<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Tyler Perini on Medium]]></title>
        <description><![CDATA[Stories by Tyler Perini on Medium]]></description>
        <link>https://medium.com/@perinita?source=rss-f44c82829945------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*jexh7xpNKVyJyf4iDpbzrg.jpeg</url>
            <title>Stories by Tyler Perini on Medium</title>
            <link>https://medium.com/@perinita?source=rss-f44c82829945------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 15:30:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@perinita/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Personalized optimal diets for Chinese breastfeeding women by linear programming models]]></title>
            <link>https://perinita.medium.com/personalized-optimal-diets-for-chinese-breastfeeding-women-by-linear-programming-models-59586383936a?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/59586383936a</guid>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Wed, 27 Apr 2022 16:50:16 GMT</pubDate>
            <atom:updated>2022-04-27T16:56:07.057Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by student </em><strong><em>Yiyi Yang</em></strong><em>, edited by Tyler Perini.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/638/1*4UBBM-TQL6YLKaOWAtpxlA.jpeg" /><figcaption>Lactating Women Diet Illustration (Feng, 2020)</figcaption></figure><p>Lactating mothers must eat a high-quality diet to replenish nutrient stores depleted during pregnancy and ensure adequate nutrition for breastfeeding. However, multiple studies in China have found that breastfeeding women’s food quality and nutritional consumption are inadequate. The goal of this study was to use a dietary modeling method to create tailored optimal meals for breastfeeding mothers in urban China that satisfied their nutrient needs.</p><p>During the Maternal Infant Nutrition Growth study, data were collected from 576 lactating women aged 0 to 240 days postpartum in three cities: Beijing, Guangzhou, and Suzhou, between 2011 and 2012. An interviewer-administered 24-hour survey was used to collect data on dietary intake. Linear programming was used to create meal plans that met the China Food Reference Intakes 2013 and the Chinese Dietary Guideline 2016 guidelines for lactating women with the least amount of variance from the observed dietary intake.</p><p><strong>Model</strong></p><p>The <a href="https://doi.org/10.1186/s12884-018-2008-6">Yu (2018) study</a> aims to develop an optimized diet plan for each lactating woman participant with least amount of deviation from their observed diet behavior. According to previous studies, the objective function is described as the sum of the absolute values of differences between the intake of each food group in the optimized diets and that in the observed diets:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/268/1*LLDcNMNvAfFlCF6jb-Wtxw.png" /><figcaption>where Y is the objective function to minimize, xOPT denotes the quantity (in grams) of food group <em>i</em> and xOBS is the quantity (in grams) of food group <em>i</em> intake in the surveyed food data.</figcaption></figure><p>Since Y is not linear, it is transformed into a linear function in the following way. New decision variables were created and defined to represent positive (P1 to P41) and negative (N1 to N41) deviation from the observed food group quantity:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/589/1*CTWOAV5ZTx-PD24lXk8y2w.jpeg" /></figure><p>The new linearized function, called Y’, is expressed as the sum of deviational variables, which is to be minimized:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/178/1*r6Ci1jqt_ekyaDAe2AkT-w.png" /></figure><p>In terms of constraints, there are two sets — <strong>nutrient constraints</strong> (Table 1) and <strong>food group intake constraints</strong> (Table 2). Nutrient constraints include energy intake per day, macronutrients (carbohydrate and fat), and other nutrients (protein, vitamins, trace elements, fiber, etc.). Lower limits are based on Recommended Nutrient Intake (RNI) and Adequate Intake (AI) or AMDR. Upper limited are based on AMDR, Tolerable Upper Intake Level (UL) or Proposed Intakes for Preventing Non-Communicable Chronic Diseases (PI-NCD). Food group intake have upper limits defined over 41 food group codes in the Chinese Food Composition Table.</p><p><strong>Results</strong></p><p>Individually optimal diets for 576 lactating mothers were established using dietary modeling. All of the food and nutrient intake constraints imposed in the linear programming models were met by the optimal diets. The huge disparity between observed and optimal diets shows that breastfeeding mothers in China may only be able to meet their nutritional needs after significant dietary adjustments. Furthermore, the investigation revealed that meeting the required intake for six elements, including vitamin A, vitamin B1, vitamin B6, calcium, selenium, and dietary fiber, was difficult. Furthermore, K-means cluster analysis showed four clusters in the improved diets. The four clusters confirmed that linear programming-developed optimal diets might define food patterns by geographical region and lactation duration.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/832/1*jEkGj5sYFgnVi_oHvhRy2Q.png" /></figure><p><strong>Sources</strong></p><p>Feng, L. (2020). <em>Lactating Women Diet Illustration</em>. When I Couldn’t Be With My Sister After Childbirth, We Practiced Zuo Yue Zi From Afar. Retrieved April 22, 2022, from <a href="https://www.bonappetit.com/story/zuo-yue-zi.">https://www.bonappetit.com/story/zuo-yue-zi.</a></p><p>Yu, K., Xue, Y., Zhao, W., Zhao, A., Li, W., Zhang, Y., &amp; Wang, P. (2018). Translation of nutrient recommendations into personalized optimal diets for Chinese urban lactating women by Linear Programming models. <em>BMC Pregnancy and Childbirth</em>, <em>18</em>(1). <a href="https://doi.org/10.1186/s12884-018-2008-6">https://doi.org/10.1186/s12884-018-2008-6</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=59586383936a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Integer Programming for Cohort Detection]]></title>
            <link>https://perinita.medium.com/integer-programming-for-cohort-detection-e6bc139d2d1c?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6bc139d2d1c</guid>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Wed, 27 Apr 2022 16:32:40 GMT</pubDate>
            <atom:updated>2022-04-27T16:32:40.036Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by student </em><strong><em>Alex Zalles</em></strong><em>, edited by Tyler Perini.</em></p><p>The Covid-19 pandemic brought into focus a topic that epidemiologists have long studied: how we naturally cluster ourselves into social cohorts, where there are many interactions within a cohort, but relatively fewer interactions between cohorts. The 2021 paper titled “Methods in Constrained Community Detection: An Integer Optimization Model and Heuristic Approach for Cohort Creation”, published in<strong> </strong><a href="https://ieeexplore.ieee.org/xpl/conhome/9732060/proceeding"><em>2021 Eighth International Conference on Social Network Analysis, Management and Security (SNAMS)</em></a><em>,</em> outlines how Operations Research can be utilized to assist in the reopening of offices in the midst of the Covid-19 pandemic. Particularly, modeling the workspace and the workers as cohorts, and seeing how one can best create and adjust cohorts to minimize the interactions between people.</p><p>However, inherent in this problem is not just one modeling question but two. We must both minimize interactions when considering solely individuals, but also minimize interactions between cohorts themselves. In this article, we will examine how the paper artfully walks this line, utilizing two different programs to attack the same problem.</p><p>This problem is focused on the creation of cohorts, specifically to battle the coronavirus when returning to work or school. In settings where people can be split up into smaller groups, this problem concerns what the sizes of the cohorts should be and how they should be created. Below is a visual, where the dots represent the people, and the circles represent the different cohorts that can be created.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/265/1*1yy2gzrauHHWVf7skejHRg.jpeg" /></figure><p>While this problem focuses on the covid pandemic, it can be applied to many situations. Given certain parameters for a group of individuals, how can we break up that group into smaller groups to minimize the interactions between people? And then, how can we also minimize interactions between groups themselves? Our model for this problem is actually two different models: one for individuals only and the second for individuals and cohorts.</p><p>Before we approach this problem, let us outline some research that has been done surrounding the topic at hand. First, this problem of cohort creation is actually quite nuanced, particularly because many of the existing programs and methods are not <em>flexible</em> enough to accommodate many different types of information. In this instance, models often restrain what constraints are included and instead use constraint programming, where a feasible set is selected out of a larger general set with constraints that were not included in the original program. So, this paper aims to confront this issue and make the solution method more generalizable and flexible.