Artificial neural network-based model for predicting compressive strength of cemented gravels

In order to study the compressive strength of cemented gravel, a compressive strength prediction model was established; with 142 groups of in-house uniaxial compressive strength test data of cemented gravel with different mixing ratios as the object of study, the 11 influencing factors in the samples were downscaled by using the principal component analysis method, and the downscaled principal components were used as the input layer of the BPNN, so as to establish the compressive strength prediction of cemented gravel based on the PCA-BPNN model. The results show that when the number of principal components is 3, the cumulative contribution rate of the principal components reaches 86.483%, and the contribution rates of the three principal components are 65.084%, 11.702%, 9.097%, and the eigenvalues of the three principal components are all greater than 1. Then the three principal components are used as the input layer to build the compressive strength model of cemented gravel based on PCA-BPNN, and the RMSEs of the training set and test set are 0.253 and 0.28, respectively. The RMSE of the training set and the test set are 0.253 and 0.286, and the R2 is 0.947 and 0.942, which are not much different from the BPNN model with the number of nodes in the input layer of 6 and 11 before dimensionality reduction, and they are obviously better than the BPNN model in the convergence speed and training time. It can be seen that the PCA-BPNN model has good prediction effect and prediction efficiency, especially drastically reducing the training time, which provides a new idea for the prediction research of the mechanical properties of cemented gravel.