ALBANY BARCELONA BANGALORE ICEM 2015 June 26, 2015 Boulder, CO A SOLAR AND WIND INTEGRATED FORECAST TOOL (SWIFT) DESIGNED FOR THE MANAGEMENT OF RENEWABLE ENERGY VARIABILITY ON HAWAIIAN GRID SYSTEMS JOHN W ZACK & KEN PENNOCK AWS TRUEPOWER, LLC ALBANY, NY jzack@awstruepower.com DORA NAKAFUJI HAWAIIAN ELECTRIC COMPANY HONOLULU, HI dora.nakafuji@hawaiianelectric.com 463 NEW KARNER ROAD ALBANY, NY 12205 awstruepower.com info@awstruepower.com
Overview Background and Motivation Four Components of a Customized Solar and Wind Forecasting Approach for Hawaii - Sense - Model - Communicate - Validate Current Status and Next Steps
Hawaiian Electric Co (HECO): Wind and Solar Penetration Oahu Molokai L: 5.0 MW S: 2.1 MW Maui L: 180 MW W: 73 MW S: 36 MW Hawaii L: 180 MW W: 30 MW S: 50 MW L: 1200 MW W: 100 MW S: 357 MW Lanai L: 5.0 MW S: 2.3 MW L: Typical daily peak load W: Wind gen capacity S: Solar gen capacity (mostly distributed)
Forecasting Goal: Minimize Costs & Reduce Risks of Integrating Renewable Energy on the Grid Problem: Minimize the cost of integrating the non-dispatchable variability of wind and solar generation into the electric grid while maintaining a very high level of reliability Potential Solutions Flexible/ lower cost backup generation Storage Reduce variability through diversity Demand response programs Forecasting the resource & production
HECO Approach to Short-Term Forecasting Atypical Utility Environment - High level of wind and solar penetration - Very high cost non-renewable generation - Vertically integrated utility; no market Most ops decisions are short-term (0-4 hrs) Minimal day-ahead decision-making - No interconnections Unique Meteorological Issues - Extremely data sparse in all directions - Variability typically driven by small scale features (island-induced and propagation from oceanic areas) - No NWS rapid update NWP products 4-Component Approach Sense Model Communicate Validate Global & Regional Met Data Wind Power Forecast Gen Facility Data Ensemble of Forecast Methods Optimized Ensemble Algorithm Power Production Models Targeted Sensor Data Solar Power Forecast
1 Sense: Targeted Sensor Network WindNET 8 SoDARs Real-time data on MADIS Side-scanning LiDAR Radiometer SolarNET 9 Pyranometers 3 Solar Monitoring Stations (SMS) LM-1/LM-2 5-watt PV panels Radiometer LiDAR SoDAR SMS LM-1 Panel
1 Sense: Estimated Solar Irradiance from Visible Satellite Image Brightness Uses image data from GOES-West Customized version of an algorithm developed by Perez et al (2002) Provides an estimate of the Global Horizontal Irradiance (GHI) at all forecast locations on the islands at 1 km / 15-minute resolution Calibrated for Hawaii using Hawaiian Electric s network of pyranometers and solar monitoring stations Visible Brightness Satellite Estimated Irradiance (W/m 2 ) PEREZ, R., P. INEICHEN, K. MOORE, M. KMIECEK, C. CHAIN, R. GEORGE and F. VIGNOLA, 2002: A new operational model for satellite-derived irradiances: Description and validation. Solar Energy, 73, 307-317.
2 Model: Multi-method Forecast Ensemble
2 Model: Multi-method Forecast Ensemble Standard Cycle NWP: Native resolution and downscaled with in-house NWP
2 Model: Multi-method Forecast Ensemble Frequent Update NWP: 3 in-house models run on a 2 hour update cycle with 3 km resolution
2 Model: Multi-method Forecast Ensemble Cloud Motion Vector Model: Pyramidal Image Matcher (PIM), a multi-scale feature advection model, with Analog Ensemble postprocessing
2 Model: Multi-method Forecast Ensemble Model Output Statistics (MOS): 3 MOS methods to reduce systematic NWP errors
2 Model: Multi-method Forecast Ensemble Optimized Ensemble Algorithm: Statistically weights AVAILABLE ensemble members to construct a composite forecast based on historical performance
2 Model: Multi-method Forecast Ensemble Power Output Models: Converts forecasts of met variables to electric power output
2 Distributed Solar Power Output Model General Approach Output model is the relationship between solar irradiance (GHI) at substation to PV generation on all circuits into that substation GHI is measured or estimated (from satellite images) at each substation PV generation is estimated from sampled rooftop production data Model Formulation Attributes (panel model, orientation etc.) of distributed PV are very diverse so bulk behavior is modeled PV capacity (MW) by circuit is available Default: theoretical power model Data-based: Substation-specific models being constructed from scaled sampled production data from individual systems and estimated/measured substation GHI Sampled PV production data Substation GHI : measured or estimated Data-based Power Output Model Ideal performance Substation GHI (W/m 2) Substation PV production model
3 Communicate: Forecast Specifications Forecast Target Entities Wind Wind Generation Facilities (WGFs) System (island)-wide aggregate Variables: wind speed, power production & 30-min ramp rate Solar Solar Generation Facilities (SGFs) Substation aggregates of distributed PV Combined substation-sgf clusters System (island)-wide aggregate Variables: GHI, power production & 30- min ramp rate Time Scales 0-6 hours 15-minute update cycle 15-minute forecast time resolution 0-48 hours 6-hour update cycle 1-hour forecast time resolution Substation Clusters Substations
3 Communicate: Forecast Product Format 0-6 Hours-ahead Example Upward 30-min ramp rate forecast Measured or Estimated Data for the Entity Color shading: forecasted probability density Solid line: 50% POE value Dashed lines: 20%, 40%, 60%, 80% POE values Downward 30-min ramp rate forecast
4 Validate: Evaluation Approaches Objectives Assess operational value Develop and guide user confidence Monitor progress of efforts to improve forecast accuracy Provide guidance for further forecast system development Approach 4 components 1. Event-based evaluation: did it help make more effective operational decisions? 2. Traditional deterministic metrics by look-ahead time applied to 50% POE forecast: MAE, RMSE etc. 3. Probabilistic metrics applied to power output and ramp rate probabilities: Brier Score, RPSS, etc. 4. Assessment of the component models & methods
4 Evaluation Feedback: Improvements to Satellite-based Cloud Motion Vector (CMV) Forecasts Problem CMV skill (MAE) vs persistence is high over ocean and low over land Trade Regime Unskilled Skilled 0.6 0.3 Solution Apply Analog Ensemble (AE) method as a post-processing method to the CMV forecasts to predict Clear Sky Index (CSI) Matching parameters: (1) CSI, (2) cloud motion vector speed and (3) direction Forecast is AE ensemble mean 60-min Forecast Skill Non-Trade Regime 0.1 0.0-0.2-0.4-0.8 60-min Forecast Skill
Current Status and Next Steps: Further Increase the Value of Forecasting Current Status Real-time development version running Transition to ops: end of 2015 Future More effective forecast utilization Further customize content/format Integration into EMS Increase user confidence in forecasts Facilitate use of probabilistic info Further improve forecast accuracy Improve distributed power output models Improve physics-based models WRF-Solar Gather additional (targeted) data More effective data assimilation Refine configuration of statistical tools