Semester of Graduation
Summer 2018
Degree
Master of Science in Engineering Science (MSES)
Department
College of Engineering
Document Type
Thesis
Abstract
River stage prediction is an important problem in the water transportation industry. Accurate river stage predictions provide crucial information to barge and tow boat operators, port terminal captains, and lock management officials. Shallow river levels caused by prolonged drought impact the loading capacity of barges and tow boats. High river levels caused by excessive rainfall or snowmelt allow for greater tow capacities but make downstream transportation and lock management risky. Current academic river height prediction systems utilize either time series statistical analysis or machine learning algorithms to forecast future river heights, but systems that combine these two areas often limit their analysis to a single station or river basin. Empirical models require excessive computational power and cannot provide up-to-the-minute projections. In this project, the United States inland waterway system is divided into 24 subnetworks with the Atchafalaya, Lower Ohio, and Lower Mississippi subnetworks given special attention. Model generation, tuning, and testing processes are documented. The generated models are able to predict river stage one week in the future with root mean square error less than 0.75 feet for all three highlighted subnetworks.
Date
7-2-2018
Recommended Citation
Rohli, Eric, "Predicting River Stage Using Recurrent Neural Networks" (2018). LSU Master's Theses. 4760.
https://repository.lsu.edu/gradschool_theses/4760
Committee Chair
Knapp, Gerald
DOI
10.31390/gradschool_theses.4760
Included in
Computational Engineering Commons, Databases and Information Systems Commons, Hydrology Commons, Numerical Analysis and Scientific Computing Commons, Statistical Models Commons