Radial Deformation Prediction For The First Impoundment Of Super-high Arch Dams Based On An Integrated Model Of Fusion Mechanism

With the advancement and development of artificial intelligence in the field of water conservancy engineering,it has become a research focus on how to use monitoring data to predict dam deformation during the first storage period of a very high arch dam in an efficient,accurate and reliable manner.As the water level rises and the time increases the amount of monitoring data on the deformation of the dam during the first storage period of a very high arch dam is small to large,and it is obvious that using a single method to complete the water level-deformation prediction for different data levels is inaccurate.In view of the complexity and uncertainty of the temperature and stress state of the dam during the first storage period of a very high arch dam and the large errors in the predicted values of dam deformation for multiple storage targets during the first storage period of a very high arch dam by existing methods,this paper proposes a set of integrated HST-NN models using a mechanism-data model fusion approach.The main research work in this paper is as follows:(1)An integrated HST-NN model is designed.The formula is chosen to be combined with the neural network to build the integrated model,so that the data-driven neural network model also has the constraints of the formula to achieve better prediction and generalisation ability.An integrated model of formulae and deep neural networks is built to explore the generalisation ability of the model,as well as the model training and prediction effects when only a small amount of data is available.The results show that when there is enough data,both the pure deep neural network model and the formulaic model can achieve good prediction results and can accurately investigate the maximum deformation value corresponding to the highest water level.However,when the amount of data is small,the integrated model shows a greater advantage.During the first storage period,as the amount of data increases from small to large,the project needs to predict the maximum deformation and the deformation trend of each measurement point during the complete first storage period in advance for a small amount of data at the beginning of the storage period.The integrated model can be better adapted to the engineering problems of the first storage period of extra-high arch dams and has better stability in engineering applications due to the combination of the deep data information mining ability of pure deep neural network models and the strong generalization ability and strong mechanics of formulas.(2)The formula for predicting dam deformation during the first storage period of a very high arch dam is summarised,and the influence of three temperature factors on the accuracy of the formula method is discussed and calculated.The results of the calculations show that the model accuracy is higher when the harmonic factor is used to express the temperature component compared to the air temperature factor and the dam temperature factor.From the results,the formula method is suitable for short-term predictions,but when dealing with long series predictions,the formula method performs poorly and the overfitting phenomenon is very serious.The formula method can make more accurate predictions when dealing with data with less hidden information.When dealing with data with more complex hidden information,the formula method does not have a good ability to dig deeper into the data information.(3)The DNN model,a deep learning model,was introduced to the problem of predicting dam deformation for multiple targets during the first storage period of a very high arch dam.After calculation and comparison with the formula method,it can be seen that the DNN model can deeply explore the internal implicit information of the monitoring data of the first storage period of the extra-high arch dam.Compared with the formula method,the DNN model can predict the trend of the dam deformation time-series process when there is a small amount of data,and the model is more stable than the formula method until the second target water level is reached.(4)The trained integrated model was directly migrated to similar projects.From the calculation results,the integrated model proposed in this paper has good model migration capability and can better handle the prediction of dam deformation during the first storage period of a very high arch dam.