Abstract
Geopolymer is a new green and environmental friendly building material. However, its preparation process involves so many variables that there is no complete and standardized preparation method so far. In this work, back propagation neural network (BPNN) was used to forecast the slump and compressive strength of composite geopolymers with its high precision and engineering applicability, the prediction results of BPNN were also compared with random forest (RF) and k nearest neighbors (KNN) algorithm model. To train the BPNN models, a total of 191 data sets were used, which collected from different researchers in the open literature, and ten input parameters were considered, then 29 data sets were randomly selected by computer to verify that the error of the model is within the acceptable range. The BPNN model was run by Matlab, and the determination coefficient of determination (R) has been used for investigating the proposed model accuracy. As a result, the R values of the BPNN model for prediction slump and compressive strength are 0.9394; the R values of the RF model for prediction slump and compressive strength are 0.9017 and 0.9274, respectively; and the R values of the KNN model for prediction slump and compressive strength are 0.9121 and 0.9059, respectively. Those results show the performance of the BPNN model was better than the latter two models to estimate the strength and slump, it can be considered that the BPNN has the advantages of high accuracy, high efficiency and strong promotion ability in the prediction of composite geopolymers.
NSTL
Elsevier