1 Impact of ICT infrastructure in measurement for distribution system state estimation Ankur Majumdar, Sara Nanchian and Bikash Pal Dept. of Electrical and Electronic Engg Imperial College London
Overview Role of ICT and smart measurements in active distribution DMS architecture Three phase state estimation in unbalanced distribution system Vulnerabilities associated with state estimation Leverage measurements and attack strategies Algorithm Diagnostic robust generalized potential (DRGP) Externally studentized residuals Test systems and results Conclusions and future work
Why state estimation and bad data? There is an increasing adoption of smart instrumentation and smart meters with ICT infrastructure. Power system is more prone to attacks from adversaries. Tampered data will affect control and computing functions - security analysis, VVC etc.- to avoid contingency and cascaded tripping SE is at the core of EMS/DMS, which plays a crucial role in control and operation of modern power systems. State estimation is a procedure of obtaining network node voltages and angles the attack ICT system will impact the quality of estimation and hence all other network functions Any tampering with data and maliciously operating switch will also result in gross errors in data. So in principle, the effect of malicious attack can be cast as bad data detection.
Distribution management system (DMS) architecture Measurements vector Interface with ICT system Attack strategy Attacker Basic EMS/DMS functions State Estimator State estimates Detection of Leverage Points and Bad Data Distribution Optimal Power Flow Notify DNO Distribution Network Operator Control set-points Power Grid
Setting up state estimation Find x that minimizes a weighted least square of errors min J = [ z meas h func ( x)] T R 1 [ z meas h func ( x)] Subject to: c c eq ineq ( x) = 0 ( x) 0 x z meas h func (x) state variable of voltage magnitudes and angles measurement function measurement function
Leverage points and bad data Measurements that significantly influence the state estimates z = H x + e each observation ( z i, Hi) is a point in the factor space of regression. Leverage measurement is an outlier in the H i space. Leverage points can be good or bad. Good leverages are useful in improving the estimates. Though bad leverages are harmful to many estimators. Exhibit masking/swamping effect Bad data-outliers in the measurement space
Attack strategies Attacking power flow measurements Attacking power injection measurements Attack strategy is based on three theorems Theorem 1 states how a successful attack can be made by changing the diagonal elements of the hat matrix Theorem 2 shows how much those diagonals has to be increased to make an attack Theorem 3 suggests how to increase the value of the diagonal
Detection of leverage points Leverage points have small residuals. Hence, difficult to identify bad data. Leverage points are identified by the unified diagnostic and robust approach. The method is called diagnostic robust generalized potentials (DRGP). The generalized potentials pot * ii ( D Kii ( = 1 Kii ( D) Kii ) D) i R i D Start Calculate RMD based on MVE or MCD Form set R and set D Compute * pot ii All * pot ii in D > cut off No No Put the data * With pot ii cut off in D sequentially Back to R Yes Leverage points identified Stop With cut-off * med ( pot ) + c. MAD( * ii pot ii )
Studentized residuals Internally and externally studentized residuals Generalised studentized residuals (GSR) a form of externally studentized residual + = D i K r R i K r r D ii R D i D ii i R D i i st ) ( ) ( ) ( ) ( *, 1 ˆ 1 ˆ σ σ
123 bus test system model 30 31 120 12 29 49 26 33 34 32 28 27 25 24 23 22 21 20 19 15 10 14 11 35 9 3 8 13 16 2 17 1 7 6 4 5 48 45 43 41 116 36 18 38 60 50 51 52 37 59 66 67 39 40 58 65 63 61 62 64 117 53 54 55 56 57 97 95 93 94 92 123 118 91 90 89 88 121 109 106 102 119 98 68 73 77 9687 82 112 111 113 114 115 110 107 108 103 104 105 99100 101122 69 70 71 72 74 75 76 81 83 78 79 85 80 86 84
L-R and DRGP-GSR plot
L-R and DRGP-GSR plot
Normalized residuals and GSR
Implication of the results An attacker can take advantage of the presence of leverage measurement for mal intention The methodology is advantageous against other residual based bad data detection technique It will assist DMS/EMS to take effective control and operational decisions
Conclusions Largest Normalised Residual (LNR) does not work on multiple interacting and conforming bad data. Masking and swamping effect render LNR unsuitable DRGP identifies the high leverage in the presence of Masking and Swamping LNR can not GSR can identifies the bad data So combination of DRGP-GSR overcomes the limitation of L-R
Future work An attacker can inject gross error into the tap position measurements. The reactive power flow through transformer will vary significantly when a tap measurement is in error. The reactive power flow being on a short line has the possibility to be a leverage measurement. Therefore, a extension of this work is to have a robust detection technique for bad tap measurements. Another aspect is to incorporate the correlations among loads and correlation between loads and the real measurements to improve the quality of the pseudo measurements and hence the estimate.
