Strategic Behavior in Energy and Reserve Co-optimizing Markets with Distributed Energy Resource (DER) Participation

Prof. Michael Caramanis (Boston University), Presentation on Day 2 (Nov.4) of Workshop on Architecture and Economics of the Future Grid

Transcript

1. Workshop on Architecture and Economics of the Future Grid An Investigation of Strategic Behavior in Energy and Reserve co optimizing Markets with DER Participation Texas A&M, November 4, 2016 Michael Caramanis mcaraman@bu.edu
2. Key Issues with T&D Marginal Cost Based Markets • DERs: – can trade energy and reserves, – have complex preferences (inter-temporally coupled), dynamics and capabilities • Losses, Voltage constraints and more are important in Distribution Networks and require AC load flow modeling • Computational and Communication Constraint complexity => Only Distributed market clearing algorithms feasible/practical 2
3. Imbalances & Prices Sub-problem Solutions ADMM, a PMP Algorithm that May Achieve Network Asset and Participant Objective maximization Consensus. Asynchronous Distributed Algorithm Issues: A. Technical: Converge? PMP based convergence Certificate? Vulnerability to Malicious Communication Interception? B. Social Welfare Issues: Can DER Strategic Behavior and Collusion Hurt Economic Efficiency (i.e., Social Welfare)?
4. Information Access to Distributed Decision Makers • Fully Distributed Algorithm: Strictly local information to participants (DERs, Loads, Generators), Busses, Lines/resources • Is there Welfare loss relative to a “know it all” Centralized Decision maker? (intractable, uni,plementable, but let us ask the question in principle) • Is there Welfare loss to Strategic Behavior of DER decisions with Network information? Under Collusion with Other DERs through Load Aggregators? 4
5. Centralized Algorithm works as Follows (Roughly) • ISO knows true preferences, Dynamics, and Capabilities of all DERs • But ISO either knows all DSO information OR Interfaces with DSO on losses Info. Distribution Network info, CN , => Centralized Decision Maker Schedules Generators and DERs while knowing Detailed Loss and marginal loss functional terms No Distribution Network info, C, => Centralized Decision Maker Schedules Generator and DERs with DSO providing Loss and Marginal Loss Estimates. (Some iterations are needed here to convergence) 5
6. Distributed Algorithm works as Follows (Roughly): • Given Feeder DLMP Estimates, each DER Schedules it self iteratively based on Information Access: DN,A Full Network information, in collution with other DERs DN Full Network info but no collusion D No network Information, DLMP estimate provided • Given DER Tentative Schedule ISO Scheduled generators, calculates Expost LMPs and provides new estimate of LMP. Updated DLMP estimated by DERs based on Network Info or DSO conversion of LMP to DLMP estimates • With appropriate DLMP Estimate update process convergence is possible. 6
7. -We Investigate Analytically and Numerically: Social Welfare, Strategic Behavior, Collusion and Information Access 7
8. Necessary to Simplify T&D Market Model while Retaining Salient Characteristics 8
9. 9 T&D Network
10. 10 Distribution Networks may be meshed but often are operated in Radial topology.(Courtesy Jiankang Wang, PhD thesis, MIT 2013)
11. 11 Meshed Transmission, Sub-transmission, and Distribution Schematic Expanded Distribution Feeder Schematic. Although depicted as a radial topology, it may be partly or even primarily meshed (ConEd). N n n     n n n n , ( ), ( ), ( ) ( ) P R i n n n n Q t t OC t   n n   ( ) ( ( ) ( ) ) , , i i i Q n i P n R n i n t t t t n v    ( ), ( ), ( ), ( ) k k k k P Q R n n n n t t t t    
