Forecasting reservoir-induced landslide deformation using genetic algorithm enhanced multivariate Taylor series Kalman filter

Given the uncertain nature of reservoir-induced landslides, developing a relatively simple forecast model with high accuracy and strong robustness is still a challenge. This paper improves the classic Kalman filter to the genetic algorithm enhanced multivariate Taylor series Kalman filter (GMT-KF) model, which takes into account the rainfall, the variation of reservoir water-level, and lagging effect. Taking the Outang landslide in the Three Gorges Reservoir of China as an example, the time series data of three GPS monitoring points and two monitoring boreholes were compared with their predicted time series obtained from the GMT-KF model. The results show that the GMT-KF model attains both high accuracy and strong robustness. Moreover, the non-training nature makes the GMT-KF model holds much potential of satisfying the early-warning requirement of the emergent deployment of automatic monitoring systems at the reservoir-induced landslide sites.