基于神经网络预测混凝土中氯离子浓度的分布

Predict chloride concentration in concrete based on neural network

现阶段对氯离子浓度预测大多基于Fick定律以及在此基础上的一些改进模型,其明显的缺点就是模型的参数较多,应用不便。而人工神经网络不需要研究各个影响因素间的关系,常用于对模糊问题的研究。基于BP神经网络与径向基函数神经网络,给出了预测混凝土中氯离子浓度分布的分析模型,并结合试验对比了两种方法的有效性,结果表明径向基函数神经网络比BP神经网络具有较好的准确性与稳定性。最后提出通过建立庞大的神经网络数据库来预测不同环境作用下不同配合比的混凝土在任意时刻的氯离子浓度分布。

Presently,predicting chloride concentration mostly based on FICK＇s Law as well as some improved model in this foundation.The obvious shortcoming of this method is so many parameters to be determined before using,which results in poor application.However,the neural network does not need to study relationships between different factors,commonly used in fuzzy question research.In this paper,Back-Propagation Network and Radial Basis Function Network are engaged to predict chloride concentration in concrete.Furthermore,we verified the effectiveness of the method with the experimental data.The result shows that Radial Basis Function Network is better than Back-Propagation Network at accuracy and stability.Finally a suggestion has been proposed that by establishing huge neural network database chloride concentration of different mixture ratio concrete under different environment can be predicted.