Research On Abnormal Diagnosis Method Of Landslide Deep Displacement Monitoring Sensor Based On Multivariate Correlation

Landslide is a frequent geological disaster in mountain areas,which endangers infrastructure and people’s life safety.Landslide monitoring and early warning is an important means to reduce and avoid disasters.The landslide remote automatic monitoring system collects the monitoring data of sensors of different types and positions in real time.Through the comprehensive analysis of these data,the deformation characteristics of the landslide are obtained.In the actual monitoring process,the sensor fault data will disturb the judgment of abnormal landslide movement.The study of data anomaly caused by sensor failure has practical application value for improving the precision of landslide early warning.The factors causing landslide mainly include landslide geotechnical composition,structure and physical and mechanical characteristics,rainfall,engineering construction,etc.These factors are interrelated.The displacement of landslide is usually caused by the joint action of multi-factors.The characteristics of landslide displacement are related to the action elements and its own geotechnical structure,which has a certain regularity.Different types of sensors are used to monitor different influencing factors,and multiple sensors of the same type are arranged at different locations for monitoring,so as to comprehensively analyze the movement characteristics of landslide mass from multiple perspectives and factors,thus achieving high-precision early warning.Deep displacement monitoring is to drill holes on the landslide mass and arrange displacement sensors at the same depth to monitor the displacement changes at different depths of the landslide mass.The monitoring data of different depth sensors are correlated.Concurrently,these deep displacement monitoring data are also related to surface displacement monitoring data,rainfall,water pressure and other monitoring data.This dissertation analyzes the variation characteristics of deep displacement monitoring time series data with human activity cycle,rainfall and water pressure,selects the corresponding time window,uses the gray correlation analysis method to calculate the similarity of each sensor,determines the change interval of similarity,forms the correlation matrix of landslide deep displacement sensor,and uses the correlation matrix to analyze the correlation characteristics of sensors at different depths in different landslide stages;According to the layout specification of deep displacement monitoring sensors,there is a sensor arranged in the rock stratum below the sliding surface,which is usually not affected by the sliding of landslide mass.The correlation confidence of other sensors is calculated based on this sensor.On this basis,the sensor failures are divided into three types: disconnection,noise and gradual damage.The data characteristics of various types of sensor failures are analyzed,and the following sensor failure judgment rules are established:(1)The correlation degree between other sensors in the correlation matrix and the reference sensor exceeds the correlation interval,while the correlation degree between other sensors is within the correlation interval,so it is judged that the reference sensor may be faulty.(2)When the reference sensor is normal,as long as the correlation degree between other sensors and the reference sensor exceeds the correlation interval,it is judged that the sensor may be faulty.(3)The sensor failure is determined according to the subsequent changes of 7 consecutive data points.The purpose of studying sensor failure is early warning.Landslide warning should be immediate,but the determination of sensor failure requires a certain observation time.Therefore,according to the different stages of landslide,the alarm rules in case of sensor failure are formulated:(1)When the landslide is in a stable stage,the sensor failure does not alarm;(2)When the landslide is in the deformation stage,if there is no rainfall in the first 15 days,the alarm will not work;if there is rainfall,the alarm will work;(3)When the landslide is in the critical sliding stage,the alarm will work.The experiment uses the monitoring data on the landslide monitoring website of Beijing Jiugan technology company.The website currently has thousands of monitoring landslides.Six landslide monitoring data were selected for method verification.The experimental results show that the method proposed in this paper has higher accuracy than kernel principal component analysis(KPCA)and LSTM depth learning method in the aspect of failure judgment of landslide deep displacement monitoring sensor.