RBF与改进BP神经网络的混凝土抗冻性预测

Prediction of RBF and improved BP neural network on concrete frost resistance

混凝土相对动弹性模量是反映混凝土抗冻性的一个重要指标,利用RBF和改进的BP神经网络,建立了混凝土相对动弹性模量的预测模型。该模型输入层为混凝土水灰比、水泥用量和冻融循环次数,输出层为混凝土相对动弹性模量。在采用建立的两种模型对试验结果进行预测时,先对输入数据进行标准化处理,同时对预测结果与试验结果进行比较。结果表明:采用改进的BP方法得到的预测结果相对误差均小于4%,比采用RBF方法得到的预测结果精度要高。

Relative dynamic elastic modulus is an important indicator of the frost resistance of concrete. Predictive models of relative dynamic elastic modulus are established by RBF and improved BP neural network. The inputting layer was water-cement ratio, ce- ment content and freeze-thaw cycles, respectively, and the outputting layer relative dynamic elastic modulus of concrete. The built models turned the inputting data into standardized data before prediction the test results by RBF and improved BP. The results showed the relative error is less than 4% by improved BP neural network. The prediction accuracy obtained by improved BP model is higher than that of by RBF method.