摘 要：为了准确分析高速公路边坡性状，确保路基的稳定性，结合高速公路边坡环境特征，以边坡监测数据为基础，利用长短期记忆人工神经网络（LSTM）方法建立了高速公路边坡稳定性预测模型。以影响边坡稳定性的边坡质量系数、边坡结构系数、坡高系数、坡角系数、工程因素等因素为评估依据，采用降噪补缺、数据变换等方法处理LSTM前端数据，利用LSTM方法计算高速公路边坡稳定系数，与递归神经网络（RNN）方法进行比较。结果表明，高速公路边坡预测稳定系数为1.69，边坡安全稳定性良好，且符合实际。新方法的最大相对误差为1.60%，绝对MAPE仅为1.80%，较传统RNN方法预测更加精准。所得结论验证了深度学习在边坡稳定性预测评估过程中的有效性，对深入研究公路边坡稳定性具有借鉴价值。

关键词：岩土力学;高速公路工程;边坡稳定性;深度学习;LSTM理论;预测

中图分类号：U416   文献标识码：A

DOI： 10.7535/hbgykj.2021yx02011

Research on highway slope stability based on LSTM method

LI Bo

（Hubei Electric Power Survey and Design Institute Company Limited， Wuhan， Hubei 430000， China）

Abstract：

In order to accurately analyze the properties of the highway slope and ensure the stability of the roadbed， combined with the environmental characteristics of the highway slope， based on the slope monitoring data， the forecast model of highway slope stability was established by using the Long and Short-term Memory （LSTM） artificial neural network method. According to the factors such as slope quality coefficient， slope structure coefficient， slope height coefficient， slope angle coefficient， and engineering factors that affect slope stability， the LSTM front-end data was processed by noise reduction method and data transformation method， and the LSTM method was used to calculate the highway slope stability coefficient and was compared with the recurrent neural network （RNN） method. The research results show that the predicted stability coefficient of the highway slope is 1.69， the slope safety and stability is good and in line with reality. The maximum relative error of this method is 1.60%， and the absolute MAPE is only 1.80%， which is more accurate than the traditional RNN method. The conclusions verify the effectiveness of deep learning in the slope stability prediction and evaluation process， and have reference value for indepth study of highway slope stability.

Keywords：

geotechnical mechanics;highway engineering;slope stability;deep learning;LSTM theory;prediction

公路建設受到地形、地质、自然环境等因素的影响，公路边坡极易发生失稳、滑坡及坍塌等地质灾害[1]。高速公路边坡稳定性评估信息具有不完整性、多重性和不确定性等特点，如果评估不准确，容易导致事故的发生[2-3]。因此，对高速公路边坡稳定性预测的研究是有效提高高速公路边坡安全稳定性、降低安全风险事故的重要举措。

对于高速公路边坡稳定性的研究，贺为民等[4]针对高速公路滑坡现场进行勘查分析，引入极限平衡理论，运用FLAC法评估了公路边坡的安全稳定性，讨论了影响……