Economic Impacts of Advanced Weather Forecasting on Energy System Operations
Victor M. ZavalaDirector’s Postdoctoral Fellow
Mathematics and Computer Science DivisionArgonne National Laboratory
Innovative Smart Grid TechnologiesJanuary 2010
Joint Work: Emil Constantinescu and Mihai Anitescu
MotivationCurrent Grid
MotivationSmart Grid
- Major Adoption of Renewable Resources (20-30%)- Highly Distributed and Uncertain Generation and Consumption
Motivation
Generation Costumer
Transmission/Distribution
System OperatorON/OFF
Power Levels Loads
Weather -Disturbance-
Forecasting is Critical to Minimize Reserves and Ensure Robustness
Two-Way One-Way
Consumer Level Generation Level
Unit Commitment and Economic Dispatch
Motivation
Weather Affects Control and Energy Management Systems at Multiple Levels
MotivationSmart Grid
Loads and Renewable Generation Exhibit Non-Intuitive Spatiotemporal Correlations
1. Weather Forecasting
2. Unit Commitment Level- Wind Power Generation
3. Generation Level- Photovoltaic-Hydrogen Hybrid System
4. Consumer Level- Building Energy Management
5. Conclusions and Future Work
Outline of the Talk
1. Weather Forecasting
Weather Forecasting with WRF
Major Advances in Meteorological Models (WRF)- Highly Detailed Phenomena- High Complexity 4-D Fields (106- 108 State Variables)
Model Reconciled to Measurements From Distributed Stations
Data Assimilation Techniques:- 3-D Var Courtier, et.al. 1998
- 4-D Var Navon et.al., 2007
- Extended and Ensemble Kalman Filter Eversen, et.al. 1998
http://www.meteomedia.com/http://www.emc.ncep.noaa.gov/gmb/ens/
Is WRF Accurate and Computationally Practical Enough for Grid Operations?
Weather Forecasting with WRF
Current Time
Data Assimilation Forecast
Computing Exact Covariance Matrix is Impractical:1) Create Empirical Distribution of Current Atmospheric State Z, Constantinescu & Anitescu, 2009
2) Propagate Samples through WRF Model
Making WRF Simulations Feasible:Targeted Resolutions and Computer Resources
Jazz Cluster at Argonne National Lab
Weather Forecsating with WRFValidation Results (Illinois, 2006) with NOAA Data
Temperature [oC] Wind Speed [m/s]
2. Unit Commitment Level
Unit Commitment-Energy DispatchStochastic Unit Commitment (UC) with Uncertain Wind Power Generation
Analyze Impact of WRF Accuracy and Uncertainty on UC Performance
Inference Analysis with WRF
Integration WRF & Stochastic Unit Commitment- Few Wind Realizations Available (~ 50) “Optimal” Cost Never Known Exactly! - How to Generate More Realizations?
Inference Analysis with Weighted Average Resampling1) Sample Weights on Hyperplane and Compute2) Solve Stochastic UC Problem with Batches of Realizations
Cost Lower Bound Distribution
Integrative Computational Framework
WRFModel
EnergyDispatch
Thermal PowerGenerators
EnsembleGenerator
DataAssimilation
Wind Farms &Meteo Stations
Wind PowerCurve
Unit Commitment
State ForecastSamples
Initial State Field Samples
ReanalyzedState Field
Measurements
Wind PowerSamples
ON/OFFStates
Power LevelSet-Points
Wind SpeedSamples
CostFunction
Wind PowerSample Batches
Cost Lower andUpper Bounds
Weighted Average
Resampling
InferenceAnalysis
Uncertainty Quantification
Stochastic Optimization
WRF Resolution and Realizations Must be Adapted According to Stochastic UC Solution
Integrative Closed-Loop StudyAnalyze Effects of Large Adoption Levels of Wind Power
- Stochastic UC for 3 Days of Operation (Adoption Level 20%)
Wind Power Profiles
Integrative Closed-Loop Study
Demand
Thermal
Wind
- WRF is Accurate with Tight Uncertainty Bounds But Excursions Do Occur- Need to Tailor Resolution of Data Assimilation Step
Analyze Effects of Large Adoption Levels of Wind Power- Stochastic UC for 3 Days of Operation (Adoption Level 20%)
