IBM Research: Energy and Utility Themes

Transcription

1 IBM Research: Energy and Utility Themes Vehicle to Grid Integration Dr. Carl Binding, Dr. Olle Sundstrom, Dieter Gantenbein, Bernhard Jansen April

2 Overview IBM Research: who we are Motivation: renewable energy in transport and distribution grid The back-ground: EDISON Virtual Power Plant concept Smart charging as optimization problem Tackling grid constraints Results Outlook

3 IBM Research ~3,000 employees Almaden Watson Zurich Haifa China Tokyo Austin I India Brazil IBM Research Lab Melbourne

4 An Evolving Environment 1970's 1980's 1990's 2000's 2010's Corporate funded research agenda Technology transfer Collaborating across IBM Shared agenda Effectiveness Work on client problems Industryfocused research Research in the marketplace Collaborative Partnerships Emerging Services markets Science Globalization Research and Services Centrally funded Joint programs Inter-disciplinary collaboration in the market and across the globe

5 Reimagining how science and technology can have impact Managing human impact on rivers by streaming information Reducing traffic jams by creating them Helping premature infants by sensing complications before they happen Reimagining the energy grid by synchronizing supply Reducing CO2 while boosting business efficiency Mapping beneath the seafloor to help reduce the risk of dry holes Fighting infectious disease by spreading data Improving communication by talking to the Web Creating drinking water by filtering oceans

6 Motivation Integration of natural power-sources into electricity grid Wind-power, photovoltaics Intermittent; require balancing power respectively load-balancing Ideally: we would like to store electrical power! Gasoline powered automobiles are a problem CO 2 emissions Depletion of oil resources Can Electric Vehicles (EV) or Plug-In Hybrids (PHEV) help? No or less pollution and oil consumption Support electricity grid as load-balancing and possibly storage for electrical power But: What are the implications of an EV fleet on existing electricity grids? Can the EV batteries be charged in practical amounts of time? Can we store and release electrical energy using fleets of EVs? GW wind 29,4% : New capacity installed: 200GW natural gas 50,0% coal 5,2% pv 4,4% others 0,8% fuel oil 3,5% nuclear 2,9% large hydro 1,9% biomass 1,7% small hydro 0,1% geothermal 0,0% EU 0,44 0,63 0,84 1,21 1,68 2,50 3,48 4,75 6,45 9,68 12,9 17,3 23,2 28,6 34,4 40,5 48,1 56,5 64,94 WORLD 1,74 1,98 2,32 2,80 3,53 4,82 6,10 7,64 10,15 13,59 17,4 23,9 31,1 39,3 47,6 59,1 74,1 93,8 120,8

7

8 EDISON project (Denmark)

9 Bornholm 400V Grid Details 60KV/10KV sub-station customers 18 60KV/10KV substations 73 km/58 km 60KV overhead/cables 247km/634km 10KV overhead/cables 91 60KV/10KV feeders, 6 per substation KV/400V transformers 60KV transmission line 10KV transmission line 400V distribution grid

10 Virtual Power Plant (VPP) Combine different types of renewable and non-renewable generators and storage devices to be able to appear on the market as one powerplant with a defined hourly output or input. (ecogrid.dk) V2G poses particular challenges: Randomness of consumption/generation Time and location Size of generating/storage capacities KWh Mobility/Roaming Inter-utilities agreements are needed

11 VPP Data Flows VPP TSO/DSO RET GenCo Trip info & charging statistics Load curve boundaries Preferred load curve Production schedule(s), Prices, etc Weather forecast Optimization Preferred load curve Grid (congestion) constraints Preferred load curve Grid (congestion) constraints EV charging schedule Load curve

12 Linear optimization for smart charging (simplified) P b = ηp int (battery efficiency) minimize t s c T p b (overall cost over p b ) subject to A s t s p b >= b s A g p b <= b g A b t s p b <= b b b l <= p b <= b u (trip energy constraints) (generation constraints) (battery constraints) (battery power constraints) m*n elements in decision variables p b for m vehicles, n timeslots.

13 Various target objectives Dumb charging Price optimized Absorb excessive wind power

14 Handling Grid Constraints We consider the grid as a graph composed by transmission lines and transformers Edges: transmission line or transformer max capacity, actual flow (of power) Vertices: grid nodes (PQ, PV) Power flows from generation to consumption (load) Can the scheduled power flow towards loads? Without overloading transmission and transformers Without voltage instability Two approaches: Net-flow/min-cut max-flow: power only Load-flow: power and voltage Limited to tree-shaped distribution grids Power Generator Power Consumer cap, flow Source Power Generator Electrical Grid Power Consumer Target cap, flow Power Generator cap, flow Power Consumer

15 Power Generator Power Generator Demand: 257 KW, Cap: 217 KW Electrical Grid Island 1 Island 2 Demand: 324 KW, Cap: 234 KW Power Generator Power Generator Demand: 217 KW, Cap: 217 KW Electrical Grid Demand: 234 KW, Cap: 234 KW Island 1 Island 2

16 Results from adding grid constraints 6 Power limitations only 3 charging modes Eager/dumb Price-based; not constrained Price-based; constrained Price is based on Pr base + α*(1 P wind,max / P wind, current ) price [cents/kwh] time [h] price-based charging eager charging power [MW] base load grid-aware price-based charging time [h]

