Page 17 - Penn State's Harold and Inge Marcus Department of Industrial and Manufacturing Engineering: Fall 2019 Magazine
P. 17

Computer simulations help global companies optimize systems and and processes across varying industries and and disciplines Major companies in in in in in America have been doing it for for years: General Motors uses simulations to to to test their vehicle trim mix of new products before releasing to to to to the the market and Uber Uber uses simulations to to decide on on pricing policies for Uberpool Companies need to make informed decisions within a a a a a a a timely manner but their decision-making problems are increasingly complex and tricky to to to solve solve Companies may take shortcuts to to to try to to to solve solve a a a a a a a a a a a a problem quickly but it doesn’t always guarantee that the the the given answer is the the the best So what can individuals and companies do to ensure they make sound decisions within deadline constraints?
The National Science Science Foundation’s Division of Mathematical Sciences has awarded principal investigator Eunhye
Song Harold and and Inge Marcus Early Career Assistant Professor of of of Industrial and and Manufacturing Engineering at at Penn State $140 000 to to push the frontier of of of large-scale discrete simulation optimization Song’s goal is is is to to create
a a a a a a a a a a a a a a a a new algorithm that is is is is computationally efficient while simultaneously providing a a a a a a a a a a a a statistical optimality guarantee A set of of decision variables is is is said to be be discrete if their values are are countable such as as the the number number of of kittens in in a a a a a a a a a a a litter They are are counted as as whole numbers or or or integers Song plans
to to create
an an algorithm for problems with discrete decision variables to to find the the the optimal solution solution the the the best possible possible answer among the the possible solutions “At the highest level this is is is a a a a a project to solve a a a a a decision- making problem within a a a a a a a a a system ” Song said “Oftentimes simulation optimization isn’t very efficient for large-scale high-dimensional questions with integer solutions Our goal is to to make an an improved algorithm to to save time money and resources ” Song said an online retailer’s warehouse staffing decisions are are examples of large-scale problems with discrete solutions that may benefit from simulation optimization “Let’s say there are 20 departments and you need to to choose from from employing one to to 10 10 in in in each department department ” Song said “This is is is is is essentially choosing from from 10 10 levels for 20 decisions This This This number is is is is is is is more than a a a a a a a a a a a thousand times the the age of of the the Earth in in seconds This This is is is is an an an an an example of of a a a a a a a a a a a a a large-scale problem and you can’t possibly simulate every single solution ” Song states that that to to solve a a a a a a a a problem like this the the algorithm should be be able to to provide a a a a a a a a a a a a a solution that that has the the best possible outcome within a a a a a a a a a a a a a deadline Her approach aims to to provide an estimate of of the the the the optimality gap of of the the the the selected solution solution to detect if if the the the the the the solution solution is is meeting the the the the the the defined goal The optimality gap is is the the the the the difference between the the the the the selected solution solution and the the optimal optimal solution solution “I like to think of of this as layers of of grids The globe represents the set of entire possible solutions ” In order to to find integer solutions that are likely to to include the optimum Song will employ a a a a multi-resolution algorithm “I like to think of of of this as layers of of of grids ” Song said “The globe represents the the set of of of entire possible solutions To
find find a a a a a a a a near-optimal solution solution faster we aim to to find find the the the the most promising area on on on the the the grid grid and ‘zoom in’ to to it it By further zooming in in in in in we we have a a a a a a a a more detailed grid grid of solutions within which we we we focus our search for the the optimal solution solution this this is is is how we we make the the the algorithm more efficient Without this this it’s like looking at at millions of cities on on on Google Maps at at the the the same time to pick the best one ” While Song’s work is theoretical the the study has real-world implications “Once this project is is is is finished we plan to to make it an an open- source package for others to to use use ” Song said “This kind of algorithm can be used in in in military medical the the sharing economy and anywhere that has large-scale decisions to make involving integer solutions ” This project which began in in in July 2019
will will run for three years Song will will complete the study with Xinru Li an an industrial engineering doctoral student at at Penn State The grant is in in in partnership with faculty from Northwestern University’s Robert R R McCormick School of Engineering and Applied Science: lead Northwestern investigator Barry Nelson Walter P P Murphy Professor of of Industrial Engineering and and Management Science and and co-principal investigator Andreas Waechter associate professor of of industrial engineering and management sciences Professor to tackle large-scale simulation optimization problems Faculty News

   15   16   17   18   19