ABSTRACT

This study presents a novel hybrid intelligent approach using Extreme Learning Machine (ELM) and Equilibrium Optimiser (EO) (ELM-EO) for predicting resilient modulus, M_r of Unbound Granular Materials (UGMs). Fourteen various blends of Recycled Concrete Aggregate (RCA) with Recycled Clay Masonry (RCM), and Electric Arc Furnace Steel (EAFS) slag with limestone aggregates were tested in the laboratory using routine and advanced tests. The laboratory M_r testing produced 224 measurements based on the average of triplicate samples for each blend. The performance of the ELM-EO approach was evaluated and compared with conventional regression, ELM-biogeography-based optimisation (BBO) (ELM-BBO) and ELM-genetic algorithm (ELM-GA) approaches using the same input properties. The inputs used for the M_r prediction are the bulk stress, percent of RCM, and/or percent of EAFS. The results demonstrate that the performance of ELM-EO and ELM-BBO approaches is better than ELM-GA and regression approaches for predicting M_r. The overall statistical measures of the proposed approaches show that the ELM-EO approach ranks first as it outperforms the other approaches with coefficient of determination (R²) of 0.924 and Root Mean Square Error (RMSE) of 37.08 MPa.

摘要

本研究提出了一种新颖的混合智能方法，使用极限学习机 (ELM) 和平衡优化器 (EO) (ELM-EO) 来预测弹性模量M _r的未结合颗粒材料 (UGM)。在实验室中使用常规和高级测试对 14 种再生混凝土骨料 (RCA) 与再生粘土砌体 (RCM) 以及电弧炉钢 (EAFS) 炉渣与石灰石骨料的各种混合物进行了测试。实验室先生_{_}根据每种混合物的三次样品的平均值，测试产生了 224 次测量结果。评估了 ELM-EO 方法的性能，并与使用相同输入属性的传统回归、基于 ELM-生物地理学的优化 (BBO) (ELM-BBO) 和 ELM-遗传算法 (ELM-GA) 方法进行了比较。用于M _r预测的输入是体积应力、RCM 百分比和/或 EAFS 百分比。结果表明，ELM-EO 和 ELM-BBO 方法的性能优于 ELM-GA 和回归方法预测M _r。所提出方法的总体统计测量表明，ELM-EO 方法排名第一，因为它的确定系数优于其他方法 ( R ²) 为 0.924，均方根误差 (RMSE) 为 37.08 MPa。