Study Of Forecasting In Tailing Pond Accidents Based On Sparsity Bayesian Learning

Tailings are a class of major hazards. Tailings dam is one of the most important forms to result in mining disasters. So whether or not the stability of the tailings dam is directly affect security of lives and properties of people living in mines and down streams. In view of that, it is meaningful and necessary to research the stability of tailings dam. This thesis relies on online safety monitoring system and applies sparse Bayesian learning practice model---relevance vector machine based on probability learning to comprehensively analyze the regression analysis. On the basis of deep analysis of factors influencing the stability of tailings dam, the importance of saturation line and dam displacement factors is established and the input vector of tailings dam seepage line and forecasts of dam displacement is constructed. It aims at features of the continuity of saturation line and dam displacement to put forward time series forecasting model on account of sparse Bayesian and obtains satisfactory results on regression and model classification. Major research findings and conclusions area as follows:1) It deeply analyzes the tailings dam construction, theories, disasters and seriously concludes the manifestation and reasons of the tailings at home and abroad. This thesis firstly confirms some security factors influencing the stability of tailings dam and gets the monitoring unit according to these factors, which combines with relevant laws, regulations and constraints within the industry issued by the countries. Then according to the actual situation of tailings of small and long ditch in the county of Luanchuan, the final unit items are determined and online safety monitoring system aiming at the mine is designed, which includes on-site monitoring and early warning systems and post-release and Data Analysis sharing systems. In the prediction model, data of each monitoring unit is determined to be the data origin of the prediction system, and a portion of appropriate historical data is selected as the training samples of the training module for training. 2) It aims at infiltration line height time series and the dam displacement time series to conduct research, and makes use of sparse Bayesian machine learning methods based on probability theory and practical learning model:relevance vector machine. It constructs prediction model of infiltration line and dam displacement model in terms of regression and state classification of dam security monitoring dada and takes these two models as examples to completely validate on regression and classification of machine learning. Under the condition of employing the same learning sample and prediction sample, the relevance sector machine achieves better results comparing with support vector machine model of similar kernel function. The relevance vector machine can obtain high accuracy in the prediction because of its characteristics including high sparseness and learning structure based on probability. With the same kernel function and training samples, compared with support vector machine, the number of kernel function involved in the calculation in the relevance vector machine greatly decreases and the forecast calculation time is shorter. In addition, the relevance vector machine could use kernel function to probabilistic forecasting. It is because of these advantages of the relevance vector machine that make it greatly promising in safety monitoring prediction of the tailings dam. Particularly, in the application of time series prediction, it is more suitable for practical conditions of tailings inspection monitoring. Together with its advantages of high sparseness and probability prediction, it is greatly practical for real time prediction of tailings systems running trends online and determining the type of dam state.