Alireza Yazdani Post-Doctoral Research Associate Department of Civil & Environmental Engineering Rice University Presented at: SAMSI Uncertainty Quantification Transition Workshop May 22nd, 2012 NETWORK TOPOLOGY UNCERTAINTY QUANTIFICATION SYSTEM PERFORMANCE EVALUATION SUSTAINABLE WATER SUPPLY MANAGEMENT 2 • Water Distribution Systems (WDS) are large complex networks of multiple interdependent nodes (e.g. reservoirs, fittings, fire hydrants) joined by links (e.g. pipes, valves, pumps). • Main system components: • • • • • Source Treatment Transmission Storage Distribution A hypothetical network representation 3 The US Water infrastructure is old, fragile and inadequate in meeting the increasing demand for water. • • Last year’s Texas drought resulted in a spike in water main breaks (CBS local, Aug 2011). • Existing centralized networks, suffer from high water age, bio-film growth, pressure loss and high energy consumption. • There is currently an underinvestment (~ $108.6 Billion). Source: (EPA, 2006 Committee on Public Water Supply Distribution Systems: Assessing and Reducing Risks, National Research Council, and 2009 Report Card for America’s Infrastructure) 4 2009 ASCE Report Card for America’s Infrastructure Aviation D Bridges C Dams D Drinking Water D- Energy D+ Hazardous Waste D Inland Waterways D- Levees D- Public Parks & Recreation C- Rail C- Roads D- School D Solid Waste C+ Transit D Wastewater D- America's Infrastructure G.P.A. = D A = Exceptional B = Good C = Mediocre D = Poor F = Failing 5 • A sustainable Water Supply System is one that supplies anticipated demands over a sensible time horizon without degradation of the source of the supply or other element’s of the system’s environment.* • Criteria: • Achieving sustainability requires integrated Reliability: •analysis adequate flow and pressure,ofavailability of the and optimization performance physical components criteria while dealing with uncertainties in • Water Quality: • Acceptable water age and chemical contents • Efficiency: • leakage management, operational efficiency and environmental impacts the data/model/natural environment * Water Distribution Systems (2011), D. Savic, J. Banyard (Eds.), ICE Press. 6 Reservoir and treatment facilities Adequacy (quality/quantity): How does water taste there? Is the pressure sufficient? Reliability: what if these pipes break together?! Efficiency: what is the cost/impacts of getting water here? A slightly reconfigured EPANET representation of Colorado Springs WDS 7 • Reducible ( epistemic) uncertainty: Resulting from a lack of information in model about the system, typically reduced through inspection, measurement or improving the analogy between the abstract model and real system • Irreducible (aleatoric) uncertainty: Natural randomness in a process, usually described by probabilistic approaches Not to be absolutely certain is, I think, one of the essential things in rationality. Bertrand Russell Image taken from: S. Fox (2011), Factors in ontological uncertainty related to ICT innovations, I. J. Manag. Proj. Busin, 4 (1), 137-149. 8 • •Model Determining the pipe size, tank diameter, network topology at design (e): inability to represent true physics of the system and itsstage behaviour • •Data Placement of sensors/control to monitor water quality (e): measurement error, valves inconsistent/inaccurate/inadequate data • • • Prediction of the physical components failure rates and evaluating failure consequences Operation (e): related to the system construction, design, equipments, deterioration, •maintenance Estimating water weekly/monthly/yearly water demand to support normal/peak consumption (a): the unpredictability of nature and its impacts on resources the system •Natural Assessing impacts of climate/demographical changes on 9 WDS Performance is largely affected by network topology Uncertainty in system performance due to the unknown/unpredictable parameters may be reduced through studying topology. Source unavailable Reservoir Reservoir Pipe Break/Contaminant Ingress Tank • • • Reliability: how often the system fails (in quantity or quality terms). Vulnerability: how serious the consequences of the failure may be. Resiliency: how quickly the system recovers from failure. 