Abstract: The deformation of an expansive soil canal slope is significantly affected by drying-wetting cycles such as precipitation,groundwater level fluctuation,and evaporation. Displacement prediction contributes significantly to the prevention and mitigation of slope failures. One canal section of a well-known water transfer project,which has a typical deep excavated expansive soil canal characteristic,was taken as the studied case. The groundwater level of this canal section was high. The deformation of the canal slope exceeded the design warning value three years after the excavation,and continued to develop. The rigid support structure of the slope had been damaged due to this severe slope deformation. The vertical displacement of the canal slope and its relationship with groundwater level and precipitation were qualitatively analyzed based on the observed time series. The time series analysis and statistical models of the vertical displacement of the canal slope were developed. An integrated approach for the vertical displacement prediction combining variational mode decomposition(VDM)algorithm,least squares support vector machine(LSSVM)algorithm,and k-fold cross-validation was proposed. The accumulated vertical displacement was decomposed into a trend,periodic,and random component by the VDM algorithm. The VDM algorithm also decomposes influencing factors into high-frequency and low-frequency factors,corresponding to the periodic and fluctuating components. The LSSVM algorithm forecasts the trend displacement using the operation time as the input dataset. The periodic and fluctuating components of the influencing factors were selected as the input datasets to forecast the periodic and fluctuating displacement components by the LSSVM algorithm. Four influencing factors including groundwater level,precipitation,canal water level,and air temperature were considered. The total displacement was obtained by superimposing the three predictive components. The grey relation analysis determines the grey relation degrees between the influencing factors and the vertical displacement. All the grey correlation degrees between the influencing factors and the vertical displacement are greater than 0.6. Among them,groundwater level,effective precipitation,and air temperature are negatively correlated with vertical displacement. Canal water level is positively correlated with vertical displacement. The results prove that the VMD algorithm can decompose the trend,periodic and fluctuating displacement components well. The VMD algorithm can also decompose the influencing factors into the periodic and fluctuating components,and can identify local fluctuations. The time-dependent effect of the vertical displacement is significant during the initial operation period. The prediction accuracy of the proposed VMD-LSSVM model is obviously better than those of the statistical model and the S-LSSVM model which directly takes influencing factors as input factors. The proposed model presents satisfactory prediction accuracy,thus,is effective and of considerable practical value in expansive soil slope engineering.