Predicting Marshall parameters of flexible pavement using support vector machine and genetic programming

Created by W.Langdon from gp-bibliography.bib Revision:1.7630

@Article{ZHANG:2021:CBM,

• author = "Weiguang Zhang and Adnan Khan and Ju Huyan and Jingtao Zhong and Tianyi Peng and Hanglin Cheng",
• title = "Predicting Marshall parameters of flexible pavement using support vector machine and genetic programming",
• journal = "Construction and Building Materials",
• volume = "306",
• pages = "124924",
• year = "2021",
• ISSN = "0950-0618",
• DOI = "doi:10.1016/j.conbuildmat.2021.124924",
• URL = "https://www.sciencedirect.com/science/article/pii/S0950061821026751",
• keywords = "genetic algorithms, genetic programming, Marshall parameters prediction, Genetic programming and SVM method, Coarse aggregate to filler percentages ratio",
• abstract = "The Marshall mixture design method of asphalt concrete pavement in Pakistan is based on Asphalt institute MS-2 respective of the general specifications of National highway authority, which significantly affects the reliability of parameters used in Marshall design. Traditional way of determining the corresponding parameters and the optimum bitumen content usually involves complicated, time consuming and cost-expensive, laboratory procedures. Therefore, this research conducted research on the applications of machine learning techniques i.e., support vector machine (SVM) and genetic programming (GP), for the prediction of Marshall parameters (i.e., Marshall stability, flow, and air voids) of flexible pavement base and wearing course. A comprehensive dataset of Marshall mix design was collected from four different road sections. The dataset includes 114, and 145, Marshall stability, Marshall flow and air voids results of the base and wearing course, respectively. The three input parameters considered for the modeling are bitumen content, percentage of coarse aggregate to filler material, and unit weight of compacted aggregates. Statistical criteria are used to evaluate overall performance of the developed models. Meanwhile, GP-based models were assessed by parametric analysis to compare the trends of the models with the practical study. The results show that both the techniques are more efficient and superior than traditional methods in terms of generalizability and prediction capability for Marshall parameters of both courses, which are proved by correlation coefficient (R) (in the case of this study > 0.85). SVM obtains outburst performance than GP by setting the optimal parameters. However, GP provided an empirical expression, which is also validated by parametric study and can be used to estimate the Marshall stability, Marshall flow, and air voids of flexible pavements base course, and wearing course, respectively",

}

Genetic Programming entries for Weiguang Zhang Adnan Khan Ju Huyan Jingtao Zhong Tianyi Peng Hanglin Cheng

Citations