Outlier Detection Of Dam Safety Monitoring Data Based On LSTM Model

With the enhancement of national economic strength and the development of water resources,a large number of dams with world-class difficulty have been built(under construction)in China,these large-scale dam projects and many technical problems bring many challenges to the safety construction and operation of the dam.In order to evaluate the safety of the project,the rapid and accurate analysis of the dam safety monitoring data has become a key link,and the abnormal value detection of the safety monitoring data is the basis of data analysis,which is of great significance.In this paper,the abnormal value of the monitoring data of a high arch dam is detected by using the long and short-term memory network(LSTM)method,and the results are compared with those of the stepwise regression model and the pauta criterion(3 σ criterion).The conclusions are as follows:(1)In most cases,the single factor LSTM model used in this paper(only based on the historical measurement values to predict the current measurement values)can better fit the measured data and has strong prediction ability;(2)When the effect of stepwise regression model is good,the detection effect of LSTM model is equivalent to that of LSTM model;when the effect of stepwise regression model is poor,the detection effect of LSTM model is better than that of stepwise regression model;(3)Because most of the monitoring data are periodic(greatly affected by the water level and temperature),and the standard deviation of the monitoring data is large,the normal value range is too large and the detection effect is very poor when using the pauta(3 σ)criterion to detect the abnormal value.The detection effect of the LSTM model and the gradual return model is better than that of the pauta(3 σ)criterion;(4)The stepwise regression model needs a large number of regression factors as the basis of model establishment.In some cases(such as the lack of environmental quantity or the monitoring data itself is environmental quantity),the stepwise regression model can not be established,while the LSTM model is not restricted by the regression factors,and can normally achieve high-precision outlier detection.