An Integrated Approach to the Prediction of Weather, Renewable Energy Generation and Energy Demand in Vermont

Transcription

1 1 An Integrated Approach to the Prediction of Weather, Renewable Energy Generation and Energy Demand in Vermont James P. Cipriani IBM Thomas J. Watson Research Center Yorktown Heights, NY Other contributors to the work: L. Treinish, A. Praino, I. Khabibrakhmanov, C. Tian, M. Sinn IBM A. Stamp, M. Coombs, R. D'Arienzo, D. Haas Vermont Electric Power Company Rutland, VT

2 2 Background and Motivation Renewable energy production and energy demand have significant sensitivity to local, short term weather conditions In Vermont, there are additional challenges due to local variations in geography, meteorology, and energy use Intermittency in renewable generation coupled with variation in demand can lead to congestion in the transmission system As a result, renewable power production is curtailed

3 3 Vermont Electric Power Company (VELCO) VELCO operates an interconnected electric transmission grid 738 miles of transmission lines acres of rights-of-way 55 substations, switching stations and terminal facilities Equipment that enables interconnected operations with Hydro- Québec 1400 miles of fiber optic communication network 52-mile 450 kv direct current line owned by VETCO

4 4 Background and Motivation What is the potential to predict periods of congestion on the transmission grid and improve its stability? Ability to accurately predict renewable power (i.e., wind and solar) and electricity demand with sufficient spatial and temporal precision, and lead time Can highly localized, NWP-based forecasts be adapted to address these problems and reduce the uncertainty in decision making? Can the link between weather and impact be quantified to improve the operation of the grid and utilization of renewable power?

5 5 Approach: Coupled Weather and Impact Modeling Weather Data Renewable Sensor Data Renewable Generation Deep Thunder Weather Model Power Data Wind Solar Electric Demand Renewable Integration Meter Data

6 6 Domain and Datasets Data assimilation (3dVAR) of nearreal-time Earth Networks WeatherBug and Meteorological Assimilation Data Ingest System (MADIS) observations Surface stations, radiosondes, aircraft, ship, profiles, satellite, ~4000 stations (gray markers on map): 9 km nest (~4000), 3 km nest (~1750), 1 km nest (~550) varies for each forecast Cold started every 12 hours Other datasets for initialization SRTM 90 m terrain (NASA) Land use (National Land Cover Database) 2 km SSTs (NASA) MODIS-based green vegetation fraction (NASA) Land surface fields (NASA) Vermont 1071x1071 km, every 9 km 564x564 km, every 3 km 300x297 km, every 1 km Example Forecasting Domain for Vermont (Gray Dots Mark Locations of Weather Sensors Used for the Forecasts)

7 7 Current NWP Configuration WRF-ARW /3/1 km nested for 48 hours (51 vertical levels (~10 below 180 m) Previous version was at 2 km horizontal resolution with 45 vertical levels Run twice daily (initialized at 00 and 12 UTC) RAP for background fields NAM for lateral boundary conditions Physics configuration for highly urbanized to rural domain o MYNN-2.5 PBL and surface layer physics o RRTMG long- and short-wave radiation o TKE-based gust estimation o Thompson double-moment microphysics o Explicit cumulus physics for innermost nests, Grell Freitas for outer nest Example Forecasting Domain for Vermont (Gray Dots Mark Locations of Weather Sensors Used for the Forecasts)

8 8 Deep Thunder Web Portal 48-Hour Forecasts for Vermont

9 9 Deep Thunder Web Portal 48-Hour Forecasts for Vermont

10 10 Deep Thunder Web Portal Site-specific 48-hour Forecasts for Vermont

11 11 Verification Point-wise continuous statistics (e.g., MAE, ME, RMSE, etc.), Categorical statistics based upon thresholds Wind speed Temperature Precipitation and 3- and 6-hour accumulation Utilize WeatherBug and MADIS data, initially within the finest resolution nest, decomposed according to the time interval of the event For precipitation: comparison against NCEP Stage-IV Implemented using the Model Evaluation Tools (MET) package Forecast cycles that cover significant events

12 12 8 July 2014 Examples of Forecast Skill by Event Temperature ( o F) bias (Deep Thunder: -0.6, NAM: -1.2) Wind speed (mph) bias (Deep Thunder: 2.9, NAM: 8.7) Precipitation CSI (Deep Thunder: 0.51, NAM: 0.43) 23 July 2014 Temperature ( o F) bias (Deep Thunder: 0.4, NAM: 0.5) Wind speed (mph) bias (Deep Thunder: 3.0, NAM: 7.8) Precipitation CSI (Deep Thunder: 0.41, NAM: 0.26) 2 September 2014 Temperature ( o F) bias (Deep Thunder: 0.6, NAM: 2.1) Wind speed (mph) bias (Deep Thunder: 1.6, NAM: 4.1) Precipitation CSI (Deep Thunder: 0.41, NAM: 0.26)

13 13 Wind Power Forecasting Predictive Statistical Models Built from Historical Weather Forecasts and Observations, Power and Other Data Data for Wind Farms NWP-derived, [u, v, w] across the blade extent, temperature, pressure and moisture Wind and temperature measurements from turbines and met tower Turbine nacelle direction Generated power at each turbine and accumulated power, including availability and operational mode Engineering characteristics of the turbines (e.g., specifications, power curves, etc.) Wind Power Models Wind Power Forecasts Per turbine Per farm Day-ahead focus

14 14 Renewable Power Web Portal

15 15 Solar Power Forecasting Predictive Statistical Models Built from Historical Weather Forecasts and Observations, Power and Other Data Data for Solar Farms NWP-derived near-surface GHI, DNI, GNI, wind speed & direction. temperature, pressure and moisture Irradiance, wind and temperature measurements at the farm, if available Generated power at each farm and accumulated power, including availability and farm operational mode Engineering characteristics of the PV panels (e.g., specifications, power curves, etc.) Solar Power Models Solar Power Forecasts Per farm Day-ahead focus

16 16 Renewable Power Web Portal

17 17 Current Status and Summary Enabled an operational capability for the weather forecasting component Demonstrated the ability to couple NWP to both renewable and demand forecast models Continuing to improve calibration of all four models, and characterization of uncertainties Operational statistics for evaluation Retrospective analysis and tuning using new events that have impact Developing additional specialized visualizations and methods of dissemination Development of an ensemble data assimilation system

18 18 Ongoing Research for Regional Application (UMD) Research progression: Breeding method (Toth and Kalnay, 1993 & 1997) Fast-growing errors inadvertently bred in the analysis cycle Includes non-linearity (full physics model) Implement LETKF for nested, regional domain Development of a computationally feasible system Additional datasets to drive the analysis Instability: (I) Method for choosing the initial perturbations - Random (WRFDA); existing forecast database (II) Perturbation amplitude (III) Breeding interval - More frequent, given short-term forecasts and regional domain Ensemble: Local bred vectors and E-dimension (Patil et al, 2001)

19 19 Thank You! Questions?