Performance assessment of basalt fibre concrete under freeze-thaw cycles using hybrid long short-term memory models

In cold regions, freeze-thaw cycles（FTC） significantly degrade the performance of concrete. The synergistic use of different admixtures is an effective anti-freezing method, but traditional testing methods are costly and time-consuming. To optimize current technical approaches, this study proposes a Ensemble Model combining Long Short-Term Memory (LSTM) networks with swarm intelligence optimization algorithms to address the adaptability of multi-parameter materials in basalt fiber concrete (BFC) under FTCs. By integrating self-conducted experimental data and referenced datasets, a diverse experimental database was constructed. Improved algorithms, namely Asynchronous Learning Particle Swarm Optimization (AsyLnCPSO) and Hybrid Genetic Algorithm-based Particle Swarm Optimization (GA-HIDMS-PSO), were paired with LSTM neural networks to systematically evaluate their adaptability and effectiveness in predicting the performance of basalt fiber-reinforced concrete under freeze-thaw conditions. The results demonstrate that swarm intelligence optimization algorithms (SIOA) significantly enhance the prediction accuracy and stability of the LSTM model. The two modified particle swarm optimization algorithms showed notable improvements: AsyLnCPSO performed well on the training set but exhibited limited adaptability to overall data variations, while the GA-HIDMS-PSO excelled in high-dimensional and complex constraint problems, demonstrating the strongest stability with a fitting accuracy of 0.938, significantly improving the prediction accuracy of concrete performance under FTCs. Further SHAP interpretability analysis revealed that basalt fibers (BF) and mineral admixtures play a positive role in frost resistance. This study combines the automation capabilities of swarm intelligence algorithms with the processing power of deep learning to develop a precise automated design model, providing a robust technical approach for adapting to complex engineering environments and advancing the application of intelligent design methods in civil engineering. The findings not only offer technical support for the efficient and sustainable design of concrete materials but also lay the foundation for their widespread application in complex engineering environments in the future.