基于奇异谱分析法和长短时记忆网络组合模型的滑坡位移预测

Landslide Displacement Prediction Based on Singular Spectrum Analysis and a Combined Long Short-term Memory Neural Network Model

为提高滑坡位移的预测精度,提出了一种基于奇异谱分析法(singular spectrum analysis,SSA)和长短时记忆网络(long short-term memory neural network,LSTM)组合的动态预测模型.首先,利用SSA将滑坡位移时间序列分解为趋势项、周期项位移子序列同时剔除噪声,减少随机波动对实验结果的影响.然后利用高斯拟合方法对趋势项位移子序列进行拟合预测;LSTM神经网络模型对其周期项位移子序列进行预测.最后,通过叠加各位移子序列的预测值,得到累积滑坡位移的预测值,并且通过在训练集中加入预测值来更新LSTM网络,实现动态位移预测.结果表明,与经典反向传播(back propagation,BP)神经网络、支持向量机(support vector machine,SVM)和差分自回归移动平均(auto regressive integrated moving average,ARIMA)三种预测模型相比,LSTM模型更优.该模型在中国新滩滑坡中得到了验证,最终预测值也表明该组合模型有较高的精度.

In this study,we propose a dynamic prediction model based on singular spectrum analysis(SSA)and long short-term memory neural network(LSTM)to improve the prediction accuracy of landslide displacement.First,we use SSA to decompose the landslide displacement time series into a trend term and a periodic term displacement subsequence.We eliminate noise to reduce the influence of random fluctuation on experimental results.Then,we use the Gaussian fitting method to fit and predict the trend term displacement subsequence.We apply the LSTM neural network model to predict the periodic term displacement subsequence.Lastly,we obtain the prediction value of cumulative landslide displacement by superimposing the predicted values of each displacement sequence.We update the LSTM network by adding prediction values to the training set to achieve the dynamic displacement prediction.Results show that the LSTM model is better than the traditional back propagation(BP)neural network,support vector machine(SVM),and differential autoregressive integrated moving average(ARIMA).Furthermore,we verify the model in Xintan Landslide in China.The final predictive value demonstrates that the combined model has a high accuracy.