Electricity market reform to enhance the energy and reserve pricing mechanism: Observations from PJM

Transcription

1 Flexible operation and advanced control for energy systems Electricity market reform to enhance the energy and reserve pricing mechanism: Observations from PJM January 7, 2019 Isaac Newton Institute Cambridge University, UK Hung-po Chao, Ph. D. Senior Director and Chief Economist PJM Interconnection, LLC

2 PJM as Part of the Eastern Interconnection Key Statistics Member companies 1,000+ Millions of people served 65 Peak load in megawatts 165,492 MW of generating capacity 176,569 Miles of transmission lines 82, GWh of annual energy 792,314 Generation sources 1,304 Square miles of territory 243,417 States served 21% of U.S. GDP produced in PJM 13 + DC As of 2/2017 2

3 Introduction For over 20 years, the PJM wholesale markets have successfully worked to produce competitive prices and attract investments to support efficient and reliable operations In recent years, the PJM markets have been undergoing a significant transition While such transitions have also occurred elsewhere, each has presented some unique challenges Opportunities exist to improve the price mechanism PJM is in the process of working with stakeholders to enhance the price mechanism in energy and reserve markets 3

4 The PJM market environment Uplift payment is persistent totaled $2 billion over ten years Nearly 20% of the time, energy price falls when demand increases Over 60% of energy is produced by self-scheduled generation Many CTs submit inflexible bid parameters Revenue has shifted from the energy market to the capacity market Slowing demand growth, flattening supply curves and penetration of renewables present new challenges 4

5 Market Revenue Distribution 5

6 Convexity Plays a Fundamental Role in Market Mechanism A competitive equilibrium is Pareto optimal, if 1) Preferences are locally non-satiated, 2) There exist no transaction costs, external effects or market power 3) Everyone acts as a price taker, not trying to influence price Any Pareto-efficient allocation can be attained by a price mechanism in a competitive equilibrium with transfers, if every production set and every preference relation is convex 6

7 Desirable Attributes for a Pricing Mechanism Efficiency Prices reflect scarcities and costs Support efficient system operations Incentive compatibility Promote truthful bids Reward efficient behavior Revenue sufficiency Sustain competitive entry and investment Minimize the use of uplift payments 7

8 Evidence and Impacts of Non-convex Conditions Energy and reserve prices do not accurately reflect scarcities or costs Lumpy commitments suppress prices Price falling when demand rises sends wrong signals Prices do not incentivize efficient behavior Units frequently submit inflexible bid parameters A large number of units chose to self-schedule Prices do not assure recovery of fixed costs Units needed to serve load may incur losses Units not needed would be asked to stay offline but may be profitable to run Reliance on uplifts and capacity payments becomes normal routine 8

9 The Current Pricing Mechanism The same schedule/dispatch model is used for pricing Allocation Model Pricing Model Maximize market surplus The dual prices of a restricted linear programming problem Need make-whole payments 9

10 Optimal Allocation is Attained for Each Pricing Mechanism V I (y) = Max B d c i x i, z i d,x i R,z i 0,1 n i=1 Demand d + y Supply n x i i=1 10

11 Dual Optimization for the Current LMP Approach V I = Min φ p + π 0 i z 0 i, p p n i=1 Consumer Surplus, φ p max B d pd d R Producer surplus π 0 i z 0 i, p max px 0 i c i x i, z x i R i 11

12 Extended LMP (ELMP): Convex-hull Pricing Pricing model is based on the convex-hull relaxation of the allocation (schedule and dispatch) model Allocation Model Pricing Model Maximize market surplus V CH y Conv V I y p CH V CH 0 12 Uplift (solution support) payments include make-whole payments and lost opportunity costs

13 Extended LMP with Lagrangian dual Pricing model is based on the Lagrangian dual optimization of the allocation (schedule and dispatch) model Allocation Model Pricing Model Maximize market surplus Uplift payments for solution support 13

14 Extended LMP with Lagrangian Dual Formulation V L Min p φ p + π i L p n i=1 Consumer Surplus, φ p max B d pd d R Producer surplus π L i p max px i c i x i, w i x i R,w i 0,1 14

15 ELMP with Integer Relaxation: The PJM Proposal Pricing model is an integer relaxation of the allocation model It produces a close, and sometimes an exact, approximation of ELMP if the cost functions are homogeneous of degree one in both commitment and dispatch variables Allocation Model Pricing Model Maximize market surplus Uplift payments for solution support 15

