应力状态下混凝土碳化深度的神经网络预测

为计算应力状态下预应力混凝土在一定条件下的碳化深度,将混凝土应力水平取为影响碳化速度的参数.在已有试验结果的基础上,分别建立了预应力混凝土碳化深度实用计算模型,以及BP网络、径向基函数(RBF)网络和广义回归(GRNN)网络的三个神经网络预测模型,并通过实例将碳化深度试验值、实用公式计算值及神经网络预测值进行了比较分析.结果表明:考虑混凝土应力水平对碳化深度的影响是合理的,试验回归得到的实用碳化模型计算误差在9%以内;同时,所建立的BP、RBF以及GRNN网络模型均具有较高的计算精度以及良好的泛化能力,仿真和预测误差基本上在5%和4%以内,均低于实用计算模型的误差值.由此可见,所建神经网络模型的仿真及预测结果是理想的,可同时考虑各种影响因素组合、行之有效的混凝土碳化深度预测方法.

Carbonation depth prediction of pre-stressed concrete based on artificial neural network

In order to calculate the carbonation depth of pre-stressed concrete under certain conditions, the stress level of concrete was regarded as an influencing factor on concrete carbonation. Based on the present test data, a practical model for calculating the carbonation depth of pre-stressed concrete was built. And three artificial neural networks (ANN): the BP network, the radial basis function (RBF) network and the generalized regression neural network (GRNN), were established to predict the carbonation depths. The predicted values of the three network models were compared with experimental values and calculated values. The results show that the carbonation depth calculation model with concrete stress level is practicable and its relative error is within 9%; and the three networks have high precision and good generalization ability, whose simulation and prediction errors are within 5% and 4%, lower than the error of calculation. Thus the results of the networks are good, which proves that ANN is an effective method in analyzing and predicting the carbonation depth.