ee392n - Spring 2011 Stanford University Intelligent Energy Systems 1 Lecture 3 Intelligent Energy Systems: Control and Monitoring Basics Dimitry Gorinevsky Seminar Course 392N ● Spring2011

Traditional Grid Worlds Largest Machine! –3300 utilities –15,000 generators, 14,000 TX substations –211,000 mi of HV lines (>230kV) A variety of interacting control systems ee392n - Spring 2011 Stanford University 2 2 Intelligent Energy Systems

Smart Energy Grid 3 Conventional Electric Grid Generation Transmission Distribution Load Intelligent Energy Network Load IPS Source IPS energy subnet Intelligent Power Switch Conventional Internet ee392n - Spring 2011 Stanford University Intelligent Energy Systems

Business Logic Intelligent Energy Applications ee392n - Spring 2011 Stanford University Intelligent Energy Systems 4 Database Presentation Layer Computer Tablet Smart phone Internet Communications Energy Application Application Logic (Intelligent Functions) Application Logic (Intelligent Functions)

ee392n - Spring 2011 Stanford University Intelligent Energy Systems5 Control Function Control function in a systems perspective

ee392n - Spring 2011 Stanford University Intelligent Energy Systems6 Analysis of Control Function Control analysis perspective Goal: verification of control logic –Simulation of the closed-loop behavior –Theoretical analysis

Key Control Methods Control Methods –Design patterns –Analysis templates P (proportional) control I (integral) control Switching control Optimization Cascaded control design ee392n - Spring 2011 Stanford University Intelligent Energy Systems 7

Generation Frequency Control Example ee392n - Spring 2011 Stanford University Intelligent Energy Systems 8 Controller Turbine /Generator sensor measurements control command Load disturbance

Generation Frequency Control Simplified classic grid frequency control model –Dynamics and Control of Electric Power Systems, G. Andersson, ETH Zurich, ee392n - Spring 2011 Stanford University Intelligent Energy Systems 9 Swing equation:

P-control P (proportional) feedback control Closed –loop dynamics Steady state error ee392n - Spring 2011 Stanford University Intelligent Energy Systems 10 frequency droop x(t)x(t) Step response frequency droop x+  0 u

AGC Control Example AGC = Automated Generation Control AGC frequency control ee392n - Spring 2011 Stanford University Intelligent Energy Systems 11 Load disturbance AGC generation command frequency measurement

AGC Frequency Control Frequency control model –x is frequency error –cl is frequency droop for load l –u is the generation command Control logic –I (integral) feedback control This is simplified analysis ee392n - Spring 2011 Stanford University Intelligent Energy Systems 12

ee392n - Spring 2011 Stanford University Intelligent Energy Systems13 P and I control P control of an integrator I control of a gain system. The same feedback loop g -k I  cl x -k p b  d x

ee392n - Spring 2011 Stanford University Intelligent Energy Systems14 outer loop Cascade (Nested) Loops Inner loop has faster time scale than outer loop In the outer loop time scale, consider the inner loop as a gain system that follows its setpoint input inner loop - inner loop setpoint output Plant Inner Loop Control Outer Loop Control - outer loop setpoint (command)

ee392n - Spring 2011 Stanford University Intelligent Energy Systems15 Switching (On-Off) Control State machine model –Hides the continuous-time dynamics –Continuous-time conditions for switching Simulation analysis –Stateflow by Mathworks offon passive cooling furnace heating setpoint

Optimization-based Control Is used in many energy applications, e.g., EMS Typically, LP or QP problem is solved –Embedded logic: at each step get new data and compute new solution ee392n - Spring 2011 Stanford University Intelligent Energy Systems 16 Optimization Problem Formulation Embedded Optimizer Solver Measured Data Control Variables Plant Sensors Actuators

Cascade (Hierarchical) Control Hierarchical decomposition –Cascade loop design –Time scale separation ee392n - Spring 2011 Stanford University Intelligent Energy Systems 17

Hierarchical Control Examples Frequency control –I (AGC)  P (Generator) ADR – Automated Demand Response –Optimization  Switching Energy flow control in EMS –Optimization  PI Building control: –PI  Switching –Optimization ee392n - Spring 2011 Stanford University Intelligent Energy Systems 18

Power Generation Time Scales ee392n - Spring 2011 Stanford University Intelligent Energy Systems 19 Power Supply Scheduling Power generation and distribution Energy supply side Time (s) 1/

Power Demand Time Scales Power consumption –DR, Homes, Buildings, Plants Demand side ee392n - Spring 2011 Stanford University Intelligent Energy Systems 20 Demand Response Home Thermostat Building HVAC Enterprise Demand Scheduling Time (s) 1001,000 10,000

Research Topics: Control Potential topics for the term paper. Distribution system control and optimization –Voltage and frequency stability –Distributed control for Distributed Generation –Distribution Management System: energy optimization, DR ee392n - Spring 2011 Stanford University Intelligent Energy Systems 21

ee392n - Spring 2011 Stanford University Intelligent Energy Systems22 Monitoring & Decision Support Open-loop functions -Data presentation to a user

ee392n - Spring 2011 Stanford University Intelligent Energy Systems 23 Monitoring Goals Situational awareness –Anomaly detection –State estimation Health management –Fault isolation –Condition based maintenances

ee392n - Spring 2011 Stanford University Intelligent Energy Systems24 Condition Based Maintenance CBM+ Initiative

ee392n - Spring 2011 Stanford University Intelligent Energy Systems 25 SPC: Shewhart Control Chart W.Shewhart, Bell Labs, 1924 Statistical Process Control (SPC) UCL = mean + 3·  LCL = mean - 3·  Walter Shewhart ( ) sample mean quality variable Lower Control Limit Upper Control Limit

ee392n - Spring 2011 Stanford University Intelligent Energy Systems 26 Multivariable SPC Two correlated univariate processes y 1 (t) and y 2 (t) cov(y 1,y 2 ) = Q, Q -1 = L T L Uncorrelated linear combinations z(t) = L·[y(t)-  ] Declare fault (anomaly) if

ee392n - Spring 2011 Stanford University Intelligent Energy Systems 27 Multivariate SPC - Hotelling's T 2 Empirical parameter estimates Hotelling's T 2 statistics is T 2 can be trended as a univariate SPC variable Harold Hotelling ( )

Advanced Monitoring Methods Estimation is dual to control –SPC is a counterpart of switching control Predictive estimation – forecasting, prognostics –Feedback update of estimates (P feedback  EWMA) Cascaded design –Hierarchy of monitoring loops at different time scales Optimization-based methods –Optimal estimation ee392n - Spring 2011 Stanford University Intelligent Energy Systems 28

Research Topics: Monitoring Potential topics for the term paper. Asset monitoring –Transformers Electric power circuit state monitoring –Using phasor measurements –Next chart ee392n - Spring 2011 Stanford University Intelligent Energy Systems 29

Optimization Problem Measurements: Currents Voltages Breakers, relays State estimate Fault isolation Electric Power System model Electric Power Circuit Monitoring ee392n - Spring 2011 Stanford University 30Intelligent Energy Systems ACC, 2009

End of Lecture 3 ee392n - Spring 2011 Stanford University Intelligent Energy Systems 31