The objective of this research is to establish a foundational framework for smart grids that enables significant penetration of renewable DERs and facilitates flexible deployments of plug-and-play applications, similar to the way users connect to the Internet. The approach is to view the overall grid management as an adaptive optimizer to iteratively solve a system-wide optimization problem, where networked sensing, control and verification carry out distributed computation tasks to achieve reliability at all levels, particularly component-level, system-level, and application level.
Intellectual merit. Under the common theme of reliability guarantees, distributed monitoring and inference algorithms will be developed to perform fault diagnosis and operate resiliently against all hazards. To attain high reliability, a trustworthy middleware will be used to shield the grid system design from the complexities of the underlying software world while providing services to grid applications through message passing and transactions. Further, selective load/generation control using Automatic Generation Control, based on multi-scale state estimation for energy supply and demand, will be carried out to guarantee that the load and generation in the system remain balanced.
Broader impact. The envisioned architecture of the smart grid is an outstanding example of the CPS technology. Built on this critical application study, this collaborative effort will pursue a CPS architecture that enables embedding intelligent computation, communication and control mechanisms into physical systems with active and reconfigurable components. Close collaborations between this team and major EMS and SCADA vendors will pave the path for technology transfer via proof-of-concept demonstrations.
A primary objective of this research on smart grids is to establish a foundational framework that enables significant penetration of renewable distributed energy sources (DERs) and facilitates flexible deployments of plug-and-play application. Under the common theme of reliability guarantees, this project is carried out in three tasks. We first explored overarching principles for the interconnecting architecture of cyber-physical system where a cyber-network overlays a physical-network. Due to the interdependence between the networks, in the event of attacks, node failures in one network may result in a cascade of failures affecting both networks -- potentially leading to the collapse of the entire infrastructure. In this project, we characterized the optimum inter-link allocation strategy against attacks with unknown topology, and quantified the robustness of interdependent systems against this sort of catastrophic failures. Another main focus of this project is on stochastic modeling and forecast of wind energy. We investigated short-term forecast of wind farm generation by applying spatio-temporal analysis to extensive measurement data from a large wind farm where a number of wind turbines are installed over an extended geographical area. Since wind ramps incurs significant uncertainty in wind power generation, we devised SVM algorithms to capture the wind ramp dynamics, based on the key observation from the measurement data that wind ramps often occur with specific patterns. Then, an SVM enhanced Markov model for wind generation forecast is then developed, which takes into account not only wind ramps but also the diurnal non?stationarity and the seasonality of wind farm generation. Further, with a high penetration of wind generation, ramp rates therein can be much larger than anticipated, causing cycling of expensive spinning reserve. We have devised new control schemes to combine existing automatic generation control (AGC) with anticipatory control to enable units to be cycled prior to forecasted generation ramp events, allowing the system to prepare for this forecasted change in generation or load. In task 3, we investigated PMU-based online dynamic security assessment (DSA), including operating condition variation, forced topology change and missing PMU data, and developed a novel adaptive ensemble decision tree (DT) based framework. In ensemble DT learning, multiple small DTs, with a smaller height than a fully-grown DT, are combined to give security classification decision via weighted voting. The models and algorithms developed in this project have been validated by extensive numerical experiments. This forecast accuracy of SVM-based Markov model has been tested by using realistic wind farm data provided by NREL, and its performance gain is pronounced in the presence of wind ramps compared to conventional approaches. Two patent applications based on these results have been filed, and have already received much interest from an investment company. The effectiveness of the DT-based DSA algorithms has been validated , by using a much larger test system in the Western Electricity Coordinating Council (WECC). These proof-of-concept demonstrations will pave the path for technology transfer.
