assessing opportunities to exploit stranded power - andrew chien
DESCRIPTION
Andrew Chien 2015-2016 Seminar SeriesOctober 29, 2015With ambitious new “renewable portfolio standards” of 50% in California by 2030 and for 20% of electricity for the entire United States by 2030, both excluding hydroelectric, the power grid faces dramatic challenges in absorbing higher levels of variability, whilst continuing to deliver reliable, cheap energy. And with the DOE’s recent Wind Power vision indicates a 35% national RPS from wind is possible by 2050, the challenge continues beyond these nearer term goals. As generation variability increases, limited transmission produces a phenomenon known as “stranded power” – energy that is generated, but goes unused because it cannot be transmitted to any point of use. We are exploring the possibility of creating “dispatchable” loads that can exploit such power and generate value. One such system is Zero-Carbon Cloud (ZCCloud) that provides cloud computing services, with the resulting revenue improving the economic viability of renewables. Such dispatchable loads can increase the efficiency and stability of the grid, as well as the economic viability of high RPS operating points. Initial studies show that ZCCloud can create high-value, computing resources with payback periods as short as several years.TRANSCRIPT
11/14/15
1
Assessing Opportunities to Exploit Stranded Power Andrew A. Chien *University of Chicago and Argonne National Laboratory w/ Fan Yang, Rich Wolski (UCSB), Victor Zavala, Kibaek Kim (Argonne) UCSB Institute for Energy Efficiency October 29, 2015
Outline • What is Stranded Power? • Understanding Stranded Power • Exploiting Stranded power • Looking Forward (Grid, RPS)
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 2
11/14/15
2
Scopes of Energy Efficiency
• Rack, Data center • Power Distribution, Power Grid • Generators
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 3
Talk Focus
Stranded Power
• Grid economic management mechanisms can cause renewable power to be “stranded” • Ramps, Transmission Congestion or Shortage, Correlated Production
• MISO “down dispatched” 2.2TWH of wind in 2014 • Equivalent to 251 MW average; ~2%, MISO: Success!
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 4
11/14/15
3
UNDERSTANDING STRANDED POWER
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 5
Mid-continent Independent System Operator (MISO)
• 15 states, serves 42M people
• 130 GW peak load, >500TWh/year
• $37B Power (2014) • 413 Market participants • >2,000 pricing nodes
• >65,000 miles transmission • $2.2B transmission
charges
• ~10% Wind generated
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 6
11/14/15
4
Transformation of the Power Industry and Power Grid
Power Grid in 1975
• Vertically integrated • Few, Large fixed
generation sites • Large loads,
geographically fixed, 2x dynamic cyclic
Power Grid in 2015
• Separation of generation and distribution
• Explosion in large-scale distributed renewables
• Highly variable Generators
• Loads same
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 7
Central Control Economic Dispatch,
“open” market, Localized Pricing
Renewable Variation and Economic Dispatch in MISO
• Technical: Generators bid “Eco_max,min”, and price curves hourly
• Economic: Global optimization, focused on “Merit order”, Day-ahead Market, Real-time Market
• Control: 5-minute dispatch (with 4-second adjustments) • Payments determined by actual deliveries
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 8
Capability Price info • Feasible
• Cheapest • Green • Fair • Fast (5-minute increments)
• + Reliable • +“As much as you want”
11/14/15
5
Locational Marginal Pricing and Local Outcomes
• Market prices based on the “incremental” cost of each MWh • Variation+Transmission+Economics: Complex, coupled behavior • Produces wide variation at the edge (generators) • LMP => the market price for the generator
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 9
Generation Offered
Transmission Constraints
MISO Market Schedule & Prices
• Feast and famine • Variable Generation • Prices from $100 to -$50 [Range is wider now]
• Generator’s Goal: a good “Average” Price
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 10
Wind Power Pricing Minnesota Hub
11/14/15
6
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 11
<add python timeline plot>
Hours
Example Site - Timeline
• Price vs. Time [2015: Jan-April]
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 12
• Price vs. Time, Unit 8238, January 2015
Example Site – Timeline (zoom)
<add python timeline plot>
Hours
$$/M
Wh
11/14/15
7
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 13
• EcoMax (potential) vs Delivered [2015: Jan-April]
Example Site – Stranded Power
Stranded Power
• Precise model: Instantaneous • Power with no economic value, for a 5-minute interval
• Define: LMP0 as LMP < $0 / MWh
• Define LMP(x) as LMP < $x / MWh
• LMP0, LMP1, LMP2, ... LMP5 • $32/MWh is average for MISO
• So, LMP5 is power 6x lower than the average price
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 14
11/14/15
8
Analysis of a Power Grid Market • MISO Market Records Properties
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 15
Period Jan 1, 2013-Apr 14, 2015
28.5 Months
Generation Sites 1,259 Total 200 Wind
5-min Intervals 76.9M Total 36.6M Wind
Power 1,188 TWh Total 89 TWh Wind
$$ Revenue $95 Billion $4 Billion
Wind Sites, SP Duty Factor: LMP0
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 16
11/14/15
9
LMP Stranded Power vs Price Threshold ($/MWh)
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 17
Interval Sizes, LMP
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 18
Best Single Site
11/14/15
10
Stranded Power, V2 • A 2nd “stranded power” model
• NetPrice (average over a longer period) • Offer a good deal to the generator for the entire period
• Idea: smooth out the fluctuations