16 Extended LMP with Integer Relaxation Formulation V R Min p φ p + π i R p n i=1 Consumer Surplus, φ p B d pd max d R V I V CH = V L V R Producer surplus π R i p max px i c i x i, w i x i R,w i 0,1 16

17 Example 1 Type 1 Plant Smokestack Type 2 Plant High Tech Capacity (MW) 1, Construction cost ($/hour)) 53,000 30,000 Fixed cost ($/MW-hour) Marginal cost ($/MWh) Average cost ($/MWh) Scarf, H. E., "The Allocation of Resources in the Presence of Indivisibilities," Journal of Economic Perspectives, Vol. 8, Number 4, pp ,

18 Example 1: Market Allocation Solution Demand (MW) Type 1 Unit Type 2 Unit Type 1 Output (MWh) Type 2 Output (MWh)

19 Example 1: Alternative Pricing Solutions Demand (MW) LMP ($/MWh) LMP Revenue/ Cost (%) LMP Uplift/ Cost (%) ELMP ($/MWh) ELMP Revenue/ Cost (%) ELMP Uplift/ Cost (%) LMP causes significant revenue shortfall, and reliance on the capacity market is necessary for new entry 19

20 Example 1a: Price Elastic Demand Demand (MW) Type 1 Unit Type 2 Unit Type 1 Output (MWh) Type 2 Output (MWh) LMP ($/MWh)

21 Example 1a: Price Elastic Demand Demand (MW) LMP ($/MWh) LMP Revenue/ Cost (%) LMP Uplift/ Cost (%) ELMP ($/MWh) ELMP Revenue/ Cost (%) ELMP Uplift/ Cost (%) % % % % With LMP, the capacity market remains necessary 21

22 Observations Under the current LMP, the market allocation may not reward efficient entry, i.e., i, s. t. π i 0 z i 0, p 0 π i 0 1 z i 0, p 0 Need make-whole payment when π i 0 z i 0, p 0 = π i 0 1, p 0 < π i 0 0, p 0 0 Induce self-schedule when 0 = π i 0 z i 0, p 0 = π i 0 0, p 0 < π i 0 1, p 0 If the cost functions are homogeneous of degree one, then V I V CH = V LD = V R 22

23 Current Pricing Mechanism Creates Distorted Incentives Units A and D are flexible Units B and C are lumpy or block-loaded Minimum runtime = 1 hour Load = 700 MW Hour 1 Hour 2 LMP = $35/MWh Total cost = 16,500 Unit D: 100MW/$ MW/$35 Unit C: 200 MW/$30 MW, Unit B: 200 MW/$25 MW, Unit A: 200 MW/$10 MW, Load = 700 MW LMP = $35/MWh Total bid-cost = 16,500 Unit D: 100MW/$ MW/$35 Unit C: 200 MW/$30 MW, Unit B: 200 MW/$25 MW, Unit A: 200 MW/$10 MW, 23

24 Optimal Dispatch with Reduced Demand Hour 1 Hour 2 LMP = $35/MWh Total cost = 16,500 LMP = $35/MWh Total bid-cost = 10,500 Load = 700 MW Units A and D are flexible Units B and C are lumpy or block-loaded Minimum runtime = 1 hour Unit D: 100MW/$ MW/$35 Unit C: 200 MW/$30 MW, Unit B: 200 MW/$25 MW, Unit A: 200 MW/$10 MW, Load = 500 MW Unit D: 100MW, $ MW, $35 Unit B: 200 MW, $25 Unit A: 200 MW, $10 Unit C: 200 MW, $30 24

25 Unit C May Bid Inflexibly and Profit Hour 1 Hour 2 LMP = $35/MWh Total cost = 16,500 LMP = $35/MWh Total bid-cost = 11,500 Load = 700 MW Units A and D are flexible Unit D: 100MW/$ MW/$35 Unit C: 200 MW, $30 Load = 500 MW Unit D: 100MW, $ MW, $35 Unit B: 200 MW, $25 Units B and C are lumpy or block-loaded Minimum runtime = 1 hour Unit B: 200 MW, $25 Unit A: 200 MW, $10 Unit C: 200 MW, $30 Unit A: 200 MW, $10 Unit C raises Minimum Runtime to 2 hours and replaces Unit B 25

