THE USE OF OPTIMIZATION MODELS IN PUBLIC-SECTOR PLANNING
By E. Downey Brill, Jr.
Management Science Vol. 25, No. 5, May 1979
Summary
This paper had many similarities to the paper about “wicked” problems that we read at the beginning of the semester. The overall jest of the paper was that optimization models are useful for public-sector planning but should not be viewed as a way of obtaining “the answer.” After outlining the shortcomings and limitations of optimization models, the author presents different uses of optimization models as a means of aiding in the planning process. He proposes that optimization models be used in conjunction with other tools such as simulation models and analytical methods as well as other models. The use of various models and methods helps create numerous different alternative solutions that can be evaluated by the decision makers. The author endorses this outlook on the use of optimization models because he believes that models cannot truly capture the nature of a complex problem and inventive and creative solutions are needed to meet all quantitative and qualitative constraints and priorities. Optimization models cannot truly create inventive solutions since this is creative human process, although they can promote creativity by proposing possible alternatives when the complexity of the problem overwhelms the planner.
Discussion
In the thirty years since this paper was published, computational speed has increased dramatically and optimization methods have evolved; however, this paper is still relevant and probably will be for at least another thirty years if not longer. The reason for its relevancy is that computers cannot be depended on to give “the answer” when solutions deal with non-quantitative variables that are often political and social in nature. That is why high-level human decision makers are needed to choose the best solution from a range of choices. Hopefully it stays this way until I’m not replace by a computer.
GA-QP MODEL TO OPTIMIZE SEWER SYSTEM DESIGN
By Tze-Chin Pan and Jehng-Jung Kao
Journal of Environmental Engineering ASCE, January 2009
Summary
The author uses a combination of genetic algorithm (GA) and quadratic programming (QP) optimization models to find alternative solutions for the design of sewer systems. The decision variables associated with the GA model were coded as chromosomes. Pipe diameters and the location of the pumping stations are decision variables which are expressed as genes within each chromosome. The QP model decision variables are the slopes and buried depths at the downstream ends of the pipes.
Alternatives were also created by using MGA. The author says, “To facilitate the analysis for the selection of an appropriate alternative, various near-optimum design alternatives were generated by applying the MGA method. The purpose of the MGA is to identify maximally different solutions which can still be regarded as good alternatives when compared to the mathematically optimal solution.”
The optimization strategy outlined above was applied to a sewer system for a residential area with a total drainage area of 260 ha, 56 nodes, and 79 links. The models gave numerous design alternatives with various costs for the planner to choose from. In situations where essential factors are excluded, the author finds that the GA-QP and DPPP alternatives may be inappropriate or infeasible and MGA alternatives provide the best solutions.
Discussion
This paper was interesting and its principles may very well be applied to a multitude of systems; however, it seems that the authors ignore many of the things that constrain real-life sewer systems. Most city sewer systems are old and often buried in unfavorable locations. Certain sites or pipe depths proposed by the model may be extremely expensive because they require special equipment to place the pipe or the land use above the pipe makes it difficult to place the pipe. For instance, if the pipe is placed in extremely rocky soil or underneath a major roadway. These addition cost are not factored into the model. In the authors defense, if these expenses were modeled as constraints it would take a lot of addition work and make the model a lot more complex.
Monday, April 20, 2009
Friday, April 10, 2009
Blog 9 billion and 63
Compromise Programming Methodology for Determining Instream Flow under Multiobjective Water Allocation Criteria
By Jenq-Tzong Shiau and Fu-Chun Wu
Journal of the American Water Resources Association
October 2006
Summary
The purpose of this paper is to quantitatively address “the problem of compromises between human water demand and instream flow requirements.” To illustrate this problem, the authors choose to model the Kaoping diversion weir in southwestern Taiwan. The diversion weir was designed to “simultaneously assure the water supply reliability and sustain the natural flow variability.” However the second objective was not meet (at least not at an acceptable standard), since the weir flow often varied significantly from the naturally flow. There are three primary demands placed on the flows of Kaoping Creek: 1) Instream flow releases; 2) Agriculture water withdraws; 3) Municipal uses. The water allocation priorities are in the same order as listed.
