Review Article
1) Article Reference
Berry, J., Fleisher, L, Hart, W. Phillips, C. and Watson, J.-P. (2005) “Sensor Placement in Municipal Water Networks”, Journal of Water Resources Planning and Management, 131(3) pp. 237-243
2) Summary
This paper presents an approach for determining the placement of contaminant sensors within a municipal water network. To increase the protection of water supply systems, the use of real time early warning systems (EWS), have been widely adopted. Usually, utilities wish to place online sensors so deployment cost is minimized and the level protection maximized.
Considering a variety of approaches, the paper assumes that a attack occurs only in a single point; consider the total population exposed; the sensors protect downstream populations; and transitions between time periods are ignored. With this assumptions the authors simplify health impacts, ignore concentration and temporal effects.
The objective of the model was to minimize the expected fraction of the population that is at risk for some attack. The attack is modeled as a release of contaminant at a single point of the network. It is assumed that any point downstream of the attack can be contaminated. The EPANET water network simulator was used to determine water flow in the network. The attack scenarios were defined by a probability distribution over all pairs of population weighted flow and attacks points (from experts and scenario development).
To solve the optimization problem it was used mixed integer programming were the objective was to minimize the expected number of exposed people. The constraint considered that if a node is directly attacked it is directly contaminated. The second relate that a sensor can cover flow in a pipe in both directions. The third propagates the contaminant along the flow. One constraint is used to limit the maximum number of sensors.
To evaluate the model it was used three data sets. Two dataset were develop in EPANET and one is real. As all three dataset have some missing data, synthetic data was used to complete the datasets. The EPANET was used to determine flow patterns during four six-hour period with a 24 hour time period. The dataset 1 was adapted from Example Network provided by EPANET 2.0 with 36 nodes, 40 pipes and 1 pump station. The dataset 2 was adapted from Example Network 3 from EPANET 2.0 with 97 nodes, 117 pipes, two reservoirs and three tanks. The dataset 3 was adapted from a real world data with 470 nodes, 621 pipes, three pumps and four tanks.
The results are presented numerically in three tables. For dataset 1 and 2 the maximum number of sensors was 7 while for dataset 3 the maximum was 300. The authors present a comprehensive discussion about the numerical results and some explanations about the unlikely or unexpected outcomes. The conclusions showed that mixed integer programming can be used effectively to solve large scale sensor placement problems. Some ideas of the model generalization are presented relating the temporal effects, placement locations, sensor costs, and performance objective.
3) Discussion
I found this article useful to better understand optimization problems and it s applications and more specifically the integer programming method. I found it very similar to the previous paper. I couldn’t find any problems with the problem proposed. Further research suggestions would be to apply this methodology to real real life problems.
The objective of the model was to minimize the expected fraction of the population that is at risk for some attack. The attack is modeled as a release of contaminant at a single point of the network. It is assumed that any point downstream of the attack can be contaminated. The EPANET water network simulator was used to determine water flow in the network. The attack scenarios were defined by a probability distribution over all pairs of population weighted flow and attacks points (from experts and scenario development).
To solve the optimization problem it was used mixed integer programming were the objective was to minimize the expected number of exposed people. The constraint considered that if a node is directly attacked it is directly contaminated. The second relate that a sensor can cover flow in a pipe in both directions. The third propagates the contaminant along the flow. One constraint is used to limit the maximum number of sensors.
To evaluate the model it was used three data sets. Two dataset were develop in EPANET and one is real. As all three dataset have some missing data, synthetic data was used to complete the datasets. The EPANET was used to determine flow patterns during four six-hour period with a 24 hour time period. The dataset 1 was adapted from Example Network provided by EPANET 2.0 with 36 nodes, 40 pipes and 1 pump station. The dataset 2 was adapted from Example Network 3 from EPANET 2.0 with 97 nodes, 117 pipes, two reservoirs and three tanks. The dataset 3 was adapted from a real world data with 470 nodes, 621 pipes, three pumps and four tanks.
The results are presented numerically in three tables. For dataset 1 and 2 the maximum number of sensors was 7 while for dataset 3 the maximum was 300. The authors present a comprehensive discussion about the numerical results and some explanations about the unlikely or unexpected outcomes. The conclusions showed that mixed integer programming can be used effectively to solve large scale sensor placement problems. Some ideas of the model generalization are presented relating the temporal effects, placement locations, sensor costs, and performance objective.
3) Discussion
I found this article useful to better understand optimization problems and it s applications and more specifically the integer programming method. I found it very similar to the previous paper. I couldn’t find any problems with the problem proposed. Further research suggestions would be to apply this methodology to real real life problems.
No comments:
Post a Comment