</p><p><strong>Variables and Constraints</strong></p><p>The first problem has two decision variables. First, <em>x(p,t,c)</em> is a binary variable representing whether an edge exists between people <em>p</em> and <em>t</em> in cohort <em>c</em>, where <em>x(p,t,c)=1</em> if it exists and <em>0</em> otherwise. The second binary variable is <em>y(p,c)</em>, which represents whether person <em>p</em> is in cohort <em>c</em> (<em>=1</em> if <em>p</em> is in it and <em>0</em> if not). Our objective function minimizes the sum the individual “weights” of each edge, in this instance the amount of interaction between <em>p</em> and <em>t</em>.</p><p>Some simple constraints for this problem is that the sum of <em>y(p,c)</em> over all the cohorts should equal <em>1</em> for all people, meaning that each person should be in exactly one cohort. One advanced constraint is <em>x(p,t,c) ≤ (y(p,c) + y(t,c)) /2</em>, which is the same as saying that when <em>p</em> and <em>t</em> are not both in cohort <em>c,</em> then there cannot be an edge between them in <em>c</em>, as <em>x(p,t,c)</em> is forced to <em>0</em>.</p><p>The second problem has only one decision variable, <em>z(i,j)</em>, which is <em>1</em> if <em>i</em> and <em>j</em> are in different clusters and <em>0</em> if they are in the same cluster Here, <em>i</em> and <em>j</em> can represent <em>either people or cohorts</em>. A simple constraint in this problem is <em>z(i,j) = 1</em> if <em>i</em> and <em>j</em> are cohorts, meaning that two cohorts cannot be in the same cluster.</p><p>One counterintuitive concept from the second linear program was the idea that the i and j in z(i,j) are both from the union of the people and the cohorts. In order to make sense of this, first observe that the set containing people (P) and the set containing cohorts (C ) are disjoint, meaning that there is no intersection between them. So, i and j being in the union of the two sets just means i and j can be in either P or C, but never both. Below is a venn diagram to illustrate this concept.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/199/1*GE0w2Wtl-nnSCjiyTQihgw.jpeg" /></figure><p>In this venn diagram, the areas of importance are the intersections of each state. So, there are 4 states, each possible state regarding one index intersecting with all the states of the other index, resulting in 4 possible intersections: <em>i</em> and <em>j</em> are both cohorts, <em>i</em> is a cohort and <em>j</em> is a person, <em>i</em> is a person and <em>j</em> is a person, or <em>i</em> is a person and <em>j</em> is a cohort. So, we can see through these intersections that our constraints on <em>z(i,j)</em>, whatever they may be, will check all of these possible cases, meaning not only will we be checking the relative locations of the cohorts in the clusters, but also the relative locations of people with other people, and even people and cohorts themselves, as long as <em>i</em> and <em>j</em> are in one of the 4 cases.</p><p>With this information we can actually reconstruct our initial diagram. Yes, people are in cohorts, but cohorts themselves can be in clusters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/229/1*Tec0DW8hNsT6eMGtfyN0jw.jpeg" /></figure><p>Here, we still have red cohorts, but now blue clusters group the clusters, too. This gives us the best visual of what this problem is trying to accomplish: not only grouping individuals into cohorts but cohorts then into clusters. Thus, the problem concerns an even more important question: given a group of individuals, how can we make groups of these people, and then subgroups of those groups, to minimize interactions between these people. This complicated question is then handled beautifully by the integer program, a great example of the power of operations research and modeling.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6bc139d2d1c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Operations Research Application for Home Health Care Scheduling]]></title>
            <link>https://perinita.medium.com/an-operations-research-application-for-home-health-care-scheduling-4654d19a64b8?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/4654d19a64b8</guid>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Mon, 18 Apr 2022 00:17:23 GMT</pubDate>
            <atom:updated>2022-04-18T00:17:23.293Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by student </em><strong><em>Peter Shih</em></strong><em>, edited by Tyler Perini.</em></p><p>According to Medicare.gov, “Home health care (HHC) is a wide range of health care services that can be given in your home for an illness or injury.” HHC boasts many benefits, which include reducing overcrowding in hospitals and lowering healthcare costs for patients. However, HHC presents logistical difficulties as it adds complexity to the supply chain of care and also adds many changing factors that are related to the quality of care. Instead of healthcare workers being very close to patients in a hospital setting, patients being treated through HHC must have HHC workers come to their homes in order to receive treatment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/524/1*pQm84Prwxgtm3z9VULhjSg.jpeg" /><figcaption><a href="https://allurehomecare.com/index.php/blog/5-benefits-recieving-care-home-health-care-agency">Source</a></figcaption></figure><p>In the 2015 paper, “A New Optimization Approach for a Home Health Care Problem,” Aianae Daouia et al. apply an operations research (OR) approach to optimize the routes that HHC workers travel to treat their patients. An optimal set of routes in this context constitutes minimizing the total time traveled by all workers. Because of the rapid growth of HHC as a health care sector, this application of OR is very relevant. The problem is formulated in the paper as a scheduling problem. In scheduling problems, the objective is to assign workers (or machines) to a task or a time slot in such a way that the result is optimized.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*crLwcnJH0WIx6tbsh3yN7g.png" /><figcaption><a href="https://www.servicetitan.com/features/service-scheduling-software">Source</a></figcaption></figure><p><strong>A Closer Look at the Mixed Integer Linear Program (MILP)</strong></p><p>The MILP model has 3 decision variables:</p><p>The first binary variable <em>x(i,j,k) </em>is equal to 1 if the resource <em>k</em> visits patient <em>i</em> before patient <em>j</em>, and 0 otherwise, constraints 1 and 2ensure that every patient is visited by the requested HHC workers. Constraint 3 ensures that the routes that each HHC worker takes end at the HHC headquarters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*232vPV1BxN-zFtStrLRq6A.png" /></figure><p>The second binary variable, <em>y(k,r,i)</em>, is equal to 1 if patient <em>i</em> is visited by resource <em>k</em> before <em>r</em>, and 0 otherwise. Constraints 4–6 ensure that each HHC worker will visit all the patients that they are assigned to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZyaR8jvPHtQ4dtTwwVDrBA.png" /></figure><p>Finally, <em>z(i,k) </em>represents the start time of the care of the patient <em>i</em> by the resource <em>k. </em>In this model, the scheduling takes place over a day, so the lower bound is 0 and the upper bound is 24. Constraint 7 ensures that all HHC workers begin at the HHC headquarters. Constraint 8 ensures that all patient care is completed within the allocated time period.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d7kINiGozTzf4LWJP2SxGQ.png" /></figure><p>Graph 1 shows the flow of constructing the mathematical model in three layers, which represents how the constraints are dependent on the problem that is being solved. Likewise, the output of the model is dependent on the constraints. Layer 1 is the problem that the model aims to solve. Layer 2 includes all the constraints. Layer 3 is the output, the optimal scheduling.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*amJj9Hntqng_PYDJnD4toQ.png" /></figure><p>Aiane Daouia, Adnen El Amraoui, Khaled Mesghouni. A New Optimization Approach for a Home Health Care Problem. 6th IESM Conference, Oct 2015, Séville, Spain.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4654d19a64b8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[School Transportation Budgets]]></title>
            <link>https://perinita.medium.com/school-transportation-budgets-dcc76b9ff035?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/dcc76b9ff035</guid>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Sat, 16 Apr 2022 03:18:56 GMT</pubDate>