26 Units B and C Are Trapped Hour 1 Hour 2 Load = 700 MW Units A and D are flexible Units B and C are lumpy or block-loaded Minimum runtime = 1 hour LMP = $35/MWh Total cost = 16,500 Unit D: 100MW/$ MW/$35 Unit C: 200 MW, MW/$30 Unit B: 200 MW, MW/$25 Unit A: 200 MW, MW/$10 Load = 500 MW LMP = $10/MWh Total bid-cost = 12,000 Uplift = $4,000 Unit C: 200 MW, $30 Unit B: 200 MW, $25 Unit A: 100 MW, $ MW, $10 Unit D: 200 MW, $35 Both units B and C raise Minimum Runtime to 2 hours and are dispatched 26

27 LOC Payment Supports Efficient Dispatch Load = 700 MW Hour 1 LMP = $35/MWh Total cost = 16,500 Unit D: 100MW/$ MW/$35 Hour 2 LMP = $35/MWh Total cost = 10,500 Uplift for Unit C = $1,000 Units A and D are flexible Units B and C are lumpy or block-loaded Minimum runtime = 1 hour Unit C: 200 MW, $30 Unit B: 200 MW, $25 Unit A: 200 MW, $10 Load = 500 MW Unit D: 100MW/$ MW/$35 Unit B: 200 MW, $25 Unit A: 200 MW, $10 Unit C: 200 MW/$30 27

28 Load = 700 MW Units A and D are flexible Units B and C are lumpy or block-loaded Minimum runtime = 1 hour Extended LMP Supports Efficient Dispatch with Minimum Uplift Hour 1 LMP = $35/MWh Total cost = 16,500 Unit D: 100MW/$ MW/$35 Unit C: 200 MW, $30 Unit B: 200 MW, $25 Unit A: 200 MW, $10 Load = 500 MW Hour 2 LMP = $30/MWh Total bid-cost = 10,500 Uplift for Unit C = $500 Unit D: 100MW/, $ MW/ $35 Unit B: 200 MW, $25 Unit A: 200 MW, $10 Unit C: 200 MW, $30 28

29 Bayes-Nash incentive compatibility (BNIC) Key Incentive Compatibility Concepts For each trader, truthful bidding is an optimal strategy that maximizes the conditional expectation of the individual s payoff Assume common knowledge of the prior distribution of the types of other participants Dominant strategy incentive compatibility (DSIC) For each trader, truthful bidding is a dominant strategy that maximizes the individual s payoff, for any types of other participants The well-known Vickery-Clark-Grove mechanism is DSIC DSIC implies BNIC 29

30 Incentive Compatibility for Market Mechanism With a finite number of traders, no market mechanism can be incentive compatible (Hurwicz 1972) Nash equilibrium differs from competitive equilibrium With a large number of traders, incentive compatibility is achievable in the limit as the number of traders grows to infinity. (Roberts and Postlewaite 1976) ELMP is dominant strategy incentive compatible in the limit as the number of traders grows to infinity It converges to a Vickery-Clark-Grove (VCG) mechanism Truthful bidding is an optimal strategy for each individual 30

31 Tradable rights play a key role in a market economy Tradable right doctrine Infra-marginal rents are a form of tradable rights or entitlements in competitive markets Infra-marginal rents compensate investments in tangible assets such as generating units as well as in intangible assets (such as research and development) that foster efficiency and innovation Under ELMP, each trader earns the infra-marginal rents (φ p, π i p ) The assurance of the inframarginal rents makes ELMP dominant strategy incentive compatible in the limit 31

32 Energy and Reserve Pricing in Day-ahead and Real-time Markets Day-ahead market is a forward market for energy and reserve Reserves in the real-time market are real options callable by the system operator when needed Synchronized reserve is an option to serve energy within 10 minutes Non-synchronized primary reserve is an option to become 10 minute synchronized reserve Co-optimized energy and reserve pricing works consistently in day-ahead and real-time markets 32

33 Notation Variables p e p r g, g(ω) s(ω) G r D(ω) VOLL PF Explanation the energy price the reserve price the generation output the amount of load shed the generation capacity online the reserve capacity the demand Value of lost load Penalty factor for reserve violation 33

34 Timeline Pre-dispatch Load uncertainty Actual dispatch Reliability contingency risk 34

35 Alternative Objective Functions In theory: V = Max g,g,r,s In practice: E H r ω VOLL d ω c g ω kg e. g., H r = e α r MRR V = Min E c g ω + VOLL s ω + PF MRR r ω + + kg g,g,r,s 35

