Partnership to Improve Solar Power Forecasting

Transcription

1 Partnership to Improve Solar Power Forecasting Venue: EUPVSEC, Paris France Presenter: Dr. Manajit Sengupta Date: October 1 st 2013 NREL is a national laboratory of the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, operated by the Alliance for Sustainable Energy, LLC.

2 2

3 3 Goals Demonstrate a state-of-the-science solar power forecasting system through applying cutting edge research Test the system with appropriate metrics in several geographically-diverse, high penetration solar utilities and ISOs Disseminate the research results widely to raise the bar on solar power forecasting technology

4 4 Application Regions SMUD MW SCE 350 Comm + 325Q Dist MW Xcel 90 MW LIPA 32 MW HECO 43 MW DeSoto Plant 25 MW

5 Project Concept 5

6 6 Metrics Development Metrics are a cornerstone of the forecast system development and application Important aspects: Standards of comparison to evaluate improvements o o Targets Baselines Engage stakeholders at all stages Emphases: 1. Weather values (e.g., cloud) - consider temporal and spatial events - Irradiance (DNI, GHI, POA) 2. Power 3. Operational relevance (value-based) a. Expert elicitation b. Conjoint experiments c. Mental Modeling Types of approaches (examples): Diagnostics (Distributions) Traditional metrics with statistical confidence Spatial methods

7 Seminal Research: Solar Forecasting System 7 To forecast clouds, aerosols, scattering, one must assimilate data and blend across scales.

8 8 Nowcasting 1. Total Sky Imaging 2. Statistical Prediction regimes and data 3. Satellite Cloud Advection Most current image Future image (estimation target) Cloud Advection Real time ingest of GOES-E and GOES-W satellite data 4. WRF Nowcasting Assimilate: -satellite data -TSI Data Cloud Type, AOD, Solar Irradiance

9 9 Empirical Statistical Prediction Create model determined by empirical fitting procedures Identify and Leverage Regime Example: empirically derived linear models of the form Methods to consider: Neural Networks Support Vector Machines Random Forests Genetic Algorithms Analog Methods ds/dt = L*s + r Also nonlinear fitting models

10 Satellite Cloud Advection Real time cloud properties from GOES: GSIP/Patmos-X 10

11 11 Hour(s) ahead forecasting: Cloud motion GSIP clouds + GFS cloud level winds Satellite observations from GOES are input to PATMOS-X algorithms to produce cloud map, including cloud-top heights. GFS model winds matched to cloud top height, then used to advect cloudy pixels No cloud evolution processed simple advection of initial cloud properties to future location due to direct advection Useful on time scales from 1-3 hours due to shearing effects

12 12 Vertical Cross Section Multi-sensor combined analysis/forecast of cloud fraction (Tom Auligné)

13 13 WRF-Solar Weather Research & Forecasting Model Solar version with improved: Satellite data assimilation Cloud physics parameterization Convective parameterization Clear-sky aerosol estimation Radiative transfer modeling Collaborate with NOAA to make relevant for operations

14 14

15 15 Uncertainty Quantification Analog Ensemble Method Statistical learning method to calibrate model output and provide probabilistic information Based on observed past modelobservation pairs Algorithm search for analogs and clusters them Shown to perform at least as well as full NWP ensemble systems Binned-spread/skill plot of power predictions from the calibrated ECMWF EPS (red), calibrated COSMO LEPS (blue) and AnEn (black)

16 Test 3 methods: Radiation to Power Conversion 1. Parametric fitting models 2. Explicit power conversion models a. PV-Watts b. System Advisor Model (SAM) 3. Empirical Power Conversion System Engineering Requires experience with building decision support systems Must seamlessly blend technologies across scales Must work with commercial partners to meet industry needs Provide reliable data streams for commercial forecasting partners 16

17 Operationalization & Validation Create predictions using new system for one full year o High resolution in regions with dense observations (e.g., Hawaii, BNL, Xcel, SMUD, SCE) Apply metrics o o o Weather models Power output Value Utilize existing tools as well as specialized capabilities Economic & value analysis MET includes tools to compare grids and points, using traditional and spatial methods, and methods for evaluating statistical significance. 17

18 18 Lasting Impact Open Source software Wide dissemination Improve decision making based on solar forecasts Advance solar energy penetration due to added value of forecasts. Make solar energy more economical in day ahead trading Improved ability to integrate solar energy into grid for reliability Stakeholder buy-in and optimal application of improved forecasting methodologies Advance the penetration of solar energy through stakeholder buy-in

19 Questions 19