汕头大学学报（自然科学版）

滑坡的发生受到自然、人为等多种因素的影响，使得滑坡隐患具有动态性，其识别和预测较为复杂困难.为了能够提高土建工程的安全性和稳定性，研究提出了基于改进双向长短期记忆网络算法的滑坡隐患识别方法.结果显示，研究模型的测试准确度随着迭代次数的增加而增加，在迭代次数为600左右达到0.98以上的准确度，而传统滑坡隐患识别评价模型在迭代次数为1650左右收敛，且准确度低于研究模型，为0.90左右.传统滑坡隐患识别评价模型的训练损失和测试损失分别为0.20和0.18，均大于研究模型的0.15和0.16.研究模型对研究区域不同乡镇的滑坡易发性进行评价，其中相比于其他乡镇，乡镇B、C的滑坡易发性小的区域更多，乡镇E、F、G的易发性较高的滑坡区域较多.可见通过识别并治理潜在的滑坡隐患，为避免因滑坡灾害导致的土地退化、生态破坏等问题提供参考，从而促进资源的合理利用和环境的可持续发展.

The occurrence of landslides is influenced by various factors such as nature and human activities, making landslide hazards dynamic and difficult to identify and predict. In order to improve the safety and stability of civil engineering, a landslide hazard identification method based on an improved bidirectional long short-term memory network algorithm is proposed. The results showed that the testing accuracy of the research model increased with the increase of iteration times, reaching an accuracy of 0.98 or above at around 600 iterations, while the traditional landslide hazard identification and evaluation model converged at around 1 650 iterations and had a lower accuracy than the research model, at around 0.90. The training loss and testing loss of the traditional landslide hazard identification and evaluation model are 0.20 and 0.18, respectively, both greater than the 0.15 and 0.16 of the research model. The research model evaluates the landslide susceptibility of different townships in the study area. Compared with other townships, townships B and C have more areas with lower landslide susceptibility, while townships E, F, and G have more areas with higher landslide susceptibility. It can be seen that by identifying and addressing potential landslide hazards. Reference can be provided to avoid land degradation, ecological damage, and other issues caused by landslide disasters, thereby promoting the rational use of resources and sustainable development of environment.