Research on landslide displacement risk prediction and avoidance method

The accuracy and reliability of landslide displacement predictions are critical to disaster early warning systems; however, most current research focuses on improving numerical precision while overlooking the potentially severe consequences of displacement underestimation. This paper proposes a landslide displacement prediction method that integrates a risk-aversion mechanism. In this approach, double exponential smoothing is employed to capture long-term trend changes in landslide displacement, and a hybrid model combining Support Vector Regression (SVR) and Extreme Gradient Boosting (XGBOOST) is developed to characterize periodic displacement fluctuations. To address the risk of underestimation, a loss function incorporating a penalty factor is designed, and a constraint mechanism is introduced to adjust periodic displacement predictions toward a higher confidence interval. The predicted trend and periodic displacement components are then superimposed to generate a complete cumulative displacement forecast. In an empirical evaluation using the Dayangshan landslide in Suzhou, the proposed risk-averse prediction method not only maintains high predictive accuracy but also significantly reduces the rate of displacement underestimation, enabling dynamic and adaptive optimization of early warning strategies and risk control.