Improved Kernel Limit Learning Machine Based on Vector Weighted Average Algorithm for Concrete Compressive Strength Prediction
Haoran Sun · Applied and Computational Engineering · 2025
In this paper, a kernel-limit learning machine method based on an improved vector-weighted average algorithm is proposed for enhancing the accuracy of concrete compressive strength prediction. Quantitative relationships between components and compressive strength are revealed by Pearson correlation analysis: cement content shows the strongest positive correlation (r=0.50), followed by superplasticizer (r=0.37) and age (r=0.33), while water content shows a significant negative correlation (r=-0.29). The prediction model based on this model showed good performance in both the training and test sets, with a coefficient of determination (R²) of 0.913 and a root mean square error (RMSEC) of 4.875 for the training set, and R² of 0.867 and RMSEP of 6.171 for the test set, It is worth noting that the model's prediction accuracy in the test set decreased by only 0.046 units of R² compared with the training set, and the error increase remained reasonable. and the error increase remains within a reasonable range, indicating that the improved algorithm has excellent generalization ability and engineering applicability. By integrating feature correlation and intelligent algorithm optimization, this study not only provides a new method for concrete material strength prediction, but also constructs a quantitative model that can provide an important theoretical support for the optimal design of concrete proportion and the assessment of structural durability of engineering structures, which is of practical application value for improving the efficiency of concrete material research and development and the level of engineering quality control.