基于神经网络在路面基层压实参数中的应用

Application of Neural Network to Compaction Parameters in Pavement Base

石灰工业废渣稳定类半刚性材料是高等级公路路面基层材料常见形式之一,根据规范和设计要求可分为含骨料类和不含骨料类。当骨料含量超过50%时,室内重型击实试验劳动量大,并且干密度和含水率曲线不稳定。在已知不含骨料的石灰工业废渣稳定类半刚性材料(即结合料)的最大干密度和最优含水率的基础上,通过结合人工神经网络理论,基于Matlab的BP人工神经网络,建立并编制了含骨料的石灰工业废渣稳定类半刚性材料压实参数(最大干密度和最优含水率)的预测网络模型,经过对网络模型的大量训练、训练函数和传递函数的调整及初始训练数据的规—化,最后建立了6→15→2的网络模型,其网络模型预测结果稳定准确,有一定实际应用价值。

The semi-rigid material of lime stabilized industry wastes is commonly used in pavement base of high-grade highway, which contains certain aggregate or none aggregate according to the requirements of standard and design. Generally, the indoor heavy compaction test is not only labor-consuming, but also unable to achieve the precise compaction parameters of the semi--rigid mixture when the aggregate is over 50%. Based on the known compaction parameters of the binder without aggregate and combined with artificial neural network theory, the simulating network model of this type of semi-rigid material＇s compaction parameters was created by BP network in Matlab. Through large amounts of training, adiustment of training function and transfer function and normalization of initial input data, eventually the 6→15→2 network was established. The results simulated by the network model are correct and stable, which show certain practical application values.