This study deals with using a neural network in generalized Group Method of Data Handling (GMDH) type successfully to model and predict the standard cube strength of SFRRCAC (steel fiber reinforced recycled coarse aggregate concrete) based on the data obtained experimentally through point-load test (PLT) and ultrasonic pulse velocity (UPV) approaches. Two methods of singular value decomposition (SVD) and genetic algorithm (GA) were used simultaneously for designing the model optimally. Some input–output data for training and examining the developed models were used where replacement level of recycled coarse aggregate, volume fraction of steel fibers, PLT and UPV results, are taken as inputs and SFRRCAC’s standard cube strength is considered as the output variables. Based on the findings, the attained model is greatly able to reliably estimate the compressive strength of SFRRCAC based on the combination of PLT and UPV results. Ultimately, using sensitivity analysis for the GMDH neural network model, the input parameters effect on the model output was investigated. Based on the sensitivity analysis, it was revealed that the output variable is considerably altered by PLT and UPV compared to the other input variables.

这项研究成功地利用了基于神经网络的通用数据处理分组方法（GMDH）类型，基于通过点载荷试验获得的数据对SFRRCAC（钢纤维增强再生粗骨料混凝土）的标准立方强度进行建模和预测（PLT）和超声脉冲速度（UPV）接近。同时使用奇异值分解（SVD）和遗传算法（GA）两种方法来优化设计模型。使用了一些用于训练和检验已开发模型的输入输出数据，其中以再生粗骨料的替代水平，钢纤维的体积分数，PLT和UPV结果为输入，而SFRRCAC的标准立方强度被视为输出变量。根据调查结果，结合PLT和UPV结果，所获得的模型能够可靠地估计SFRRCAC的抗压强度。最终，通过对GMDH神经网络模型进行敏感性分析，研究了输入参数对模型输出的影响。根据灵敏度分析，发现与其他输入变量相比，PLT和UPV大大改变了输出变量。