Thesis (Ph. D.)--Ohio State University, 2001. / Title from first page of PDF file. Document formatted into pages; contains xiv, 169 p.; also contains graphics (some col.). Includes abstract and vita. Advisor: B. Chandresekaran, Dept. of Computer and Information Science. Includes bibliographical references (p. 165-169).

Thesis (M.B.A.)--Naval Postgraduate School, June 2008. / Thesis Advisor(s): Cuskey, Jeffrey. "June 2008." Description based on title screen as viewed on September 2, 2008. Includes bibliographical references (p. 75-80). Also available in print.

No abstract / Decision Sciences / M. Sc. (Operations Research)

No abstract / Decision Sciences / M. Sc. (Operations Research)

The investigation reported in this dissertation attempts to determine the feasibility of using a distance-based approach like compromise programming for discrete-time dynamic programming problems with multiple objectives. In compromise programming, a function measuring the distance from a generally infeasible ideal solution to the feasible set of the problem is the single objective acting as a surrogate for the set of multiple objectives. Since, in general, there is no single best solution to a multiple objective problem, a framework to generate a family of compromise solutions interactively on a computer is proposed. Various quantities relevant to dynamic compromise programming are defined in precise terms. Dynamic compromise programming problems are computationally difficult to solve because in order to make the distance function decomposable over stages, dimensionality of the state-space must be increased by the number of objectives. To generate compromise solutions, quasi-Newton differential dynamic programming (QDDP), a recently developed variable-metric method for discrete-time optimal control, was employed. QDDP is attractive because no second order or Hessian information is required as input. Instead, Hessian matrices are approximated by first order or gradient information. Since very little is known about its numerical properties, computational experiments were conducted on QDDP. A new strategy for updating Hessian matrix approximations was computationally tested. A constrained QDDP algorithm is proposed, computationally tested, and applied to solve a multiobjective dynamic programming problem with inequality constraints at each stage. The algorithm has the potential for application to the more general discrete-time optimal control problem with stage constraints. The framework for generating compromise solutions interactively was implemented for prototype problems. Because decision maker interaction is crucial in a multiple objective situation, special attention was paid towards developing a man-machine interface using on-screen windows. All implementation and computational testing were done on a UNIX based personal computer.

In modern organizations where managers must constantly be dealing with an overload of information, it is often observed that participants in group decision processes either are not clearly aware of their specific preferences or that they are not capable of properly formulating those preferences. When this happens, inconsistent or incomplete expression of personal preferences and their use in decision making may lead to an unjustifiable outcome for the group. Due to this problem, the strengths and effectiveness of GDSS-supported group meetings may, in some situations, not be apparent. This dissertation develops a new approach to supporting group decision making, focusing on preference knowledge of individual participants in a group. A system architecture for the design of an MCDM (Multiple Criteria Decision Making) GDSS which facilitates the process of eliciting, formulating, utilizing, aggregating, and analyzing preferences for individuals within groups is presented. The architecture integrates multi-criteria decision making paradigms with a group decision support environment. A prototype has been developed in order to demonstrate the design feasibility of an architecture that centers around four phases of choice making: alternative generation, preference specification, alternative evaluation, and preference aggregation. The prototype is designed to support managerial choice and judgment processes in collaborative meetings. The intended problem domain of the model is semi-structured managerial decisions for which decision variables (attributes) can be represented in quantitative terms to some extent, yet for which evaluation of alternatives requires a high degree of intuition and personal analysis. The process of prototyping the proposed architecture and the results from a qualitative study have provided some instructive conclusions relating to MCDM GDSS design: (1) support for human choice strategies can be integrated into a GDSS, (2) appropriate management of preferences of group participants will facilitate collaborative decision processes, (3) hierarchical decomposition of a decision problem can provide structure to a problem and thereby reduce problem complexity, and (4) managerial decisions are appropriate problems to which the current approach can be applied.

A new method for multi-objective optimization of linear and mixed programs based on Lagrange multiplier methods is developed. The method resembles, but is distinct from, objective function weighting and goal programming methods. A subgradient optimization algorithm for selecting the multipliers is presented and analyzed. The method is illustrated by its application to a model for determining the weekly re-distribution of railroad cars from excess supply areas to excess demand areas, and to a model for balancing cost minimization against order completion requirements for a dynamic lot size model.

Thesis (Ph. D.)--University of Washington, 2001. / Vita. Includes bibliographical references (leaves 160-171).

Thesis ( Ph.D. ) -- University of Texas at Arlington, 2008.

The study analyses various methods to solve multiple criteria optimization problems of different kinds and defines principles of parallel computing. A multiple criteria optimization problem has been solved applying a computer network and a new strategy that analyses and uses intermediate results in the calculation process has been suggested. The optimization problem has been solved applying a computer network and parallel computing software MPI (Message Passage Interface). Numerous experimental trials have been carried out to investigate efficiency of the designed strategy in the solution of multiple criteria optimization problems. A computer network with different number of computers solved a single problem of different duration and final results of various strategies have been compared. The experiments have proved the designed strategy to be more precise in results and more economical in computing time.