Risk-limiting dispatch for the smart grid: some research problems

Transcription

1 Risk-limiting dispatch for the smart grid: some research problems Janusz Bialek, University of Edinburgh Pravin Varaiya, University of California, Berkeley Felix Wu, University of Hong Kong 1

2 Outline 1. Current system operations: worst-case dispatch 2. Future electric power system: smart grid 3. Renewables and demand response increase uncertainty 4. Smart grid increases information and control 5. Risk-limiting dispatch 6. Preserving the interface: bundling unreliable supply/demand 7. Preserving the interface: unreliable energy transactions 2

3 Current electric power system operations day ahead market balancing market scheduling recourse operating time emergency ISO schedules generators/loads s.t. constraints are met: Power balance Operating limits (N-1) contingencies Objective Min cost s.t. constraints Uncertainty Peak load demand Forced outage (fault) Recourse Balance Emergency Load shedding 3

4 Operating risk and worst-case dispatch Operating risk Not meeting constraints Constraints Power balance g( x( t), u ) = 0 Operating limits h( x( t), u) 0 u is such that for each contingency x(t) is stable Stochastic uncertainty Outage, peak demand (N-1) contingencies Dispatch reliable power S i ( 1+ r) D r ~ 0.05 = reserve margin D = peak demand Reserve margin increases capacity cost and carbon emissions 4

5 Future electric power system Renewables Wind Solar Storage microgrid Smart grid infrastructure Smart meters, sensors Intelligent appliances Communication Demand response Wind power Solar power Micro-Grid Load Substation Low emission central plant storage 5

6 Renewable generation increases uncertainty 1500 Generation availability 1200 thermal S(p) kw n = 20 n = 1 power supply exceeds S(p) for p % hours/year Rated capacity = 1500KW Average capacity = 700KW Reliable capacity = 150KW (reliable capacity ~ thermal) Hours in a year (%) p 6

7 Wind power and worst-case dispatch day ahead market balancing market scheduling operating time emergency kw Hours in a year (%) Rated capacity = M Reliable capacity = M If M is scheduled need reserve capacity of 0.8M If 0.2M is scheduled it should displace 0.8M of thermal power, which becomes reserve 7

8 California example1/2 Source: Y. Makarov and D. Hawkins, CALISO 8

9 California example2/2 Source: Y. Makarov and D. Hawkins, CALISO 9

10 Demand response increases uncertainty Demand response Demand reduced by D(h) for at least h hours/year D(80) = - 300W Worse-case dispatch cannot use demand response to reduce reserves 10

11 Case against worst-case dispatch Wor st-case dispatch designed for system with reliable power transactions in which Generators have 0-1 failure characteristic Short-term peak demand is predictable Information is scarce and decisions cannot be refined Wind power generation and demand response are highly uncertain Consequently, renewable generation and demand response are unfairly treated by worst-case dispatch 11

12 Greater information and control More accurate information Smart meters, sensors More refined control Intelligent appliances Demand response Tighter feedback Communication Enable risk-limiting (vs. worst-case) dispatch 12

13 Risk-limiting dispatch: scheduling scheduling operating time t T t σ Scheduling Decision : Generation/demand u σ Max objective such that the risk of not meeting operating constraints is less than (1-p*) based on information at scheduling time yt T σ 13

14 Risk-limiting dispatch: recourse scheduling recourse operating time t T t σ t T ρ Recourse Decision : Generation, demand response u ρ Max objective such that the risk of not meeting operating constraints is less than (1-p*) based on information at recourse time yt T ρ 14

15 Risk-limiting dispatch: emergency scheduling recourse emergencyoperating time t T σ t T ρ t T ε t Emergency Decision : Generation, interruptible load u ε operating constraints must be satisfied based on information at emergency time yt T ε 15

16 Risk-limiting dispatch: summary scheduling recourse emergencyoperating time t T σ t T ρ t Tε t Optimization for system operation: Impossible stochastic sequential dispatch problem * See: Bouffard and Galliana, Trans Power Syst, 2008; Morales et al. Trans Power Syst,

17 Wind power offer in reliable-power-only market c < 1 < c2 c3 t T σ W (t) t T ρ t T P( W ( t) Y t T ) σ Buy s 1 Buy s 2 Buy s 3 Sell reliable D ε t Deliver D = {W(t),s 1, s 2, s 3 } Assume P ( W ( t) > x Y ) = 1 Max E{ c1d c1s1 c2s2 c3s3 D, s1, s2, s3} D c c 1 2 M s = D s W ( )] s 0 2 [ 1 t s1 = D + c c = t T M σ x M c c M W(t) 0 unused wind power D buy s M used wind power buy recourse s 2 0 reliability 1 17

18 unused wind power reliable wind power c c 1 2 M M W(t) 0 Comparisons unused wind power used wind power buy recourse s 2 0 reliability 1 c 1 < c 2 0.8M reserves used wind power Worst-case dispatch Revenue = c 1 0.2M Reliable-equivalent dispatch Net revenue = 1/2 c 12 / c 2 M PIRP dispatch Revenue = 1/2 c 1 M Subsidy = 1/2 c 1 0.8M 19

19 Interruptible power transactions Interruptible power contracts parameterized by (c, ρ) with per MW price c delivers power with probability ρ, and no power with probability (1- ρ) Suppose market creates contracts (c 1, ρ 1 ),, (c k, ρ k ) with 1= ρ 1 > > ρ k and c 1 > > c k Note ρ 1 is reliable power How will wind generator react? Take k = 4 20

20 t Wind generator example T σ t T ρ t Tε sell s deliver i w 1 w 2 w 3 w 4 w 1 w i At scheduling time, supplier sells w i units of ρ i reliability power t Assume availability state s i is known at recourse time At time t deliver contracts ρ 1... ρ i Revenue of c i wi with no subsidy w 4 w 3 w 2 w 1 s 4 s 3 s 2 generation availability s 1 ρ 4 ρ 3 ρ 2 ρ 1 21

21 How will customers behave Recall that less reliable power is cheaper: c 1 > > c k Customers may select bundle to Max i [ ρ U ( D ) (1 ρ ) L( D ) c D ] U(D) is utility of consuming D L(D) is loss of not consuming D i i i i i i 22

22 Incremental deployment Permit interruptible power in distribution system and reliable power in bulk power system 23

23 Conclusion Federal programs are deploying smart grid elements on large scale Current practice of worst-case dispatch requires subsidies for renewable sources and demand response More accurate and timely information and more refined control suggest shift to risk-limiting dispatch, which does not require subsidies Risk-limiting dispatch can be introduced incrementally in distribution system 24