Felix Wu Distinguished Lecture in Power Systems · Felix Wu Distinguished Lecture in Power Systems The University of Hong Kong, 10th November 2014 ... • Control theory ... 2.5GW,

||

Göran AnderssonPower System Laboratory, ETH Zürich

Felix Wu Distinguished Lecture in Power SystemsThe University of Hong Kong, 10th November 2014

The Future Electric Power SystemDevelopments and New Analysis Tools

||

Outline

• Power System Lab (PSL), ETH Zurich

• New Challenges and Recent Developments

• Power Node Modeling Framework

• Cyber- security in Power Systems

• Concluding Remarks

2

||

PSL Research (and Teaching)

• Power System Operation and Control

• Power Markets

• Future Energy Systems

Main Methods & Tools used• Control theory• Optimization• Simulation• ...

3

||

The PSL Team

4

||

Complexity of Power SystemsComplexity along several dimensions

Time (milli)seconds (e.g. frequency inertia, frequency&voltage control),minutes (e.g. secondary/tertiary frequency&voltage control),hours/days (e.g. spot market-based plant/storage scheduling),months/years (e.g. seasonal storage, infrastructure planning).

Space 1‘000+ km, e.g. interconnected continental European grid(Portugal – Poland: 3‘600 km, Denmark – Sicily: 3‘000 km).

Hierarchy from distribution grid (e.g. 120/240 V, 10 kV) tohigh-voltage transmission grid (220/380/500/… kV, AC and DC).

5

||

49.88

49.89

49.90

49.91

49.92

49.93

49.94

49.95

49.96

49.97

49.98

49.99

50.00

50.01

50.02

16:45:00 16:50:00 16:55:00 17:00:00 17:05:00 17:10:00 17:15:00

8. Dezember 2004

f [Hz]

49.88

49.89

49.90

49.91

49.92

49.93

49.94

49.95

49.96

49.97

49.98

49.99

50.00

50.01

50.02

16:45:00 16:50:00 16:55:00 17:00:00 17:05:00 17:10:00 17:15:00

8. Dezember 2004

f [Hz]

Frequency Athens

The grid frequency – A key indicator of the state of the system

f - Setpoint

Frequency Mettlen, Switzerland

PP - Outage

PS Oscillation

Source: W. Sattinger, Swissgrid6

||

Source: A new frequency control reserve framework based on energy-constrained units (Borsche, Ulbig, Andersson, PSCC 2014)

Spectrum of the system frequency and the AGC signal

7

||

Power Flow

Control(e.g. line

switching)

PAST – Traditional view

Storage[+/–]

Loads (assumed non‐controllable)

Power Grid(Line rating & Voltage/Frequency constraints)

[+/–]: Power regulation up/down possible.

Coal Nuclear Gas Hydroconventional

Generation [+/–]

Complexity of Power Systems

(DE capacity values of aroundyear 2000)

Fully dispatchable~99% of all generation

Hydro Storage only

Fully dispatchable (energy constraints)~10% of peak load

Observable & well predictableInterruptible loads by manual or static control means

(large industrial loads, hot water boilers)

8

||

Increasing fluctuating RES deployment = Fluctuating power in-feed Germany 2012: 63.9 GW power capacity ≈ 75% of fully dispatchable fossil generation.

(Wind+PV) 77.1 TWh energy produced ≈ 15.2% of final electricity consumption.

Wind+PV: Still mostly uncontrolled power feed-in – Hydro: «well»-predictable power feed-in.

Mitigation Options Improvement of Controllability: Implementation of Wind/PV curtailment in some countries. Improvement of Observability: More measurements and better predictions of PV and

wind power feed-in (state estimation & prediction).

Sources: BaSt2012, IEA ElectricityInformation 2011, BMU AGEE 2013, own calculations

PPV > PWind

PWind > PHydro

EWind > EHydroPPV > PHydro

EPV > EHydro

Trends and Challenges

9

||

Transmission Grid

[+/–]: Power regulation up/down possible.

