Simulation Optimization for Threat Management in Urban Water

Simulation – Optimization for Threat Management in Urban Water Systems Sarat Sreepathi North Carolina State University Internet 2 – SURAgrid Demo Dec 6, 2006

Our Team � North Carolina State University � Mahinthakumar, Brill, Ranji (PI’s) � Sreepathi, Liu (Grad Students) � Zechman (Post-Doc) � University of Chicago � Von Laszewski (PI) � University of Cincinnati � Uber (PI) � Feng (Post-Doc) � University of South Carolina � Harrison (PI) Greater Cincinnati Water Works 2

Water Distribution Security Problem 3

Water Distribution Problem 4

Why is this an important problem? Potentially lethal and public health hazard Cause short term chaos and long term issues Diversionary action to cause service outage Reduction in fire fighting capacity Distract public & system managers 5

What needs to be done? �Determine �Location of the contaminant source(s) �Contamination release history �Identify threat management options �Sections of the network to be shut down �Flow controls to � Limit spread of contamination � Flush contamination 6

DDDAS Aspects �Dynamic Data Driven Application Systems �Dynamic �Data �Optimization �Simulation �Workflow �Computer Resources �Data Driven and Vice Versa �Water Demand Data �Water Quality Data 7

Key DDDAS Developments �Algorithm and Model Development � Dynamic Optimization � Bayesian Data Sampling and Probabilistic Assessment � Model Auto Calibration � Model Skeletonization � Network Assessment using Back Tracking �Middleware Development � Adaptive Workflow Engine � Adaptive Resource Management � Controller Designs �Cincinnati Application Scenario Development � Source Identification � Sensor Network Design � Flow control design 8

Water Distribution Network Modeling Solve for network hydraulics (i. e. , pressure, flow) Depends on Water demand/usage Properties of network components Uncertainty/variability Dynamic system Solve for contamination transport Depends on existing hydraulic conditions Spatial/temporal variation time series of contamination concentration 9

Source Identification Problem �Find: L(x, y), {Mt}, T 0 �Minimize Prediction Error t t �∑i, t || Ci (obs) – Ci (L(x, y), {Mt}, T 0) || �where � � � � L(x, y) – contamination source location (x, y) • unsteady Mt – contaminant mass loading at time t • nonlinear T 0 – contamination start time • uncertainty/error t Ci (obs) – observed concentration at sensors Cit(L(x, y), {Mt}, T 0) – concentration from system simulation model i – observation (sensor) location t – time of observation 10

Interesting challenges �Non-unique solutions �Due to limited observations (in space & time) Resolve non-uniqueness �Incrementally adaptive search �Due to dynamically updated information stream Optimization under dynamic environments �Search under noisy conditions �Due to data errors & model uncertainty Optimization under uncertain environments 11

Resolving non-uniqueness �Underlying premise �In addition to the “optimal” solution, identify other “good” solutions that fit the observations �Are there different solutions with similar performance in objective space? Search for alternative solutions 12

Where we are now… �Optimization Algorithms for Source Characterization � Dynamic optimization (ADOPT) – WDSA 06 � Non-uniqueness (EAGA) – WDSA 06 �Implementation � Coarse-grained parallelism � Real-time visualization � Seamless job submission on Teragrid � Simple workflow � Demo at I 2 meeting �Project Website: � www. secure-water. org 13

Preliminary Architecture Sensor Data Parallel EPANET(MPI) EPANET-Driver Optimization Toolkit Middleware EPANET Grid Resources 14

Graphical Monitoring Interface 15

Challenges �Problem complexity �Improved search algorithms for � multiple sources, non-uniqueness, dynamic source characteristics �Using Grid resources �Adaptive resource query and allocation �Adaptive work migration �Integration into workflow engine 16

What’s Next? �Dynamic optimization for determining optimal location of sensors and optimal sampling frequency �True integration of workflow engine into the cyberinfrastructure �Backtracking to improve source identification search efficiency 17

Our Cyberinfrastructure Static RF AMR Sensor Network Mobile RF AMR Sensors & Data Dat a Portal Static Water Quality Sensor Network Adaptive Workflow Adaptive Wireless Data Receptor and Controller Deci sions Adaptive Optimization Controller Optimization Engine Resource Needs Model Param eters Bayesian Monte. Carlo Engine Adaptive Simulation Controller Model Outputs Algorithms & Models Resource Availability Simulation Model Grid Resource Broker and Scheduler Middleware & Resources Grid Computing Resources 18

Questions? 19