Bridge Temperature And Strain Prediction Based On Meteorological Data Driven And Deep Learnin

Bridge temperature and strain are important mechanical indicators to evaluate the performance of bridges.Accurate bridge temperature and strain data can help to detect possible bridge diseases in time and design protection plans scientifically in advance,thus ensuring the structural safety of bridges.The bridge temperature and strain are closely related to the atmospheric environment in which the bridge is located,and meteorological data is an important parameter to measure the atmospheric environment,therefore,there is also a strong connection between meteorological data and bridge temperature and strain.Meanwhile,deep learning,as a machine learning method that has proven its superiority in several fields,provides a new way to build up a prediction model between meteorological data and bridge temperature and strain.Based on this,this paper uses deep learning neural networks to carry out research on the prediction of bridge temperature and strain from meteorological data,and the main work is as follows:(1)Firstly,the bridge health monitoring data of the Hedong Bridge and the meteorological data of Guangzhou city were collected,and the original data were manually screened,and the data from 2014 to 2016 were determined to be used for the mid-span and 1/4 span(L3,L4,L5sections)of the main span of the bridge and the meteorological station data of Baiyun Airport in Baiyun District,Guangzhou city for that time period.Secondly,in terms of data preprocessing,the outliers present in the original data were identified using the quadratic distance method and replaced with reasonable values;the wavelet 4-layer decomposition adaptive threshold processing was used to remove the noise in the original data as well as the vehicleinfluenced part of the strain data.(2)In the bridge temperature prediction,based on the correlation analysis results of the distance correlation coefficient,the LSTM model is established and Bayesian optimization is introduced to predict the temperature of all measurement points of the bridge cross-section with the meteorological temperature as the input,and the bridge temperature prediction from the meteorological temperature is realized,and the optimal historical observation time suitable for the temperature prediction model of L3,L4 and L5 cross-sections is determined to be 18 hours,and the number of LSTM layers is 2 layers.After that,the prediction accuracy was further improved by input feature combination analysis,and the optimal meteorological combination for temperature prediction was determined to be meteorological temperature and dew point temperature.(3)In the bridge strain prediction,based on the correlation analysis results of neural networks,the meteorological temperature was determined as the main controlling meteorological factor affecting the strain.Based on the LSTM model and Bayesian optimization,the strain prediction at each measurement point was performed by the meteorological temperature,and the number of layers of the LSTM model and the historical observation time were determined to be suitable for the strain prediction at each measurement point;the meteorological temperature was combined with other meteorological factors as the input to continue training the model,and the prediction accuracy of the model was observed under the multiple meteorological factors input,and the optimal combination of input meteorological factors for each measurement point was obtained to achieve The prediction of bridge strain by meteorological data was achieved.The prediction results show that for most of the measurement points,the prediction effect of the combination of multiple meteorological factors is better than that of the single meteorological temperature input;for most of the measurement points located on both sides of the section,the prediction accuracy is further improved by adding the meteorological wind speed input.