Step-type Landslide Deformation Prediction Based on Multivariable Self-optimizing Dynamic Neural Network
The traditional cumulative deformation prediction methods are diverse in curve structure decomposition meth-ods and characterization model selection,which brings about the problems of large workload of prediction methods,low predic-tion accuracy,and restricted applicability objects.To address the above problems,a multivariate self-optimizing dynamic neural network is established based on a nonlinear autoregressive model considering the effects of rainfall,reservoir water level and reservoir water level changes on the cumulative deformation of landslides.The neural network is applied to the prediction of the cumulative displacement of the typical stepped landslide in the Baijiabao landslide of the Three Gorges Reservoir.By analyzing the time series of the cumulative curve of landslide deformation,a nonlinear autoregressive neural network is composed by using a neural network to solve the full curve model.The parameters and structure of the neural network are optimized and trained u-sing multiple swarm genetic algorithms,and the mean square error of the fitness function is used as the prediction model error deviation criterion.The results show that the self-optimized dynamic neural network proposed in this paper has high accuracy in fitting the cumulative displacement of multiple measurement points of landslides.Its error can be controlled within 1%,and the prediction process reduces the error caused by subjective factors.The neural network can provide a reference for the prediction of cumulative displacement of such step-type landslides.
NETL
NSTL
万方数据