In an effort to optimize this multiobjective system, the authors utilize the Range of Variability Approach (RVA) to evaluate the hydrologic alterations. The RVA resulted in 32 Indicators of Hydrologic Alterations, which are integrated into a single index that allows for the optimization of the multiple conflicting objectives. The optimal operation scheme was “then determined using the compromise programming among multiple conflicting objectives.” The objective function was to minimize both hydrologic impacts and water supply shortages.
The model determined “that the current minimum flow release of 9.5 m3/s does not effectively serve to restore the natural flow variations.” It was found that if the water releases were increased the overall stream flow variations would be reduced; however, the increase in flow releases would simultaneously increase the water supply shortage ratios. If equally weighting was given to the natural flow variability and water supply reliability, the optimal instream flow is 26 m3/s.
Discussion
Did anyone else find it a little peculiar that the registered agricultural water withdraws remained nearly constant for the entire year but the diversions for municipal use was extremely variable? Typically it is the opposite. Do these people find it beneficial to irrigate all year around but consider bathing and drinking only important from May to December? I’m assuming they have some other water source, but still.
By Jenq-Tzong Shiau and Fu-Chun Wu
Journal of the American Water Resources Association
October 2006
Summary
The purpose of this paper is to quantitatively address “the problem of compromises between human water demand and instream flow requirements.” To illustrate this problem, the authors choose to model the Kaoping diversion weir in southwestern Taiwan. The diversion weir was designed to “simultaneously assure the water supply reliability and sustain the natural flow variability.” However the second objective was not meet (at least not at an acceptable standard), since the weir flow often varied significantly from the naturally flow. There are three primary demands placed on the flows of Kaoping Creek: 1) Instream flow releases; 2) Agriculture water withdraws; 3) Municipal uses. The water allocation priorities are in the same order as listed.
In an effort to optimize this multiobjective system, the authors utilize the Range of Variability Approach (RVA) to evaluate the hydrologic alterations. The RVA resulted in 32 Indicators of Hydrologic Alterations, which are integrated into a single index that allows for the optimization of the multiple conflicting objectives. The optimal operation scheme was “then determined using the compromise programming among multiple conflicting objectives.” The objective function was to minimize both hydrologic impacts and water supply shortages.
The model determined “that the current minimum flow release of 9.5 m3/s does not effectively serve to restore the natural flow variations.” It was found that if the water releases were increased the overall stream flow variations would be reduced; however, the increase in flow releases would simultaneously increase the water supply shortage ratios. If equally weighting was given to the natural flow variability and water supply reliability, the optimal instream flow is 26 m3/s.
Discussion
Did anyone else find it a little peculiar that the registered agricultural water withdraws remained nearly constant for the entire year but the diversions for municipal use was extremely variable? Typically it is the opposite. Do these people find it beneficial to irrigate all year around but consider bathing and drinking only important from May to December? I’m assuming they have some other water source, but still.
Monday, March 30, 2009
Blog 8
Neural Network-Based Simulation-Optimization Model for Reservoir Operation
By T.R. Neelakantan and N.V. Pundarikanthan
Summary
In many reservoir operations studies - such as that of Chennai, India, which is the case used for this article – complex simulation models combined with optimization models are computationally infeasible when trying to model the problem. In an effort to circumvent this problem, the authors propose a way to develop a “planning model for reservoir operation that uses a simulation-optimization approach.” To do this they use a neural network-based simulation model, which was developed for reservoir system operation, as a submodel in a Hooke and Jeeves unconstrained nonlinear programming model. The optimization model minimized the operation policies.
That’s it in a nutshell… if you want more you’ll have to wait until Wednesday when Michelle and I present. The suspense is killing you I’m sure.
Discussion
I thought this article was a little more difficult to follow than the previous few articles. Or it could be that I’m already creating excuses for why I might be clueless come Wednesday. Either way, I still don’t feel I have a great grasp on all the terms and concepts explained in the paper. In particular, I don’t really understand how the exemplars are selected or how a neural network simulation model is formed.
