Enhanced Dung Beetle Optimizer-Optimized KELM for Pile Bearing Capacity Prediction

The safety associated with the bearing capacity of pile foundations is intrinsically linked to the overall safety, stability, and economic viability of structural systems. In response to the need for rapid and precise predictions of pile bearing capacity, this study introduces a kernel extreme learning machine (KELM) prediction model optimized through a multi-strategy improved beetle optimization algorithm (IDBO), referred to as the IDBO-KELM model. The model utilizes the pile length, pile diameter, average effective vertical stress, and undrained shear strength as input variables, with the bearing capacity serving as the output variable. Initially, experimental data on pile bearing capacity was gathered from the existing literature and subsequently normalized to facilitate effective integration into the model training process. A detailed introduction of the multi-strategy improved beetle optimization algorithm (IDBO) is provided, with its superior performance validated through 23 benchmark functions. Furthermore, the Wilcoxon rank sum test was employed to statistically assess the experimental outcomes, confirming the IDBO algorithm’s superiority over other prevalent metaheuristic algorithms. The IDBO algorithm was then utilized to optimize the hyperparameters of the KELM model for predicting pile bearing capacity. In conclusion, the statistical metrics for the IDBO-KELM model demonstrated a root mean square error (RMSE) of 4.7875, a coefficient of determination (R2) of 0.9313, and a mean absolute percentage error (MAPE) of 10.71%. In comparison, the baseline KELM model exhibited an RMSE of 6.7357, an R2 of 0.8639, and an MAPE of 18.47%. This represents an improvement exceeding 35%. These findings suggest that the IDBO-KELM model surpasses the KELM model across all evaluation metrics, thereby confirming its superior accuracy in predicting pile bearing capacity.