            <atom:updated>2022-04-16T03:18:56.549Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by students </em><strong><em>Matthew Anderson and Franco Gomez</em></strong><em>, edited by Tyler Perini.</em></p><p>It’s no secret that public schools over America can and have faced issues with their budget. In lieu of the ongoing pandemic, it has become even more apparent that job security is not a guarantee, and unforeseen circumstances can cause schools to scramble to find ways to allocate their budget effectively. The <a href="https://arxiv.org/pdf/1803.09040.pdf">research paper by Zeng, Chopra, and Smilowitz</a> focuses on how to help public schools have more control over their budget by reducing their transportation costs. To state the obvious, fewer school buses are cheaper than more. The purpose of the article is to create an integer linear program (ILP) formulation by which public schools can reduce costs by minimizing the number of required school buses by staggering start times and utilizing multiple routes. The savings from effective optimization algorithms can be profound; by implementing the biobjective algorithm, the Boston Public Schools reduced public school transportation costs by $5 million (Bertsimas 2019).</p><p>There are two related problems of interest to study. The school bus scheduling problem (SBSP) is a subproblem of the school bus routing problem (SPRP), and SBSP aims to “optimize school start time and bus operation times to minimize transportation cost.” In this research paper, Zeng, Chopra, and Smilowitz present a time-indexed ILP with the aim of “minimizing the number of buses to serve all bus routes such that each route arrives in a time window before school starts.” The hope of this study is to start with a set of routes and buses to generate an ILP meant for a small set of public schools and then implement a randomized rounding algorithm in order to apply the findings to a larger scale scale to larger problems.</p><p>Shifting our focus to the model for the SBSP , there are two binary decision variables included. The first is <em>x(i,t)</em>, which is equal to 1 if route <em>i</em> arrives at time <em>t</em>, and 0 otherwise. A set of routes and a school start times are given. These times are not continuous and are contained as intervals. (For instance, a time of 7:12:43 AM is unreasonable, but 7:15 AM is.) Say a school bus takes a particular route; then the binary decision variable is “on” (equal to 1) if that school bus will arrive at its school destination at a specific time given a particular route. Otherwise, we say that variable is “off” (equal to 0).</p><p>Another binary decision variable in this study is the binary variable <em>y(s,t)</em>, which is on if school <em>s</em> starts at time <em>t</em>, and off otherwise. These variables are indexed by the same set of time intervals but also an additional set of public schools.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/650/1*71gDyVQ7QCqFYxPWtjXm5A.png" /><figcaption>Binary decision variables are like a light switch! The lightbulb is turned on when the value equals 1. Conversely, when the value equals 0, the lightbulb is turned off. When their conditions are met, binary decision variables turn “on,” and they can not only control lights but also other mechanisms in the model. However, when their conditions are not met, they are turned off, and they don’t directly impact the model — there’s only darkness.</figcaption></figure><p>One final decision variable <em>z</em> represents the maximum number of school buses needed. This is more representative of the objective for this model, but since it’s linked with a constraint we will discuss, it’s worth a brief mention. The objective of the model is to minimize the maximum number of school buses needed, i.e., to minimize <em>z</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/934/1*Ejr2AoZGqaUPs6q9Dqj4kQ.png" /></figure><p>Now, to get into the ILP model:</p><ul><li>The first two constraints are assignment constraints. (1a) makes sure that each route is assigned to a start time, and (1b) makes sure that each school is assigned to a start time. Intuitively, this makes sense because both <em>x </em>and <em>y </em>are binary variables, and the constraints make sure that that exactly one of each variable is turned on for all routes and all schools in the program. It does this by setting the sum of the binary variables across all the times equal to one for each route and for each school. Notice that the binary decision variable cannot be on for multiple times, so each unique route and school have unique start times.</li><li>(1c) is a time constraint that guarantees each bus route will reach the school at a time before school starts. In the summation indices of (1c), t’≥t presents an arbitrary time and <em>l(s) </em>is the length of the time window it takes for a route to arrive at school. For example, it is good if a school starts at a time t’ when <em>x(i,t) </em>is turned on because it ensures that the bus will reach the school before school starts. If that school doesn’t start at <em>t</em>, then the school should start at <em>t+1</em>, or <em>t+2</em>, etc. up until <em>t+l(s)</em>. On the flipside, if <em>x </em>equals one but an appropriate <em>y </em>is never turned on because the school doesn’t start at any of the particular times, then the constraint will not be satisfied, which will be helpful in determining whether the school needs to shift its start time or add more buses.</li><li>The fourth constraint (1d) links the decision variables to the objective <em>z</em>. This constraint introduced the variable <em>r(i)</em>, which denotes the length of route <em>i</em>. With this in mind, we can see that the left side of (1d) sums the total number of routes in operation within the feasible time interval. This constraint is what enables the model to minimize the <em>maximum</em> number of busses required.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/592/1*Meob4H-dcdU7Oze7kGGA4Q.png" /><figcaption>Essentially, constraint (1d) sets the maximum number of buses required equal to the maximum number of routes in operation during the same time interval. In this example, three buses are assigned to three routes. Each bus can only travel that specific route. In other words, we want at least one bus to be assigned to one route.</figcaption></figure><p>This ILP was the first one presented in the article and thus is the simplest version of the model. It incorporates restrictions on school start times and route operation times by fixing part of the decision variables, sets appropriate bounds, and accounts for school start times. However, the complexity of the model makes it difficult to solve. To account for this, the authors present two more complex ILP models. The second version of the ILP makes the model easier to solve and the third ILP uses a randomized rounding algorithm that generalizes the model and implements it at a larger scale.</p><p>Although this paper focuses on minimizing the maximum number of school buses needed, there are certainly other objectives that can help reduce costs, such as optimizing routes and bus stop locations. Nevertheless, the authors provide an in-depth look at multiple possible models to reduce costs by optimizing the number of buses needed. School districts that are struggling with their budget, especially in light of the ongoing pandemic, should consider the methodology discussed in this research paper.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dcc76b9ff035" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Optimizing hydroelectric power production in an environment plagued by climate change]]></title>
            <link>https://perinita.medium.com/optimizing-hydroelectric-power-production-in-an-environment-plagued-by-climate-change-27a9800b7d2a?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/27a9800b7d2a</guid>
            <category><![CDATA[hydroelectric-power]]></category>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[mathematical-modeling]]></category>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Tue, 30 Nov 2021 03:47:44 GMT</pubDate>
            <atom:updated>2021-11-30T03:47:44.606Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by student </em><strong><em>Ben Smiley</em></strong><em>, edited by Tyler Perini.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*661HGA7SSjavvzUCatijmQ.png" /></figure><p>The global energy market has been heavily reliant on the burning of coal and oil for over a century, and they have become one of the largest contributors to climate change. The serious threat posed by a one degree increase in average global temperatures has initiated an international search for methods of power production that are less harmful to the environment. Many of these methods, such as hydropower, utilize the naturally occurring systems of the earth. However, the results of climate change thus far have already disturbed such systems, making traditional methods of hydroelectric production unreliable and inefficient. In order to continue a global shift towards renewable energy, hydroelectric production must adapt to the changes in the hydrological cycle using mathematical optimization.</p><p>The article “Hydropower plant operation rules optimization response to climate change” by <a href="https://www.sciencedirect.com/science/article/pii/S036054421831363X">Jianxia Chang et. al. (2018)</a> focuses on building an optimization model for the Hanjiang River hydropower plant system. Subject to the conditions of climate change and the design of the respective plants, the objective of the model is to maximize energy production. The plant designs include cascade reservoirs, which contain a certain amount of water that is strategically used to run the plants and produce energy. The water levels in these reservoirs depend directly on the amount of inflow experienced from the river, which has become increasingly unpredictable due to glacier retreats and abnormal snowfall.</p><h3>Model</h3><p>The linear programming (LP) model uses the following four decision variables:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GvRNBwNUO8bDFnrSzoe_hQ.png" /></figure><p>The following are three constraints in the model:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/660/1*EtasMVvqPl-Pxtb3FVIsLA.png" /></figure><p>First, the energy output of plant <em>i </em>during period <em>t </em>must meet a base level energy production required by the population to function normally (<em>Nmin</em>). However, energy production cannot surpass the maximum level the plant was designed to produce (<em>Nmax</em>). Second, Second, the plant must maintain a minimum discharge flow of <em>Qmin</em> but is unable to surpass the maximum potential discharge of the plant, <em>Qmax</em>. Third, the amount of water within the reservoir at the end of period <em>t</em> is equal to the amount of water present that the beginning of the period plus the difference between the inflow and outflow of the reservoir (q-Q) multiplied by the difference in periods.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*Copf9NwtWtq21UQLUZYjtg.png" /></figure><p>Chang, J., Wang, X., Li, Y., Wang, Y., &amp; Zhang, H. (2018, July 17). <em>Hydropower Plant Operation Rules Optimization response to climate change</em>. Energy. Retrieved November 23, 2021, from <a href="https://www.sciencedirect.com/science/article/pii/S036054421831363X.">https://www.sciencedirect.com/science/article/pii/S036054421831363X.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=27a9800b7d2a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Improving Healthcare by Reducing Patient Handoffs]]></title>