12. 12 Incurred Cost Distribution, Congestion, Reserves, Voltage Control, Losses, Transformers, Deliverability Generation Costs 50% Transmission Costs 5% Distribution Network Costs ~ 10xTransm. Costs , , ℜ ഫ ≤ ≤ ത ℜ Locational Marginal Price of Real Power at bus n, during hour t Locational Zonal Price of Reserves at bus n one Z, during hour t t n t n LMPP LMP z    
13. 13 , , t E t n n t t n n LMPP ZMP       2 2 { [( ) (Q ) ]} i i i i t t t t n n n n n i P P R P     2 2 {Q [( ) (Q ) ]} i i i i t t t t n n n n n i Q X P     ,Q , i i i i i i t t t n n n t n n n P V V V    { 2 } i i i t t t n n n n i R P      2 2 TransfLossOfLife [( ) ( ) ] t t t n n P Q    Interface between Transmission & Distribution
14. Simple -- though not too simple -- model -Model Energy and Secondary Reserves but disregard Reactive Power. -Model Quadratic Losses in Distribution Network. -Model Meshed T Network with Energy and Reserve Capable Conventional Generators -Aggregate Distribution Networks to Bundles of Single Radial Lines. -Model DERs connected to n(i) for all n and i 14
15. Simplified T&D Network Relations C D DN DN,A CN -Feeder Losses -Marginal Losses -Translation of DER offered reserves to Root node LMP and DLMP relationship and 15 ( ) ( ) ( ) ( ) ( ) 2 ( ) 2 2 2 ( ) ( ) n i n i n i n i j n i n i t t t t j L d p       ( ) ( ) ( ) n i n i n i t t m   ( ) ( ) (1 ) n n n i n i t t i s t r m r     ( ) ( ) (1 ) n i n i n t t t m     ( ) ( ) (1 ) n i n i t t t m    
16. Information Access to Centralized, C, and Distributed, D, Algorithm Market Clearing Process Distributed Cases (Iterative): -D: provided by DSO -DN : and But each DER j, knows its own schedule only. -DN,A : as above except the Load Aggregator knows all DER loads. Centralized Cases: -C: Centralized Decision maker receives Loss and Marginal Loss information. -CN : Centralized Decision maker knows functional form of losses and marginal losses. 16 ( ), ( ), ˆ ˆ and n i k n i k t t t    ( ) ( ), ( ) ( ) ( ), 1 , ˆ ˆ (1 ( )) n i n i k n i n i j n i k n k t t t t j s d p          ( ) ( ), ( ) ( ) ( ), 1 ˆ ˆ (1 ( )) . n i n i k n i n i j n i k k t t t t j s d p         
17. Centralized Decision Making Subject to With either obtained from the DSO, Case C, or derived from the detailed information about the Distribution Network, CN . 17 , 1 , 1 , 1 , 1 , , , , , ( ) min ( ) g n k g n k t t g g n k t g g n k p r t n g n t g t c p r r              , 1 ( ), 1 ( ), 1 , ( ) ( ) 0, g n k n i k n i k t t t t g n n i p L t            , 1 ( ), ( ), , ( ), (1 ) , g n k n i k j n i k t t t t t g n n i j t r m r R          g n g n g n t t t g n g n g n t t t p r g p r g         ( ), 1 ( ), 1 and n i k n i k t t L m  
18. Recall Simplified T&D Network Relations C D DN DN,A CN -Feeder Losses -Marginal Losses -Translation of DER offered reserves to Root node LMP and DLMP relationship and 18 ( ) ( ) ( ) ( ) ( ) 2 ( ) 2 2 2 ( ) ( ) n i n i n i n i j n i n i t t t t j L d p       ( ) ( ) ( ) n i n i n i t t m   ( ) ( ) (1 ) n n n i n i t t i s t r m r     ( ) ( ) (1 ) n i n i n t t t m     ( ) ( ) (1 ) n i n i t t t m    
19. Distributed Decision Making Subject to where Depend on type of information access considered. The objective function is minimized either over individual DER decisions, D and DN, with no knowledge of other DER schedules, or over all DER decisions, DA and DN,A cases. 19 ( ), 1 ( ), 1 ( ), ( ), 1 ( ), ( ), 1 1 , , ˆ ˆ min ( ) ( ) ( ) d j n i k j n i k t t n i k j n i k n i k j n i k j k t t t t t p r t t p r U x            ( ) ( ) ( ) ( ) ( ) 1 [1 ] j n i j n i j n i j n i j n i t t t t t j p x x p C        ( ) ( ) ( ) j n i j n i j j n i t t t p r C     ( ) ( ) ( ) j n i j n i j n i t t t r p    ( ), ˆn i k t  ( ), ˆn i k t 
