关键词: 滑坡位移预测　, GNSS位移监测　, 时间序列分解　, 双指数平滑法(DES)　, 粒子群优化极限学习机 (PSO-ELM)

Abstract: In order to improve the prediction accuracy of landslide displacement,a combined prediction model combining double exponential smoothing (DES) and particle swarm optimization extreme learning machine (PSO-ELM) is proposed. Based on the time series decomposition theory,the model uses DES to predict the trend term displacement controlled by geolog ical conditions,uses ELM to learn the periodic term displacement caused by external factors such as rainfall,and demonstrates the ignorability of the random term through the significance test.The GNSS monitoring data of a typical landslide in Chongqing is used as a case to verify.The results show that the mean absolute error,root mean square error and mean absolute percentage error of the proposed DES-ELM model in the total displacement prediction of GB07 monitoring points are 1.982 0 mm,2.679 4 mm and 1.372 3%,respectively,and the coefficient of determination is 0.909 1.The prediction accuracy is significantly bet ter than the comparison models such as polynomial-gated recurrent unit (GRU).Furthermore,after introducing the particle swarm optimization (PSO) to optimize the ELM hyperparameters,the mean absolute error and root mean square error of the pe riodic term prediction are further reduced by 76.33% and 76.70%,which effectively alleviates the phase lag of the original ELM model and improves the stability and goodness of fit of the model.This study provides an effective and reliable solution for high-precision prediction of landslide displacement.

Key words: landslide displacement prediction,GNSS displacement monitoring,time-series decomposition,double exponential smoothing(DES),particleswarmoptimization-extremelearningmachine(PSO-ELM)