            <link>https://perinita.medium.com/mp3-ad9bab4fca46?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/ad9bab4fca46</guid>
            <category><![CDATA[mathematical-modeling]]></category>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[healthcare]]></category>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Tue, 30 Nov 2021 03:25:38 GMT</pubDate>
            <atom:updated>2021-11-30T18:51:24.078Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by student </em><strong><em>Kritin Shanmugam</em></strong><em>, edited by Tyler Perini.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xIVXmbKWuT6IUx_WTEdZrg.jpeg" /></figure><p>A 1999 report in the <em>Institution of Medicine</em> [1] declared that over 100,000 deaths and nearly 15 million injuries occur each year due to medical errors. More recent reports across multiple journals account for almost triple the number of deaths and injuries. Medical errors act as one of the leading causes of death and illness in the United States, ahead of breast cancer, and they cost tens of billions of dollars to the healthcare system. An influential source of these errors are communication errors between healthcare providers. Furthermore, these errors are further related to fatigue, as a consequence of residents’ working hours. In an effort to reduce the fatigue in residents and fellows, the Accreditation Council for Graduate Medical Education, eleven years ago, executed regulations on their hourly length of shifts, amount of weekly working hours, off time between shifts, and the regularity of successive days and nights of duty. The regulations on the shifts dramatically increased the number of communication mistakes and patient handoffs between providers. A patient handoff constitutes the transfer of a patient and any necessary information from one provider to another. Even though improved patient handoffs can slightly decrease communication errors, a reduction in handoffs themselves will immensely prevent dangerous mistakes from being made. A research study by <a href="https://doi.org/10.1007/s10729-013-9237-z">Kazemian, P., Dong, Y., Rohleder, T.R. et al. (2014)</a> is unique in that it focuses on redesigning physician schedules using an integer linear program (ILP) to reduce the number of handoffs. Therefore, the objective function of the researchers’ ILP is to minimize the number of patient handoffs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zxcIWBoOE4A1fo3VIX4nJg.jpeg" /></figure><p>There are a three primary decision variables in this model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wvPf6jdT_wNDLIP9M_uPrA.png" /></figure><p>First, X is a binary variable with a purpose of assigning shifts to the trainees. The variable is relevant to the idea of limiting the weekly working hours and the length of shifts for the residents and fellows to prevent fatigue. An upper bound of X requires its summation to be at most 80/(the length of each time block in hours). This upper bound limits the weekly duty hours to 80 hours, and it is important because a trainee is likely to become dangerously fatigued if they work more than that amount. The lower bound of the summation must be greater than or equal to the minimum number of trainees needed to be in the hospital at time block <em>k </em>of day <em>j</em>. So, this constraint is ensuring that necessary coverage in the hospital is always met.</p><p>Second, W is a binary variable that is relevant to scheduling trainees to the night shift for specific days. The upper bound prevents the trainees from being scheduled for more than 4 consecutive night shifts. The parameter in this constraint is a decision of the researchers; the official restriction from the ACGME is 6 nights, but the researchers believe that extreme fatigue and insomnia is better averted by a limit of 4 nights. A large stretch of night shifts also limits the healthcare providers from properly resting, which is essential to prevent communication and fatigue-related errors. The lower bound states that W must be greater than or equal to X, for all k in the set of night time blocks available. This constraint is used to ensure that W is equal to 1 when trainee <em>i </em>works at night on day <em>j</em>.</p><p>Third, Z is a binary variable that corresponds to the days off of each trainee. There is no upper bound on this variable because an increasing quantity of days off does not lead to fatigue or insomnia. However, the summation of Z over the length of an entire month must be greater than or equal to 4, requiring that each trainee must have at least one day off per week.</p><p>The researchers also include numerous constraints in the ILP to produce implementable solutions. For instance, the maximum shift length is 16 hours for residents and 24 hours for senior trainees, and these lengths are decided by the ACGME. The first simple constraint ensures that a trainee is not scheduled for more than the maximum hours of continuous duty. The second simple constraint prevents shift changes at late nights or early mornings in an effort to allow healthy sleeping schedules for the trainees. An advanced constraint ensures that when Z(i,j) is 1, or when trainee <em>i </em>is off-duty day <em>j, </em>the trainee <em>i </em>has no scheduled shifts on that day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*In1DNurw6dD6HkpYB_4DzQ.jpeg" /></figure><p>The graphic illustrates the reasoning behind limiting the number of consecutive night shifts to 4. As the number of consecutive night shifts increases, the level of fatigue dangerously rises. In order to maintain an error-free medical environment, the physician schedule should aim to keep the number of consecutive night shifts under 4.</p><p>Integer linear programs can be implemented to improve the state of healthcare. Tens of thousands of lives and millions of injuries will hopefully be prevented through optimization strategies. As healthcare continuously grows in scale and complexity, it is essential to use operations research to ensure its efficiency and safety.</p><p>[1] Kohn LT, Corrigan J, Donaldson MS, Institute of Medicine (U.S.). Committee on Quality of Health Care in America (1999) “To err is human building a safer health system.”</p><p>[2] Kazemian, P., Dong, Y., Rohleder, T.R. et al. An IP-based healthcare provider shift design approach to minimize patient handoffs. Health Care Manag Sci 17, 1–14 (2014). <a href="https://doi.org/10.1007/s10729-013-9237-z">https://doi.org/10.1007/s10729-013-9237-z</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ad9bab4fca46" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Optimizing Evacuation of Disaster Victims]]></title>
            <link>https://perinita.medium.com/miniproject-3-c0628f49f880?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/c0628f49f880</guid>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[disaster-response]]></category>
            <category><![CDATA[mathematical-modeling]]></category>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Tue, 30 Nov 2021 02:56:31 GMT</pubDate>
            <atom:updated>2022-01-10T20:08:53.578Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by students </em><strong>Arielle Sanford and Arjun Sethi-Olowin</strong><em>, edited by Tyler Perini.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*ZeLs1K9W_XRnUXkw8Rosrg.png" /><figcaption><em>Figure 1:</em> Recent years have seen a rise in the total number of combined tropical and subtropical storms as well as hurricanes with record yearly highs reached twice in the past two decades <a href="#_ftn1">[2]</a>.</figcaption></figure><p>Natural disasters have plagued humanity for all of history. However, in recent years with the rise of climate change, major weather events such as hurricanes have come with even increasing intensity <a href="#_ftn1">[1]</a>. The number of hurricanes has also seen a slightly trended upward in recent years (see Figure 1), which is thought to be in part due to global warming. Higher intensity and greater frequency of natural disasters results in more loss of life and property damage, including destruction of critical infrastructure. To specifically address the loss of life, studies have been conducted to optimize the transport of injured civilians between disaster zones and treatment facilities, be they hospitals or temporary shelters.