Missing Data Imputation In Bridge Health Monitoring System Base On Hybrid Model Of SARIMA And Neural Network

Missing data exists in various surveys and engineering fields. Missing data may cause incomplete information so that it can bring a very negative impact to the following data analysis and treatment. Bridge health monitoring system evaluated the health status of the bridge according to the structural information of sensors installed in the key parts of bridge. Then it can provide a very important evidence for bridge owners and users to determine the safety of the bridge. However, monitoring systems long-term work in the harsh environment, a large number of missing data occurred because of the wear and damage of sensors and monitoring equipments, and seasonal lack of power supply and so on. The missing data may greatly affect the evaluation of the bridge health status. In this thesis, aimed to the missing data occurred in the bridge health monitoring system, a missing data imputation method based on a small sample population with a low error is presented for the accurate assessment of health status of the bridge. So the study of this thesis is of practical significance.In this thesis, the reasons why the missing data in the bridge health monitoring system appeared are analyzed, and several data missing cases are introduced. As the example, the data in the health monitoring system of Chongqing Dafosi Yangtze River Bridge data are processed. The characteristics of temperature and deflection and the correlations of themselves are analyzed. The missing data processing methods in other research fields are summarized. Based on the advantages and disadvantages of these methods, aimed to the characteristics of Dafosi Yangtze River Bridge's actual monitoring data, a hybrid model of SARIMA (Seasonal Auto Regressive integrated Moving Average) and Artificial Neural Network (ANN) is presented to the missing data imputation.In order to compare the advantage of different methods, missing data in bridge health monitoring system are processed respectively by the time series SARIMA method, the traditional linear regression method and the hybrid Model of SARIMA and Neural Network method which presented in this paper. The results show that the hybrid model method has the best imputation precision, lowest error and best effect. The SARIMA model is worse than the hybrid model.The actual missing data imputation results show that hybrid model of neural network and SARIMA methods which presented in this paper can achieve quick imputation effect of missing data based on small sample population with a lower error.