为实现沥青混合料性能的快速可靠预测，从材料组成设计角度出发，提出了一种基于遗传算法（genetic algorithm, GA）优化反向传播（back propagation, BP）神经网络的沥青混合料性能预测方法。首先运用灰关联分析方法对多维输入变量进行降维处理，确定了沥青混合料性能的核心影响因素，然后结合遗传算法（GA），构建了以核心影响因素为输入层、沥青混合料性能为输出层的GA-BP神经网络预测模型，再对模型进行训练验证分析与预测泛化应用，同时与BP神经网络的训练效果和预测精度进行对比，验证GA-BP神经网络模型的准确性。研究结果表明：空隙率、油石比、公称最大粒径、4.75 mm通过率、沥青种类、软化点、针入度、延度等8项性能特征的灰关联度r > 0.6，对沥青混合料性能影响显著；相比于BP神经网络模型，经过GA优化后的BP神经网络模型的均方根误差（root mean square error, RMSE）降低了16% ~ 31%，平均绝对误差（mean absolute error, MAE）降低了15% ~ 24%，R²值提升了0.01 ~ 0.27，说明其具有更好的学习拟合能力；在对沥青混合料动态模量、动稳定度、残留稳定度、劈裂抗拉强度比和极限弯拉应变的预测精度上分别提高了35.26%、47.78%、23.13%、31.92%、35.75%，说明GA-BP神经网络模型具有更强的泛化应用能力。研究成果为实现沥青混合料性能的快速预测、指导沥青混合料材料组成设计提供重要参考。

To achieve rapid and reliable prediction of asphalt mixture performance, a method for predicting asphalt mixture performance by optimizing the back propagation (BP) neural network with a genetic algorithm (GA) from the perspective of material composition design was proposed. Initially, a grey relational analysis method was employed to reduce the dimensionality of multidimensional input variables, identifying the core influencing factors of asphalt mixture performance. Subsequently, integrating the GA, a GA-BP neural network prediction model was constructed with the core influencing factors as the input layer and asphalt mixture performance as the output layer. The model underwent training, validation analysis, and prediction generalization application. A comparison with the training effectiveness and prediction accuracy of the BP neural network was conducted to verify the accuracy of the GA-BP neural network model. The research results indicate that the grey relational degrees of eight performance characteristics, including air void, asphalt-aggregate ratio, nominal maximum aggregate size, 4.75 mm passing rate, asphalt type, softening point, penetration, and ductility, are all greater than 0.6, signifying their significant impact on asphalt mixture performance. Compared to the BP neural network model, the GA-BP neural network model reduces the root mean square error (RMSE) by 16% to 31%, decreases the mean absolute error (MAE) by 15% to 24%, and improves the R² value by 0.01 to 0.27, indicating that it has better learning and fitting capabilities. The prediction accuracy for dynamic modulus, dynamic stability, residual stability, splitting tensile strength ratio, and ultimate bending strain of the asphalt mixture is respectively enhanced by 35.26%, 47.78%, 23.13%, 31.92%, and 35.75%, revealing the superior generalization application capability of the GA-BP neural network model. The research findings provide essential references for the rapid prediction of asphalt mixture performance and guidance in the design of asphalt mixture material composition.