</p><p>The research paper<em> Integer Linear Program Approach for Evacuation of Disaster Victims with Different Urgency Levels — </em>authored by Byung Hoon Oh, Kwangyeon Kim, Han-Lim Choi, and Inseok Hwang — aims to tackle the question of transporting patients with degrees of injury varying from severe to light, which generalizes the assumption that all patients require hospital care with equal urgency. Their novel approach to this feature uses the following intuition: “different urgency levels can be posed as time constraints that vary among victims.” In a broad sense, they use previous work on the pickup and delivery problem with time window (PDPTW) to deliver all demand (patients in this case) within each demand’s time window — think of the classic problem faced by Uber Eats. The PDPTW problem has been explored previously by Baldacci et al. (2011)<a href="#_ftn1">[3]</a>. Baldacci provides a solution to minimize costs in transportation of goods by customer requests while satisfying time windows and capacity constraints. Building off this work, Hoon Oh et al. are able to expand said model to include priority.</p><p>The authors maximize the number of patients transported to safety in time, including constraints to model: the carrying capacity of each helicopter (as the mode of transportation); the endurance in terms of fuel of each helicopter; the varying survival times of patients with varying degrees of injury; and routing constraints to avoid unnecessary (i.e., suboptimal) helicopter flights that might refuel or pick-up/drop-off patients.</p><p>The primary decision variable is defined as x, which represents the number of trips of a vehicle from node i to node j to deliver the patients of a given urgency-category in a given cycle. This <em>cycle</em> concept is key to understanding the model: it includes all the patients serviced at various locations before having to refuel the vehicle. One cycle can be modeled as a max flow network. Then the <em>total number</em> of cycles represents the total number of refueling stops for a helicopter.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8_CT1rjIb_0w6oPYFgjAVg.png" /><figcaption>Primary functional decision variable (general integer).</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*JM6qqzmdfLtkJCNAQJBiKA.png" /><figcaption><em>Figure 2:</em> A flow network for a cycle. The B node(s) represent the base for the vehicle (duplicated for clarity). The D nodes represent locations of patients (i.e., demands), and the R nodes represent destinations for patients.</figcaption></figure><p>This network diagram illustrates an example cycle for some vehicle. The vehicle must start and end at its base, B (the node has two copies for illustration purposes). The vehicle will depart from its base and arrive at one of the demand sites, labeled D (denoted by the blue arrows). Then it must move to the demand site’s corresponding destination, labeled R (denoted by the bidirectional orange arrows). From a destination, the vehicle can move to any of the demand sites or back to its base B (denoted by the green arrows). Notably, a vehicle is <em>not allowed</em> to pick up a patient from a demand site and then head directly to its base without delivering the patient to their destination, first.</p><p>As an example, we demonstrate one potential cycle which maps the flight of a helicopter as it delivers and patients to hospitals before returning to base to refuel and embark upon its next cycle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*GbyjpMWoVQw_eq7yc7yM7w.png" /><figcaption>The helicopter starts at the base to pick up the first patient(s) at D2 and transport them to R2. Then the helicopter makes back-to-back trips to D4 to pick up patients. The helicopter’s last transportation leg is from D4 to R3, after which it returns to the base to refuel. This <em>cycle</em> traverses 7 arcs from beginning to end.</figcaption></figure><p>With this network and these decision variables, the model can now be outlined with an objective function and constraints. The objective is to minimize the total number of unserviced demands, represented by W, summed over all categories and all demand sites. The first constraint represents the number of flights into and out of a base must be the same. The second constraint represents a vehicle is only allowed to depart and arrive at a base once per refueling cycle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NrqVhrAoq92jwKVePyvaLQ.png" /><figcaption>Objective function and two functional constraints of the model.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/650/1*sxOBiThHYAR_CQBrWdWmpQ.png" /><figcaption>Remaining constraints, including an either-or condition.</figcaption></figure><p>The remaining constraints represent the conditions on the serviced, unserviced, and “excessively serviced” demands. To solve the either-or constraint, the two binary decision variables, y and z, and two sufficiently large positive constants (M) are used.</p><p><em>Figure 4</em> depicts example proportions of demand types within different demand sites. Here, the “type” of demand is a categorial measure for urgency of medical attention.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*-mse1gCM3q87RcQOdYHQdA.png" /><figcaption><em>Figure 4:</em> The proportion of demand types of differing categories at 4 demand sites (D1 through D4).</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*Y1U25R430gz4f2u5gX8BDA.png" /><figcaption><em>Figure 5:</em> Example breakdown of serviced, unserviced and excessively serviced demand by each urgency type at the first demand site.</figcaption></figure><p><em>Figure 5 </em>illustrates a single demand site with different demand types, each with serviced, unserviced and excessively served demands. The purpose of these constraints is to assign the excessive service to the next underserviced category. In this example the excessively serviced portion of demand type 2 would instead be used to service the unserviced portion of demand type 3.</p><p>Further constraints are also used to prevent redundant flight pathing of helicopters — as explored in <em>The Vehicle Routing Problem</em> by Paolo Toth and Daniele Vigo <a href="#_ftn1">[4]</a> — and to prioritize completing patients of a certain priority within their given time window before proceeding to the next priority group.</p><p>With this IP, Hoon Oh et al. provide a model in which the degree of injury is considered when saving lives. However, it ignores some crucial, practical features of disaster zones; for example, delivery sites, hospitals or temporary shelters would almost certainly have capacity constraints. Naturally, whether we transport a patient to a hospital or not, if the hospital cannot treat them, we have failed in our endeavor to save them. As a result, further research is essential, including potentially a <em>multiobjective</em> approach.</p><p>As a potential secondary objective, the capacity of delivery sites could be considered. In addition to minimizing the unserviced demand, we would also minimize the patients delivered to a site (proportional to its capacity). This way, the <em>biobjective nondominated frontier</em> would give us a clearer picture of not only how many patients are transported from danger zones, but also how many are then saved in hospitals — and, crucially, how many <em>aren’t</em>. This could give disaster responders information on how best to proceed instead of blindly transporting as many people as possible that meets fuel budgets, they instead invest excess resources to transport people to hospitals <em>more sparsely:</em> rather than overwhelming a few hospitals, this strategy would aim to transport patients to a treatment site with the space to service them.</p><p>Nevertheless, Hoon Oh et al. shows how optimization can be used to save lives in natural disasters while deadly catastrophes become more common in the face of global warming.</p><p><a href="#_ftnref1">[1]</a> <a href="https://news.sciencebrief.org/cyclones-mar2021/">https://news.sciencebrief.org/cyclones-mar2021/</a></p><p><a href="#_ftnref1">[2]</a> <a href="https://noaanhc.wordpress.com/2021/06/30/was-2020-a-record-breaking-hurricane-season-yes-but/">https://noaanhc.wordpress.com/2021/06/30/was-2020-a-record-breaking-hurricane-season-yes-but/</a></p><p><a href="#_ftnref1">[3]</a> <a href="https://pubsonline.informs.org/doi/pdf/10.1287/opre.1100.0881">https://pubsonline.informs.org/doi/pdf/10.1287/opre.1100.0881</a></p><p><a href="#_ftnref1">[4]</a> <a href="https://epubs.siam.org/doi/book/10.1137/1.9780898718515">https://epubs.siam.org/doi/book/10.1137/1.9780898718515</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c0628f49f880" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nature, Society, and Mathematical Models]]></title>
            <link>https://perinita.medium.com/nature-society-and-mathematical-models-3c0c5018e6e4?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c0c5018e6e4</guid>
            <category><![CDATA[flooding]]></category>
            <category><![CDATA[economic-models]]></category>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[economics]]></category>
            <category><![CDATA[mathematical-modeling]]></category>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Mon, 29 Nov 2021 16:09:10 GMT</pubDate>
