基于GA-BP神经网络的UHPC抗压强度预测与配合比设计

开展了不同配合比条件下超高性能混凝土(UHPC)的制备与抗压强度试验,并结合已有数据形成了神经网络训练样本;根据UHPC原材料组成和性能需求设计了包含神经网络输入层(7节点)、隐层(8节点)和输出层(1节点)的拓扑结构,并引入遗传算法(GA)优化了UHPC抗压强度预测网络的初始权值和阈值;采用试验样本模拟训练了不同配合比条件下的UHPC抗压强度预测GA-BP神经网络,并以此为基础建立了基于不同性能需求的配合比设计方法.对比试验数据和传统BP神经网络方法计算结果发现,GA-BP神经网络能更好地指导UHPC抗压强度预测和配合比设计.

Compressive Strength Prediction and Mix Proportion Design of UHPC Based onGA BP Neural Network

Preparation and compressive strength test of UHPC with different mix proportions were carried out, and neural network training samples were formed by combining the existing experimental data. The topology structure of input layer(7 nodes), hidden layer(8 nodes) and output layer(1 node) of the neural network was designed according to the raw material composition and performance requirements of UHPC. The genetic algorithm(GA) was introduced to optimize the initial weight and threshold of the UHPC compressive strength prediction network. The GA BP neural network for compressive strength prediction of UHPC with different mix proportions was simulated and trained with the experimental samples, and the mix design method based on different performance requirements was established. By comparing the experimental data with the results of traditional BP neural network method, it is confirmed that the proposed GA BP neural network can better guide the compressive strength prediction and mix design of UHPC.