基于人工神经网络算法的沥青路面结构可靠性预测

Reliability Prediction of Asphalt Pavement Structures Based on Artificial Neural Network Algorithms

摘要:

为了探究人工神经网络在沥青路面结构可靠性预测中的应用效果，提升工程计算效率与精度，研究采用2-3-1结构的BP神经网络为主要分析方法，运用trainlm算法及Sigmoid传递函数，将学习误差、学习率与训练次数分别设定为10^-2、0.2与10⁴。结合西咸新区二级沥青路面实体工程，选取路表弯沉、上面层层底拉应力及基层层底拉应力为基础指标，引入温度调整系数与应力、弯沉调整因子，建立沥青路面可靠度计算公式及综合可靠度指标，进而构建基于BP神经网络的沥青路面结构可靠性求解方法。结果表明：与传统可靠性指标计算方法相比，BP神经网络预测结果的误差可控制在5.0%以内，符合工程精度要求；在实际工程中，该方法能够通过直接输入采集数据快速获得可靠性指标，避免复杂计算过程，显著提高工程效率。

Abstract:

To explore the application effects of artificial neural networks in predicting the structural reliability of asphalt pavements and to improve the efficiency and accuracy of engineering calculations, a BP neural network with a 2-3-1 structure is studied and used as the primary analytical method. The “trainlm” algorithm and the “Sigmoid” transfer function are adopted to set the learning error, learning rate and training times to 10^-2, 0.2 and 10⁴, respectively. Combined with a case of Xixian New Area Class II Asphalt Pavement Project, the road surface deflection, and the tensile stresses at the bottoms of both the upper layer and the base layer are selected as key indicators. By introducing the temperature adjustment coefficients, and stress and deflection adjustment factors, the formulas for calculating asphalt pavement reliability and a comprehensive reliability index are established, and then a BP neural network-based method for solving asphalt pavement structural reliability is constructed. The results indicate that compared with traditional reliability index calculation methods, the error of BP neural network predictions can be controlled within 5.0%, meeting engineering accuracy requirements. In practical applications, this method can quickly obtain the reliability indicators through direct input of collected data, avoiding complex computational processes and significantly improving engineering efficiency.