By T.R. Neelakantan and N.V. Pundarikanthan
Summary
In many reservoir operations studies - such as that of Chennai, India, which is the case used for this article – complex simulation models combined with optimization models are computationally infeasible when trying to model the problem. In an effort to circumvent this problem, the authors propose a way to develop a “planning model for reservoir operation that uses a simulation-optimization approach.” To do this they use a neural network-based simulation model, which was developed for reservoir system operation, as a submodel in a Hooke and Jeeves unconstrained nonlinear programming model. The optimization model minimized the operation policies.
That’s it in a nutshell… if you want more you’ll have to wait until Wednesday when Michelle and I present. The suspense is killing you I’m sure.
Discussion
I thought this article was a little more difficult to follow than the previous few articles. Or it could be that I’m already creating excuses for why I might be clueless come Wednesday. Either way, I still don’t feel I have a great grasp on all the terms and concepts explained in the paper. In particular, I don’t really understand how the exemplars are selected or how a neural network simulation model is formed.
Monday, March 9, 2009
Blog 7
Optimal Location of Infiltration-Based Best Management Practices for Storm Water Management
By Cristina Perez-Pedini, James F. Limbrunner, and Richard M. Vogel
Summary
The purpose of this study was to introduce a methodology to determine the optimal number and location of infiltration-based BMPs on a watershed to reduce peak flow flood flows at the watershed outlet (Perez-Pedini 441). Although the research can be applied to numerous watersheds, the Aberjona River watershed northwest of Boston, Massachusetts was the focus of this study. This small highly urban catchment was modeled in a spreadsheet and optimized using a genetic algorithm (GA) to determine areas within the watershed where the application of infiltration-based BMPs would be most effective in decreasing flood flows at the catchment outlet (Perez-Pedini 442). A distributed, event-based hydrologic model, along with the SCS curve number method, was employed to determine the runoff and infiltration for each of the 4,533 hydrologic response units (HRUs)
During the optimization process, the overall goal was to locate the HRUs which, if BMPs were applied, would lead to a maximum reduction in peak stream flow at the watershed outlet (Perez-Pedini 444). The end result revealed that the optimal location of the BMPs was a complex function of HRU characteristics and locations. The authors summarize the results by creating a Pareto frontier depicting the number of BMPs, which is synonymous to the project cost, and peak flow reduction.
Discussion
The research conducted by Perez-Pedini has great applicability because (1) it can be applied to nearly any watershed; (2) the analysis could be used to inform policy decisions regarding future storm water management investments; (3) the optimal number of BMPs can be constructed in stages as funds become available, while still achieving optimal reduction during each stage (i.e. When only a few BMPs are located, their optimal locations are subsets of the optimal locations of a much larger set of optimal BMPs.).
By Cristina Perez-Pedini, James F. Limbrunner, and Richard M. Vogel
Summary
The purpose of this study was to introduce a methodology to determine the optimal number and location of infiltration-based BMPs on a watershed to reduce peak flow flood flows at the watershed outlet (Perez-Pedini 441). Although the research can be applied to numerous watersheds, the Aberjona River watershed northwest of Boston, Massachusetts was the focus of this study. This small highly urban catchment was modeled in a spreadsheet and optimized using a genetic algorithm (GA) to determine areas within the watershed where the application of infiltration-based BMPs would be most effective in decreasing flood flows at the catchment outlet (Perez-Pedini 442). A distributed, event-based hydrologic model, along with the SCS curve number method, was employed to determine the runoff and infiltration for each of the 4,533 hydrologic response units (HRUs)
During the optimization process, the overall goal was to locate the HRUs which, if BMPs were applied, would lead to a maximum reduction in peak stream flow at the watershed outlet (Perez-Pedini 444). The end result revealed that the optimal location of the BMPs was a complex function of HRU characteristics and locations. The authors summarize the results by creating a Pareto frontier depicting the number of BMPs, which is synonymous to the project cost, and peak flow reduction.
