. There are several benefits and costs associated with performing analytics at different spatial and temporal granularities.
In our research, we perform analytics for the USC campus microgrid to develop reliable electricity consumption forecasting models that would work for different spatial (building-level and campus-level) and temporal granularities (15-min and daily intervals). Our goal is to provide the facility managers more insight into the consumption patterns on the campus and enable them to plan consumption and curtailment activities for Demand Response optimization in the campus microgrid.
The USC campus microgrid is a test-bed for the DOE-sponsored Los Angeles Smart Grid Demonstration Project (SGDP). The USC campus has over 170 buildings and 45,000 students and staff population with diverse demographics. The Facilities Management Services (FMS) on campus collects data on power usage at 15-minute intervals from 170 smart power meters on campus. The goal of the SGDP is to demonstrate DR optimization in the campus microgrid.
Optimizations for Real-time Grid Operations
Mitigating supply-demand mismatch under dynamically changing load and supply curves is critical to ensure smooth grid operations. Net load balancing decisions which manipulate the synthetic reserves i.e. storage, demand curtailment, solar curtailment etc. need to be made in real time under tight grid operational constraints.
In our research, we focus on developing optimization framework that enables grid net-load smoothing through scheduling, regulation, and control of synthetic reserves obtained from incentivized prosumers. The optimization framework dynamically integrates information about the current state of the distribution network and prediction from our data-driven models and computes a range of solutions to ensure supply-demand balance. We also focus on developing approximation algorithms which are computationally tractable and provide bounded error guarantees.