Dam Safety Monitoring Research Based On Blind Source Separation And Relevance Vector Machine

Dam safety is the top priority of social,national defense and ecological security.It is an important means to ensure the safety of the dam by carrying out the safety monitoring of the dam and grasping the safety state and the development trend of the dam.With the development of dam automatic monitoring system,the amount of data is growing geometrically and the data scale is expanding.How to deal with these data efficiently and scientifically becomes a new research hotspot.Based on the massive data processing and analysis demand as the starting point,dam integrity analysis as the main train of thought,research and analysis of dam monitoring data using related method of blind source separation theory and Relevance vector machine,in order to find scientific,reliable and efficient monitoring data processing method.The main research work of this paper is as follows:This paper studies the interpolation method of dam missing monitoring data,and puts forward a method based on Kernel Independent Component Analysis and relevance vector machine.According to the correlation between the monitoring data of the dam,this paper puts forward a method of dam missing monitoring data interpolation,in which method the relevant measured points are used to estimate the target points.The kernel independent component analysis(ICA)is used to extract the statistical independent feature information from the relevant measured points.The method has the advantages of high precision,strong adaptability and easy operation.In order to improve the efficiency of monitoring data analysis,a multi-point monitoring model based on fast independent component analysis and relevance vector machine(FastICA-RVM)was established.Fast independent component analysis algorithm is utilized to extract the characteristic information of the multipoint,and the algorithm is applied to remove the correlation dimension among independent variable factors for forecasts impact factor,and then use the good function approximation ability of the relevance vector machine to predict the characteristics of multi measured points.The simulation results show that the model can achieve the same prediction accuracy level of single point model.At the same time,because the relevance vector machine can not only output the deterministic forecasting results,but also can output the probability distribution of the predicted value,the prediction interval under a certain confidence level is given,which is helpful to guide the decision-making of the monitoring system.For the monitoring data of the physical quantities such as the crack opening of the hydraulic structures are generally non-stationary,a multi-point time-varying dam monitoring model based on time-frequency distribution blind source separation and multiple output relevance vector machine(TFBSS-MRVM)was established.Non-stationary feature information of multi measured points was extracted by the time-frequency distribution blind source separation algorithm which can effectively extract non-stationary signal time-varying feature.And multi-output relevance vector machine is used to predict multiple characteristic information simultaneously,which can further improve the monitoring efficiency.The example shows that the model has high accuracy for the prediction of the crack opening and other non-stationary monitoring.