Impact of ICT infrastructure in measurement for distribution system state estimation Ankur Majumdar, Sara Nanchian and Bikash Pal Dept. of Electrical and Electronic Engg. Imperial College London
Lessons Learned in the Real World: SCE s DMS Implementation Mike Kohler, Brenden Russell, Anthony Johnson
DMS SCADA Overview DMS Phase 1 project scope was to replace the home grown Distribution SCADA & Control System (DCMS) with a XA/21 DMS solution that leveraged the existing EMS UI familiarity with Operators and existing PSC support staff. Project Challenges System Architecture DMS User Interface DMS Application Integration
Project Challenges - Explore current and near term system implementations and capabilities within existing enterprise environment. Do not plan on system implementations that are only conceptual or not in progress. - Leverage existing business capabilities SCE mapping organization, current engineering tools, data. - Leverage existing EMS SCADA Security & Infrastructure installations (Domain infrastructure, RSA servers, Password Vault, Backup Solution, Network Support). - Realization that Grid Operations, Distribution Automation, Field Engineering, and Field Apparatus can not absorb large iterations of change. There will need to be intermediate steps to achieve the final long term visions.
Project Challenges - SCE Grid Operations desire a minimum impact with continued use of CGI OMS as system of record and use current operating environment configuration. - EMS and future systems (DMS, PMU, CRAS) will need to leverage common infrastructure where feasible to minimize costs and insure maintenance processes are consistent. - Existing Distribution Control and Monitoring System (DCMS) is home grown, uses obsolete technology, and software/database is not documented. System uses a custom Scada interface driver to interface to SCE private radio network. - New Business Processes and Interfaces will need to be developed to handle the new data requirements for the DMS Advanced Applications. - Opportunities would be sought to improve any shortcomings of the current tools and processes used for Distribution Automation.
System Architecture
System Architecture System Architecture Challenges Most systems are a point based systems. The scale of SCE s distribution infrastructure would potentially crush any point based solution 900 substations, 4500 feeders, 25,000 Scada Devices, 2 M potential Scada points, 70,000 switches, 6 Million customer loads, 1 M transformers, 10 M busses/nodes. Maintenance in a traditional point based model for UI and Scada would be a challenge. Auto creation of graphics was a must. There would be no way to hand build and maintain graphics using AutoCad or any proprietary editing tool. SCE s radio network communications is rather custom, loosely based on DNP 3.0, leverages report by exception as much as possible.
DMS User Interface The solution is built an a device based, database defined, auto generated user interface copied from the SCE DCMS. The user interface addressed 2 of SCE s key requirements 1) Reduce shock factor for the SCE user base 2) Generate displays/interface with no display maintenance required. 3) Can be launched from alarms, events, device summaries, device finder, bookmarks. The solution provided a DMS Device Summary to support device based rather than a point based view of equipment to allow operations and maintenance to view/report the relationship between many pieces of equipment at various electrical groupings (feeder, sub, system) and a geographical grouping (district). Our vendors collaborated to allow URL interaction between client applications to enable navigation between the DMS and OMS applications. The DMS naming convention would follow the global convention used by the OMS, GIS, and Asset Management Systems. Use CIM and GML exchange format to auto build DMS network model and graphics.
DMS Application Integration Data is extracted from SCE GIS systems. Data is augmented with other sources to complete total model definition. Data is imported into the Model Exchange Platform in Distribution CIM format and GML format. DMS Adapter defines batch input files and uses GML files to initiate graphics build. Data is committed on DMS and creates application database definition files to create runtime database. Data is not converted into points. As shown before device definitions include linkages to scada values used by applications.