17 Limiting grid overload Overload is cut back Squeezed to below 100% utilization Figure shows effect in time-slots with highest grid utilization 4 slots of 15 min Grid constraints are respected! However, price based charging is very bursty. share of grid [-] ~grid-aware price-based charging ~eager charging ~price-based charging loading [-]

18 Results Bornholm (a Danish island) electricity grid simulated. See EDISON project Traffic mix (commuters, taxis, family cars) integrated into simulation. We can evaluate the state of the grid and EVs. Thermal load on transmission lines Wind energy Power load on transformers Optimization of interaction EVs and electrical grid has been implemented We use real-world wind-data & prices Electrical grid partially synthetic Traffic pattern still sketchy Quality of prediction is important We are experimenting with different minimization functions Cool animation available : Impact of managed EVs Typical electrical power consumption & generation curve

19 Outlook Handling of voltage stability Power and length of lines, tree-shaped distribution grid assumed Data availability is an issue! Additional constraints on LP optimization to guarantee max. voltage drop: the nosecurve Planning and control flexibility Planning interval and duration: from 96*15 to n*m Simulate effect of controls during plan execution vis-à-vis true execution Modeling flexibility for power & energy Power varies over time, energy has min/max envelope Aggregation/dis-aggregation of energy storage devices Additional control possibilities Further manageable loads (heat-pumps, boilers, cooling/freezing, etc) Moving towards an MPC paradigm for the electrical power grid? Improved load and EV traffic prediction Need temporal and spatial prediction Analysis of load and traffic traces to validate prediction models Performance is always the computer scientist s issue Sparse matrix packages, Tinney schemas Newton-Raphson instead of Gauss-Seidel for load-flow Distribution, parallelism

20

21 Backup

22 Grid Constraints as Max-Flow Problem Source Ford-Fulkerson/Edmond & Karp is O(n 2 m) With nodes, edges we get 28 secs (Java) or 18 secs (MSFT VC++) on Intel Dual Core 2GHz (TP61) Goldberg and Tarjan (JACM 1988) have O(m 3 ) Down to 8 secs (MSFT VC++) Using Sleator & Tarjan dynamic tree data structure we get O(nm log(n 2 /m)) Expected, theoretical, speed-up of for our problem size! Overhead of dynamic tree structure may be non negligible Another approach: use LP formulation of maxflow ILOG/CPLEX yields < 1 sec! n variables, m constraints Power Generator Power Consumer cap, flow Power Generator Electrical Grid Power Consumer cap, flow Power Generator cap, flow Power Consumer Target

23 Battery characteristics Battery = voltage source + resistance ( state-of-energy) State-of-energy ζ ζ dot = f(ζ,p b ) = P int (ζ,p b )/E int0 Internal battery power P int P int (ζ,p b ) = -η I b (ζ,p b ) V oc (), P b >= 0 = -1/η I b (ζ,p b ) V oc (), P b < 0

24 Charging behavior Wind-power vs. time-of-day: 96 slots, 15 minutes each Price is inversely co-related to wind-power Max. charging when max wind-power == min. price Charging plan for each EV Minute difference between linear vs. quadratic optimization

25 Non linear battery optimization for smart charging P b = f( P int ) P int + α P 2 int (second order Taylor approximation) power loss for externally supplied power P b minimize t s c T p int + t s p T int D(ά)D(c)p int subject to A s p int <= b s A g,j p int + p T intd(ά)d(a g,j )p int <= g j A b p int <= b b j = 1..n p int >= b l q r p int + p T intd(ά)d(q r )p int <= p u,r r = 1..mn m*n elements in decision variable p b for m vehicles, n time-slots.

26 Solution Approach: Simulation ZRL have built a large scale simulation for An electrical grid using real EE formulas! A fleet of electrical vehicles with some kinky behaviour In order to Compute smart-charging plans for EVs Study the impact of EV traffic on the electrical distribution grid Evaluate the battery charging behaviour of EVs Verify if we can integrate EVs as micro power plants (μchp) into the grid using their batteries as storage device Simulation is based on electrical engineering and computer science Electrical power grid and load flow (EE) Optimization theory: linear programming, optimal control methods (EE/OR) Discrete and continuous systems simulation (CS) Power Generator Power Consumer Power Generator Electrical Grid Power Generator Power Consumer Power Consumer

27 10KV/400V feeder: Location, power 10KV transmission line: Thermal load 60KV/10KV sub-station: Location, power Electric Vehicle: State-of-charge, identification, distance travelled 60KV transmission line: Thermal load Wind-turbine(s): Power, wind-speed Ronne Power Station: Power

28 Charging Schedule Computation with Grid Constraints Loop Compute a charging schedule for EVs (based on energy requests, constraints) Check if grid constraints are met If not then add additional charging schedule constraints Else break End loop The loop does eventually converge, but it takes some iterations. Divided between Grid: checking grid constraints, adding charging schedule constraints VPP: computing the charging schedules based on energy requests

29 Grid constraints detection and constraining Computed max-flow through grid Identify saturated edges Results in non-overlapping sub-grids (islands) Max-flow/min-cut theorem implemented Limit the load on the islands Since the requested power cannot be delivered! Additional constraint in charging schedule computation Sum of power requested on island <= power deliverable through saturated edges Non EV loads are not affected. Requirement: Prediction of timely availability of EV for charging Plus location of that availability!

30 Constraint violations A123Systems Saft VL45-E

31 Battery constraint violations Violation of boundaries of battery state-of-energy. Range 20-90%