Prediction and Application of the Steel-Fibre-Reinforced Concrete Compressive Strength: Hybrid Methods with Stacking Ensemble Learning

As modern engineering has increasingly high requirements for material performance, steel-fibre-reinforced concrete (SFRC), as a composite material with excellent mechanical properties and durability, has been widely used in engineering. The compressive strength of SFRC, especially the uniaxial compressive strength (UCS), is a key indicator of its performance. Testing the strength of SFRC through indoor tests often requires a lot of manpower and material resources and has a long maintenance period. On this basis, this study proposed an SFRC compressive strength prediction model based on stacking ensemble learning. Using 211 collected sets of different SFRC mix proportion data, five single models were selected for stacking ensemble learning: support vector machine (SVM), decision tree (DT), K-nearest neighbor (KNN), random forest (RF) and back propagation neural network (BP). Moreover, six optimization algorithms were used to optimize the five single models. Then, the OP-Stacking hybrid model was obtained and used to predict the 7-day compressive strength of the SFRC, with MSE and R2 values of 86.9167 and 0.9398, respectively, which are better than those of the other five single models. The compressive strengths of SFRC at 7 days and 28 days were linearly fitted, and empirical formulas for the 7-day and 28-day strengths were obtained. Finally, the OP-Stacking hybrid model was encapsulated with 7-day and 28-day strength empirical formulas to establish an SFRC strength prediction system and the intelligent proportioning design, providing important support for the rapid construction of the new support design of the Central Yunnan Water Diversion Project.