Prediction interval estimation of landslide displacement using adaptive chicken swarm optimization-tuned support vector machines

Abstract

Accurate and reliable displacement prediction is vital for developing landslide early warning system, since displacement variations directly reveal the evolution of landslide. The past decade has seen the rapid development of displacement prediction technology. Most of the existing studies are based on deterministic point prediction models, with the goal of obtaining high-precision predictions. However, these pointwise predictions do not give any indication of credibility, which makes it difficult to make accurate decisions for subsequent landslide treatment. To address this problem, a prediction interval estimation method, instead of point prediction, for landslide displacement is proposed. In the methodology, double exponential smoothing is employed to deal with the nonlinear cumulative displacement, while two support vector machines are proposed to directly generate the displacement lower boundary and upper boundary. To obtain optimal model parameters, the adaptive chicken swarm optimization using chaotic mapping and adaptive inertia weight strategies is proposed to minimize a modified prediction interval-based objective function. A real-world landslide displacement data set is used to demonstrate the proposed method. Experimental results show that it is a promising tool for the construction of high-quality prediction intervals, which can inform the landslide treatment-related decision-making.