DMS Advanced Applications Overview DMS Phase 2 & 3 of the project expands capabilities to incorporate future advanced Distribution Grid Control and Analysis Technologies that provide foundation to support future vision of Distribution Grid. Business Drivers Distribution State Estimation DVVC (Immediate Solution) IVVC (Long Term Solution) Contingency Load Transfer (CLT) Fault Detection, Isolation and Restoration (FDIR) DMS Dispatch Training Simulator (DTS)
10 Business Drivers: Today s Grid Reliable Safe Safety Established standards and practices for safe design and operation Delayed/Limited situational awareness Unsophisticated protection schemes Reliability System restoration via automated equipment Antiquated equipment & wire capacity (Asymmetrical system design) Delayed customer communication Reactive system operation based upon numerous tools Anti Islanding DG design preventing microgrids Resiliency Independent voltage control can tolerate low penetration of DG
11 Business Drivers: Grid of The Future Reliable Safe Resiliency Expandable Safety Dynamic protection systems to precisely detect and isolate hazards RFI ES RFI RFI Reliability Self healing ability to adjust to system abnormalities Operation of intentional islands (Microgrids) Fully integrated real time system simulator for Grid Operations (State Estimator) ES RFI ES RFI Resiliency Automated resource control Fully capable of adapting to dynamic system conditions Comprehensive system monitoring, forecasting, and operation ES Expandability Seamless equipment and resource addition as technology expands Fully developed backbones (e.g. telecomm, automation, operations control) Uniform, accessible marketplace grid
SE Functionality: The Distribution State Estimator was built for mainly radial distribution networks. It is very robust and it can handle complex network topology Demand Estimation - Adjusts feeder loads to match SCADA measurements DMS State Estimator Distribution Power Flow - Ladder algorithm to solve for a unbalanced three-phase voltages and current flows Radial Feeders Loops & Parallel Basic Measurement Validation Non consistent SCADA with DPF results Challenges: Data Quality Challenges with GIS Bad Phasing Data OMS Graphical representation CIM / FME Development concurrent with DMS development Investments: Requires a comprehensive DA and Substation investment program to support accuracy requirements of SE. Field Data Verification and Business Process Changes
DMS Big Data Solution!
Benefits: Distribution Volt Var Control (DVVC) Assist maintenance organizations in the identification of failed PCCs Help Operations and Field Engineering identify and mitigate Voltage/VAR issues Targeted reduction in system voltage of 1% Approximate 490GWhs annual reduction Better management of VAR flow Reduction in subtransmission VAR demand (Energy saving mode) Objectives: Meet Volt/VAR requirements Minimize Average Circuit Voltage Utilize existing PCCs as a backup system Minimize Capacitor Bank Switching Automatically detect circuit re-configuration (longterm) Control limits will be definable by customer, Engineering
IVVC is a network model Voltage - Var control solution. Integrated Volt Var Control (IVVC) IVVC has 4 Objectives: Minimize MW Demand Minimize MW Losses Flatten feeder voltage Minimize control actions Voltage and/or VAR control: Switches capacitors on/off Transformer taps are raised/lowered Benefits: Adaptable to circuit reconfigurations Utilizes all voltage/var control resources Penalty and Indexing Factors allow for Flexible & powerful optimization control of voltage & var resources
Contingency Load Transfer / Restore Application Functionality: Provide switching recommendations to reduce overload on a selected device. Provides the list of switching operations to be performed in order to restore power in de-energized island Operator can view current loading and set desired loading as input parameters Option to use telemetered switch control only Business Use: Load Rolls / Transfers Emergency Loading Studies Automatic Load Restoration
Application Functionality: Fault Detection Isolation and Restoration FDIR utilizes fault-passage detection and overcurrent relay data to determine the location of fault. The faulted feeder section is automatically highlighted on graphic display Isolation FDIR will attempt to identify and recommend switching actions that will isolate the faulted section and restore power to un-faulted feeder sections upstream and downstream of faulted section
Fault Detection Isolation and Restoration (Cont) Benefits: Allows SCE to do more with less Reduce outage size and duration. Improve SAIDI Locate faults faster with less driving time. Reduce crew size to isolate and restore. Reduce windshield time, particularly with long distribution lines / big service area. Increase billing revenue through fewer and smaller outages. Improve customer service Resolve outages before customer calls. Provide the ability to service a larger territory with fewer linemen. Challenges: Few Remote Fault Indicators. A manual FDIR option is needed! Interfacing Existing Intelligent local protection schemes with FDIR Approx 6000 Automated Switches across 4500 distribution circuits. More automated switches needed! Parallel DA Investment Programs: Installation of Remote Fault Indicators 5,000 15,000 / Year starting 2015 Expansion of switch automation program
