摘要
Abstract
To address the limitations of displacement prediction accuracy for reservoir bank landslides,a hybrid GA-CNN-LSTM model is proposed.The model utilizes CNN to extract spatial features from multi-source data,LSTM to capture temporal dynamics,and genetic algorithms(GA)to optimize LSTM hyperparameters(including neuron count,learning rate,and dropout rate).Validation using landslide monitoring data from the Three Gorges Reservoir area indicates that the proposed model outperforms both CNN-LSTM and plain LSTM models,reducing RMSE by 21.7％and 63.7％,respectively,and achieving an R2 of 0.9874.The results demonstrate that GA effectively mitigates overfitting and significantly enhances prediction performance,providing a reliable tool for early warning of landslides along reservoir banks.关键词
库岸滑坡/位移预测/CNN-LSTM/遗传算法Key words
Reservoir bank landslides/displacement prediction/CNN-LSTM/genetic algorithm分类
天文与地球科学