20. Recall Simplified T&D Network Relations C D DN DN,A CN -Feeder Losses -Marginal Losses -Translation of DER offered reserves to Root node LMP and DLMP relationship and 20 ( ) ( ) ( ) ( ) ( ) 2 ( ) 2 2 2 ( ) ( ) n i n i n i n i j n i n i t t t t j L d p       ( ) ( ) ( ) n i n i n i t t m   ( ) ( ) (1 ) n n n i n i t t i s t r m r     ( ) ( ) (1 ) n i n i n t t t m     ( ) ( ) (1 ) n i n i t t t m    
21. Comparison of Optimality conditions in each of the 6 information access cases => mismatch. Extra Terms in Optimality condition Comparison 21 Market Clearing Process C CN D - DA - DN DN,A N N N N D D D D j j t t t t p r     ' N N N N C C D D j j t t t t j j r p       , , , , N A N A N A N A D D D D j j t t t t j j r p      , , , , N A N A N A N A D D D D j j t t t t j j r p      N N C C j t t j r    N N C C j t t j r   
22. EV Group data Used in Numerical Experiments 22 EV Group Arrival time Departure time Charging demand (MWh) Charging capacity (MW) 1 2 9 18 4.5 2 8 20 18 4.5 3 9 21 18 4.5 4 6 20 18 4.5 5 7 22 18 4.5 6 8 23 18 4.5
23. Numerical Experiment input # EV groups Total # of EVs Total charging demand Arrival time range Departure time range 6 1800 108 MW 2-9 9-23 23 24 hour inelastic demand (MWh) 24 hour system reserve requireme nt (MWh) Range of hourly demands (MWh) Hourly generation capacity (MW) 8760 560 215-522 700
24. Conventional Generation Energy Supply Curve: $/MWh versus MW (Similar for reserves) 24
25. Extra Term values for a selected hour in a numerical Experiment 25 Centralized C CN Distrib. D 0 0.280 DA 0 0.280 DN -0.012 -0.291 DN,A -0.060 -0.336
26. Numerical results across six clearing processes 26 EV cost ($/day) Social cost ($/day) Centralized Generators producer surplus ($/day) Inelastic demand charges ($/day) Case Effective cost Charging cost Reserve revenue Generation cost Reserve provision cost Total From generation From providing reserves Total D 1604.269 9199.329 7595.06 449684.181 24968.398 474652.579 266466.801 1512.130 267978.931 760475.727 DA 1604.269 9199.329 7595.06 449684.181 24968.398 474652.579 266466.801 1512.130 267978.931 760475.727 DN 1604.216 9199.405 7595.189 449684.143 24968.443 474652.587 266466.784 1512.009 267978.793 760475.669 DNA 1604.065 9199.638 7595.574 449684.062 24968.550 474652.612 266466.608 1511.672 267978.280 760475.377 C 1604.269 9199.329 7595.06 449684.181 24968.398 474652.579 266466.801 1512.130 267978.931 760475.727 CN 1605.051 9198.141 7593.091 449684.632 24967.886 474652.518 266465.665 1513.755 267979.420 760475.079
27. Energy and reserve DLMPs in hour 12 27 D,DA ,C DN DN,A CN 75.945 87.998 75.944 87.997 75.942 87.992 75.950 88.007 62.300 72.187 62.302 72.189 62.307 72.194 62.289 72.177 12 12 t t      12 12 t t     
28. Impact of System Losses, Relative DER Loads and EV Utility function across DN,A and CN Market Clearing Processes RECALL that: D, DA and C are identical! 28 β Flex. demand magn.* SoC utility % Avg. losses % Total reserve provision % Max reserve provision % Diff. in EV net cost 4 1.2% Fixed 8 22.6 62.3 0.06 7.2 1.2% Fixed 14 24.9 68.4 0.11 7.2 1.4% Fixed 14 27.6 76.0 0.13 7.8 1.4% Fixed 15 28.2 77.3 0.15 7.8 1.5% Fixed 15 31.0 85.1 0.22 4 1.2% Decr. 8 21.1 62.3 0.14 7.2 1.2% Decr. 14 20.0 67.8 0.33 7.2 1.4% Decr. 14 21.9 74.8 0.47 7.8 1.4% Decr. 15 21.5 74.2 0.59 7.8 1.5% Decr. 15 23.5 81.1 0.70
29. Aside on Reactive Power Market Size 29 Indicative Estimates of Average Price of Reactive Power against Power Electronics Capacity Penetration as a % of Maximum Hourly Reactive Power Consumed. PF .8 .88 .92 .95 Q/P .75 .54 .426 .33 2 1 / PF Q P PF    0 5 10 15 20 25 30 35 40 0% 30% 60% 90% PF=0.8 PF=0.88 PF=0.92 PF=0.95 $/MVarh Power Electronics Capacity as % Of Max Q consumption
30. Conclusion • Information access and value to DER of strategic Behavior. • Impact pf aggregators on competitive markets? • BUT What does tractability in terms of both computation and communication constraint tell us? Detailed Distribution Network Information Acces not Practical!!=> OK!! Aggregaors or Not!! • However, Reactive Power capacity withholding Certain under Load Aggregator Facilitated Collusion! 30