Storage[+/–] (Line rating & Voltage/Frequency constraints)

var‐RESGeneration [+/–]

Coal Nuclear Gas Hydro Biomass Wind Solar PVconv./firm‐RESGeneration [+/–]

Power Flow

Control(incl.

FACTS)

controllable Loads [+/–](price-responsiveness: Demand Response)

(control signal-driven: Demand Side Participation)non‐controllable Loads

No strict borderline

PRESENT & FUTURE – high RES shares & Smart Grid Vision (DE capacity values of year 2011)

Time-varying

dispatchable~40% of all generation

Hydro Storage, Batteries, Flywheels,

…

Soon >10% of peak load

Increase of controllable loads(faster response times, automaticcontrol)

Fully dispatchable~60% of all generation

10

||

Power Nodes Framework

Kai Heussen (DTU)Stephan KochAndreas Ulbig...

11

||

Consideration of interactions of power system units and the electricity grid, i.e. power injected into the grid or power demanded from the grid, is not giving the whole picture.

Which of these units are storages (and thus energy-constrained)?

Which of these units provide fluctuating power in-feed?

What controllability and observability (full / partial / none) does an operatorhave over fluctuating generation and demand processes?

Better Modeling of Power Systems –Motivation for Power Nodes Modeling Framework

Energy provided / demanded

Storage

?

?Controllability?Observability?

Flexibility?

GridPin

Pout

Power System Unit

12

||

Power Node Modeling Approach

1load load gen geni i i ii i i i iC x u u w v

,

State-Descriptor Form

13

||

Power Node Modeling Approach

Power feed-in to grid

Efficiency factors

Storage capacity

×state-of-charge

(SOC)

Provided / demanded power

Shedding term

Internal storagelosses v(x)

Power feed-out from grid

1load load gen geni i i ii i i i iC x u u w v

,

State-Descriptor Form

||

Examples of Power Node Definitions

Fully dispatchable generation No load, no storage (C) Fuel: natural gas (ξ>0)

Combined Heat/ Power Plant(CHP), Berlin-Mitte Offshore Wind Farm, Denmark

Time-dependent dispatchable generation, if wind blows, ξ ≥0, and if energy waste term w≥0

No load, no storage (C) Fuel: wind power (ξ>0)

General formulation: 1load load gen geni i i ii i i i iC x u u w v

1g en g e ni i i iu w 1

g e n g e ni i iu 15

||


Time-dependent dispatchable load (heating element)

Constrained ”storage” (C ≈ 10 kWh) Demand: hot water, daily pattern (ξ< 0),

internal heat loss (v > 0)

Residential electric water heaters

Emosson (Nant de Drance)

Dispatchable generation & load Battery storage (C ≈ 10-20 kWh), very small

losses (v ≈ 0) Demand: driving profile (ξ< 0), EV: (w = 0) PHEV: Substitute electricity by fuel (w ≥ 0)

Plug-In (Hybrid) Electric Vehicle (PHEV/EV)

Charging only:

Full V2G support:


lo a d lo adi ii i i iC x u v load loadi ii i iC x u

1load load gen geni i i ii i i iC x u u w

16

||


Fully dispatchable generation (turbine) and load (pump)

Constrained storage (C ≈ 8 GWh) Fuel: almost no water influx (ξ≈0)

Goldisthal Hydro Pumped Storage, Germany

Fully dispatchable generation (turbine), but no load (pump)

Large storage (C ≈ 1000 GWh) Fuel supply: rain, snow melting (ξ>0)

