1 Predic've Analy'cs for Energy Systems State Es'ma'on Presenter: Team: Yingchen (YC) Zhang Na/onal Renewable Energy Laboratory (NREL) Andrey Bernstein, Rui Yang, Yu Xie, Anthony Florita, Changfu Li, ScoD Carmichael NREL is a na/onal laboratory of the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, operated by the Alliance for Sustainable Energy, LLC.
2 Mo'va'on Increased Amount of Data in Power Systems
3 Data Nonpervasive Heterogeneous Highly variable Different resolu/on Mo'va'on Distribu'on DER Topology data Transmission Opera/ons Op/miza/on Control Planning
4 Data Nonpervasive Heterogeneous Highly variable Different resolu/on How to use these data? Mo'va'on Distribu'on DER Topology data Transmission How? System Opera'ons How?
5 Mo'va'on Predic/ve System Opera/ons Power Systems Situa/onal Awareness Training 10 Forecasting 3 1 6 2 5 Training data Target data Forecast data Errors 0.93 7 11 4 8 0.9 12 0.89 0 hour 1 PM 9 (a) Voltage (p.u.) 0.94 0.92 0.91 1 hour 2 PM B C 0.04 0.02 0 3 hour 4 PM 4 hour 5 PM 5 hour 6 PM A B C (b) 6 hour 7 PM 0.06 A 1 2 3 4 5 6 7 8 9 10 11 12 Bus Number (c) Voltage Violation (p.u.) Voltage Violation (p.u.) 0.06 2 hour 3 PM 0.04 0.02 0 1 2 3 4 5 6 7 8 9 10 11 12 Bus Number (d) 7 hour 8 PM
6 Resource Forecas'ng Load Forecas'ng Future? Nodal Voltage Load State-of-the-Art DER Distribu'on Future Transmission Future DER Future DER Future
7 Objec'ves Integrate the look-ahead state es/ma/on method with short-term resource and load forecas/ng Develop a robust grid es/ma/on and forecas/ng pla\orm Develop a novel comprehensive deep learning method for mul/modal knowledge discovery Reliably forecast grid condi/ons in 5-minute resolu/on with 30-minute look-ahead window Predic've Analy'cs for Grid Es'ma'on (PAGE)
PAGE PlaMorm 8
Overview of Sky Imager Forecast source: Google Earth Sky Imager Forecast Procedures: 1. Identify clouds; 2. Position clouds; 3. Track cloud movement; 4. Predict GHI; source: Bryan Urquhart
Sky Imager (SI) forecast Sky images Cloud category and position forecast Determine Cloud Transmissivity Solar irradiance SI with Radiative Transfer Model (RTM) Input Processing Forecast Meteorology measurements GHI measurements RTM Solar irradiance Sky images Cloud fraction and position forecast DERs
Fast All-sky Radia'on Model for Solar applica'ons (FARMS) AOD, θ, g, ω, PWV, P, ozone, REST2 Cloud transmioance and reflectance of irradiance Clear-sky transmioance and reflectance All-sky broadband Surface albedo irradiances Xie et al., Solar Energy (2016)
rmae_si = 19.2% rmae_si_rtm = 8.7% rrmse_si = 28.9% rrmse_si_rtm = 12.2% min GHI Forecast for a Cloudy Day
Error Metrics
14 Model-Based Load Forecas'ng Load Forecast User Preference Weather Forecast weights weather System Iden/fica/on models Model-Predic've Control Target Func'on paderns Sta/s/cal Learning data data data data data data data data BaDery Storage PV Inverter HVAC Water Heater Mechanical Loads Data Center
15 Data-Driven Load Forecas'ng Machine Learning-Based Load Forecas/ng Short-term High-resolu/on Using support vector regression Hybrid parameter op/miza/on Load forecas/ng demonstra/on
16 Grid Forecas'ng Input Individual power injec/ons and withdraws Individual forecasts given as mul/dimensional polytope Model Linear approxima/on between state variables (voltage angle and magnitude) and withdrew/injected powers to compute a polytope that a forecast for grid-state Mul/-dimensional deep learning for system model and parameters
17 Grid Forecas'ng Clustering Method Clustering buses according to the electric distance Linear approxima/on of voltage magnitudes Similarity metric Distance
18 Mul/-Kernel Learning Vector-valued func/on Grid Forecas'ng Regularized leas-squares problem Solu/on
19 Conclusion Integrated Resource and Load Forecaster (IRLF) provide es/mates on DER opera/on and customer loads for both current states and forecasts Grid Es'mator and Forecaster (GEF) With the informa/on produced by the IRLF and using the grid measurement data, the GEF will employ machine learning techniques to determine the interrela/onship of state variables and will (1) es/mate the current system states and (2) forecast the near-future system states
20 Thank you! Q & A NREL is a na/onal laboratory of the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, operated by the Alliance for Sustainable Energy, LLC.