36 Pricing Principles Scarcity pricing: p e ω = Option pricing: Reliability pricing: c if s ω = 0 VOLL if s ω > 0 p r = E p e ω c = p e c p r = E H r ω VOLL c g ω 36

37 Example 3 Parameter Assumption Load forecast D ω, MW Uniform 120,000 ± 2500 Minimum reserve requirement (MRR), MW Value of lost load (VOLL) = Penalty Factor (PF), $/MWh 1500 Marginal cost (c), $/MWh 50 Commitment cost (k), $/MW 10 Unit size for G, MW 300 a) 2,000 b) 5,000 37

38 VOLL = PF ($/MWh) Under LMP Example 3 Results Under ELMP Energy (MWh) 2, , ,000 Reserve (MW) 3,900 3,900 Energy price ($/MWh) Reserve price ($/MW) Energy (MWh) 5, , ,000 Reserve (MW) 4,200 4,200 Energy price ($/MWh) Reserve price ($/MW)

39 Comparison of Alternative Pricing Methods Criteria Promote efficient allocation Support system operations Current LMP Method Extended LMP with Integer Relaxation High High High Medium High High Incentive compatible Low High High Revenue sufficiency Low High High Minimized uplift payments Computationally feasible Low Medium High High Medium high Low Extended LMP with Convex-hull pricing 39

40 Market evolution? Year Event 1978 Public Utility Regulatory Policy Act 1997 FERC Orders 888, California Electricity Crisis 2005 Energy Policy Act with demand response mandate 2012 FERC issued Order 745 on Demand Response Compensation 2016 Order 745 was settled by the U.S. Supreme Court 2018 PJM received FERC Orders to reform its energy and capacity markets 40

41 Thank You 41

42 Sketch of the Proof for a Simple Case Theorem: ELMP methods, including convex-hull pricing, Lagrangian dual and integer relaxation, are dominant strategy incentive compatible in the limit Consider n identical replications of market with independent traders 1) V L p n Min nφ p + i=1 π L i p p n n 2) nφ p + i=1 π L i p = 0 nd p + ns p = 0 p n p L n n 3) V j p j Min nφ p + i=1,i j π L i p p L 4) nφ p + i=1 π i,i j p = 0 D p = 1 1 S p p n j Result A: p n j n p n = p as n Result B: j N, V L p n L V j n p j n p n = p π j L p as n 42

43 Economic Commitment and Dispatch If the cost function c i x i, z i is homogeneous of degree one in x i, z i, then for any given p R, π i z i, p is linear in z i Assume that for every i N, there is a private constraint set H i z i R l h i z i 0 Any function f(z) that is homogeneous of degree one is linear in z (z 1,, z l ), if and only if f(z) is linear in (z 1,, z l 1 ) Then if π i z i, p is linear in z i H i which is unimodular polyhedron, then v R = v L = v CH. 43

44 Example 2 1 Bus, 2 Units, 3 Periods Parameter Energy cost Start-up cost Value $40/MWh $500/start Unit 1 Unit 2 Parameter Energy cost Start-up cost Value $20/MWh $0/start No-load cost $1000/hour No-load cost $0/hour Max output 100 MW Max output 50 MW Eco Min 40 MW Eco Min 0 MW Ramp rate 100 MW/hour Hour Demand (MW) Ramp rate 50 MW/hour x Price x Price x Price 44

45 Dispatch Solution Hour Unit 1 (MW) Unit 2 (MW) Load (MW) LMP ($/MWh) Unit 1 and Unit 2 are both marginal units The LMP in Hour 1 is below the marginal cost of Unit 1 LMPs do not reflect commitment costs 45

46 Pricing Solution Hour Unit 1 (MW) Unit 2 (MW) Load (MW) LMP ($/MWh) In this case, ELMP allocates the no-load cost within each hour in ways similar to AIC (average incremental cost) pricing In this case, ELMP allocates the start-up cost across all hours in ways similar to peak-load pricing The unit dispatch in the pricing solution is discarded 46

47 Settlement results Dispatch solution Pricing solution Hour Unit 1 (MW) Unit 2 (MW) Load (MW) LMP ($/MWh) ELMP Uplift $1,200 $360 $61 Total: $1,621 The uplift payment for Unit 1 covers its losses during the commitment period The uplift payments for Unit 2 and Load represent the lost opportunity costs 47