A hybrid approach based on reservoir computing for landslide displacement prediction

Time series prediction approaches are studied in our research of landslide displacement prediction. First, the ideas of the two different types of time series prediction approaches are discussed. Reservoir computing, the algorithm for training recurrent neural networks into predictors, is expanded into a general form of establishing dynamic models that can predict the target time series. Then following the expanded concept of reservoir computing, a hybrid approach is proposed. By combining the considerations of different prediction strategies, this hybrid approach reflects both the impacts of internal and external factors on landslide displacements, and therefore can produce reliable predictions. Effectiveness of the proposed approach is validated in our experiments implemented on practical landslide displacement recordings.