Please use this identifier to cite or link to this item:
標題: Pavement performance prediction through fuzzy regression
作者: Pan, N.F.
Ko, C.H.
Yang, M.D.
Hsu, K.C.
關鍵字: Pavement performance;Pavement maintenance;Fuzzy sets;Fuzzy regression;analysis;Predictions;linear-regression;neural-networks;maintenance;models;systems
Project: Expert Systems with Applications
期刊/報告no:: Expert Systems with Applications, Volume 38, Issue 8, Page(s) 10010-10017.
Accurate predictions of future pavement conditions are essential for determining the most cost-effective maintenance strategy. The current methods for assessing pavement conditions involve either equipment measures or visual inspections. Equipment measures are not extensively implemented because of high cost; thus, subjective evaluations by road inspectors are often used as a replacement. Nevertheless, visual inspections could draw in errors and variations due to subjectivity and uncertainty. The present serviceability index (PSI), one of the most common indicators used to evaluate pavement performance, is incapable of transforming one's imprecise judgment into an exact number between 0 (the worst) and 5 (the best). Conventional regression cannot deal with visual inspection data that are linguistic or non-crisp. In contrast, fuzzy regression is capable of handling such fuzzy data. In this paper, pavement conditions are exemplified by five membership functions and estimated by using fuzzy regression to better account the uncertainties of the traditional method. Also, a similarity indicator is applied to measure the goodness of fit. A case study using pavement inspection data is presented to establish estimated fuzzy regression equations. The results demonstrate the capability of the model, which is able to assist road administration units to determine desirable repair actions regarding the predicted pavement conditions. (C) 2011 Elsevier Ltd. All rights reserved.
ISSN: 0957-4174
DOI: 10.1016/j.eswa.2011.02.007
Appears in Collections:土木工程學系所

Show full item record

Google ScholarTM




Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.