Stochastic optimization based decision support system for strategic planning in process industries

Abstract

Increased competitive pressure due to globalization has forced Indian companies to look for scientific methods of planning their operations, and efficient utilization of available resources.Recent surveys and literature on stochastic programming substantiate the fact that it is possible to make flexible and robust .near optimal. solutions using stochastic optimization models. The amount of data associated with large scale optimization planning models is usually bulky with respect to the nature of storage, handling, and manipulation. Also, the size of the stochastic optimization programs rises exponentially with an increasing numbers of scenarios, and so does the amount of data. Recognizing a wide gap between the Operations Research/Management Science developers and practitioners, we conducted this study of database optimization interface and stochastic optimization.

The objective of the research is to study and develop a method of planning which employs a user friendly, generic, multiple period, multiple scenario, integrated approach and is based upon optimization. The primary goal of the study is to derive and elucidate fundamental issues related to relational database design, reporting of optimal solution and user interface for a stochastic optimization based decision support system (DSS). In this dissertation we discuss four research issues.

First, we demonstrate that with user friendly, menu driven software, complex technologies like steel, aluminum, pharmaceuticals and polymers can be modeled in a multiple-period optimization based DSS with little or no knowledge of optimization techniques. We demonstrate that not only can multiple period planning be done, but a two stage stochastic programming can also be done using this user friendly, menu driven DSS. The research contribution is on the interface of the two disciplines of computer science and optimization, and not on algorithms.

Second, we design relational database structures for three planning models for process industries. Fourer (1997), Dutta and Fourer (2004, 2007a) primarily discussed a hierarchical database structure. We compared two variants of database structures for a manufacturing planning model. The PROCESS1 is a hierarchical database structure, while PROCESS2 is a relational one. We find that relational database structures are more suitable in large scale optimization.

Third, we identify products that have high marginal values. We demonstrate the impact of optimization by developing a set of practical experiments in the DSS. With the help of real data from four companies, we demonstrate that contribution per unit can be increased. The increase in contribution per unit in a steel company is 1.67%, an aluminum company is 1.17%, a pharmaceutical company is 0.66%, and a polymer company is 3.47%.

Fourth, we extend the multiple-period planning model of Dutta and Fourer (2004) to a two stage stochastic programming model with recourse. We model the demand uncertainty in four companies by generating scenarios and designing a set of experiments for demand variability and shift in skew-ness of probability distribution. We demonstrate that the planning using stochastic optimization provides a better solution over deterministic planning.

We further extend the multiple-period manufacturing planning model for simultaneous production, distribution and transportation. First, we extend this to multiple-period, single scenario integrated supply chain planning. Second, we extend this to multiple-period, multiscenario integrated planning for supply chain having eight fundamental elements namely Source of raw materials, Materials, Facilities, Activities, Storage Areas, Warehouses, Scenarios, and Times.We demonstrate the application using real data of a zinc company.

The main contribution of this research is a study at the interface of databases and optimization,stochastic modeling of production and supply chain planning for a typical process industry, designing of relational database structures, and demonstration of a different method of planning using a generic, menu driver, user friendly, stochastic optimization based DSS. Using this DSS, we study issues on interface of database management and optimization modeling systems for a class of stochastic optimization models.