31. Appendix • Some relevant Material follows that can be used to elaborate during D&A 31
32. Sufficient Statistic/Minimum Useful Information Exchange as Enabler of Distributed Decision and Control Challenge: # Estimate Sufficient Statistics that Represent Uncertainty and “Rest of the System” Details/Complexities #Do so in Context of Cascaded time Scales -Planning -Re-planning/slow control -Control/fast re-planning 32
33. The Planning to Operation Architecture Encountered in Today’s Power Markets is Surprising Useful (and Adaptable?) • Generation Capacity and Transmission Congestion (FTR) Markets – Years to Months • Forward Energy Commodity Markets – Months • Energy and Reserve Co-Clearing Markets: – Day Ahead: Multiple Hours – Hour Ahead/Adjustment Market – Hour • Reserve Management: – Operating: 5 min., – Regulation Service (AGC Centralized): 2-4 sec – Frequency Control (Decentralized): Real-Time 33
34. 34 RESERVES Primary (Decentr.) Secondary 1 faster (AGC- Centr.) Secondary 2 Slower (AGC- Centr.) Tertiary (Operating) Signal Frequency ωi (t) real-Time ydyn(t) 2-4 sec yslow(t) 2-4 sec ELD* 300 sec Gain 100% per 30 sec 100% per 150 sec 100% per 300 sec 100% per 600 sec -Participant Provision -System Requirements 1 1 s i i ys     2, 2, i d dy yn n sys i     2,slo 2, w slow i sys i     3 3 s i i ys     ISO Determines quantities in Red. Market Clears Quantities in Blue. is cleared in DA and HA Markets based on individual bids and System Req. Energy/Reserve Clearing Prices, used to Debit/Credit Participants *System also determines Line Flow Constraints and Contingencies, which Economic Load Dispatch (ELD) must satisfy. r i  , E   
35. 35 Bi-directional Primary and Secondary Reserve Dynamics Automatic Primary/Freq-resp. Reserves Provided as a function of Δω. Provision Scheduled at constant ramp rate (100% per 30 sec) Secondary/Regulation Reserve Provision Schedule as a function of ISO requested % target y(t) (reset in 2 or 4 sec intervals) Δω, mHz Automatic provision % of Total Procured | | | | -200 -20 +20 +200 +100% -100% t, sec ISO requested % of total procured | 300 +100% -100%
36. Dynamic Market Extension to Distribution Network Connected DERs/Prosumers • Demand Response Distributed and Commoditized • May Capture DER/Renewable Generation integration synergies • Can Improve System and DER Capacity Utilization Question: Are there Significant Social Welfare Related Drawbacks to a Distributed Market Clearing Process? 36
37. Current Multi-period Market Bidding Rules – uniform price quantity bids-- Motivate Flexible Loads to guess forward prices and Self Schedule. • Flexible Loads, typically, do not have Time Additive Utility of Period-Specific Consumption (what does this mean?) • Flexible Loads such as EVs, HVAC, Duty Cycle Appliances, and Storage have Inter- temporally Coupled Utility 37
38. Examples of Inter-temporally Coupled Flexible Loads: State Dynamics Determine Preferences 38 • Distributed PHEV Charging: State of Charge • Centralized Pumped Storage Hydro Units (Reservoir Level) 1 deptime deptime ( ) 0 if 0 i i i i i i t t t t n n n n n n x x D x U x x        1 ( 0 24 ) , , t t p n n n n n p t g t r t n n n t t t t n n n n n n n x x D G x x G G D               
39. Examples of Inter-temporally Coupled Flexible Loads: State Dynamics Determine Preferences, Continued 39 • HVAC and Duty Cycle Appliances 1 1 , ( ) , i i i i i i i i i i t t h t h t t ambient n n n n n t t t t t n n n n n D D                   
