Intelligent Algorithm Optimization of BP Neural Networks for Prediction of Compressive Strength of 3D Printed Concrete

In this study, the BP neural network prediction model was used for predicting the compressive strength of 3D printed concrete (3DPC). A dataset was created from the literature and experiments. The BP neural network topology (ANN7-8-1) was designed based on the correlation between raw material quantities and compressive strength. Intelligent algorithms such as the grey wolf algorithm (GWO), differential evolution (DE), and hybrid optimization algorithm (DE-GWO) were used to optimize the weight thresholds for the BP neural network. Training and prediction were performed using the dataset. The BP neural network prediction model for 3DPC compressive strength, optimized using intelligent algorithms, exhibited improved global search and convergence performance. GWO exhibited high convergence accuracy, while DE exhibited fast convergence speed. DE-GWO yielded significantly improved accuracy and convergence speed compared to GWO and DE. The correlation coefficient R2 was 0.9087, and the absolute percentage error MAPE was 7.25%. The DE-GWO-BP neural network can provide guidance for optimizing mix proportions and controlling the performance of 3DPC.