(440c) Effect of Different Objectives in Optimal Sensor Placement on Water Network System with Better Optimization of Non-Linear Uncertain Systems (BONUS) Algorithm
Effect of Different Objectives in Optimal Sensor Placement on Water Network System with Better Optimization of Non-Linear Uncertain Systems (BONUS) Algorithm
AIChE Annual Meeting, Salt Lake City, Utah, Nov. 8-13, 2015
For submission to “Water Sustainability and Integrated Water Resource Management”
Rajib Mukherjee and Urmila M. Diwekar
Vishwamitra Research Institute, Center for Uncertain Systems:
Tools for Optimization and Management (VRI-CUSTOM),
2714 Crystal Way, Crystal Lake, IL 60012
International Clean Water Institute
South Carolina State University
Distributed water networks in cities and metropolitan areas comprise large networks of pipes, nodes, reservoirs etc. A contamination at a location or group of locations can spread in the whole network causing large-scale devastations. It is essential to proper detection and timely alerts any probable contamination so that the impact can be minimized. This can be achieved through the deployment of sensor in the water network that can detect any incident of contamination.
Sensors cannot be located at every possible node with nonzero demand. The cost constrains the number of sensors that can be used. It is essential to locate the sensors optimally to minimize the effect from an incident. The impact from a contamination incident can be assessed in various ways; the number of people affected, extent of contamination or the length of the pipe contaminated, mass or volume of contaminant consumed from an attack incident etc. The optimal location of the sensor to minimize the effect from an attack incident can vary with the different impact assessment methodologies. In the present work, we will show the effect of different objectives in optimal sensor placement.
Water network system has various uncertainties associated with it. The demand of water at the nodes is generally uncertain or the source of the contaminant in the network and doses of contaminant can also be uncertain. The demand uncertainty at different nodes will vary the flow pattern and the uncertainty in source and concentration of contaminant in the network will lead to variation in contaminant concentration at different nodes. These uncertainties will affect the net impact even if we assess it in various ways. In the present work uncertainties are included in the analysis to make the optimal sensor placement more robust.
A model network from EPANET (Rossman, 1993) is used to simulate the system’s steady state behavior for different base cases. Even for the best possible sensors, there is always an expected delay in information processing from the sensor. This delay will allow the population to reduce their demand if contamination is noticed by a sensor or sensors upstream. The effectiveness of sensors is included in the formulation as demand uncertainty changes. This work uses information theory to determine the most cost-effective network of on-line sensors by formulating a stochastic mixed integer nonlinear programming problem. In a stochastic non-linear optimization method involves repeated EPANET runs for large number of samples for every optimization iteration. In this work, we avoid repeated EPANET runs by using the BONUS algorithm and the reweighting scheme. The results show the importance of objective selection in optimal sensor placement under uncertainty and show the advantages of the BONUS algorithm.
Rossman, L. _1993_. EPANET users manual, Risk Reduction Engineering Lab, Environmental Protection Agency, Cincinnati.