1 ELECTRIC POWER GRID INTERDICTION Javier Salmeron and Kevin Wood, Naval Postgraduate School Ross Baldick, University of Texas at Austin Sponsored in part by Department of Homeland Security, Office of Domestic Preparedness

2 What is VEGA? VEGA determines the worst possible disruption that could be caused by a terrorist attack, Compares multiple attack plans terrorists might undertake under different resource-constrained assumptions, Assesses security enhancement through preemptive measures, and VEGA is based on powerful optimization techniques. VEGA is a tool for analyzing the vulnerability and defense of electric power systems under threats posed by terrorist attacks.

3  “One can hardly imagine a target more ideal than the U.S. domestic energy” (A.B. and L.H. Lovins, 1983)  “Any U.S. region could suffer lasting and widespread blackouts if three or more substations were targeted.” (OTA, 1990)  “The U.S. is at, or is fast approaching, a crisis stage with respect to reliability of transmission grids.” (NERC, 2001)  “The U.S. electric power systems must clearly be made more resilient to terrorist attack.” (Committee on Science and Technology for Countering Terrorism, NRC, 2002) Vulnerability of Electric Power Grids: A Long-Recognized Issue

4  “And the threat isn't simply academic. U.S. occupation forces in Afghanistan discovered Al Qaeda documentation about the facility that controls power distribution for the eastern U.S., fueling fears that an attack on the power grid may one day become a reality.” (Energy Pulse, 2003)  (On Ahmed Ressam) “They were specifically trained to attack critical infrastructure, including electric power plants.” (CNN, 2002)  (On Colombian FARC) “They are leaving entire regions without service. We can’t post a soldier at every tower” (ISA spokesman, 2002) Terrorist Threat?

5 Potential targets:  Generating plants  Transmission and distribution lines  Substations Easy disruption + Widespread damage + Difficult recovery Terrorist Threat? (cont.)

6 U.S. systems are operated so that a single failure does not disrupt the system (N–1 security) We investigate vulnerability to multiple, coordinated failures (N – m). Our approach uses optimization theory to: - Mathematically represent a power grid and power flows - Identify worst-case attacks to the grid (most “disruptive”) - Provide insight into physical vulnerabilities, and help guide protective plans that will mitigate disruptions should attacks occur (We maintain the assumption of information transparency) Modeling Assumptions

7 What are the best -New investments (e.g., new facilities, lines, spare transformers) -Upgrades (e.g., replacing conductors) -Protective measures (e.g., hardening, surveillance...) in -Generating systems and/or -Transmission and distribution systems that substantially reduce system vulnerabilities? “Best”= Improve Security + Affordable (+ Market Benefits???) Key Questions

8  To defend an electric grid, first learn how to attack it! -Optimal power flow model (minimize load shedding) -Interdiction model (maximize disruption, i.e., load shedding)  Additional features of the problem are: -Time scale: Very short-, short-, medium- and long-term -Customer types; ability to “share the pain” -Uncertainty about terrorist resources -(Protective measures???) Mathematical Analysis

9 -Level 1: Optimal power flow model to minimize “disruption”: (disruption = load shedding + increased costs) Data: Power grid data Integrating Three Levels of Optimization - Level 2: Interdiction model to maximize “Level-1 disruption” Data: Power grid data and terrorist resources - Level 3: Protective model to minimize “Level-2 interdiction” Data: Power grid data, terrorist resource and counter- terrorist resources (budget for expansion, spares, upgrades, hardening) (See mathematical details at the end of this presentation)

10 t (Attack) MW shedding Other Factors: System Restoration, Demand Curves... (months) Slow repair Grid ComponentInterdictableResources (number of terrorists) Outage Duration (h) Lines (AC/DC) (overhead) YES172 (or 48) Lines (underground)NON/A TransformersYES2768 (or 168) BusesYES3 (or 2)360 (or 168) GeneratorsNON/A SubstationsYES3768 (or 360) One to several days No Repair Days to one week Lines Weeks Trafos with Spares

11 Salmeron, Wood and Baldick (2004), IEEE Transactions on Power Systems Total Load: 2,850 MW IEEE Reliability Test System A t MW +72h Attack AA A A A A A B B +768h B B B B C +360h C C C

12 VEGA VEGA: Vulnerability of Electric Power Grids Analyzer Potential Users:  Utilities, ISOs...  Government analysts

13 VEGA: Main Menu File mgmt. Grid data Optimization Results One-Line GUI Help

14 Power Grid Data

15 One-Line GUI: Power Flow After Optimal Interdiction

16 DC-OPF: i: bus, l: line, g: generator, c: customer sector P Line, P Gen : power (MW) S: power shed  : bus phase Power Flow Model (DC Approx.) s.t.

17 Interdiction Model I-DC-OPF: s.t. I-DC-OPF: Interdiction max-min problem Can be converted into a standard mixed-integer model 

18 Solve the DC-OPF Power Flow Model given the current grid configuration (  ) Based on present and previous flow patterns, assign a “Value” (V) to each interdictable asset Interdiction Model Heuristic Interdict the assets that maximize “Total Value”

19 Exact (Mixed-Integer) Linearization of I-DC-OPF

20 Results for the Linearized MIP Case/AlgorithmDirectly Interdicted ComponentsTime Period Power Shed (MW) Energy Shed (MWh) RTS-Two-Areas (M=24) HEURISTIC Buses: 116, 118, 215, 218 Substations: Sub-A1, Sub-A2, Sub-B1, Sub- B h2,693969, h1,416577,728 Total: 1,547,208 RTS-Two-Areas (M=24) MIP Lines: A30, A33-2 Transformers: A7, B7 Buses: 115, 118, 215, 218 Substations: Sub-A2, Sub-B h3,164227, h2,716782, h1,416577,728 Total: 1,587,744 Case/AlgorithmDirectly Interdicted ComponentsTime Period Power Shed (MW) Energy Shed (MWh) RTS-Two- Areas (M=12) HEURISTIC Substations: Sub-A1, Sub-A2, Sub-B1, Sub- B ,4161,087,488 Total: 1,087,488 RTS-Two- Areas (M=12) MIP Lines: A23, B23 Transformers: A7, B7 Substations: Sub-A2, Sub-B h1,804129, h1,416985,536 Total: 1,115,424

21 The VEGA Team Javier Salmeron, NPS Ross Baldick, UT Austin Kevin Wood, NPS