Emossion Storage Lake, Switzerland


1gen geni ii i iC x u 1

load load gen geni i i ii iC x u u

17

||


1 water inflowgen geni ii i i iC x u w

Dispatchable generation and load Constrained storage (C ≈ GWh range) Fuel (ξi,k): water influx from upper

basin and other inflows (ξi,k≥2) Waste (w): water discharge into lower

basin (or river) Loss (v): evaporation from bassin

Dispatchable generation, but no load Storage function dependent on

geography, C ϵ [0, … , GWh, TWh] Fuel (ξ): water influx from river, (ξ>0) Waste (w): water flow over barrage (high

water-level) or intentional water diversion

Run-of-River Plant, Zurich

1load load gen gen ,i i i ii i i k i i

k

C x u u w v


0%

100%

xi

Ci

vi wi

ξi,1

η-1geni

· ugeni

ηloadi·

uloadi

Hydro Cascade – one stage

ξi,k≥2

18

||

Unit Commitment (UC) or Optimal Power Flow (OPF) including energy storage units

Demand and RES power in-feed forecasts (perfect or imperfect)

Optimisation based on marginal generation costs (+ ramping costs)

UC: Copperplate simplification OPF: Grid constraints included In addition: Representation of

transmission and distribution grid constraints (line capacity, voltage)

Implementation: Matlab, Yalmip

Power Nodes Simulations –Predictive Power Dispatch

19

||

Source: Swiss energy strategy 2050 and the consequences for electricity grid operation – full report (Comaty, Ulbig, Andersson, ETH 2014)

Verification of the Power Node approach, 1

20

||

Verification of the Power Node approach, 2

Source: Swiss energy strategy 2050 and the consequences for electricity grid operation – full report (Comaty, Ulbig, Andersson, ETH 2014)

BfE statistics: Import 32.9 TWhe/aExport 30.9 TWhe/a

Power Node approach: Import 30.2 TWhe/aExport 36.6 TWhe/a

21

||

Possible scenario for Switzerland 2050

Zeit [ in h]

Elek

tris

che

Leis

tung

[in

MW

]

Pumped hydro CH1.7GW, 50-100GWh

η = 75–80%

Heat pumps0.7 GW, 0.6 GWh

Coolingequipment1GW, 0.1 GWh

Hot water boilers2.5GW, 1GWh

P(H)EV0.3 GW, 1.2 GWh

η = 80–90%

22

||

Storage saturation

Curtailment ofWind or PV Power Infeed

Simulation Period May 2010 (30% Wind, 50% PV, no DSP) High Temporal Resolution Tpred. = 72h, Tupd. = 4h, Tsample= 15min. Calculation Time ≈ 1min.

Simulation Results –Predictive Power Dispatch (Case Study Germany)

23

||

Assessment of FlexibilityGeneral Simulation Approach

(Wind Energy Share = 15%, PV Energy Share = 15%)




Scenario Creation

(10…1000)

Yearly Power Dispatch Simulations

(high temporal resolution)

Balance Term Assessment

GridTopology

Time Series(Wind, PV, Water

In-feed, Load)

Power Systems Specs(Power Plants, Merit Order,

Control Reserve Requirements)

Assessment & Analysis Data(Fossil Fuel Usage, Storage Cycling,

Curtailment, …)

Data Mining Challenge!

GraphicalRepresentation

Predictive power dispatch for full-year simulations High temporal resolution (15min.) – 1 year / 15min. ≈ 35’000 sim. steps Parallel calculation of 10-1000s scenarios (perfectly parallel task)

24

||

Assessment of Flexibility –Curtailed Renewable Energy in Germany

PV Power Deployment

Win

d Po

wer

Dep

loym

ent C

urtailedR

ES Energy(in %

oftotal availableR

ES Energy)

0-50% Wind Energy, 0-50% PV Energy, Full-Year 2011 simulationsonly existing hydro storage, copperplate grid model, no export, no DSP

25

||

20% Wind Energy, 10% PV Energy (EU-NREAP Goals), Full-Year 2011 simulations only existing hydro storage, copperplate grid model, no export, no DSP

Cur

taile

dR

ES E

nerg

y(in

% o

ftot

al a

vaila

ble

RES

Ene

rgy)

