Modeling of road surface condition data for the prediction of road icing

Panda, Rupali

Name:

Panda_R_2009.pdf

Size:

17.37Mb

Format:

PDF

Download

Abstract

"Road icing is a common problem in cold regions, such as Alaska, where it poses serious threat to drivers and result in the disruption of transportation facilities. Road surface temperature (RST) is considered as the most crucial factor for icing conditions. The objective of this research is to predict RST in Multiple Linear Regression (MLR) and three-layer back propagated Neural Network (NN) models using an optimum number of input variables (air temperature, dew point, relative humidity, wind gust, wind speed average, wind gust and wind directions). The data were analyzed using both randomized and chronological schemes. The results obtained from different models were compared to find the most suitable model for predicting RST. The performance of both MLR and NN models were very comparable. Therefore, in the interest of reducing modeling complexity the MLR models could be preferred instead of the complex neural network models for the aimed accuracy levels in RST prediction. It was also observed that models developed on the chronological data provided better prediction accuracy as compared to models developed on the randomized data indicating RST should probably be predicted from models that honor the time sequence"--Leaf iii

Description

Thesis (M.S.) University of Alaska Fairbanks, 2009

1. Introduction -- 1.1. Road icing forecasting -- 1.2. Models for prediction of road ice/surface temperature -- 1.2.1. Physical-based models -- 1.2.2. Statistical models -- 1.3. Contribution of the thesis -- 1.4. Data and location of the site -- 2. Data description and division schemes -- 2.1. Surface temperature in road icing models -- 2.2. Data division schemes -- 2.3. Data division in randomized scheme -- 2.3.1. Maximum surface temperature limit -- 2.3.2. Range of surface temperatures -- 2.3.3. Similar variable patterns: kohonen grouping -- 2.3.4. Year-wise grouping -- 2.4. Analysis on sequential time series data -- 2.5. Analysis methods -- 2.5.1. Multiple linear regression model -- 2.5.2. Neural network (NN) model -- 3. Multiple linear regression -- 3.1. Multiple linear regression on randomized data -- 3.1.1. Maximum surface temperature limit -- 3.1.2. Range of surface temperatures -- 3.1.3. Results with similar variable patterns (kohonen grouping) -- 3.1.4. Year-wise grouping -- 3.2. Multiple linear regression on time sequence data -- 3.2.1. Year 2003 -- 3.2.2. Year 2004 -- 3.2.3. Year 2005 -- 4. Neural network design -- 4.1. Neural network analysis on randomized data -- 4.1.1. Maximum surface temperature limit -- 4.1.2. Range of surface temperatures -- 4.1.3. Similar variable patterns (kohonen grouping) -- 4.1.4. Year-wise grouping -- 4.2. Recurrent neural network -- 4.2.1. Year 2003 -- 4.2.2. Year 2004 -- 4.2.3. Year 2005 -- 5. Discussion results -- 5.1. Results and discussion -- 5.2. Results from the randomized groups -- 5.3. Results from the time sequence data -- 5.4. Comparison between results from randomized and time sequence data -- 5.5. Comparison between different variable sets for MLR and NN models -- 6. Summary, conclusion and future directions -- 6.1. Results summary -- 6.2. Results summary for randomized data -- 6.3. Results summary for time sequence data -- 6.4. Conclusion and future directions -- References -- Appendix A: Definition of statistical parameters -- Appendix B: Figures for multiple linear regression (chapter 3) -- Appendix C: Tables for multiple linear regression (chapter 3) -- Appendix D: Figures for neural network (chapter 4) -- Appendix E: Tables for neural network (chapter 4) -- Appendix F: Tables for results and discussion (chapter 5).

Date

2009-12

Type

Thesis

Collections

Engineering