            <atom:updated>2022-01-03T16:09:35.029Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Combining dynamic optimization with macro-economic theory to study the interaction between flooding and economic growth.</strong></p><p><em>A guest article written by student </em><strong><em>Natasha Patnaik</em></strong><em>, edited by Tyler Perini.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1ENWMzR6neEOUQDVVA7VuQ.jpeg" /><figcaption><em>Mathematical modeling can help us understand the relationship between flooding and economic growth. Photo from </em><a href="https://wallpaperaccess.com/most-beautiful-rain"><em>WallpaperAccess</em></a><em>.</em></figcaption></figure><p>The 2016 paper “Modeling the Interaction Between Flooding Events and Economic Growth” [1], published in <a href="https://www.journals.elsevier.com/ecological-economics"><em>Ecological Economics</em></a>, is a great example of how mathematical models can be used to precisely describe complex relationships. This Austrian study takes an interdisciplinary approach, drawing upon principles of macroeconomics and operations research to design a model that describes how flooding events and economic growth relate to one another.</p><p>Taking a minute to think about all the underlying natural phenomena, societal activity, and time-dependent components of such a large-scale system, we can appreciate how incredibly complicated it is to try and capture all of this nuance in just a couple of equations! With this in mind, we’ll look through some highlights of the paper, with specific attention given to the formulation of the model and its most important components.</p><h3><strong>Background and Motivation.</strong></h3><p>Firstly, before we delve into the context of our model of interest, we should take a moment to appreciate how operations research and macroeconomics have been used in-tandem here, thus giving rise to a mathematical model that draws upon standard practices from both fields. In framing <em>exogenous</em> (meaning “explicitly controlled by the researchers”) economic variables and parameters within an objective function and set of constraints, we end up with an optimization problem. Then as a direct consequence, finding the optimal solution to this problem means that we have also solved for our <em>endogenous</em> (meaning “implicit components within the model”) economic variables. With this in mind, we can move onto exploring the model presented in the current paper of interest.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/824/1*91G-RwJ3j4rEkZFFizRcyQ.png" /><figcaption>Figure 1. Combining dynamic optimization and macro-economic theory into a single mathematical model.</figcaption></figure><p>Economic growth can be thought of as an increase in the value of goods and services that an economy produces over time; this is often measured as a percentage change of “real” gross domestic product (GDP), where “real” means accounting for inflation overtime. Economic growth is traditionally considered to be an important and desirable objective to pursue, because it can pave the way for increased standards of living and development. Of course, a prevalent threat to such goals of community well-being, stability, and advancement are natural disasters or environmental shocks, such as flooding events.</p><p>Socio-hydrology models are mathematical representations of how community development, flooding damage, and long-term economic recovery relate to one another. To connect the environment and the economy in this manner, dynamic optimization models and standard macroeconomic growth models have been creatively combined in order to analyze these relationships. This operates under the assumption of some “decision maker” or “social planner” deciding how much investment should be allocated towards traditional physical capital (machines, tools, buildings, infrastructure, etc.) versus defense capital (flood control basins, levees, dams, warning systems, etc. to protect from future flooding). In this allocation process, our social planner is ultimately guided by the canonical economic objective of maximizing “utility”- an idea that encapsulates the pleasure, satisfaction, or happiness derived from consumption. Here, consumption refers to the usage of any sort of good or service (for example: the food you ate for lunch and the apartment you currently rent are both examples of economic consumption).</p><p>The catch here is that there is a tradeoff between choosing to invest more in physical capital versus defense capital. By taking the pre-emptive measure of investing in flood prevention/ mitigation strategies, an economy is better positioned to minimize flood damage, thereby protecting its ability to produce more output and at least maintain the current standard of living. But on the other hand, investment in traditional physical capital is in-and-of-itself necessary to produce a greater quality and quantity of total output. So which is best: investing in production and consumption that will make us better-off now in the current time period, or investing in protection against a potential threat to the standard of living in a future time period? How do we strike a n optimal balance? And how might this balance change as time goes on, depending on factors such as our current capital stock levels, or the frequency and prevalence of flooding events?</p><p>We can summarize this problem as follows:</p><blockquote>What is the optimal combination of consumption and investment for maximizing utility, subject to shocks and damage from flooding events?</blockquote><h3>Construction of a Model</h3><p>Below is a schematic representation of the paper’s model (modified from a diagram in the paper itself), with additional labelling, color-coding, and connections to aid in our discussion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/838/1*RGgWHn6bwB1MptwlEDN6rg.png" /><figcaption>Figure 2. Schematic diagram of the model (adopted and modified from Grames 2016), which emulates the interaction between flooding events and economic growth.</figcaption></figure><p>Before we can specifically discuss the objective function, decision variables, and key constraints in mathematical terms, we stand to benefit from first considering a big-picture view of how this model works. Let’s now deconstruct the schematic from above into its component parts:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/661/1*MwnSmhT1mu_u4zeaCggGUA.jpeg" /></figure><p>The social planner chooses the level of consumption <strong>cons </strong>[referred to in figures as c(t)], the amount of investment into physical capital <strong>inv_phys </strong>[referred to in figures as i_y(t)], and the amount of investment into defense capital <strong>inv_def</strong> [referred to in figures as i_d(t)], guided by the objective of maximizing utility. Note that consumption <strong>cons</strong> is considered a direct input of the utility-based objective function. The economy is assumed to be closed, such that all the produced economic output <strong>ec_out</strong> [referred to in figures as Y(t)] will be used up (no trade with foreign economies).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/663/1*olwZyzeCKx3TSzU02qlUMg.jpeg" /></figure><p>Physical capital <strong>cap_phys</strong> [referred to in figures as k_y(t)] includes implements such as machines or infrastructure. Economic output <strong>ec_out</strong> depends on the amount of physical capital <strong>cap_phys</strong> available for production. In turn, this output can be used either for consumption <strong>cons</strong> or investment <strong>inv_phys</strong> and <strong>inv_def.</strong> Physical capital <strong>cap_phys</strong> is augmented by investment <strong>inv_phys.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/658/1*zaRCttd9z_odbzlTzX2Y5w.jpeg" /></figure><p>Defense capital <strong>cap_def </strong>[referred to in figures as k_d(t)] includes flood prevention technologies and mitigation measures, such as dams, levees, and flood control basins. The defense capital <strong>cap_def</strong> can reduce the damage caused by flooding events, and is augmented by investment <strong>inv_def.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/747/1*3IPvXoeBspJWjwiwotkkwg.jpeg" /></figure><p>Although consumption <strong>cons</strong> is a direct input of the utility function, investment <strong>inv_phys</strong> and <strong>inv_def</strong> also affects utility via augmenting/ diminishing the capital stock <strong>cap_phys</strong> and <strong>cap_def</strong>. This impacts the economy’s ability to produce more economic output <strong>ec_out</strong> for consumption <strong>cons</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/741/1*qbrq5wr7xFgHct8WNhET_g.jpeg" /></figure><p>The occurrence of flooding events depends on the water level <strong>W(t)</strong>, which is assumed to be an exogenous function. Thus, damage is a function of both water level <strong>W(t)</strong> and defense capital <strong>cap_def</strong>, and can be denoted <strong>d(W(t), cap_def)</strong>. Flooding events damage both types of capital <strong>cap_phys</strong> and <strong>cap_def</strong>.</p><h3><strong>Decision Variables, Objectives, and Key Constraints</strong></h3><p>Our three decision variables (written in pink in Figure 2) are as follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RFkV4rAi268-mTTeqdJzfQ.png" /><figcaption>Decision variables.</figcaption></figure><p>(I) Measured in billions of dollars, <strong>cons</strong> ∈ [0, ec_out]</p><p>(II) Because <strong>inv_def</strong> is originally measured in meters (for compatibility with units used in the damage function), we need to apply transformation <strong>Q</strong> to get a cost measurement in billions of dollars. Thus, <strong>Q(inv_def)</strong> ∈ [0, ec_out − cons].</p><p>(III) Measured in billions of dollars <strong>inv_phys</strong> ∈ [0, ec_out — cons − Q(inv_def)].</p><p>As mentioned previously, our objective is to maximize utility. It should be noted that the actual objective function to be maximized contains an exponential expression, which means our mathematical program is nonlinear:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*qHThZqAGaBPZreI27eCcMQ.png" /><figcaption>Nonlinear objective function. Parameter rho<strong> </strong>denotes a time preference rate, indicative of whether utility of consumption in the current time period is, or isn’t, preferred over utility of consumption in the future. <strong>U(c(t))</strong> denotes the utility function, which takes consumption as input.