40. Observation: Flexible Loads are Better off By Self Scheduling Since Current Bidding Rules Do Not allow them to Represent their Actual Utility! How? 40 * ( ) * ( ) Bid Energy at a very high price Offer Regulation Service Capacity at low price t n i t n i D R However: ISO Clearing Prices may Differ from Those Assumed in Self Scheduling. Nash Equilibrium Exists to this DGame Under Mild Conditions (Cullaway et al)
41. 41 In Multi-period, DA Market, Current Uniform Price Quantity Bidding Rules lead PHEV Charging Loads, to Strategic Behavior PHEV (aggregator?) decide Hourly demand, Dt, and Regulation Service, by estimating/forecasting clearing prices and and solving the following problem. Strategic behavior leads to Divergence of Forecasted and Market cleared Prices! Allowing revised bids removes incentives for Strategic Behavior.     , , , , 1 { ] ( )} . . EV State of Charge dynamics up/dn nature of Regulation Service D D Charging Capacity min i i i i i i t t n n i i i i i i i i i i i t E t t t t t t t n n n n n n n n t D t t t t t n n n n t t n n t t n n n m D m U x s t x x D x D                    t  t    E 
42. PHEV Participating in Multi- period Day Ahead Market • Numerical Results indicate Market Clears optimally for both Society and the PHEVs when the Inter-temporal (albeit linear) PHEV preferences are provided to the market maker by enhanced bidding rules • Provision of Reserves results in smoother Effective Load profiles. 42
43. 43 Revised Bidding Rules that Include Inter- temporal state dynamics and Associated Utility, Allow ISO/“DNO” to Clear Market. “Under Competitive Conditions” Market clears with the same schedule as the Nash Equilibrium of the Hierarchical Game. This removes Incentive for Strategic Behavior Consensus/Nash Equilibrium Achievable Through Centralized Market clearing. But is this computationally feasible? Yes under approximate representation of Load Preferences
44. Revised Bidding Rule Benefits • Flexible Loads at Distribution Level may participate in Expanded ISO/DNO Centrally Cleared Power Market bringing significant benefits, particularly w.r.t. Sustainable Renewable Generation Integration to the Grid • Expanded ISO/DNO-Operated Power Market Clearing is Practical from Information and Computational Tractability Point of view. • Inclusion of Other Important Distribution Network Costs, such as Reactive Power Compensation and Voltage Control is also Practical. 44
45. Real and Reactive Power Prices During each of 24 Hours Enable • PV location (Max. income from real and reactive power sales) • Management of Power Electronics • Cost Effectiveness of Equipping Power Electronics with controllers and capability to Provide Reactive Power Compensation. This is an Investment Decision… How Does it Compare with Imposition of Standards? • Optimal Real Reactive Power Tradeoffs in PV, Variable Speed Drives in modern HVAC, … • Provision of Reserves at Distribution Net. 45
46. Decentralized Decisions for the Provision and Consumption of Distributed Real/Reactive Energy and Reserves 46
47. Complex Sub-problems • Flexible Loads In Multi-period Markets: EVs with Deadline, Non-Ideal Battery, V2G, Reserve offering • HVAC with Variable Speed Drives & PV • Data Centers providing Reserves • ???? Distribution Network Topology Control??? i.e. Connectivity Switching??? 47
48. Ex. of Var. Speed HVAC - PV Collaboration: Action in the small by Distr. Flex. Loads 48
49. Data Centers Example (3% of US Consumption and growing!); • Cyber Layer is a Discr. Ev. Dyn System: Given Pr(job arrival), Pr(y(t)), QoS contractual req.develop optimal policy for tracking Regulation Service signal y(t). – State: Queues, Servers up/asleep/in Transition, y(t), Cumulative QoS achieved by time t. – Allowable control: Dynamic Voltage Frequency Scaling (DVFS) –continuous-, Server state transition –discrete-, assignment of jobs to servers/VMs, Inter-Center Routing 49
