Debris flow disaster are sudden and can cause significant property damage and casualties. In response to the problem of delayed prediction of debris flow disasters, this paper proposes a debris flow disaster time series prediction model based on an improved LSTM algorithm. Firstly, we obtained the dataset through on-site investigation and cleaning. Then, based on the PCA algorithm, the influencing factors of debris flow disasters were orthogonalized and dimensionally reduced, and 8 key factors affecting debris flow disasters were identified; Next, we established a time series prediction model for debris flow disasters using the LSTM network, trained, tested, and validated it with a dataset, and then conducted comparative experiments with traditional prediction models. The experimental results showed that the training error of the model was less than 0.06, the testing error was less than 0.12, and the validation set prediction error was less than 0.19. Compared with traditional prediction models, the model has high prediction accuracy and strong generalization ability.Overall, the method studied in this article has strong predictive ability and can provide certain decision-making support for relevant departments.