</figcaption></figure><p>In observing the upper bounds of our three decision variables, we see that their sum is bounded by the value of aggregate economic output <strong>ec_out</strong>. Logically, because the economy is closed (no foreign trade) and all output is used up, the money flows generated from production of goods and services must equal total spending on consumption and investments (this principle is commonly referred to as circular flow of income). Thus arises one of the key constraints of this model: the budget constraint.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/574/1*aSU0-8GX8mEKfjdPA3P5Qw.png" /><figcaption>Budget constraint of the model.</figcaption></figure><p>Two other key constraints of interest include the equation which constrains the level of physical capital <strong>cap_phys</strong>, as well as the equation which constrains the level of defense capital <strong>cap_def</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*uU93Phsp4i3CCYE50Vu96g.png" /><figcaption><em>Constraints for the level of physical capital (above) and the level of defense capital (below). The dots above the variables on the left-hand-side of the equations are Newton’s notation for derivative with respect to time. </em><strong><em>Delta_y</em></strong><em> and </em><strong><em>delta_d</em></strong><em> refer to a constant depreciation rate for physical capital and defense capital respectively. For more detail, please refer to the research article itself.</em></figcaption></figure><p>Essentially, these two constraints form a dynamical system (constantly changing until a steady state or equilibrium is reached). For further reading on the implication of this, we can refer to other texts on the standard Ramsey–Cass–Koopmans [2] model for economic growth.</p><p>Along with an exogenously determined water function and damage function, all of these constraints complete the formulation of our optimization problem. A long run optimal solution for this problem can then be calculated analytically or solved numerically. By achieving this, we can better conceptualize and discuss relevant policies for flood mitigation strategies. Of course, no mathematical model is perfectly representative of any given real-world scenario or community’s specific circumstances. Yet, through using both dynamic optimization and a macroeconomic framework, we can still arrive at meaningful and useful conclusions. As such, this paper and its findings are a great example of how to formulate a mathematical model that illuminates and clarifies complex relationships between natural phenomena and societal activities.</p><p><a href="#_ftnref1">[1]</a> Grames, J., Prskawetz, A., Grass, D., Viglione, A., &amp; Blöschl, G. (2016). Modeling the interaction between flooding events and economic growth. <em>Ecological Economics</em>, <em>129</em>, 193–209. <a href="https://doi.org/10.1016/j.ecolecon.2016.06.014">https://doi.org/10.1016/j.ecolecon.2016.06.014</a></p><p><a href="#_ftnref1">[2]</a> Wikimedia Foundation. (2021, May 2). <em>Ramsey–Cass–Koopmans model</em>. Wikipedia. Retrieved November 24, 2021, from <a href="https://en.wikipedia.org/wiki/Ramsey%E2%80%93Cass%E2%80%93Koopmans_model#:~:text=The%20Ramsey%E2%80%93Cass%E2%80%93Koopmans%20model,the%20work%20of%20Frank%20P.&amp;text=The%20Ramsey%E2%80%93Cass%E2%80%93Koopmans%20model%20differs%20from%20the%20Solow%E2%80%93,so%20endogenizes%20the%20savings%20rate.">https://en.wikipedia.org/wiki/Ramsey%E2%80%93Cass%E2%80%93Koopmans_model#:~:text=The%20Ramsey%E2%80%93Cass%E2%80%93Koopmans%20model,the%20work%20of%20Frank%20P.&amp;text=The%20Ramsey%E2%80%93Cass%E2%80%93Koopmans%20model%20differs%20from%20the%20Solow%E2%80%93,so%20endogenizes%20the%20savings%20rate.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c0c5018e6e4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Providing Access to Clean Water in Bangladesh]]></title>
            <link>https://perinita.medium.com/providing-access-to-clean-water-in-bangladesh-57bad0dca7c8?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/57bad0dca7c8</guid>
            <category><![CDATA[water-and-sanitation]]></category>
            <category><![CDATA[operations-research]]></category>
            <category><![CDATA[mathematical-modeling]]></category>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Mon, 29 Nov 2021 15:21:28 GMT</pubDate>
            <atom:updated>2021-11-29T15:21:28.561Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A guest article written by student </em><strong><em>Calvin Nilson</em></strong><em>, edited by Tyler Perini.</em></p><p>Over one third of the world’s population currently does not have access to clean water or a toilet. While people in developed countries do not need to think about where the water comes from to wash food or take a shower, to others attaining water is a day-long task that requires miles of walking and no guarantee that the water is adequately sanitary. Lack of access to clean water most directly affects public health, often leading to diarrhea and cholera outbreaks in communities. The struggle to obtain clean water impacts education and livelihoods, for which women are disproportionately effected. In short, clean water is a significant asset problem for development of poor communities around the world. Using operations research and mathematical modeling will help guide decisions about how to improve accessibility with minimal cost.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/732/1*UYEvMk0djC39wUTsuvNlNA.png" /><figcaption><a href="https://water1st.org/wp-content/uploads/2019/11/f-diagram.jpg">(Source)</a></figcaption></figure><p>Bangladesh, in particular, differs from other countries because of population density. Bangladesh has a population density of 3,277 people per square mile, the largest among countries with more than ten million people. This means that more people can be served by water systems that are near their homes, but hygiene is harder to address. <a href="https://rsaiconnect.onlinelibrary.wiley.com/doi/10.1111/j.1435-5957.2007.00114.x."><em>Supplying safe water in Bangladesh</em>, written by Akhmam and Higano (2007)</a>, employs a multi-objective mixed integer program to analyze the water crisis in Taranagar, a medium-sized village in Bangladesh with population of 2,023 people. Notably, arsenic poisoning is a large concern in this village because about 80% of the tested wells had arsenic above the permitted amount.</p><p>The main objectives of the optimization problem are to (1) minimize the cost of building the water systems, (2) minimize the distance that the people in Taranagar must travel to access water, and (3) maximize the total number of people that will be getting arsenic-free water. The decision variables Y are nonnegative integers that represent the number of safe water options to be implemented (per safe water option and location). The cost is defined by this decision variable as well as a coefficient to represent the depreciation of each safe water option. The other decision variables, X, are nonnegative integers that represent the number of people living in zone <em>i</em> that are receiving water from zone <em>j</em> from the safe water option type <em>k</em>. This decision variable is used to determine the optimal distance, which takes the sum of the distances for each person using the water systems, as well as a binary variable determining if the trip to access the water is “acceptable” (a measure determined by surveying the Taranagar community).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/1*6VGSv75P6VNQ7vZImhwInw.png" /><figcaption>Three constraints for the model.</figcaption></figure><p>The first constraint of the model states that the total number of people living in a zone must be less than or equal to the number of people with access to water in another zone. This constraint turns on when the trip from the people to the water system is deemed acceptable and is turned off when the trip is unacceptable. A second constraint restricts the number of people using a certain water system from exceeding that system’s capacity. The alpha coefficient represents the capacity of each type of safe water option. Because the number of people with arsenic-free water takes priority over the distance and cost, the model includes it as a constraint in a biobjective optimization problem. The third constraint requires that the total number of people with arsenic-safe water must be at least some proportion (gamma) of the total population of the village (P).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/806/1*ZCtE-j2qNCDjCa7wm9o6OA.png" /><figcaption>Computational results from the model.</figcaption></figure><p>The paper introduced multiple cases. Case 1 describes the optimal cost and distance of water systems where the constraint on the standard for arsenic levels was not considered. Case 2 shows the optimal cost and distance of the model where all the water options are safe and below the maximum arsenic levels permitted by the World Health Organization (WHO). From the results, there is a clear difference in cost between cases. The filter used to get rid of the arsenic in the water is a more expensive option, however the community will see more long-term benefits in public health. Also, as the graph shows, if the acceptable distance from household to water system is decreased, the cost only rises once the average distance reaches below 70 and 80 meters for Cases 1 and 2, respectively.</p><p>The model allows decision makers to adjust their preferences, whether that be paying a higher cost for shorter travel distance or sacrificing quality guarantees for cheaper access. Ideally, all communities would have access to arsenic-free water; however, some households might not have the money to immediately pay the higher cost. Therefore, a cheaper water system may be the only feasible option. We can see through this study the benefits of optimization on suggesting strategies for water accessibility.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=57bad0dca7c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bringing Better Care in Better Time]]></title>
