Optimization of prediction models for the compressive strength of recycled aggregate concrete using artificial neural networks

Recycled coarse aggregate is increasingly utilized as a substitute for natural coarse aggregate in concrete production. The mechanical properties of recycled aggregate concrete (RAC) are largely influenced by the qualities of recycled coarse aggregate. Typically, the development of a strength prediction model for RAC emphasizes the effect of each component's content on strength while overlooking the influence of recycled coarse aggregate qualities on the compressive strength of RAC. This paper investigates the significance of input variables, particularly key properties of recycled coarse aggregate, such as water absorption, particle size distribution, and crushing index, and identifies the optimal combination of input variables and hidden layer nodes, while determining the best ratio of training to test sets. Furthermore, the backpropagation (BP) neural network was optimized, and the performance of various machine learning models for predicting compressive strength was evaluated and compared. The results indicated that the prediction performance of the BP neural network improved significantly under the optimal combination of input variables and the optimal number of hidden layer nodes. Moreover, the BP neural network outperformed other commonly used machine learning methods, including the radial basis function (RBF) neural network, support vector machines (SVM), and linear regression.