MSL has teamed with New Mexico State University and The National Renewable Energy Laboratory (an MSL Member) for a proposal to the U.S. Department of Energy’s Established Program to Stimulate Competitive Research (EPSCoR) program, for a Building EPSCoR-State/DOE-National Laboratory Partnerships grant. The team’s proposal, “Building a federated learning framework for trustworthy and resilient energy internet of things (eIoT) infrastructure,” focuses on Federated Machine Learning (FML). It was selected by DOE at the pre-proposal stage for full proposal submission, and it approaches several FML-related aspects of smart grid data architectures and applications.
The “energy internet of things” (eIoT) is a promising technical aid for the energy-management change drivers, such as rising demand for electricity, the prominence of clean and distributed energy resources (DERs), the emergence of electrified transportation, deregulation of power markets, and innovations in smart grid technology. The convergence of cyber, physical, and economic frameworks in the energy sector depends primarily on the distributed heterogeneous eIoT devices and their collaborative management. These eIoT devices will increasingly get integrated into the power grid providing unparalleled visibility into energy generation, use, and operations. This proliferation will create a deluge of operational and information technology data, from geographically diverse and heterogeneous sources, necessitating use of data-driven machine learning (ML) applications. The ML applications will digest the data volume, preferably close to the devices themselves, and help increase grid scalability and efficiency. Given the data volumes, the need for resilience, and user/data privacy needs, federated machine learning (FML) can be used as a foundation to improve not only visibility and optimization, but also security of the eIoT infrastructure.
NMSU’s Professor Jay Misra will serve as Principal Investigator. MSL will provide subject matter expertise and organize a seminar for the NMSU graduate student researchers on smart grid and microgrid trends and technologies, and host a site visit at the research microgrid operated by (MSL Member) the University of NM. NREL will provide simulation and emulation expertise and environments, featuring its Advanced Research on Integrated Energy Systems (ARIES) platform. ARIES represents a substantial scale-up in experimentation capability from existing research platforms, allowing for research at the 20-MW level. The scale of the platform is amplified by a virtual emulation environment powered by NREL’s 8-petaflop supercomputer.
This project aims to be the first to propose a foundation blueprint for secure, verifiable, and privacy-preserving FML frameworks that can be used to build the trustworthy and resilient eIoT infrastructure of tomorrow. It targets basic innovation on five research dimensions. The first is the creation of a distributed public key infrastructure, utilizing the certificate transparency standards, to handle the diversity of the devices (from low-capability legacy devices to capable computing devices), while also enabling seamless onboarding/offboarding of devices and resilient key management. The second dimension, builds a verifiable and trusted execution framework for device operations, which enables trustworthy insight exchanges under a zero-trust setting. In the third, the first two dimensions are utilized as foundations to develop a use-case inspired FML framework for eIoT, to be studied in the context of short-term energy forecasting. The framework accommodates parameters exchange within the ambit of communication infrastructure constraints while preserving user/data privacy through differential privacy (DP). The framework will be generalizable to other use cases. The project addresses anomaly detection and coordinated attack mitigation in the fourth dimension by significantly enhancing the generic FML framework built in the third dimension. In the fifth, the team will develop a co-simulation framework and also emulation/testbed functionalities to validate algorithms, protocols, and the consequent overall eIoT infrastructure.