Assessment of frost heave in coarse-grained soil: a novel application of multi-strategy enhanced dung beetle-optimized KELM model

Coarse-grained soils are typically classified by engineering standards as non-frost heave soils. Nonetheless, significant frost heave can still occur in the bases of canals composed of coarse-grained soil, leading to increased maintenance and repair costs for water conservancy projects. Consequently, accurately estimating the frost heave ratio of coarse-grained soils is crucial for mitigating canal freezing and safeguarding agricultural safety. Given the discontinuous and nonlinear nature of the frost heave processes, traditional experimental and theoretical approaches exhibit limitations in their capacity to accurately characterize these phenomena. This study leverages the strengths of the kernel-extreme learning machine (KELM) in terms of generalization capability and learning efficiency, further optimizing the KELM model's structure through the application of the multi-strategy improved dung beetle algorithm (MIDBO). This optimization yields enhanced convergence accuracy, accelerated convergence speed, and increased robustness. The model's input variables encompass initial water content, fine grain content, relative compaction, and overburden pressure, while the frost heave ratio is designated as the output variable. Consequently, an innovative prediction model based on MIDBO-KELM for the frost heave ratio of coarse-grained soil has been developed. The predictive performance of both the MIDBO-KELM model and the KELM model was rigorously assessed using various metrics, including RMSE, MAE, correlation coefficient (r), coefficient of determination (R²), variance accounted for (VAF), mean absolute percentage error (MAPE), residual standard deviation (RSR), Nash–Sutcliffe efficiency (NS), weighted mean absolute percentage error (WMAPE), index of agreement (IOS), a20 index, and index of agreement (IOA). The findings indicate that all evaluation metrics for the MIDBO-KELM model surpassed those of the KELM model, thereby demonstrating a superior predictive accuracy for the frost heave ratio of coarse-grained soil. The high accuracy and practicality of this method for forecasting frost heave in coarse-grained soils have been substantiated. The outcomes of this research provide a foundational basis for the design, maintenance, and investigation of frost heave mitigation strategies in seasonal frozen soil canals.