            <link>https://perinita.medium.com/bringing-better-care-in-better-time-addbd729b5d7?source=rss-f44c82829945------2</link>
            <guid isPermaLink="false">https://medium.com/p/addbd729b5d7</guid>
            <category><![CDATA[integer-programming]]></category>
            <category><![CDATA[mathematical-modeling]]></category>
            <category><![CDATA[mental-health]]></category>
            <dc:creator><![CDATA[Tyler Perini]]></dc:creator>
            <pubDate>Sun, 28 Nov 2021 19:27:16 GMT</pubDate>
            <atom:updated>2021-12-14T16:50:38.671Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>An Integer Programming Approach to Tele-Psychotherapy Scheduling</strong></p><p><em>A guest article written by student </em><strong><em>Amey Maley</em></strong><em>, edited by Tyler Perini.</em></p><p>The ongoing COVID-19 pandemic has illuminated many to the silent mental health crisis taking place worldwide. From the loss of loved ones to job security, myriad pandemic-related stressors have exacerbated poor mental health across the nation. According to <a href="https://www.kff.org/coronavirus-covid-19/issue-brief/the-implications-of-covid-19-for-mental-health-and-substance-use/">2020 data collected by the Kaiser Family Foundation</a>, nearly half of one-thousand surveyed Americans reported experiencing negative mental health effects as a result of the pandemic. Moreover, the <a href="https://emergency.cdc.gov/han/2020/han00438.asp?ACSTrackingID=USCDC_511-DM44961&amp;ACSTrackingLabel=HAN+438+-+General+Public&amp;deliveryName=USCDC_511-DM44961">CDC reports</a> the number of opioid overdose deaths in the United States increased by 18.2% from June 2019 to May 2020, a period which directly captures the start of the COVID-19 pandemic. Under these circumstances, the demand for mental health services has exploded, driving the expansion of tele-psychotherapy services to provide remote mental health support.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/906/1*b-Y0OxHAqVJbmSWEflFQWA.png" /><figcaption>Social isolation is one of many drivers of declining mental health during the COVID-19 pandemic, resulting in greater demand for mental health services. (Source: <a href="https://doi.org/10.1016/s1473-3099(20)30797-0">The <em>Lancet Infectious Disease</em>s</a>)</figcaption></figure><p>Challenges in providing tele-psychotherapy stems from two areas: (1) scarce personnel with the appropriate training and technical skills to conduct sessions; and (2) highly specific needs among diverse patient populations which have propelled the uptick in mental healthcare demand during the pandemic. More specifically, in their <a href="https://doi.org/10.1007/s10729-020-09543-z">2021 publication in <em>Health Care Management Science</em></a><em>,</em> Andrés Miniguano-Trujillo et. al. characterize divergent psychological needs between three mental health patient populations in Ecuador: (i) frontline healthcare workers treating COVID-19, (ii) individuals grieving COVID-19 loss, and (iii) risk-of-suicide patients. Given these challenges, Miniguano-Trujillo and colleagues investigated how an integer programming model could be implemented to both meet the needs of patients and the capacity of providers via the strategic scheduling of these virtual appointments.</p><p>Under these circumstances, an integer program (IP) provides some unique benefits for optimizing schedules. For instance, since scheduling utilizes a discrete number of therapists, patients, and sessions slots, it is appropriate to model the associated decisions via integer and binary-valued variables. Miniguano-Trujillo et al., label their IP model as the Multi-periodic Patients Assignation to Therapists Model (MPATM).</p><p>The MPATM decision variables are summarized by x, y, and z. The general integer variable x represents the number of patients in each category who are assigned to a given therapist in a given period. Another binary decision variable y represents if a given therapist treats any patients in a given period. Finally, z is a general integer variable which represents the number of patients who cannot be assigned a qualified therapist during a given period due to limited supply of qualified therapists.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M3ryHWdBl1q87qUSZfpWwQ.png" /><figcaption>Decision variables for MPATM.</figcaption></figure><p>Some of the simple constraints for this model are as follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hXIxdOurfx9-Dls_u0geMQ.png" /><figcaption>Simple constraints for the MPATM.</figcaption></figure><p>The first constraint imposes an upper bound on the total number of patients assigned to each therapist across the time periods. The second constraint supplements this upper bound by providing a restriction on the number of patients assigned to each therapist within a given period.</p><p>The model additionally includes advanced constraints, an example of which is shown to the left. This constraint promotes a more even caseload among the therapists within a given period. It achieves this by ensuring a bound on the proportion of patients assigned to each therapist in a group (represented by q). Each therapist is constrained to be within a group where the number of patients scheduled to that group divided by the total number of therapists in the group (minus one) is a minimum for the number of patients assigned to that therapist.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_L7_x2Iq_Wv5uShkCWMUjw.png" /><figcaption>Constraint on the proportion of patients assigned to therapists within a group.</figcaption></figure><p>One of the most intriguing components of the MPATM is the objective itself. While framed as a single objective model, the two summations reflect two distinct goals. The former aims to maximize the fit between patient-therapist assignments, and the latter seeks to maximize the total contribution by therapists. Thus, c and d can be viewed as hyperparameters that may be adjusted accordingly to dictate how these goals are prioritized. In fact, Miniguano-Trujillo et al., identifies two algorithms for setting these hyperparameters and subsequently analyzed the results through the course of this paper.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1010/1*HY6XRhmPvL7yy_cLZVg2iA.png" /><figcaption>Objective for MPATM.</figcaption></figure><p>Two different algorithms were used to set the objective function parameters for the MPATM, algorithm 1 which utilized a period-based heuristic, and algorithm 2 which utilized a category-based heuristic. The model was then tested using these two algorithms against six different combinations of patients and therapists. To simplify interpretation, the combinations of patient-therapist numbers were represented via the associated number of scheduling periods. The objective value and percent of unscheduled patients produced via the MPATM utilizing these algorithm outputs were subsequently plotted against scheduling periods in order to compare the success of the two algorithms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*uwBmKreEH7TKM05BfBuj7A.png" /><figcaption>Simulated results from the MPATM algorithms; source: <a href="https://doi.org/10.1007/s10729-020-09543-z">Miniguano-Trujillo 2021</a>. Note that the y-axis of the first graph represents percentage (of patients unscheduled) with an upper limit of 3%. The second graph plots the objective value, and since the problem is to maximize the objective, Algorithm 2 tends to outperform Algorithm 1 as the number of periods increases.</figcaption></figure><p>Two key takeaways can be learned from the simulated MPATM results. First, regardless of what heuristic is being used, the model is highly successful at scheduling patients, with the percentage of unscheduled patients never exceeding 3% among the simulated instances. Second, the period-based algorithm for setting parameters garners greater success with scheduling patients despite generating lower objective values. This speaks to the broader idea that in mathematical programming, an objective is only as good as its composite parameters, and when applicable, experimenting with hyperparameters not only provides an opportunity to prioritize different issue areas but is a necessary means to isolate vulnerabilities and strengths within a given model.</p><p>As we look towards a future where telehealth becomes more prevalent, IP approaches such as MPATM hold a lot of potential as an expeditious means to align patients with unique needs to providers capable of catering to those needs via strategic scheduling. As this pandemic has demonstrated, healthcare providers are not immune to burnout, thus optimizing scheduling beyond just patients’ needs but also prioritizing provider capacity may alleviate some of the pressures on the mental healthcare system.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=addbd729b5d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>