Energy Rating of Storage (ε)

Storage Capacity todayπ ≈ 7 GW (8% of peak load)ε ≈ 40GWh (~6h)

Assessment of Flexibility –Curtailed Renewable Energy in Germany

26

||

Why is a predictive dispatch optimization necessary? Strong impact of prediction horizon length (Tp) on dispatch performance visible. Example German power system (with varying wind/PV energy shares). Simulation parameters full-year 2010, 15min sampling time, artificial pumped hydro

storage capacity of 50x nominal values (7GW/42 GWh nominal power/energy) Full-year simulations of 25 setups with varying wind/PV share

Figure description– x-axis: [0, 5, 10, …, 50%] of PV energy share of total yearly load demand.– y-axis: [0, 5, 10, …, 50%] of wind energy share of total yearly load demand.– color coding: Curtailmentof Wind&PV energy (dark blue: ≈0%, dark red: ≈50%).

Tp = 1h Tp = 12h Tp = 24h

27

||

Why is a predictive dispatch optimization necessary?

But Longer prediction horizons are computationally expensive (same for detailed models )

x-axis: Prediction Horizon Length (6h – 240h). y-axis: Computational Effort (Solver Run-Time). 28

||

Time Aggregation – Heterogeneous Sampling

Pumped hydro storage usage for different optimization approaches (MA S. Deml)

Time aggregation greatly enhances performance for predictive power dispatch problems. Here at least 2x less computational effort.

Heterogeneous-sampling MPC (HSMPC) significantly outperforms uniform-sampling MPC and is close to the Performance Bound (PB) dispatch.

29

Quantitative Insights

||

Change of Load Flow Patterns in European Power System

North To

South

South To

North

North To

South

South To

North

Year 2010

Year 2010(50% RES)

30

A Comment on Volatility

||

Cyber-Security in Power Systems

Maria VrakoupouloKostas MargellosPeyman EsfahaniJohn Lygeros

31

||

From security requirements to societal cost

Cyber attack

SCADA system

Power network

Societal cost

32

||

Attacks on Power Systems

• SCADA and EMS are complex monitor and control systems for the transmission grid

• Many attack opportunities Sensor and actuators Communication systems Software systems (e.g. control) Human operators Physical infrastructure

• How can we protect these systems against cyber-attacks?

33

||

Which signals could be manipulated by a cyber-attack?

Generator set points (AGC, etc)

Load tap changers

Status of switches

Configuration changes (macros)

Can we find an attack signal that is able to lead our nominal state in unsafe operation?

The AGC is one of very few automatically closed loop controllers of the SCADA system. (The only one?)

34

||

Frequency controlSecondary (AGC)Plant outage Primary Tertiary frequency control

source: swissgrid

35

||

Frequency control

AGC

PI

Ptie-line, f

Control area

d1

d2

dn

36

||

Test system: Two-Area Power Network

tie lines

Area 1Area 2

AGC AGC

measurementsmeasurements

generation setpoints

generation setpoints

Communication lines

37

||

Two-Area Power System modelingTowards lower complexity

‘Full’ model

• 567 dynamic states

• algebraic states

• voltage + frequencydynamics

• AVR, PSS, governor, AGC

Multimachinemodel


• no algebraic states

• voltage + frequencydynamics

• governor, AGC

• node elimination

Multimachineclassical model



• frequency dynamics

• governor, AGC

• node elimination

Two machine frequency model



• frequency dynamics

• governor, AGC

• center of H aggregation

38

||

Part I : Attack on the Automatic Generation Control (AGC)Outline

• Problem setup

• Impact analysis

• Intrusion detection and mitigation

39

||

Test system: Two-Area Power Network

What can attacker do with access to AGC signal in one area?