Application Functionality: DMS Dispatch Training Simulator Instructor GUI similar to EMS Instructor GUI Drag and drop Devices to built various simulation scenarios. Direct interact with instructor model though circuit display Supports 7 types of events Business Impacts: System Parallels Load Transfers Closed Box Scenarios. Underground switches Train System Operators on SCADA device behaviors. Simulate RTS device failures, operating devices and telemetered devices N-1 conditions Coordinate Transmission Distribution Switching Scenarios
Thank you
1 Requirements for Distribution State Estimation for Quasi-Static Time Series Analysis of Distribution Network Dr. M. Kemal Celik U.S. Department of Energy Office of Energy Efficiency & Renewable Energy Solar Energy Technologies Office (SunShot)
2 Introduction US Department of Energy, Solar Energy Technology Office SunShot Groups & Systems Integration Systems Integration Projects Operations & Operational Planning Requirements New Operations/Information Technologies State Estimation State Estimation Definition Distribution State Estimation Standard Data Interaction Outline Study 2014 Confidential
3 SunShot Program Structure $1/Watt SunShot 2020 Goal
Systems Integration DISPATCHABILITY GRID PERFORMANCE AND RELIABILITY POWER ELECTRONICS COMMUNICATIONS 4
SunShot Systems Integration Funding Opportunities
Enabling Extreme Real-Time Grid Integration of Solar Energy (ENERGISE) 6
Where We Want to be in D 7
8 Operations & Operational Planning Present Mins into future Hours/days into future Operations Operational Planning 1. Short-term load forecasting 2. Solar forecasting 3. Distribution state estimation 4. Power flow/what-if analysis 5. Optimization/VVO&CVR&&Markets
Operational/Information Technologies Traditional GIS DMS SCADA OMS CIS Smart AMI & Smart meters - New generation SMs with more features - Easier access to MDM/database Renewable & Distributed - Solar, Wind, DR Electrical vehicles, Smart appliances, Home EMSs Community storage - Residential/Commercial Smart inverters Solar forecasting Confidential
State Estimation State estimation is a statistical analysis More equations than # of unknown variables Random small errors in equations (measurements) Small # of gross errors (bad data/measurements) z h( x) e r z h( xˆ ) Using measurements, z, solve for ˆx minimizing r z are measurements: P/Q flows & injection, voltages, phasors, etc. x are system states: Voltage phasors at each node e are random measurements errors (assumed to follow some probability distribution function) r are measurement residuals Confidential 10
State Estimation State estimation is a statistical analysis Weighted Least Squares, Weighted Least Absolute Value, Weighted Least Median Mathematical representation of the equations are exact Only if there are no topology errors Random errors are present in measurements Assumed to follow a known probability distribution Redundancy (# of meas/# of variables) vary across the network model Observability some parts of the networks may not have enough measurements Confidential 11
Confidential State variables includes each phase of the bus Requires multiphase network models 12 Distribution State Estimation (DSE) T c n b n a n c b a c n b n a n c b a V V V V V V x ] [ 1 1 1 1 1 1
Numerical Analysis for Distribution Networks Short-term forecast Granular load/solar forecasting/small scale PV generation Multiphase-phase topology processing Multiphase-phase topology estimation Multiphase-phase state estimation Multiphase-phase power flow What-if simulations & prescriptive analysis Multiphase-phase switching analysis Multiphase-phase voltage optimization Optimization (trans-active energy) 13 Study 2014 Confidential
14 Challenges in DSE (& numerical network analysis) Accurate network models (as-built vs as-operated) Low (near zero) measurement redundancy Very few traditional SCADA measurement points Need to incorporate other sources of data (µ-)pmus, smart meter data, smart inverter data, pseudo-m PMUs good or bad to have? Synchronization of measurements Multi-phase network topology processing Multi-phase observability analysis How to use measurements that are deleted during observability Intermittency of DERs Need for QSTS for operations/operational planning Robust prescriptive analytics to provide operators with a set of actions Filtering and visualization of significant results Alerting to problematic areas, equipment, snapshots, etc. Confidential
SunShot Projections 15
Importance of Operational Planning & DSE 16
Importance of Operational Planning & DSE 17
18 Operations & Operational Planning Present Mins into future Hours/days into future Operations Operational Planning 1. Short-term load forecasting 2. Solar forecasting 3. Distribution state estimation 4. QSTS analysis
Need for Standards for Data Exchange DIFF Reverse Mapping Actions/ Alerts/ Processes EQ_DIFF 3-φ Load Flow Volt/VAR & CVR Time-series Analysis + MS 3-φ State & Load Forecast Volt/VAR & CVR Enhanced + MS/SSH 3-φ Topology & State Asoperated + EQ/CN 3-phase TP 3-phase PFL As-built networks CIM 19
Cybersecurity & Interoperability 20