Prediction of interface yield stress and plastic viscosity of fresh concrete using a hybrid machine learning approach

摘要

The interface yield stress and the plastic viscosity of concrete mixes critically influence their pumpability. This study constructs and verifies a data-driven method for predicting these two important parameters. The proposed method is a hybridization of Least Squares Support Vector Machine (LSSVM) and Particle Swarm Optimization (PSO). The LSSVM is employed to infer the mapping function between the two concrete mix's parameters and their influencing factors. Moreover, in order to overcome the challenging task of fine-tuning the LSSVM model hyper-parameters, the PSO algorithm, a swarm intelligence based metaheuristic, is utilized to optimize the LSSVM prediction model. A data set including 142 experimental tests has been collected in this study to construct and verify the proposed hybrid method. Experimental results supported by the Wilcoxon signed-rank test point out that the hybridization of LSSVM and PSO (with coefficients of determination = 0.71 and 0.77 for interface yield stress and plastic viscosity predictions, respectively) can deliver predictive results superior to those of benchmark models. Hence, the hybrid model of PSO and LSSVM can be a promising alternative to assist engineers in the task of concrete structure construction.