Interactive multiple objective programming using Tchebycheff programs and artificial neural networks

A new interactive multiple objective programming procedure is developed tha t combines the strengths of the interactive weighted Tchebycheff procedure (Steuer and Choo. Mathematical Programming 1983;26(1):326-44.) and the inte ractive FFANN procedure (Sun, Stam and Steuer. Management Science 1996;42(6 ):835-49.). In this new procedure, nondominated solutions are generated by solving augmented weighted Tchebycheff programs (Steuer. Multiple criteria optimization: theory, computation and application. New York: Wiley, 1986.). The decision maker indicates preference information by assigning "values" to or by making pairwise comparisons among these solutions. The revealed pr eference information is then used to train a feed-forward artificial neural network. The trained feed-forward artificial neural network is used to scr een new solutions for presentation to the decision maker on the next iterat ion. The computational experiments, comparing the current procedure with th e interactive weighted Tchebycheff procedure and the interactive FFANN proc edure, produced encouraging results. Scope and purpose Artificial neural networks have been shown to possess an ability to learn a nd represent complex mappings, and have been applied to pattern recognition problems. The authors of the current paper believe that a decision maker's preference structure map be viewed as a pattern. and thus should be amenab le to artificial neural networks. In a previous work (Sun, Stam and Steuer. Mangement Science 1996;42(6):835-49.), the authors developed a neural-netw ork-based interactive solution method for multiple objective programming pr oblems. The current paper extends that earlier work, combining elements of traditional interactive method with neural networks. In the computational e xperiments, the new method produced better results for most problems than b oth the traditional interactive method and the earlier neural-network-based method, thus providing an attractive alternative to existing interactive m ultiple objective programming procedures. (C) 2000 Elsevier Science Ltd. Al l rights reserved.