Wide-area blackouts: why do they happen and how can modelling help

1. Wide-area blackouts: why do they happen and how can modelling help Professor Janusz W. Bialek Durham University

2. Outline • Modelling of electrical networks • Overview of recent blackouts and their causes • How can modelling help in preventing blackouts

3. Overview of recent blackouts • Only wide-area blackouts, not local ones • Local ones a majority • Interconnected system blackouts in 2003: US/Canada, Sweden/Denmark, Italy • UCTE “disturbance” 2006 • May 2008 disturbance in the UK

4. Modelling of electrical networks • A network is a planar graph with nodes (buses, vertices) and branches (lines, edges) • GB high-voltage transmission network is meshed and consists of 810 nodes and 1194 branches • UCTE and US interconnected networks consist of several thousands nodes • For most analyses, the network is described by algebraic equation (Current and Voltage Kirchhoff’s Laws) • Electromechanical stability of rotating generators is described by differential equations

5. Preventing blackouts • We can’t live without electricity so the power system has to be designed and operated in a robust manner • Should ride through “credible” disturbances • Trade-off between cost of keeping reserves and security • A proxy to probabilistic risk assessment: (N-1) contingency – a deterministic criterion • N-1 contingency: a single disturbance (generation/line outage) should not cause problems • it is unlikely that 2 or more units will be lost simultaneously • generation reserve: the loss of the largest infeed (a nuclear reactor of 1320 MW at Sizewell B) • Transmission reserve: loss of double-circuit line (N-D)

6. When do blackouts happen? • ... when (N-1) contingency analysis has not been done properly (Italy 2003, UCTE 2006) • ... or when more than 1 thing went wrong (US/Canada 2003, Sweden 2003, GB 2008) • ... or hidden mode of failure (London 2003) • The new world of renewables and Smart Grids may require the use of probabilistic risk assessment • Briefly today, more Thursday 3.30pm “Mathematical modelling of future energy systems”

7. Classification of blackouts • Transmission inadequacy: a failure in a transmission network causes a cascading overloading of the network (a majority) • Generation inadequacy: failures of power plant(s) cause a deficit of generation (GB 2008 disturbance) • Usually a mixture: an initial network fault causes a separation of the network into parts with deficit/excess of generation

8. Major transmission failures in late summer/autumn 2003 • 7 blackouts affecting 112 million people in 5 countries • 14 August 2003, USA/Canada • 23 August 2003, Helsinki • 28 August 2003, south London • 5 September 2003, east Birmingham • 23 September 2003, Sweden and Denmark • 28 September 2003, whole Italy except Sardinia • 22 October 2003, Cheltenham and Gloucester

9. The Oregonian, 24 August 2003, after C. Taylor

10. NE of USA/Canada: before

11. NE of USA/Canada: after

14. Where it all began: Ohio and surrounding areas Source: US/Canada Power System Outage Force

15. How it all started: tree flashover at 3.05 pm Source: US/Canada Power System Outage Force

16. Bad luck? • Alarm and logging system in FirstEnergy (FE) control room failed 1 hour before the cascade started • Not only it failed, but control room engineers did not know about it • When lines started to trip they could not take corrective action: the system was not (N-1) secure after first trips

17. Cascading tripping: an initial line trip casues overloading on other parallel lines Source: US/Canada Power System Outage Force

18. Effect of line trips on voltages: depressed voltage (Ohm’s Law) Source: US/Canada Power System Outage Force

19. Source: US/Canada Power System Outage Force

20. Source: US/Canada Power System Outage Force

21. Speed of cascading Source: US/Canada Power System Outage Force

22. Normal load, big margins Denmark self-sufficient, southern Sweden supplied from central/northern 1.2 GW Oskarshamn nuclear plant trips due to a feed-water valve problem 5 min later double busbar fault trips 4 lines at Horred substation (N-5) contingency 1.8 GW Ringhals nuclear plant shuts down Southern Sweden and western Denmark blacks out Danish/Swedish blackout: 23/09/03, 5 M people

23. Italy

24. Frequency as real power balance indicator • Power generated must be equal to power consumed • Frequency is the same at any part of interconnected network • If there’s a sudden loss of generation, energy imbalance is made up from kinetic energy of all rotating generators • The speed (frequency) drops triggering all turbine governors to increase generation automatically (feedback control) • If frequency drops too much,automatic load shedding isactivated • generation deficit => frequency drops, generation surplus => frequency increases Source: National Grid

