Optimizing ultra-high-performance concrete with recycled fine and CO₂ reduction strategies: a machine learning and swarm intelligence approach

The growing environmental impact of cement-based construction materials, particularly ultra-high-performance concrete (UHPC), necessitates innovative approaches to optimize mechanical properties while reducing carbon emissions. This study explores the environmental impact of UHPC and demonstrates how machine learning and advanced optimization techniques can improve its mechanical properties while reducing carbon emissions by analyzing UHPC formulations containing recycled concrete powder (RCP). The Random Forest (RF) and Long Short-Term Memory (LSTM) models, enhanced with Particle Swarm Optimization (PSO) and Chicken Swarm Optimization (CSO), were used to predict and optimize performance. The dataset consisted of 380 instances combining real and AI-generated techniques using a Variational Autoencoder, and includes the key UHPC components like cement, fly ash, silica fume, and water-to-binder ratio, with CO₂ emissions as the output. Results showed that RF models, particularly RF-CSO and RF-PSO, achieved high accuracy (RMSE: 7.04, MAE: 4.16), while LSTM-PSO excelled with temporal pattern (RMSE: 46.22, MAE: 4.53), highlighting its ability to handle non-linear patterns. Standalone LSTM performed poorly, emphasizing the need for optimization. The optimized UHPC formulations achieved up to 20.83% CO₂ reduction, aligning with the Global Cement and Concrete Association (GCCA) and demonstrating a meaningful step toward low-carbon concrete solutions. The research underscores the potential of machine learning and optimization in enabling sustainable, data-driven UHPC design, with RF models requiring minimal hyperparameter tuning and LSTM models needing more extensive refinement to ensure reliability and reduce errors.