The application of neural network in dam safety monitoring

Dam safety monitoring is an important means for remaining the dam safe, while stress-strain monitoring has been an extremely important part in the dam monitoring. Sometimes the traditional forecasting methods are not high accuracy, so, in order to improve the accuracy of prediction. This paper presents a dam strain prediction model based on Least Squares Support Vector Machines(LS-SVM). Applied in one dam, LS-SVM shows the advantages of good robustness and high prediction accuracy. The strain prediction accuracy improves a lot than using the traditional stepwise regression method, so it provides reliable and effective ways and means in dam strain analysis. The objective of dam safety data analysis is to analyze the security state comprehensively for normal operating projects with the aid of the measured data, and seek the insecurity which may exist or will exist in actual project, so that we can take precautions against possible trouble [1]. Stress-strain monitoring of dam has been an extremely important part of dam safety monitoring, in order to understand variation of the dam strain with the variation of the external load, and predict the trends to make an important basis for security monitoring, mathematical models will be built by the known strain data, which will compare the measured value and predicted value, and determine whether the abnormal situation to the development trend of the anomalies and make the appropriate countermeasures. Currently, the prediction methods of the strain analysis of concrete dams are regression, deterministic model, hybrid model, neural network model, the gray model and so on. Statistical methods improve the factor of multicollinearity problems to some extent, but there are still some ineffective and inconvenient to use and other issues. Neural Network and Grey Theory model ignore some physical meaning, generally focus on the time trend of data, the environmental impact of the amount is few considered [2]. This paper uses LS-SVM method based on Matlab to analyze the strain data and forecast, which make a conclusion that the model based on this algorithm is superior through comparing the precisions of two methods.