Discussion
The research conducted by Perez-Pedini has great applicability because (1) it can be applied to nearly any watershed; (2) the analysis could be used to inform policy decisions regarding future storm water management investments; (3) the optimal number of BMPs can be constructed in stages as funds become available, while still achieving optimal reduction during each stage (i.e. When only a few BMPs are located, their optimal locations are subsets of the optimal locations of a much larger set of optimal BMPs.).
Sunday, March 1, 2009
Blog 6
OPTIMIZATION OF REGIONAL STORM-WATER MANAGEMENT SYSTEMS
By Pradeep Kumar Behera, Fabian Papa, and Barry J. Adams
Summary
This paper presents dynamic programming optimization methodologies which seek to minimize the cost associated with detention storage. The objective function that is minimized is constrained by two environmental constraints that the regional outlet must satisfy: 1) environmental regulations for runoff quantity and 2) environmental regulations for runoff quality. The primary cost associated with storm-water detention ponds – and the cost which are used in the objective function - is the land which the detention pond occupies and the initial construction, operation, and maintenance costs. It should be noted that an optimizing methodology is presented for determining the design parameters of a single storm-water management pond and is then expanded (using dynamic programming as mentioned earlier) to a multiple parallel catchment system (Behera 107).
Numerous inputs are entered into the model (you’ll have to read the paper if you want to know what they all are because I’m sure not going to list them all) but there is only one thing minimized and that is the cost of all detention ponds for the desired levels of runoff and pollution control.
Discussion
I really enjoyed this paper because the basic methodologies and techniques used to solve the optimization problem can be easily applied to any other real-world system with any number of catchments (although the paper only used three). In addition, the decision variables and constraints of this model can be easily adjusted to meet different requirements of either the developer, the engineer or various government regulations – something that adds even more flexibility and applicability to the model. My understanding of DP is still incomplete but I hope by the end of the semester we will all have enough expertise in the subject to be able to implement research such as this.
By Pradeep Kumar Behera, Fabian Papa, and Barry J. Adams
Summary
This paper presents dynamic programming optimization methodologies which seek to minimize the cost associated with detention storage. The objective function that is minimized is constrained by two environmental constraints that the regional outlet must satisfy: 1) environmental regulations for runoff quantity and 2) environmental regulations for runoff quality. The primary cost associated with storm-water detention ponds – and the cost which are used in the objective function - is the land which the detention pond occupies and the initial construction, operation, and maintenance costs. It should be noted that an optimizing methodology is presented for determining the design parameters of a single storm-water management pond and is then expanded (using dynamic programming as mentioned earlier) to a multiple parallel catchment system (Behera 107).
Numerous inputs are entered into the model (you’ll have to read the paper if you want to know what they all are because I’m sure not going to list them all) but there is only one thing minimized and that is the cost of all detention ponds for the desired levels of runoff and pollution control.
Discussion
I really enjoyed this paper because the basic methodologies and techniques used to solve the optimization problem can be easily applied to any other real-world system with any number of catchments (although the paper only used three). In addition, the decision variables and constraints of this model can be easily adjusted to meet different requirements of either the developer, the engineer or various government regulations – something that adds even more flexibility and applicability to the model. My understanding of DP is still incomplete but I hope by the end of the semester we will all have enough expertise in the subject to be able to implement research such as this.
Monday, February 23, 2009
Another one...
SENSOR PLACEMENT IN MUNICIPAL WATER NETWORKS
By Jonathan W. Berry; Lisa Fleischer; William E. Hart; Cynthia A. Phillips; and Jean-Paul Watson
Summary
The findings presented in this paper were to address the concerns of the U.S. Environmental Protection Agency that most of the U.S. water supply is highly susceptible to contamination (either accidental or intentional) and that any contamination would go largely undetected, which would place a high risk on public health. The authors present a model that optimizes the placement of sensors in municipal networks to detect maliciously injected contaminants (Berry 237). The LP method utilized is a mixed-integer program that seeks to minimize the fraction of the population at risk. Three separate networks were tested (two fictional networks taken from EPANET and one real-life network). The result of the research is an MIP model that effectively solves large-scale sensor-placement problems (so claims the authors, although the effectiveness and validity of the findings can always be disputed). It was also demonstrated that noise or uncertainty in the data had very little impact on the results of the analysis.