10 • Centralized treatment/operation • water quality deterioration • cost of wastewater collection • high energy loss • Decentralized treatment • shorter pipe lengths • improved water quality? • more efficient? Image from D. Kang, K. Lansey, Scenario-based Robust Optimization of Regional 11 Water/Wastewater Infrastructure,doi:10.1061/(ASCE)WR.1943-5452.0000236 Metric Spectral Graph Theory Proxy for •Fault-tolerance (design) •Rate of contaminant spread Centrality measures •Component criticality analysis •Network vulnerability to random failures/targeted attacks Path length/distances •Friction losses •Design/Operation Cost •Access between source and nodes •Water residence time Loops •Redundancy • Reliability 12 • Random networks: • • • • Small worlds: • • • Random degree distribution (equal connectivity likelihood) Network equally vulnerable to failures/attacks (typical nodes) Examples: spatial networks (no hubs, large diameter) Gaussian or exponential degree distribution Large networks with low path lengths and high clustering Scale free networks: • • • Scale-free networks/power law degree distribution Many low degree nodes with very few highly connected hubs Robust against random component failures yet fragile under targeted attacks on the hubs 13 Image: Albert, Barabasi and Bonabeau, (2003), Scale-free Networks, Scientific American, 288, 50-59. 14 Colorado Springs (CS), USA City of Houston (COH), USA Richmond Yorkshire Water (RYW), UK 15 Colorado Springs City of Houston Richmond Nodes 1786 3926 872 Links 1994 5801 957 Total pipe length (km) 117.01 3166.15 75.61 Average pipe length (m) 187.12 574.2 633.09 2.43 e-4 2.26 e-4 6.09 e-5 Average node degree 2.23 2.96 2.19 Average path length 27.23 25.94 51.44 Central-point dominance 0.42 0.34 0.56 Critical ratio of random breakdown 0.57 0.42 0.32 Graph diameter 69 72 135 Maximum node degree 4 9 4 Meshedness coefficient 0.0586 0.239 0.0495 Node (link) connectivity 1 (1) 1 (1) 1 (1) Topological efficiency 5.2 % 2.4 % 3.4 % Metric Algebraic connectivity 16 W1=1 i W1=0.5 W2=0.5 i W3=0.3 W3=0.3 i W3=0.3 W3=0.3 i W3=0.6 d=0.4 W3=0.2 Demand-adjusted entropic degree (DAED)* combines topology and physics by incorporating the number of links attached to a node, the capacity of the link connections and the way they are distributed while taking into account the demand for water at each node. * A. Yazdani, P. Jeffrey (2012), Water Resour. Res., doi:10.1029/2012WR011897, in press 17 Node ID Node ID 900 800 0 700 0 600 100 500 1 400 200 300 300 200 2 Node Degree 400 100 3 DAED 4 80 2 60 0 DAED 4 0 1800 1600 1400 1200 1000 800 600 400 200 0 Node Degree CS RYW 500 120 3 100 1 40 20 0 18 RYW CS Colorado Springs’ top three most important nodes Richmond’s top three most important nodes ID Degree DAED Normalized DAED ID Degree DAED Normalized DAED 144 4 542.28 1 153 2 94.37 1 1229 3 354.36 0.65 20 2 75.59 0.80 1373 3 192.27 0.36 219 2 64.38 0.68 19 100% Richmond 80% Colorado Springs 40% 20% 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0% 0.0 Pr { > f } 60% Normalized DAED 20 • The analysis of WDS topology: Reduces model uncertainty and offers a computationally inexpensive and less datadependent simplified approach Helps quantifying vaguely understood qualities such as redundancy, optimalconnectivity and fault-tolerance Supports development and comparison of the alternative design and operation (e.g. Decentralized) scenarios • The UQ via studying interactions of system topology and performance (hydraulic reliability, energy use, water quality) provides theoretical support for finding sustainable solutions for water infrastructure systems planning and management (rehabilitation/design/expansion problems). • Due to the WDS specifications, data and model uncertainties, and hydraulic complexities, advanced UQ techniques (e.g. spectral methods, multiple regression and survival analysis and non-parametric statistics) have a special place in the realistic analysis of WDS vulnerability/sustainability. 21 • Performance analysis and comparison of the centralized, decentralized and hybrid layouts in terms of water quantity and quality • Analysis of historical failure data to develop component/system failure rate models serving reliability analysis • Investigating the role of network topology (in the presence or absence of shut off valves) in facilitating mass transport/preventing the spread of contaminants within the system validated by the EPANET models 22 • Rice University Shell Centre for Sustainability • SAMSI for the travel support • Dr. Leonardo Duenas-Osorio and Dr. Qilin Li of Rice University Civil and Environmental Engineering 23