Can he cause frequency or power exchange range violations ? Load shedding or generator tripping ( Black out)

40

||

Synthesizing an Attack Signal a) Random switching signal

Naïve attacker cannot violate frequency constraints ! More intelligent policy is needed …

uwxgwxfx ),(),(

Nonlinear System Dynamic

Random bang-bang signal

frequency dynamics

41

||

Attacker Policy

uwxgwxfx ),(),(

Nonlinear System Dynamic

MCMC

),(),(),( 000 wtuwxgwxfx

Theoretically does the job !

(Perfect model)

Practically not ! (With parameter

uncertainties)

Imperfect Model

],,[ 021 HHw

frequency dynamics

Synthesizing an Attack Signalb) Open-loop policy

42

||

Synthesizing an Attack Signalc) Feedback policy (Feedback Linearization + MCMC)

Achieve almost the same

performance with imperfect

information

Imperfect Model

Perfect Model

frequency dynamics

- not applicable to large-scale networks !43

||

Synthesizing an Attack Signald) Feedback policy (gain scheduling)

Gain scheduling

Constant linear feedback

Select K so as to make the linear system unstable

+ applicable to large-scale networks !

frequency+ voltage dynamics

44

||

Up to now…

• Can attacker cause problems by manipulating AGC?

Yes he can!

With fairly sophisticated feedback controllers

• What can we do about it?

45

||

Part I : Attack on the Automatic Generation Control (AGC)Outline

• Problem setup

• Impact analysis

• Intrusion detection and mitigation

46

||

∆P

fYmeasPower

Network

FDIFilter

residual

Fault Detection & Isolation

47

||

Filter Construction a) Linear filter

Set of Linear Problems

z rN(s)L(s) Residual

FDI Filter

residual

known (measurements, input signal)

unknown(states, disturbance)

fault or attack signalto be detected

48

||

(a) Test on the linear model (b) Test on the nonlinear model

Resultsa) Linear filter


49

||

Filter Construction b) Nonlinear filter

∆P χ

Set of QuadraticProblems

nonlinear part

50

||

Resultsb) Nonlinear filter

Test on the nonlineal model


51

||

Linearization based seems more sensitive in ideal setup, but not robust with respect to nonlinear terms

Filter based on family of excitation signals is robust to load deviations at all nodes in the network

Linear filter Nonlinear filter

Case study 1a) Filter based on a simplified model of the IEEE 118-bus system

frequency dynamics

52

||


Despite some obvious uncertainties, the filter works fairly well

‘Full’ model

53

||


Simple idea to improve filter performance

‘Full’ model

54

||

Disconnect AGC after detecting the attack !

Case study 2Attack – Detection – Resolution


AGC signal (cyber attack) Tie line power

Filter response (residual)

AGC signal (cyber attack) after detection Tie line power

55

||

The challenges of integrating renewables are manifold but – in principal –managable.

Accurate modeling, simulation and analysis tools necessary for studying power systems and derive adaptation strategies from such decision support tools.

Hard Paths – Solve problems simply by oversizing everything.(= oversized, expensive, inefficiently operated power system)

Soft Paths – Solve problems via more control & optimal operation.(= right sized, less expensive, efficiently operated power system)

Control Based Expansion

Computation and communication is cheap (and getting cheaper),(physical grid investments are expensive)

Also other challenges (power markets, consumption growth, …)

Some Conclusions (1)

56

||

With the introduction of more (wide-area) control loops, the risks for cyber-attacksincrease

Methods to study cyber-attacks will grow in importance

Means to prevent and identify cyber-attacks will be an essential part of futurepower systems analysis

Some Conclusions (2)

57

|| 58

A general reflection on research

In the middle of the forest there is an unexpected gladethat can only be found by someone who is lost.

Tomas TranströmerNobel Prize Laureate in Literature 2011

||

Building an Energy System is a Team Work

59

||

Thank you for your attention

60

Documents

Felix Wu Distinguished Lecture in Power Systems · Felix Wu Distinguished Lecture in Power Systems The University of Hong Kong, 10th November 2014 ... • Control theory ... 2.5GW,