Wind Power Output Forecasting

Wind Power Output Forecasting

Loading Social Plug-ins...
Language: English
Save to myLibrary Post page as Slide Download PDF
Go to Page # Page of 15

Description: Wind forecasts become essential for effective grid management with “high” wind penetrations (>5%). Wind fluctuations increase requirements for spinning reserves and raise electricity system production costs. Large ramp events can affect electricity system reliability.

State-of-the-art forecasts have high economic value compared to their cost. Wind conditions are site-specific and time/height variable. Accuracy is crucial.

Wind resource assessment programs must be designed to maximize accuracy.

Author: Arvinder Singh (Fellow) | Visits: 2917 | Page Views: 2934
Domain:  Green Tech Category: Wind/Water/Geo Subcategory: Wind Power 
Upload Date:
Short URL:


px *        px *

* Default width and height in pixels. Change it to your required dimensions.

Maxims of Tech: Rules of Engagement for a Fast Changing Environment
Implementing Real-time Wind Power Output Forecasting in India
Assocham: Mainstreaming Green Energy Chennai April 29, 2010
Arvinder Singh India Business Head +91-99450-58233

� � � � � � � Company Introduction Forecasting Experience & Core Competence Integrating forecasting in India TM How it works? eWind Forecasting System Approach Forecasting Skill Measurement Data input required and sources Summary

Company Background & Core Competence
Atmospheric Sciences, Modeling and Measurement, Validation & Engineering Firm
lEstablished lIndustry

in 1983; 90+ employees Leader & Consultant for 40,000 MW on- and offshore lProject roles in over 60 countries lExtensive technical & broad experience in the wind industry l Model developers - not just model users l 22 years of atmospheric model development and customization l Ongoing internal R&D program

A World Leader in Forecasting

� 10 years in wind & solar forecasting � 55% US-wind market, 15,000MW � PhD atmospheric scientists backed by a cluster of 150 processors � Cutting-edge research
Market Share USA

Forecasting Key Clients
� California Independent System Operator: � 18 wind farms; 1,000+ MW � Hour and Day-Ahead Forecasts

Electric Reliability Council of Texas: � 75 wind farms; 8,900 + MW � Hour and Day-Ahead Forecasts, Ramp Detection Forecast New York Independent System Operator: � 13 wind farms; 1,200 + MW � Minutes, Day, and Hour-Ahead Forecasts Each client has unique forecasting needs that require unique forecasting solutions

� � � � � � The wind resource drives project viability. Wind conditions are site-specific and time/height variable. Accuracy is crucial. Wind resource assessment programs must be designed to maximize accuracy. Combination of measurement and modeling techniques gives the most reliable result. Know the uncertainties and incorporate into decision making. Good financing terms depend on it.

Why Forecast for India?
� Wind forecasts become essential for effective grid management with "high" wind penetrations (>5%)... � Wind fluctuations increase requirements for spinning reserves and raise electricity system production costs � Large ramp events can affect electricity system reliability � State-of-the-art forecasts have high economic value compared to their cost

Investor confidence in Indian Wind Farms
...identifying the true wind power potential in India?
... Capacity Factors at hub-height?

1. 2. 3. 4. 5.

Met tower verification and data validation Long term reference data for M-C-P Spatial distribution and Climate Adjustment Wakes Performance Gap (energy estimates vs. plant output)

Influences on Uncertainty
l l l l l

Measured Speed Shear Climate Resource Model Plant Losses

Sensor Types, Calibration & Redundancy, Ice-Free, Exposure on Mast, # of Masts Height of Masts, Multiple Data Heights, Sodar, Terrain & Land Cover Variability Measurement Duration, Period of Record @ Reference Station, Quality of Correlation Microscale Model Type, Project Size, Terrain Complexity, # of Masts, Grid Res. Turbine Spacing (wakes), Blade Icing & Soiling, Cold Temp Shutdown, High Wind Hysteresis, etc.

Investor confidence in India
Engaging with the lending community on 3rd party energy assessments
� � � � � � � � � � � � � � SBI (State Bank of India) SBI-Caps (State Bank of India Capital Markets) IREDA (Indian Renewable Energy Development Agency) PFC (Power Finance Corporation) stand DEG (German Development Bank) DBS (Development Bank of Singapore) IL&FS (Infrastructure Leasing and Finance) Rabobank ADB (Asian Development Bank) IFC (International Finance Corporation) ICICI Bank Yes Bank Proparco (French Development Bank) IDFC (Infrastructure Development and Finance Corporation).

Typical Forecasting Services

Typical Forecasting Services
� Hour-Ahead Forecasts
� � � Delivered hourly, covers next several hours in hourly increments 1-8 hours (24 updates per day) Next-operating-hour unit commitment

� Day-Ahead Forecasts
� Delivered 2-4 times per day, covers next several days in hourly increments � � � � Unit commitment and scheduling Market trading Operations and maintenance Construction scheduling

Forecast Time Horizons
� 5 - 60 minutes � Uses: Regulation, real-time dispatch decisions � Phenomena: Large eddies, turbulent mixing transitions � Methods: Largely statistical, driven by recent measurements 1-6 hours ahead: � Uses: Load-following, next-operating-hour unit commitment � Phenomena: Fronts, sea breezes, mountain-valley circulations � Methods: Blend of statistical, NWP models Day-ahead � Uses: Unit commitment and scheduling, market trading � Phenomena: "Lows" and "Highs," storm systems � Methods: Mainly NWP with corrections for systematic biases Seasonal/Long-Term � Uses: Resource planning, contingency analysis � Phenomena: Climate oscillations, global warming � Methods: Based largely on analysis of cyclical patterns