Discussion
Although several assumptions made in the paper were a little unrealistic (constant flow path and velocity, no variance over time and space of contaminant concentration, etc.), I would deem the overall work both viable and reasonable considering the purpose of the research. I found the paper to be a good follow-up to the previous discussion paper.
The authors speak on how this knowledge will be easily transferable to real-world large-scale sensor-placement problems. I wonder how many public water supply companies 1) are aware of this research; and 2) would know how to go about implementing this work if they did… maybe they could hire us!
By Jonathan W. Berry; Lisa Fleischer; William E. Hart; Cynthia A. Phillips; and Jean-Paul Watson
Summary
The findings presented in this paper were to address the concerns of the U.S. Environmental Protection Agency that most of the U.S. water supply is highly susceptible to contamination (either accidental or intentional) and that any contamination would go largely undetected, which would place a high risk on public health. The authors present a model that optimizes the placement of sensors in municipal networks to detect maliciously injected contaminants (Berry 237). The LP method utilized is a mixed-integer program that seeks to minimize the fraction of the population at risk. Three separate networks were tested (two fictional networks taken from EPANET and one real-life network). The result of the research is an MIP model that effectively solves large-scale sensor-placement problems (so claims the authors, although the effectiveness and validity of the findings can always be disputed). It was also demonstrated that noise or uncertainty in the data had very little impact on the results of the analysis.
Discussion
Although several assumptions made in the paper were a little unrealistic (constant flow path and velocity, no variance over time and space of contaminant concentration, etc.), I would deem the overall work both viable and reasonable considering the purpose of the research. I found the paper to be a good follow-up to the previous discussion paper.
The authors speak on how this knowledge will be easily transferable to real-world large-scale sensor-placement problems. I wonder how many public water supply companies 1) are aware of this research; and 2) would know how to go about implementing this work if they did… maybe they could hire us!
Monday, February 16, 2009
Clean Water Monitoring Stuff
OPTIMAL LOCATIONS OF MONITORING STATIONS IN WATER DISTRIBUTION SYSTEM
By Byoung Ho Lee and Rolf A. Deininger
Summary
The passing of the Safe Drinking Water Act in 1974 required monitoring to be instituted in water distribution systems around the country; however, no guidance was given on how the sampling was to be executed. The authors address the best methods to locate monitoring stations in a water distribution network. The paper presents these methods by 1) defining new or relevant concepts 2) providing two examples that demonstrate the methods and concepts in real-world practice. The authors claim that the best set of stations is one that maximizes the coverage (as defined in the paper).
Discussion
This paper seems like it has lots of real-world applicability – after all, nearly every water distribution system in the US is required to monitor and I would bet that most of them are not optimizing their coverage. It has been nearly 17 years since this article was written and I would be interested if any more work has been done optimizing monitoring stations – not just research but have communities actually used this knowledge to expand their system’s coverage. If this information has merely lain on the shelf unused, I would be curious as to why. Is it too costly to implement? Are communities unaware that their systems are relatively ineffective?
By Byoung Ho Lee and Rolf A. Deininger
Summary
The passing of the Safe Drinking Water Act in 1974 required monitoring to be instituted in water distribution systems around the country; however, no guidance was given on how the sampling was to be executed. The authors address the best methods to locate monitoring stations in a water distribution network. The paper presents these methods by 1) defining new or relevant concepts 2) providing two examples that demonstrate the methods and concepts in real-world practice. The authors claim that the best set of stations is one that maximizes the coverage (as defined in the paper).
Discussion
This paper seems like it has lots of real-world applicability – after all, nearly every water distribution system in the US is required to monitor and I would bet that most of them are not optimizing their coverage. It has been nearly 17 years since this article was written and I would be interested if any more work has been done optimizing monitoring stations – not just research but have communities actually used this knowledge to expand their system’s coverage. If this information has merely lain on the shelf unused, I would be curious as to why. Is it too costly to implement? Are communities unaware that their systems are relatively ineffective?
Subscribe to:
Posts (Atom)
