Chaotic Characteristics Identification and Prediction Using PSO-LSSVM Model of Reservoir Landslide Groundwater Level Time Series

Reservoir landslide groundwater level time series prediction in Three Gorges Reservoir area is of great significance for landslide stability analysis.The groundwater level time series may be of chaotic characteristics under the impact of external factors such as seasonal heavy rainfall and periodic reservoir water level fluctuation.Based on the phase space reconstruction of reservoir landslide groundwater level time series,saturation correlation dimension method and maximum Lyapunov exponent method were used to verify the existence of chaotic characteristics of groundwater level time series.Then Least Squares Support Vector Machine(LSSVM)model with high prediction accuracy was proposed for groundwater level time series prediction.The Particle Swarm Optimization(PSO)was applied to select the optimal combination for the parameters of LSSVM model.The proposed PSO-LSSVM model can resolve the difficulty in parameters selection of LSSVM model.Daily average groundwater level series of STK-1hydrology hole on the Sanzhouxi landslide in Three Gorges Reservoir area was taken as an example to verify the existence of chaotic characteristics.PSO-LSSVM model was compared with BP Neural Network model.The results show that the groundwater level series is of obvious chaotic characteristics,the Root-Mean-Square Error and goodness of fit of PSO-LSSVM model are 0.193 mand 0.815,respectively.In addition,prediction accuracy of the proposed model is higher than that of BP Neural Network model.The proposed model reflects the evolution law of reservoir landslide groundwater level time series effectively and thus is greatly practicable.