Lead Performer: GE Research – Niskayuna, NY
Partner: Pacific Northwest National Laboratory – Richland, WA
DOE Total Funding: $2,973,087
Cost Share: $1,093,534
Project Term: June 1, 2020 – May 31, 2023
Funding Type: BENEFIT 2019 Funding Opportunity Announcement
Project Objective
This project builds upon GE Research’s (GER) existing cyber-fault (cyber-attacks and system faults) detection and isolation expertise coupled with the commercial building systems modeling expertise of PNNL to develop a reinforcement learning (RL)-based adaptive model predictive control (MPC) architecture. This project aims to detect and localize cyber-faults at 99% accuracy (considering false positives and false negatives, where these levels of accuracy have been obtained in other Department of Energy (DOE) programs. The proposed resilient building energy management system (BEMS) architecture (MPC plus actor critic RL estimator–ACRE) will ensure safe and near-optimal closed-loop operation under all identified cyber-fault scenarios, first identifying the current state of the building under the detected and localized cyber-fault, then proceeding to accommodate the cyber-fault with a bumpless transfer to more resilient operation mode.
Project Impact
Integrating the expertise of GER and PNNL in resilient cybersecurity and BEMS technologies, this project pushes the state of the art by using a reinforcement-learning based estimator, which not only incorporates information about cyber-faults but also adapts an MPC to achieve uninterrupted operation with desired occupant comfort and productivity to sustain the BTO goal of 30% energy savings.
Contacts
DOE Technology Manager: Erika Gupta
Lead Performer: Dr. Mustafa Dokucu, GE Research