点击次数：

论文类型：会议论文

发表时间：2005-01-01

收录刊物：CPCI-S

页面范围：1269-1272

摘要：The relationship between the influence elements of freezing-thawing durability and the durability factor (DF) was very complicated. Though there were some methods to predict the DF of concrete, yet till now few of them are acceptable. The influence elements of freezing-thawing durability of concrete are mainly as follows: S(mean spacing), F(fraction of total paste volume within the distance of S from the edge of the air voids), A (hardened concrete air content), P/A (hardened concrete paste air ratio), a (air voids specific surface), and all the air-void system parameters could be obtained through test based on ASTH C666, Procedure A. This paper put forward a new approach-Modified Back Propagation (MBP) Neural Network based on MATLAB to predict the DE Through altering the MBP Neural Network structure, learning rate, and the target error, some sets of weights and threshold values can be achieved. About 80% of the data were utilized to train the MBP Neural Network, and 20% to predict. On comparing the prediction results, the MBP Neural Network with 5 neural elements in the input layer, 15 in the hidden layer, and I in the output layer (5-15-1) proved to be the best one. The values of the learning rate and the target error are 0.05 and 0.01 respectively. From the MBP Neural Network prediction results, it can be seen that the relative errors between the experimental DF and the prediction values are all in the range of 4%. The calculation results show that the DF of concrete can be relatively precisely predicted with the air-void system parameters (S, F, A, P/A, a) and MBP Neural Network presented in this paper, which will be applied to concrete engineering.