Font Size: a A A

Neural network models for prediction, estimation, and optimization: Algorithms and applications

Posted on:1996-05-01Degree:Ph.DType:Dissertation
University:University of Maryland, College ParkCandidate:Kwon, OhseokFull Text:PDF
GTID:1468390014986091Subject:Electrical engineering
Abstract/Summary:
Neural network models are useful for a wide variety of prediction, classification, optimization, and pattern recognition problems. A network of neurons can try to learn the complex relationships between input (independent) variables and output (dependent) variables. In this dissertation, we apply neural network models to important prediction, estimation, and optimization problems. First, we use feedforward neural networks to predict Atlantic hurricane activity. Second, we construct neural network models that estimate the optimal length of a traveling salesman tour through 10 to 80 points located in a rectangular region. Third, we construct neural network models that predict initial returns for initial public offerings. Fourth, we model several kinds of knapsack problems using a modified Hopfield neural network. In the first three applications, we use a backpropagation algorithm with enhancements and, in the fourth application, we use a modified Hopfield network.;This dissertation has four objectives. (1) Model four important prediction, estimation, and optimization problems using neural network models. (2) Develop and implement a neural network code that is capable of solving each problem quickly and accurately. (3) Compare the outcomes of neural network models to traditional models, such as regression models, when appropriate. (4) Discuss how the neural network models can be used in practice.;The results of our modeling efforts reveal that neural networks are useful tools for prediction, estimation, and optimization problems. The adaptive nature of neural network models makes them very appealing for prediction and estimation problems. When applied to prediction and estimation problems, most of our neural network models outperformed regression models. We also applied Hopfield network models to several knapsack problems and most of the Hopfield models produced good results.;In Chapter 2, we provide a general description of the neural network models that we will use to model the four problems. In Chapter 3, we describe how we model hurricane activity. In Chapter 4, we present the traveling salesman application and, in Chapter 5, we present the initial public offerings project. In Chapter 6, we describe the application of neural networks to several knapsack problems. The major research contributions of this dissertation and the work that lies ahead are discussed in Chapter 7.
Keywords/Search Tags:Neural network models, Prediction, Optimization, Estimation, Knapsack problems, Chapter, Application
Related items