EFFECT FACTORS’ SELECTION AND PREDICTION OF COMPRESSIVE STRENGTH OF SCC USING A HYBRID NETWORK BASED ON GA

Abstract

Compressive strength is the most important evaluation index for concrete. In order to predict the compressive strength of self-compacting concrete, two kinds of artificial neural networks (ANNs), including the BP (Back-propagation) networks and the hybrid networks DRGA-BP based on GA (Genetic algorithm), were designed and applied in this study. With DRGA-BP, the most representative variables were selected out from many initial inputs to reduce data dimensions and also the weights and thresholds of BP model were optimized. The results showed that the hybrid model presented better prediction accuracy with the R2 (coefficient of determination) of 0.9602, and appeared to well agree with the experimental data and was quite reliable. Finally, a mix ratio design method based on DRGA-BP model was proposed for reducing material waste and saving time in the process of concrete production with continuous adjustment.