Intelligent Analysis And Prediction Of Mine Slope Deformation Monitoring Data Based On Gaussian Process

The exploitation of mineral resources can cause geological disasters, such as ground deformation and subsidence, mine slope instability, landslide, collapse, etc.,which seriously threaten the safety of lives and property. Automation and intelligent multi-sensor integrated monitoring systems are the current research direction of mine ground disaster monitoring. In this dissertation, the automatic monitoring system from No. 2 mine slope monitoring project of Pingshuo open pit mine was selected as a case study. Gaussian process(GP) theory is used to study intelligent analysis methods and prediction models, and provide a scientific basis for prevention and control of ground disaster in mining area. This study combined with automatic monitoring technology can be applied in disaster monitoring, and achieve to remote and online intelligent real-time monitoring, which has broad possibilities.The quality of monitoring data impacts the effect of the subsequent analysis and predictions. Therefore, the first step in this thesis is to analyze the reliability of the monitoring data in terms of how the gross error affects posterior variance. For this purpose, an algorithm is designed, which locates multiple gross errors and generates location matrix during the data processing. Next, reliability theory and the least-squares method are used to determine the estimation and correction equations of the gross error for use in a Gauss–Markov model. The algorithm, the estimation and correction equations are collectively known as full search estimation(FSE). Geo Robot gross errors detecting and coordinate conversion parameters reliability solving as two different experiments were performed to demonstrate the effectiveness of FSE. Experimental results show that FSE can provide better ability to resist gross error and its performance is superior to its rivals.Through using the FSE method, the reliability of monitoring data can be guaranteed, and the interference of gross error can be eliminated to identify the so-called ?true anomaly‘ of the deformed region. Furthermore, FSE creates favorable conditions for the intelligent analysis of deformation data using GP.Data interpolation problems often occur during deformation monitoring data analysis. In this thesis, a one-dimensional(1D) interpolation method in the time domain is proposed based on Gaussian process regression(GPR). GPR can adapt to linear and nonlinear interpolation in the time domain through selecting sample sizes. In the spatial domain, interpolation methods also based on GPR are developed according to the general spatial interpolation principle for selecting sample data.The posterior variance calculated as a weighting factor from the GPR is used for the spatiotemporal correlation of the monitoring data, and a cross validation method is used to demonstrate its feasibility.The analysis of the space–time displacement characteristics in the deformation region is the focus of the data analysis. The common absolute index describes the 3D space–time displacement characteristics of monitoring points, but it is not sufficient for determining the stable state of monitoring points. However, the cumulative displacement rate ratio(CDRR), the ratio of the short-term displacement rate to cumulative displacement rate, is a relative index that measures the relative stability of monitoring points. Furthermore, the magnitude and sign of the CDRR can provide simple and intuitive information on the deformation state of monitoring points. Through repeated calculation and analysis the CDRR over a period of time, according to the 3-sigma rule, the steady states of monitoring points can be divided into four levels: stable, weakly stable, unstable and extremely unstable.It is difficult to capture the entire space–time evolution trend and deformation law of the deformation region by only analyzing the deformation characteristics of monitoring points. Therefore, the absolute index(cumulative displacement) of all monitoring points is the subject of analysis here. Modeling methods are developed using GPR space–time interpolation to study the deformation trend surfaces of the No. 2 mine slope. Using the relative index(CDRR) of all monitoring points, the CDDR outliers are extracted using FSE and the results determine the Gaussian process classification(GPC). Next, a local stability analysis method is carried out based on GPC, which gives the local stability of the entire No. 2 mine slope.Monitoring points often show obvious nonlinear characteristics during deformation, therefore, the monitoring points deformation prediction models are carried out with GPR theory. Regarding the update and accumulation of monitoring data, a dynamic update mode of the super parameter ―progressive~truncation‖ and a method for optimization of the sample training set are developed. To ensure the prediction accuracy using GPR, the combined kernel function Matern32+SE is used, which fits the deformation characteristics by summing two kernel functions. To predict the deformation of the No. 2 slope, the GPR time-driven deformation intelligent prediction model(GPR-TIPM) and the data-driven deformation intelligent prediction model(GPR-DIPM) are used. The numerical experiments show that GPR-TIPM and GPR-DIPM are both satisfactory, GPR-TIPM performs better. Finally, using the same data, GPR-TIPM was compared with two classic prediction models, GM(1,1) and AR(p). The results show that the prediction accuracy of GPR-TIPM is the highest in short- and middle-term predictions.In the last part of this thesis, the design is presented for a prototype system architecture for deformation data monitoring and processing software. From this architecture, the server-side near-real-time data processing program and the geographic information system client-side online visualization analysis system were developed according to the models and algorithms presented in this thesis using Matlab and C# programming languages.