25. 3 am: import 6.7 GW, 25% of total demand, 300 MW over agreed CH operated close to (N-1) security limit but Italy didn’t know about it 86% loaded internal Swiss Lukmanier line trips on a tree flashover 3.11 am: ETRANS informs GRTN (disputed, no voice recordings) GRTN reduces imports by 300 MW as requested Source: UCTE

26. two more CH lines trip and Italy loses synchronism with UCTE • Island operation: import deficit leads to a frequency drop and load shedding • Until 47.5 Hz, 10.9 GW of load shed but 7.5 GW of generation lost • Frequency drops below 47.5 Hz and remaining units trip • Blackout 2.5 minute after separation: whole Italy, except of Sardinia. Source: UCTE

27. UCTE disturbance in 2006 • UCTE: Union for the Co-ordination of Transmission of Electricity – association of TSOs now renamed ENTSO-E (European Network of Transmission System Operators for Electricity) Source: UCTE

28. Flows just before the blackout • Generation 274 GW including 15 GW of wind (5.5%) • Strong east-west power flows, i.e. the West depends on imports • strong wind generation in northern Germany Source: UCTE

29. Timeline • 18 Sept: a shipyard request EON for a routine disconnection of double circuit 380 kV line Diele-Conneferde in northern Germany on 5 Nov • 3 Nov: the shipyard request to bring forward the disconnection by 3 hours. Late announcement could not change exchange programs • 4 Nov 9.30 pm: EON concludes empirically, without updated (N-1) analysis, that the outage would be secure. Wrong! • 9.38: EON switches off of the line • 10.07: Alarms of high flows. EON decides, without simulations, to couple a busbar to reduce the current • Result: the current increases and the line trips • As the system was not (N-1) secure, cascading line tripping follows • separation of UCTE into 3 regions with different frequencies Image: http://www.cruise-ship-report.com/News/110506.htm

30. 10 GW surplus 51.4 Hz 0.8 GW deficit 49.7 Hz 8.9 GW deficit 49 Hz Source: UCTE

31. Western Europe: 8.9 GW deficit • Frequency drop to 49 Hz triggered automatic and manual load shedding (17 GW) and automatic tripping of pump storage units (1.6 GW) • However the frequency drop caused also tripping of 10.7 GW of generation – more load had to be shed • DC connection UK-France: continued export from France despite the deficit!

32. Resynchronisation • A number of uncoordinated unsuccessful attempts made without knowledge of the overall UCTE situation • Full resynchronisation after 38 minutes Source: UCTE

33. GB May 2008 event: a near miss • Cockenzie and Sizewell B were lost within 2 mins: (N-2) event, 1714 MW • Loss of Sizewell B is the largest infeed loss planned for (1320 MW) • Further 279 MW of wind tripped due to frequency drop (total 1993 MW) • Automatic load shedding of 546 MW triggered at 48.8 Hz • Voltage reduction caused reduction of demand by 1200 MW • More generation was connected and supply restored within 1 hour Source: National Grid

34. Will there be more blackouts? • People tend to learn from the past • ... but generals are usually prepared to the last (rather than future) war • Lessons learned from the blackouts - improvements in communications and coordination in Europe and USA • ... but new challenges are looming ahead

35. Generation adequacy issues • Possible problems after 2015 (Ofgem Discovery report) • Regulatory uncertainty

36. Increased penetration of renewable generation • Wind already a contributing factor to UCTE 2003 and GB 2008 disturbances • “Any feasible path to a 80% reduction of CO2 emissions by 2050 will require the almost total decarbonisation of electricity generation by 2030” (Climate Change Committee Building a Low Carbon Economy 2008)

37. Smart Grids • Comms-enabled responsive demand, electric cars etc • Highly stochastic generation and demand: (N-1) contingency criterion may become obsolete soon – new probabilistic risk assessment tolls required • Dependence on comms networks is a new mode of failure

38. Example of new modelling techniques: preventive network splitting • EPSRC grant started January 2010 (Complexity Science call) • Exciting collaboration between graph theorists from Southampton (Brodzki, Niblo) , OR experts from Edinburgh (Gondzio, McKinnon) and power engineers from Durham (Bialek, Taylor) • Split the network in a controlled manner before it partitions itself • Initial main challenge: speaking the same language, mutual education

39. Conclusions • (N-1) contingency criterion has served us well in the past but there were a number of wide-area blackouts in 2003, 2006 and 2008 • New challenges of increased wind penetration and Smart Grids • New mathematical modelling tools required to prevent future blackouts