Research and application for early warning of landslides based on hierarchical clustering coupling weighting Markov Chain model

One of the difficulties in landslide research can be always classified into two main categories:the nonlinear motion law of landslide and its stochastic state in the process of evolution. However,some existing landslide early-warn- ing methods and theories have low credibility,high error rate prediction,and untimely early warning. As can be seen, the displacement velocity progress in the landslide evolution can be regarded as a Markov process based on Markov chain theory and the basic idea of hierarchical clustering. Hierarchical clustering can convert GPS velocity observation signals to state signals. The model proposes a new description of acceleration larger than 0 of landslide displacement from the perspective of stochastic process. It is considered that when the monitoring displacement velocity state of the day,the day before,and the predicting displacement of the next day are all abnormal,and the mean and standard devi- ation of the dynamic sample displacement velocity continue to keep increasing,it heralds that slope is about to slide and then form a landslide. Considering the anti-interference and reliability of early warning model with respect to time- liness requirement,the paper presented an evaluation function to test the effectiveness of landslide early warning model by early warning sensitivity,early warning accuracy rate,and prediction consensus rate. Also,the landslide early warn- ing method of weighted Markov chain coupling hierarchical clustering has been put forward. The early warning model was tested with Pingzhuang open-pit mine engineering example. GPS surface displacement monitoring data was taken as the training dataset and validation dataset. The disquisition demonstrates that the early warning method proposed in the paper not only determines landslide early warning time,but also much reflects the promoter,occurrence,and pro- gress of landslide in the region of dynamic evolution. For the 4. 17 landslide engineering example in Pingzhuang open- pit mine,the optimal sample size is 20,the average 7-day landslide early warning sensitivity is 84% ,average early warning accuracy rate is 93% ,and the average prediction consensus rate is 76% . The parameter can meet the engi- neering requirements.