Simultaneous Generation of Optimum Pavement Clusters and Associated Performance Models

Mukesh Khadka,David K Hale,Alexander Paz,Cristian Arteaga

doi:10.1155/2018/2159865

Abstract

With regard to developing pavement performance models (PPMs), the existing state-of-the-art proposes Clusterwise Linear Regression (CLR) to determine the pavement clusters and associated PPMs simultaneously. However, the approach does not determine optimal clustering to minimize error; that is, the number of clusters and explanatory variables are prespecified to determine the corresponding coefficients of the PPMs. In addition, existing formulations do no address issues associated with overfitting as there is no limit to include parameters in the model. In order to address this limitation, this paper proposes a mathematical program within the CLR approach to determine simultaneously (1) an optimal number of clusters, (2) assignment of segments into clusters, and (3) regression coefficients for all prespecified explanatory variables required to minimize the estimation error. The Bayesian Information Criteria is proposed to limit the number of optimal clusters. A simulated annealing coupled with ordinary least squares was used to solve the mathematical program.

Highlights

Pavement performance models (PPMs) are developed using a two-step approach
In order to avoid prespecifying the number of required clusters, this study proposes a mathematical program to simultaneously determine an optimal number of clusters, the assignment of segments into clusters, and regression coefficients for all explanatory variables
Given the constraints for feasible partitions defined in the problem formulation and the minimum number of observations required in a cluster, n = 800, the proposed algorithm determined 16 as the maximum

Summary

Introduction

Pavement performance models (PPMs) are developed using a two-step approach. Pavement segments with similar characteristics are grouped into clusters using a few critical factors, such as pavement type, age, and traffic volume. The objective of clustering is to group the pavement segments that perform over time. In practice, the performances of pavement segments within a cluster differ significantly because clusters are formed using only a few critical factors [1, 2]. If inappropriate characteristics are used, clusters may include homogeneous segments with different performance behavior or heterogeneous segments with similar performance behavior [4]. The prediction accuracy of PPMs can be improved by subdividing the pavement segments into more uniform clusters. This subdivision is not always possible due to limited information [1]

Methods

Results

Conclusion