基于基因表达式编程的NSM FRP-混凝土粘结强度预测模型

纤维增强复合材料（fiber-reinforced polymer，FRP）已被广泛应用于混凝土加固工程中。FRP与混凝土界面间的粘结性能是影响加固效果的重要因素之一，为准确预测FRP嵌入式加固（near-surface-mounted，NSM）NSM FRP-混凝土的粘结强度，运用基因表达式编程（gene expression programming，GEP）方法，选取混凝土抗压强度、粘结长度、槽深宽比、FRP轴向刚度、FRP抗拉强度及环氧树脂抗拉强度等6个参数作为粘结强度的影响因素，建立了NSM FRP与混凝土粘结强度的预测模型，提出了具体的计算公式。通过比较粘结强度预测值与实验值，发现二者较为接近，说明该模型具有一定的可靠性。对该模型进行敏感性分析，发现其可以反映粘结强度与单因素之间的内在关系，即粘结强度随着粘结长度、混凝土抗压强度、槽深宽比及FRP轴向刚度等因素的增大而增大。将该GEP模型与经验模型及小波神经网络模型进行比较，并选取6个统计指标对模型进行评价。结果表明，GEP模型与小波神经网络模型的精度较高，各项误差指标均较小，决定系数分别为0.793和0.787。总体而言，GEP模型的精度略优于小波神经网络模型，二者的精度均远高于经验模型。

Bond Strength Prediction Model of the Near-surface-mounted Fiber-reinforced Polymer Concrete Based on Gene Expression Programming

Fiber-reinforced polymer (FRP) has been widely used in concrete reinforcement projects. The bonding performance between the FRP and concrete influences the reinforcement effect well. Numerous studies were conducted on the bond strength of FRP externally bonded concrete, and a lot of empirical models were developed, while there were few empirical models of the bond strength of FRP near-surface-mounted (NSM) concrete. In order to accurately predict the bond strength between near-surface-mounted FRP and concrete, the gene expression programming (GEP) method was employed to develop a bond strength prediction model of NSM FRP bonded to concrete, and a specific calculation formula was established. The model was developed using six parameters including the concrete compressive strength, bond length, groove depth-to-width ratio, FRP axial rigidity, FRP tensile strength and epoxy tensile strength. The predicted values calculated by the proposed formula agreed well with the experimental values, indicating that the model was reliable. Through the sensitivity analysis of the model, it was found that the GEP model can reflect the internal relationship between bond strength and single factor, i.e., the bond strength increased with the increase of the bond length, concrete compressive strength, the groove depth-to-width ratio and FRP axial rigidity. The GEP model was compared with the empirical model and the wavelet neural network model. Six statistical indicators were selected to evaluate the prediction performance of these models. It can be found that the accuracy of the GEP model and the wavelet neural network model was high, and the errors were low where the coefficients of determination were 0.793 and 0.787, respectively. In general, the accuracy of the GEP model was slightly better than the wavelet neural network model, and the accuracy of the two models was much higher than the empirical models.