Aggregated Power Profiles
3. Generation Level
PowerLosses
SolarRadiation
Storage
Hybrid Photovoltaic-H2 System
MPPT
• Operating Costs and Loads Driven by Uncertain Radiation Ulleberg, 2004
•Multiple Power Losses
Load Demand
PowerLosses
PowerLosses
• Candidate for Distributed Power Generation (Buildings, Homes)
Hybrid Photovoltaic-H2 SystemProactive Energy Management Z. Krause & Anitescu, 2009
Minimize Power Losses
Power Balances
State-of-Charge, Fuel Cell and Electrolyzer Limits
• Energy Management Usually Reactive (e.g.; Fuzzy Logic) Ulleberg, 2004
• Proactive Manager with Forecast Horizon of 1hr, 1 Day, …,14 Days
Radiation Forecast
Chicago, IL 2004
1 hr 4 hr 6 hr 12 hr 1 day 3 day 7 day 14 day0
50
100
150
200
250
300
350R
elat
ive
Cos
t [%
]
Horizon
Reactive (Steady-State)
Proactive 1Day
• Costs Reduced By 300% From 1-Hr to 14-Day Forecast
• Close-to-Optimal Profit Achieved with Short Forecasts
14 Day
1Hr
Hybrid Photovoltaic-H2 System
Profiles of Fuel Cell Power
Short Forecasts = Aggressive Actions
Long Forecasts = Smooth Actions
Hybrid Photovoltaic-H2 System
0 50 100 150 200 250 300 350-1
0
1
2
3
4
5
Time [Days]
Cos
t [$]
DeterministicStochastic
Load Satisfaction Deterministic vs. Stochastic
Deterministic Fails
Hybrid Photovoltaic-H2 System
Proactive Energy Manager Forecasts Grid Load
Uncertainty Modeling is Critical for Robust Load Satisfaction
4. Consumer Level
Thermal Management of Building Systems
Minimize Annual Heating and Cooling Costs
0 50 100 150 200 250 300 350
-10
0
10
20
30
Time [Days]
Tem
pera
ture
[o C]
Time-Varying Electricity Prices Peak & Off-Peak
Pittsburgh, PA 2006
BuildingEnergy Balances
www.columbia.edu/cu/gsapp/BT/LEVER/
1 hr 3 hr 6 hr 9 hr 12 hr 16 hr 24 hr0
20
40
60
80
100
Horizon
Rel
ativ
e C
ost [
%]
Energy Management of Building Systems
1hr
24 hr
Weather Forecast Leads to 10-40% Cost Reduction (Depends on Insulation)
0 2 4 6 8 1015
20
25
30
Time [Days]
Tem
pera
ture
[o C]
Proactive Energy Manager Shifts and Forecasts Load Profile
Comfort Zone
24hr Forecast1hr Forecast
Effect of Forecast on Energy Costs
Energy Management of Building SystemsEmpirical Model Validation Results, Pittsburgh Area 2006 Z, Anitescu, et.al. 2009
One Hour Ahead
5 Days Ahead
Empirical Model Cannot Capture Weak Periods, Inconsistent Uncertainty Bounds
Energy Management of Building SystemsWRF Validation Results, Pittsburgh Area 2006 5 Days Ahead Forecast Z, Anitescu, et.al. 2009
Confidence Interval
Forecast
Measurement
Hours (August 1st-5th )
Temperature Correlation Field
Proactive Energy Manager with Different Forecast Models
Perfect Forecast
EmpiricalModel
WRF
Energy Management of Building Systems
5. Conclusions and Future Work
Smart Grid- Weather Forecasts with Detailed Uncertainty Information (e.g. Correlation Fields)
Preliminary Analysis of Advanced Weather Forecasting in Energy Operations- Tight Uncertainty Bounds for Lower Costs and Robustness
Exploit Advances in Numerical Weather Prediction Models- Predictive Capabilities Astonishing - Temperature, Wind Speed, Radiation
Open Questions for Smart Grid- How to Generate Low Cost Forecasts for Power Companies and Costumers?- Targeted Resolution and Assimilation Constrained by Computational Resources- How to Aggregate Highly Distributed Loads and Generations? - Integration of Hierarchical Levels (Effects of Uncertainty in Markets)
Conclusions and Open Issues
Economic Impacts of Advanced Weather Forecasting on Energy System Operations
Victor M. ZavalaDirector’s Postdoctoral Fellow
Mathematics and Computer Science DivisionArgonne National Laboratory
Innovative Smart Grid TechnologiesJanuary 2010
Joint Work: Emil Constantinescu and Mihai Anitescu