Distribution system operations (DSO) are designed to maintain reliability in the presence of predictable variability. Future distribution systems will operate in a dramatically different environment with deep penetration of distributed energy resources (e.g. solar, EVs, storage, smart loads) and widespread adoption of novel devices for network management resulting in increased variability. Unless DSO can be adapted to these conditions, performance and revenues of future utilities will be severely impaired. How to adapt Future systems will generate a wealth of data from consumers, line sensors and network equipment. Utilizing this data for learning, prediction and resource coordination is challenging and not fully understood. This proposal seeks to connect "bits to watts" utilizing modern data analytics in order to enable scalable and cost effective DSO for future distribution networks. In particular the proposal explores new approaches in machine learning, optimization and behavior learning and their applications in power systems. The methods will be implemented in a software platform: Visualization and Insight for Demand Operations and Management (VISDOM). The research component of the proposal will enable emissions reductions and massive scaling of the management of behind the meter resources. It contributes to the budding smart grid data analytics industry expected to reach a $6 billion market size by 2020. The education component of the proposal will create a novel curriculum and online education in data thinking to prepare the data analytics workforce of the future.
The project will make use of large spatial and temporal data sets from industry and utilities to explore new approaches in machine learning, stochastic control & optimization and behavioral economics to address problems in power systems. The central problems that will be addressed are: (i) Build an adaptive consumer behavior learning framework that scales to large numbers of consumers; (ii) Investigate probabilistic demand forecasting and pricing methods at multiple scales ranging from individual residential consumers to communities; (iii) Develop a novel network reconstruction and monitoring framework to learn the power distribution network from data; (iv) Create data and simulation driven placement and coordination mechanisms for residential demand-side resources; and (v) Utilize an interactive platform that engages consumers in real-time to develop novel randomized trial approaches and apply it to innovative behavioral programs. Impacts such as increasing the value of consumer demand flexibility by more than 50% are expected. The resulting methods will be made available in open-source in the VISDOM platform. VISDOM can support a thriving community of academics and industry partners that experiment with demand side management. Currently, every project develops non-transparent and limited analysis mechanisms that consume time and resources. More broadly, the time-series data based approaches developed in this proposal are applicable to other fields such as marketing, healthcare and e-commerce. The education component will advance concepts from data thinking into power systems. The proposed curriculum includes a new hands-on course in data analytics for energy systems for undergraduate and masters students; online adult education courses directed at utility professionals and a broader audience and a K12 experimental practicum prepared with high school teachers visiting the PI's lab in a summer program. In addition, a smart grid seminar involving distinguished speakers from academia and industry will be supported and made available online.
