Objectives  The inadequate utilization of multisource monitoring data and the unstable results of displacement prediction are often caused by inaccurate extraction of random components, uncertain optimal training data set and timeliness in the comprehensive landslide displacement prediction study.

  Methods  On that account, a new landslide prediction model is proposed by integrating the variational mode decomposition with the gated recurrent unit recurrent neural network on the basis of landslide displacement time series analysis.

  Results  Taking Baishuihe landslide in the Three Gorges Reservoir Area as an example, the monitoring data including displacement and reservoir water level and rainfall data from July 2003 to December 2012 are selected for analysis and research. The root mean square error of the predicted value of the model is 9.715 mm and the coefficient of determination is 0.967. The results show that the model guarantees high prediction accuracy and has obvious advantages in effectiveness and timeliness as well.

  Conclusion  Therefore, it has a strong application and popularization value in reservoir bank landslide displacement prediction research.