Integration of Behind-the-Meter PV Fleet Forecasts into Utility Grid System Operations

Transcription

1 Integration of Behind-the-Meter PV Fleet Forecasts into Utility Grid System Operations Adam Kankiewicz and Elynn Wu Clean Power Research ICEM Conference June 23, 2015 Copyright 2015 Clean Power Research, L.L.C

2 Acknowledgements Research funded under DOE SUNRISE award Demonstrate improved net utility load forecasts by incorporating behind-the-meter PV forecasts for CAISO and all PV in California Collaborations with: California ISO Jim Blatchford and Rebecca Laferriere Itron Frank Monforte, Ph.D 2

3 SOLAR PREDICTION Founded in 1998 with the mission to power intelligent energy decisions Most widely used solar resource database PROGRAM OPTIMIZATION Research Consulting Software ~6.0 GW of renewable incentives processed ENERGY VALUATION >30 million solar estimations performed

4 CPR Solutions by Market Segment Utility / ISO / Energy Agency Customer Engagement WattPlan Incentive & Interconnection Processing PowerClerk Grid Planning & Operations SolarAnywhere FleetView Lead Generation & Quoting PowerBill APIs Solar Industry Incentive & Interconnection Processing PowerClerk API Performance Benchmarking for Fleet O&M SolarAnywhere SystemCheck API 4

5 Today s Presentation Background and Motivation Advantages of Explicit Behind-the-Meter Forecasting Preliminary Results Observations and Future Work 5

6 Load and Generation Used to be Separated 6

7 Behind-The-Meter PV Challenge: Load and Generation are Now Comingled 7

8 Load Forecasting Challenges Many factors influence load forecasting accuracy, such as weather, geographic diversity, data quality, forecast horizon, human behavior, time of week, etc. Increasingly influenced by behind-the-meter PV energy production! 8

9 Behind-The-Meter PV Changes Load Shapes Hour of day Source: California Independent System Operator (CAISO) 9

10 Load (MWh) Behind-The-Meter PV Changes Load Shapes BTM PV generation is influencing CAISO operations now! /7/2007 1/14/2008 2/19/2009 1/14/2010 1/27/2011 2/6/2012 1/9/2013 1/27/2014 1/22/ Hour of Day Source: California Independent System Operator (CAISO) 10

11 Output Calculations Data FleetView PV Simulation Methodology PV Specifications PowerClerk Solar Irradiance Data SolarAnywhere Satellite & NWP PV Simulation Engine PV Production Forecasts FleetView methodology also supports historical simulation needs 11

12 California Metered PV System Simulations Accurate utility-scale PV simulations require detailed site specifications (PV modules, inverters, orientation, row spacing, etc.). Same methodology as pre-construction PV energy simulations (PVsyst, SAM, etc.) Aggregated CAISO Fleet PV production 120+ Metered Utility-Scale PV Systems in the CAISO region 12

13 CA Behind-the-Meter (BTM) PV Fleet Simulation PowerClerk links administration to fleet simulation Explicit PV site simulations using same methodology as employed with utilityscale PV simulations Schneider Electric Inverter (GT PG) 352 SunPower 327 W SPR-327NE-WHT-D 10 Tilt, 181 Azimuth XX.XXX N, XX.XXX W Commissioned Jan Distribution system to regional and system-wide BTM PV simulation capabilities 13

14 Simulated PV Power (MW) Utility Load (GW) Advantages of Explicit PV Fleet Simulations Example of fleet system specs Significant variance in PV energy production by orientation Wide distribution of PV system specs can alter fleet PV energy output :00 8:00 12:00 16:00 20:00 Time of Day

15 PV Fleet Power Forecasting Aggregate PV fleet forecasts provided to CAISO. The challenge is delivering a forecast in a meaningful way to the customer!

16 ALFS Historical Training Test Evaluated BTM PV fleet historical simulation and forecasts as training input into CAISO s Automated Load Forecasting System (ALFS) BTM PV production in CAISO load sub-region was modeled from Jan 2010 through Feb

17 Historical ALFS Day-Ahead Training Results Hour of Day-Ahead Forecast 9 am 12 pm 3 pm CPR BTM Dataset? No Yes No Yes No Yes Load Forecast Error (MW) Load Forecast Error (%) 1.11% 1.10% 1.18% 1.12% 1.27% 1.21% BTM Coefficient T-test (significant if < -1.64) A BTM coefficient of -1 is the ideal result (i.e., for every predicted MW of BTM generation, one MW of load is shed) Results are close to the theoretical value of -1 BTM production datasets offer a statistically significant improvement for mid-day and late-afternoon load forecasting BTM production datasets offer a smaller impact for morning load forecasting Believed to result from higher morning load variability 17

18 CAISO Load (MW) CAISO Solar Generation (MW) Morning Load Variability Challenges Example CAISO solar generation and load profiles (7:45 AM 6/04/2015) Day-Ahead Demand Forecast Hour-Ahead Demand Forecast Actual Demand Morning CAISO load variability is much larger than PV solar generation Hour Total aggregate load shown. Variability is often higher at the zonal load level 18

19 Day-Ahead Load Forecasting Observations ALFS Day-Ahead (DA) Forecasting Results: Evaluated CAISO ALFS day ahead forecast metrics side-by-side with and without BTM forecast inputs (127 days (Jan-May)) Hourly ALFS DA Load Forecast Error Metrics Results suggest benefits when BTM PV forecasts are incorporated into the ALFS DA forecasting framework 19

20 Example Cloudy Day ALFS DA Load Forecasts Feb 5 12:00 LST Feb 6 12:00 LST Feb 7 12:00 LST Feb 8 12:00 LST Feb 9 12:00 LST Significant day-ahead load forecast improvement observed on cloudy days with inclusion of BTM solar forecasts! 20

21 Observations Accurately forecasting behind-the-meter PV is growing in importance. BTM forecasts can be used as input into load forecasts and also provide explicit power prediction (for planning/scheduling/spinning reserve purposes) Preliminary results demonstrate that detailed PV fleet simulations can improve load forecast accuracy (especially under challenging cloudy day conditions) Other considerations exist beyond just forecast accuracy: Maintaining the system (e.g., where will the PV system spec data come from? Is it scalable as number of PV systems grows?) Integrating forecasts into existing tools 21

22 Future Work Expand effort to include additional CAISO zones and hour-ahead forecasting timeframes during critical summer load season Additional work implementing IR satellite-based forecasts for improved pre-dawn solar forecasts Follow-on work: CEC EPIC funded (CPR Teamed with Itron) Address cost-effective strategies for high penetration of PV into distribution systems by integrating PV modeling into utility planning and operation tools 22

23 Thank you Please feel free to contact us for any details or clarification related to presentation Skip Dise SolarAnywhere Prod. Manager Adam Kankiewicz Solar Research Scientist Elynn Wu Solar Analyst The information herein is for informational purposes only and represents the current view of Clean Power Research, L.L.C. as of the date of this presentation. Because Clean Power Research must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Clean Power Research, and Clean Power Research cannot guarantee the accuracy of any information provided after the date of this presentation. CLEAN POWER RESEARCH, L.L.C. MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.