50. 50 Recall that under limited bidding rules in multi-period markets (i.e. under today’s uniform/myopic price quantity bids) individual flexible loads “gues” clearing prices and behave according to: Strategic behavior leads to Divergence of Forecasted and Market cleared Prices! Allowing revised bids removes incentives for Strategic Behavior.     , , , , 1 { ] ( )} . . EV State of Charge dynamics up/dn nature of Regulation Service D D Charging Capacity min i i i i i i t t n n i i i i i i i i i i i t E t t t t t t t n n n n n n n n t D t t t t t n n n n t t n n t t n n n m D m U x s t x x D x D                    t 
51. Under New Bidding Rules allowing Flex Load to Express True Utility, ISO/DNO can solve the central social welfare optimization problem WITH each flexible load’s inter-temporal state dynamics, i.e., 51 ( ) ( ) ( ) , ( ) , ( ) ( ) ( ) ( ) ( ) , , , , , , , , , , ( ) ( ) ( ) ( ) ( ) ( ) ( ) , ( ) ( ) ( ) ( ), ( ) ( ) 2 , ( ) [ ( )] . . ( ) 2 max c t t R t t t W t n i n n j n i j n i n j j c c F F R c t n i n t t n i n i d g g D R g t i t t t t t t t n n n n n i n i t t c t n j n i n i n n i j n i n i t j n i j d u d c g r g U x s t g D d D                            ( ) , R, ( ) ( ) , ( ) ( ) ( ), 0, , & other System and Local ( ) constr. &dyns i E x t n R t t t c t x t n n i j n i n n n i j t g m R R t n i              
52. 52 Numerical results ERCOT Data for Inter-temporally Coupled Utility Model 0 50 100 150 200 250 300 350 400 450 500 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm 12am 1am 2am 3am 4am 5am 6am 7am 8am 9am 10am 11am Time Period Total KWh Current Load Updated Load w/RS Updated Load w/o RS Updated Load w/Dump
53. 53 Numerical results CAISO Data for Inter- temporally Coupled Utility Model 0 50 100 150 200 250 300 350 400 450 500 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm 12am 1am 2am 3am 4am 5am 6am 7am 8am 9am 10am 11am Time Period Total KWh Current Load Updated Load w/RS Updated Load w/o RS Updated Load w/Dump
54. Shared Objectives on Smart Grid, but Potentially Alternative Platforms • Objective: Environmental + Economic + Energy Reliability Sustainability • Platforms Described In terms of two Poles: – Centralized Utility Control and Standards – Dynamic Price Discovery in Markets • Advantages/Challenges of Market Platform: – Flexible and adaptable to unforeseeable change – Implementation: Tech. Feasibility, Consumer Acceptance, Regulatory Issues (Competitive Markets?) 54
55. • Whole Sale Markets Evolve Fast to Use of Cascaded Markets (day ahead, hour-ahead, 5 min and 4 sec for Reserve Management) and Joint Clearance/Optimization of Reserves • Extended Experimentation with Advanced Communication and Metering at Distribution/Retail (LV) Network (AM<10M in US today) • Utilities support Direct Utility Control RATHER THAN MARKETS to guide/coordinate Demand Response. Major Washington Peninsula Experiment by PNNL-IBM-Westinghouse-Utilities-Universities • FERC Pushes with various levels of enthusiasm Parity of Loads and Generation in Existing Wholesale (HV) Markets (including Reactive Power Transactions, FERC order 890). Distribution Markets (?) under State NOT FERC Jurisdiction! • On April 25, 2015, New York PSC leapfrogs the “as of yet most progressive California PSC”. NYPSC announces intention to enable Market-based deployment of Distributed Resources and Load management. Substantive Regulatory Reform Policy Decisions have been announced for December 2014 and March 2015. 55 Evolution of Cyber Smart Grid Use in the US
56. NY PSC REV Intention Objective: Optimization of Power System Societal Benefit How to Achieve: Collaborative/ Coordinated, Distributed Decision and Control in the Presence of Flexible Loads and Smart Grid CPSs with Fast Cyber Layer Growth. Platform: Independent Distr. Net Operator vs Vert. Integrated Distr Utility56