Research On Multi-source Heterogeneous Sensor Information Fusion For Landslide Monitoring

Landslide as a common geological disasters with great harm usually occurred in the nature, which often result in significant losses to people’s lives and property. China has a vast territory and complex geological conditions, and the geological disasters occurs frequently, which cause great damage. According to preliminary statistics, china lost 180 to 300 billion RMB caused by slope deformation such as landslide, collapse, etc. With the continuous and deep study about the evolution of landslides, more and more researchers concentrate on the early warning and forecasting of landslide disasters, and proposed plenty of techniques to prevent and control the landslide disasters.Landslide has the characteristics of multi field evolution under the control of the leading factors, and the evolution process is divided into the gradient type, the sudden type and the stable type. Confirming the evolution process and the stage of landslide are the basis of the landslide prediction, and the multi field characteristic information of landslide under complex environment is an important basis to confirm the evolution process and the stage. So monitoring the characteristics of landslide, tracking the dynamic changes of landslide, processing the monitoring data, promptly discovering and predicting the landslide threshold are the effective technique to make the landslide prediction.With the rapid development of the Internet of things technology, the application of multi-source heterogeneous sensor networks is often used in landslide monitoring, which gradually replacing the traditional manual monitoring, and play an important role in monitoring and early warning. Effective establishment of multi-source and heterogeneous sensor integrated monitoring system, extracting the multi field feature information of the landslide, and multi field information fusion processing and decision analysis are the main methods for landslide prediction.Through the research on the data processing method in the existing landslide monitoring system, the effect of diverse information fusion methods on the landslide prediction is different, so researchers begin to study the data fusion algorithm and the forecasting model. At present, the research of information fusion is mainly focused on the fusion algorithm, the fusion structure and the specific information fusion system modeling and implementation.We have found that there are lots of key issues and difficulties in information fusion applications based on the landslide monitoring system with multi source and heterogeneous sensor Research, Mainly for:1) The Lots of data conversion contents:Because the data form of multi sensor output, the description of the environment is different, therefore, the primary task of information fusion is to convert these data into the same form and description, and then carry out the relevant processing. The contents of which include the conversion of different levels of information, as well as the conversion of the same level of information, which can form the same description, resulting in the transformation of the calculation process is more complex, the system overhead is too large.2) The difficult of data correlation:In the fusion processing, the accuracy of sensor measurement and the interference of various environment are easy to cause the relevant two sense, which leads to the difficulty of the correlation, and the difficulty of the correlation coefficient is difficult to be solved.3) The complexity of data management:In information fusion system, the real-time database and non real time database are usually built to manage the information. The role of real-time database is to provide the results of the current sensor to the fusion center in time, and to provide all kinds of other data needed by the fusion computation. The intermediate results and the analysis results are also stored in the fusion processing. The non real time database stores the historical data of each sensor, the auxiliary information about the monitoring object and environment, and the historical information of the fusion calculation. In the process of information fusion, the capacity of the database is large, the search speed is quick, the open connection is good, and it can provide a good user interface to other heterogeneous platforms.4) The single of the prediction model:The core of the information fusion system is to verify, analyze, add, modify, and finally tracking and estimate the landslide trend based on the fusion model. But in the present system, the fusion model is too simple, and the accuracy of the landslide is not obvious, the probability of successful prediction is too small.Based on the above problems, and widely consulting the domestic and foreign relevant to landslide prediction method, the multi sensor information fusion theory and monitoring method based on the literature. This paper first analyzes the history and present situation of evolution, prediction of landslide, then summarizes the information fusion theory in landslide prediction, application characteristic, the development of trend, and combines with my participation in the research project three gorges of landslide monitoring information system development, proposed the ideas to solve the problem:l)Based on the problem of the conversion of the excessive content, the method of multi-source heterogeneous data characteristics is studied. The G-CWT algorithm for multi-source data unsteady efficient parallel computing is proposed to improve the fusion efficiency. The fusion efficiency is improved; Pixel level fusion processing is performed on the underlying data formed by the formation of multi-source heterogeneous sensors, using the theory of multi source sensor to realize the abnormal values between the information, and reduce the error caused by the nonlinear transformation between the data;2)Based on the problem of data correlation, combined with the analyzes of the multi source heterogeneous data in correlation monitoring system, the multiple regression analysis model is proposed to discover significant correlation factor and multi-factor correlation coefficient between calculated by the least squares method to improve the status of landslide monitoring multi-source data fusion process related determines missing quantify mechanisms;3)Based on the data organization and management, the organization and management model of multi-source heterogeneous data is proposed, which is based on point source data warehouse technology, and the information processing platform is formed, which can be shared and integrated in the landslide monitoring data;4)Based on the problem of the single model, the feasibility study is carried out to predict the feasibility of the combination of multiple model combination. The combination model is suitable for the characteristics of multi field monitoring system. Breakthrough the limitations of a single fusion system, improve the probability of successful prediction.The works of this paper includes:(1) based on the current situation of geological disasters in the Three Gorges Reservoir area, and the research center of geological disasters in the Three Gorges Reservoir area, the theoretical research of landslide disaster monitoring and data management platform is carried out. (2)Time series data for monitoring system performance formed the diversity of features, such as large samples and small sample data, linear and nonlinear data, stationary and non-stationary data coexistence of the system of multi-sensor data in a time-domain sequence method was analyzed, using Xiao Weina test algorithm determines outliers and reasonably excluded, the method proved to be an effective outlier detection methods, combined with examples demonstrate the fusion of multi-source sensor valuation method prevents false outliers; and intuitive scatter diagram method determines the distribution of stationary monitoring data, cubic spline interpolation method of polynomial fitting and data smoothing and achieve pre-filled and other issues, feasibility through simulation experiments prove the method described above; for more information Fusion conversion processing overhead big problem, GPU-based wavelet transform parallel computing method (G-CWT), the GPS data were non-stationary massive parallel calculations and simulations, improving the computational efficiency of the fusion system. (3) Landslide for highly nonlinear characteristics of complex systems, and the introduction of small samples can be a better solution, nonlinear regression analysis of time series forecasting problems fitting method for multi-source heterogeneous monitoring data fusion process, the establishment of the landslide by multivariate regression analysis multi-factor correlation model between variables, multiple sources and then monitoring samples carried out experiments and tests show that the fusion method for reducing the prediction error is valid. (4) After analyzing the factors affecting the deformation of landslide, the main components of the model are analyzed, and the input variables of the neural network model are determined by genetic algorithm. The model parameters are optimized by genetic algorithm and the BP-GM model of the landslide deformation is established by combining the characteristics of GM gray prediction model. The combined model is used to analyze and predict the deformation of the Loess Slope in Badong County, and the results are better in the simulation experiments. Theoretical and experimental results show that the combined forecasting model is better than the single model.The work of this paper is to provide reference for the design of modern landslide monitoring system based on Internet of things, emphatically introduces the abnormal value determination and eliminating method,and analyzes the function of the fusion method of multi-source data fusion in the prevention of miscarriage of justice. which aims to improve the efficiency of the integration of data from multiple sources and heterogeneous sensor data and provide a new theoretical method for landslide prediction using G-CWT parallel processing method. conducts forecasting and processing of multi-source data and experimental result analysis by using multiple stepwise regression analysis. At last the paper introduces the BP-GM combined model prediction, numerical analysis indicate that the prediction accuracy of the combined model is higher than that of a single model, which is proved to be a desirable method to optimize the results,and provide a new theoretical method for landslide prediction.