• NetPrice(C): NetPrice < C (C = $0,$1,...,$5)
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 19
Wind Sites, SP Duty Factor: NetPrice0
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 20
11/14/15
11
NetPrice Stranded Power vs Price Threshold ($/MWh)
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 21
NetPrice Stranded Power: Intervals (by Count), NetPrice5
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 22
Best Single Site
11/14/15
12
NetPrice Stranded Power: Intervals (by time), NetPrice5
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 23
Best Single Site
Lots of Stranded Power
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 24
11/14/15
13
Dispatchable Loads: Approach • Idea: Create on-demand computational loads that can
exploit the stranded power • Reduce grid waste and increases stability • Provides greater financial return for renewable generators • A means to achieve higher RPS and lower grid costs
• Two types • Spot Instances, Preemptible instances, Peer to Peer, Desktop Grid • High-Performance Computing
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 25
DISPATCHABLE COMPUTING LOADS: CLOUD SERVICES�
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 26
11/14/15
14
Zero-carbon Cloud (ZCCloud)
• Build distributed computing servers that act as a “dispatchable load” • Start on demand when there is stranded power • Consume the power, continuously while available • Shutdown gracefully when the power is no longer available
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 27
Amazon’s Spot Instance Model
• Execute a “virtual machine” or “container” at a time when the computing is below a bid price
• Start time + End time: dictated by the system • Queue of “persistent” and “new” spot requests
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 28
(Courtesy Amazon)
11/14/15
15
DISPATCHABLE COMPUTING LOADS: HIGH-PERFORMANCE COMPUTING�
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 29
Batch High-Performance Computing Model
• Execute a parallel job (array of machines or VM’s), 1-16,384 • Start time + End time: dictated by the system (jobs request runtime) • Queue of “new” parallel job requests
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 30
Large Parallel jobs
Batch Scheduler Parallel
Computing Resources
Results
11/14/15
16
Traditional Data Center System Limits
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 31
Augmenting Mira with ZCCloud
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 32
10 Petaflops, 4MW, 786K Cores 2nd System of the Same Size 10 Petaflops, 4MW, 786K Cores
11/14/15
17
ALCF Workload Traces • Mira: #5 Supercomputer in the World • IBM Blue Gene/Q System, 768K Cores, 10 Petaflops
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 33
Period Dec 31, 2013 – Dec 30, 2014
# of Jobs 78,795
Runtime (Hours) Range: 0.004 to 82 Average: 1.7, StDev: 3.0
# Nodes Range: 1 – 48K Average: 1,975, StDev: 4,100
Basic Performance, by Workload Shape
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 34
• Understanding Intermittent Resources • Adding resources translates into Improved Performance • Even better than basic capacity increase (1.5x) • Shape doesn’t matter
11/14/15
18
Performance, by Job Size
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 35
• Understanding Intermittent Resources • Performance improvement Overall, but varies with size • Benefit Weighted heavily to “large, capability jobs”
• Insights for system mgmt?
Performance, by “Ontime” or “Late”
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 36
• Ontime jobs can complete in the current “volatile resource” available window
• Late jobs cannot (arrived too late) • Ontime jobs benefit more! But Late jobs benefit as
well...
11/14/15
19
Performance, by Stranded Power Model
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 37
• LMP models lower duty factor produces less benefit • Some loads are infeasible • Netprice high duty factors provide growing benefit
Performance, Equal Duty Factor
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 38
• LMP models provide lower duty factor • LMP models perform worse than periodic model with
equal duty • NetPrice models can provide higher duty factors • Can provide better performance
11/14/15
20
Related Work • Integration of Renewables into the Power Grid
• Wind Vision, CAISO, Grid expansion, Power Markets
• Energy-efficiency in Data Centers • Power Distribution, Dynamic Management, Energy Proportional
Computing, “Drowsy” management
• Solar Augment and LT-Purchase for Wind, Colocation • A good thing “generally”, support the finances of renewable power • Create covariation with renewables in the grid: counter-variation of
load (bad)
• Follow the Sun • Closest, create co-varying load (good)
• ZCCloud: New “volatile” computing resource model
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 39
Summary • Stranded power is a growing challenge for renewables
• Computational Dispatchable loads are promising • Benefits grids and renewable generators
• Cloud Services Evaluation • Initial Economic Evaluation
• High Performance Computing Evaluation • Intermittent resources can be gracefully integrated • Stranded Power models are important
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 40
11/14/15
21
Looking Forward • What other uses of stranded power are feasible?
• Lightweight manufacturing • Opportunistic tasks: Carbon removal? Reduction?
• What opportunities are there to create higher value computing services from intermittent energy? • Combining sites • Intelligent migration
• What benefits can “dispatchable loads” accrue to the grid? • Increased stability, reduced operating margin • Increased ability to exploit renewable power • Partial Load shifting
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 41
BENEFITS OF DISPATCHABLE LOADS FOR GRID
October 29, 2015 Opportunities Stranded Power (UCSB-IEE) 42