Study On Dam Monitoring Statistic Model Based On Kalman Filtering

In order to meet the requirement of dam safety monitoring, according to previous statistic method and focus on the defects of least square regression statistic theory, that is it needs large numbers of observation datum, model's precision is not great and it can't real-time observe, the statistic models for dam monitoring were put forward in the dissertation based on Kalman filtering, wavelet and BP neural network. The correctness and feasibility of the models were validated through the engineering practice.The contents and major results are as follows:(1) This paper reviewed previous statistic method, and compared with these methods and found out respective merits and faults.(2) Based on Kalman filter theory, the Kalman filtering regression statistic model was built. Set up state equation and observation equation based on statistic model, which transformed least square estimation to Kalman filtering state estimation. It was shown by study that this model provided a recursion solution for estimation with the Kalman filter, which can simplify the modeling process as on-line refreshing state conveniently through the previous time state value and present measure value, so it was a new efficient on-line modeling method. It was shown by comparison between actual project data and modeling results that the structure of Kalman filtering statistic model was stable when the parameters were chose within reasonable extent. Similarly, the change of model regression coefficient also can used in evaluation of dam safety state.(3) As a result of the instrument failure and other complex factor, the dam measured data often appear the singular value and much noise. Sometimes the useful information even can be submerged by the noise. In view of this, this paper combined the Kalman filtering with the wavelet multiresolution theory, and established wavelet multiresolution Kalman filtering regression statistic model. The two have the complementation in eliminating noise of the dam observation data, so this model could apply to the high noise, expanded the Kalman filtering's application scope, also increased the forecast precision.(4) Because the Kalman filtering regression statistic model belongs to linear model, in consideration of non-linear characteristics of dam, the outstanding non-linear mapping ability ofBP neural network model was combined with the ability of Kalman filter's recursion solution for estimation successfully, built the dam monitoring model of BP neural network and high-order nonlinear Kalman filtering. The presented algorithm can overcome the disadvantage of long training time for BP neural network and slowly convergence speed and overlap iterate, and be high degree. It was demonstrated that the method was a new and efficient neural network algorithm with the on-line training feature.