Economic MPC of Thermal Storage for Demand Response American - - PowerPoint PPT Presentation

Economic MPC of Thermal Storage for Demand Response

American Control Conference, July 1, 2015 Kevin Kircher, kircher.mae.cornell.edu

1 / 19

Background Model Simulation Discussion

2 / 19

Background Model Simulation Discussion

3 / 19

Power system peaks are expensive

2 4 6 8 10 12 50 100 150 200 250 300 350 400 450 500

Hourly average load (GW) Number of hours

2013 NYC load histogram

• 95th percentile: 8.5 GW
• maximum: 11.5 GW
• 1 GW of new peaking

capacity: $1 billion

4 / 19

Depeaking with thermal storage

• peaks happen on hot summer days, driven by AC
• curtailing cooling on hot days risks bothering occupants
• storage eliminates this risk
• why thermal storage?

⋄ electrochemical storage:1 500-600 $/kWh ⋄ thermal storage:2 14-20 $/kWhth (equivalent to 35-60 $/kWh with chiller COP of 2.5-3)

• 1R. Hensley et al., “Battery Technology Charges Ahead.” McKinsey Quarterly 3 (2012):

5-50.

• 2A. Arteconi et al., “State of the Art of Thermal Storage for Demand-Side Management.”

Applied Energy 93 (2012): 371-389. 5 / 19

Can MPC handle the incentives that real buildings face?

A challenging case study

• ConEd’s default rate plan3 for large commercial buildings

⋄ hourly energy prices determined by wholesale market ⋄ three-tiered demand charge

• a ConEd demand response program4

Main result: yes, but it’s important to include true incentives, particularly demand charge, in MPC objective function

3Rider M - Day-Ahead Hourly Pricing. General Rule 24: Service Classification Riders. ConEd, 2014. 4Commercial System Relief Program. Demand Response Program Details, ConEd, 2014. 6 / 19

Background Model Simulation Discussion

7 / 19

Physics

• DOE “large office” prototype5 (3 floors, 14,000 m2)
• quasi-steady model extends seminal work6 to include

⋄ two chillers ⋄ temperature-varying COPs ⋄ non-ideal tank and heat exchanger efficiencies

5Commercial Building Prototype Models: “Large Office.” Building Energy Codes Program, U.S. Department of Energy. (2011) 6Henze, G. et al. “Development of a Predictive Optimal Controller for Thermal Energy Storage Systems.” HVAC&R Research 3.3 (1997): 233-264. 8 / 19

Physics (continued)

x(k + 1) = Ax(k) + B(k)u(k) + Gw(k)

• states

⋄ tank charge (x1, kWhth) ⋄ cooling deficit (x2, kWhth)

• controls

⋄ ice chiller power (u1, kW) ⋄ cooling from ice (u2, kWth) ⋄ main chiller power (u3, kW)

• disturbances (Gaussian, white)

⋄ cooling demand (w1, kWth) ⋄ electrical demand (w2, kW)

9 / 19

MPC optimization

• 24-hour horizon, half-hour time steps
• minimize

+ energy cost + increase in demand cost + occupant discomfort + terminal cost (tank depletion) − demand response revenue

• subject to

⋄ chiller capacity and ramping limits ⋄ tank limit

• solved in CVX, driving SDPT3

10 / 19

Background Model Simulation Discussion

11 / 19

Simulation day

6 12 18 24 20 30 40

Temperature and Coefficients of Performance

Temperature ( ◦C) 6 12 18 24 1 2 3 4 COP Main Chiller Ice Chiller Temperature 6 12 18 24 100 200 300 400 500

Expected Loads and 95% Confidence Intervals

Cooling Load (kW t h) Time (hours) 6 12 18 24 100 200 300 400 500 Other Electric Loads (kW)

12 / 19

Prices

6 12 18 24 0.2 0.4

Energy and Demand Response Prices

c e(k) ($/kWh) 6 12 18 24 2 4 c dr (k) ($/kWh) 6 12 18 24 5 10 15 20

Demand Prices

c d(Ti) ($/kW) c d(T1) c d(T2) c d(T3) 6 12 18 24 2 4 6 x 10

−3

Prices of Under- or Over-cooling

Time (hours) c u(k ) ($/kWh2)

13 / 19

A typical Monte Carlo run

6 12 18 24 1000 2000 Tank Charge State x1 (kWht h) 6 12 18 24 −500 500 Cooling Load Deficit x2 (kWht h) 6 12 18 24 50 100 Power to Ice Chiller u 1 (kW) 6 12 18 24 200 400 Cooling from Ice Melt u 2 (kW t h) 6 12 18 24 50 100 Power to Main Chiller u 3 (kW) Time (hours)

14 / 19

A typical Monte Carlo run (continued)

6 12 18 24 50 100 150 200 250 300 Total Power Consumption kW MPC Baseline No Storage 6 12 18 24 −80 −60 −40 −20 20 40 60 Costs Time (hours) $ Energy Demand Response Under-cooling

15 / 19

Background Model Simulation Discussion

16 / 19

Important to model demand charge

6 12 18 24 50 100 150 200 250 300 Demand Charge Ignored Total Power (kW) Time (hours) 6 12 18 24 50 100 150 200 250 300 Demand Charge Included Total Power (kW) Time (hours) Energy Demand Response Demand Under−cooling Tank Depletion −1000 1000 2000 3000 4000 Costs with and without Demand Charge in Objective Function $ Without gd With gd

17 / 19

Lots of extensions

• optimal tank size?
• simulate for a month, study demand charge in depth
• other economic incentives

⋄ critical peak pricing ⋄ ancillary services ⋄ contracts with aggregators

All code is available by email or at kircher.mae.cornell.edu.

18 / 19

Thanks to. . .

• the Consortium for Electric Reliability Technology

Solutions (CERTS) for funding

• Max Zhang for advising
• Santiago Naranjo Palacio, Brandon Hencey, and Eilyan

Bitar for ideas and feedback

• you for listening!

19 / 19