Study and verification on an improved comprehensive prediction model of landslide displacement
Tianlong Wang,Rui Luo,Tianxing Ma,Hao Chen,Keying Zhang,Xu Wang,Zhaowei Chu,Hongyue Sun
DOI: https://doi.org/10.1007/s10064-024-03581-5
IF: 4.2
2024-03-02
Bulletin of Engineering Geology and the Environment
Abstract:Currently, many models are extensively employed for landslide displacement forecast; however, the potential impact of structural parameters on the accuracy of model predictions remains inadequately addressed. Additionally, the optimization algorithm commonly utilized in the realm of landslide displacement forecast still faces certain challenges, including a restricted search area and a propensity to converge on local optimal solutions. In this study, a new comprehensive prediction model is proposed. To begin with, the conventional Aquila optimizer algorithm (AO) has undergone improvement in three key areas: population initialization in Tent chaotic mapping, global optimal individual update on adaptive weight factor, and population update based on opposition-based differential evolution. The improved Aquila optimizer algorithm (IAO) and AO have been evaluated through benchmark function comparisons to verify the effectiveness of three improvement strategies. The results demonstrate the effectiveness of it, especially in terms of its enhanced convergence speed, optimization accuracy, and robustness. This led to the utilization of IAO for parameter optimization of variational mode decomposition (VMD) and bidirectional long short-term memory neural network (BiLSTM), ultimately constructing the IVMD-IAO-BiLSTM prediction model. Then, taking the Qili landslide in Zhejiang Province of China as an example, combined with two years' worth of independent monitoring data of rainfall, groundwater level, and surface displacement, the displacement series is predicted based on IVMD-IAO-BiLSTM. Under the same prediction conditions, it is compared with the multi-factor-multi-scale model proposed by Xiong C and ten commonly used prediction models. The results validated the superiority of the IVMD-IAO-BiLSTM model, which had the lowest root mean square error and mean absolute error among all prediction results at 0.14 and 0.09, respectively. Therefore, IVMD-IAO-BiLSTM effectively harnesses the enhancing effect of structural parameters, providing a new approach for early warning and risk assessment of landslides with its notable prediction accuracy and performance.
engineering, environmental,geosciences, multidisciplinary, geological