Research on the prediction method of debris flow susceptibility based on PSO-LSTM-SAM

¹ Aneng Tibet Construction and Development Co., Ltd, 850000 Lhasa, China
² School of Information Science and Technology, Southwest Jiaotong University, 610000 Chengdu, China

^* Correspondence author's e-mail: xx_zxhua@swjtu.edu.cn

Abstract

To mitigate the losses caused by debris flow disasters, effective prediction methods are essential. However, traditional approaches often suffer from low accuracy and poor generalization in predicting debris flow events. To address these issues, this paper proposes a debris flow risk prediction method based on a Particle Swarm Optimization (PSO) and Long Short-Term Memory (LSTM) neural network, enhanced with a Self-Attention Mechanism (PSO-LSTM-SAM). By conducting slope debris flow simulation experiments, data were collected and processed from soil moisture content sensors, earth pressure sensors, and soil shear parameter measurements. The PSO-LSTM-SAM model was developed to improve the extraction of critical features and achieve accurate predictions of debris flow susceptibility. Finally, the results show that the proposed method can predict mudslide susceptibility with an accuracy of 88.4%, which is a significant improvement in mudslide prediction accuracy compared with the traditional algorithm, and verifies the effectiveness and feasibility of the proposed method.