A novel hybrid interval prediction method of reservoir colluvial landslides with step-like displacements is proposed to solve two significant difficulties in traditional methods, i.e., the inaccurate physical meaning of displacement components and the low computational efficiency of interval prediction models. According to the mechanism of landslides, the critical level model was adopted to establish the mathematical function to achieve the prediction of trend displacement components at first. And then, based on the trend components, the periodic displacement components were obtained by applying the time series decomposition model. After that, the Multi-Objective Grey Wolf Optimization-based Support Vector Regression model under the Upper and Lower Bound Estimation was adopted to construct the prediction interval of periodic displacements. Finally, the prediction results of trend displacement and periodic displacement components were combined to build the cumulative displacement prediction interval. Through the verification of the Baishuihe landslide case, the hybrid model has high computational efficiency and accuracy with high-quality prediction intervals.