Regulatory & Market Conditions

� Demand for forecasting will not emerge until the regulatory and market conditions favor it � Wind industry should embrace forecasting � makes wind a "better customer" � Factors favoring forecasting:
� Increasing wind penetration � Market participation rules, e.g. imbalance charges � Participation of wind in power trading

Considerations for Implementing in India

� Centralized or Decentralized? � Integration with Grid Operations � Data Requirements

Typical System Configurations

System Operator
Project A
Scheduling Dispatch Regulation
Load Forecasts

System Operator
Scheduling Dispatch Regulation Load Forecasting

Project A Project B Project C eWind eSolar

eWind eSolar
Forecast s

Project B Project C

Centralized System Grid Operator

Decentralized System Plant Owner

Centralized v. Decentralized
� Centralized systems
� � � � Owned or contracted by the grid operator Lower total cost for multiple plants Easier to set and enforce standards, maintain consistent quality Potential to aggregate data from different plants and improve forecast quality

Decentralized systems
� Forecasts supplied individually by wind projects � Standards can be set, but enforcement may be difficult

Forecast System Configuration

� Centralized systems managed by grid operator currently predominate in US and Europe � Imposing forecasting on individual plants less successful � Relevant time horizons depend on grid flexibility, operating norms
� Typically, load following (0-6 hours) and plant scheduling (next day) are most important

Integration with Grid Operations
� � The forecasts may be fine, but will they be used? Forecasts should be customized to the real needs of the grid operators
� Confidence levels on routine forecasts � Focus on critical periods, e.g., times of maximum load or maximum load swing � Ramp forecasts � Severe weather forecasts

� �

Dedicated staff to monitor forecasts Other steps to make integration more effective: training, visualization tools, plant clustering

� Forecast performance varies with many factors
� Forecast time horizon (especially for short-term) � Amount and diversity of regional aggregation � Quality of generation & met data from the plant � Distribution of wind speeds relative to the power curve � Type of wind and weather regime � Shape of the plant-scale power curve � Amount of variability in the wind resource � Sensitivity of a forecast to initialization error

Typical Errors in an Individual Farm

MAE (% of rated capacity) �1-hour: 2%-6% �6-hour: 8%-15% �24-hour: 12%-18% �48-hour 13%-22%

20% 15% 10% 5% 0% 0



9 12 15 18 21

24 27 30


36 39 42 45 48

Forecast Time Horizon (Hours)

� Errors increase rapidly in first 6 hours (statistical domain) � Errors grow slowly from 12 to 120 hours (NWP domain) � Real-time on-site data from site is required to outperform persistence from 0 to 3 hours

Data Needs
� Accuracy of forecasts depends on frequency, quality, and types of data received from wind farms � Meaningful forecasts can be done without any plant data, but accuracy will be poor � Ideally:
� � � � � Hub-height met tower at each plant, 2 levels of sensors Turbine and total plant output Wind data from nacelle anemometers Turbine availability Data delivered at least as frequently as forecasts (i.e., daily for next-day forecasts, hourly for next-hour forecasts)

� Data transfer by secure FTP, web services, e-mail...

"eWind" Technology

� � � � � � �

Secure and reliable Unmatched accuracy Custom numerical weather models Adaptive learning Forecast ensembles Ramp forecasts Other specialized forecasts

Rapidly Updating Numerical Weather Prediction (NWP) Models

Physical Models

� � � � �

Physical equations of the atmosphere are solved on a grid Initial conditions are obtained from government center weather observations Models typically run 2x or 4x per day out 1-5 days Some forecast providers rely solely on government-run models AWST runs its own � Why? Higher resolution Can be customized Create larger ensembles

Statistical Models
� � � Correct for systematic NWP errors (high/low biases) Incorporate recent data from the site or nearby locations � improves accuracy Can include conversion of forecasted winds to plant output AWST employs several different statistical models: linear regression, neural networks, support vector machines...
Predict ors Predict and

P1 ,P2 ,...

Training Algorithm

F = f ( P1 ,P2 ,...)

Plant Output Models
� Establishes a relationship between meteorological variables and power production for a specific plant Implicitly or explicitly account for Turbine power curves Topographic effects Wake losses Other losses Availability

Aggregating Forecasts Reduces Errors
j y
Farm Avg (12- 52.8 MW) Regional Avg (4-158.5 MW) System (634 MW) 35 30 25 20 15 10 5 0
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42

� Example: Alberta Wind Forecasting Pilot Project
� 12 wind farms divided into 4 geographic regions of 3 farms each

� RMSE for regional dayahead forecasts was 15%20% lower than for the individual farms � RMSE for system-wide day-ahead forecasts was 40%-45% lower than for the farms

Look-ahead Period (hrs)

Ramp Events
� Large ramp events gaining increasing attention � Optimizing forecasts to MAE or RMSE tends to reduce ramp-forecasting skill � Attempting to maximize ramp-specific skill scores may solve this problem



� Wind forecasting is becoming ever more important in India as wind penetration grows � Considering current wind penetration in some states, it is already late for India to be implementing forecasting � Current forecasting technology is imperfect but very cost effective compared to no forecast at all � Benefits of aggregation and need for large investments (e.g., observational networks) support centralization of forecasting operations � Utilities/developers should undertake I-C-B route to hire a wind forecast provider, undertake pilot tests

Thank you
Arvinder Singh India Business Head AWS Truewind, LLC +91-99450-58233