| There are many monitoring projects for concrete dams.Deformation,as a typical effect that characterizes the trend change or even transformation of the structural state,is an important non-engineering measure to monitor and control the operation performance of concrete dams.The diagnosis of deformation singular values,the prediction of deformation behavior evolution,the formulation of early warning indexes,and the development of analysis platform are the key to the safety monitoring of concrete dam deformation behavior,and they are also the research hotspots and difficulties of dam engineering.To improve the intelligence level of concrete dam safety monitoring and promote the deep application of artificial intelligence in the field of concrete dam safety monitoring,the key theoretical methods,such as the intelligent diagnosis method of deformation singular values,the high-performance intelligent prediction model,the formulation of spatiotemporal joint early warning indexes,and the development of intelligent analysis platform are deeply studied,and the theoretical system and application platform of intelligent monitoring for deformation behavior of concrete dam are constructed in this paper.The main research contents are as follows:(1)The characterization mode of concrete dam deformation singularity is revealed,and the intelligent diagnosis method of concrete dam deformation singular values is proposed.The spatiotemporal evolution characteristics of concrete dam deformation and its influencing factors are deeply analyzed,and the common singular modes of concrete dam deformation are summarized.Give the complex distribution characteristics and prominent local singular characteristics of the concrete dam deformation monitoring data,and the defects of normal distribution prior assumption,blunt local singular characteristics,and poor detection performance of the commonly used singular value detection methods,a temporal and spatial correction of local outlier factor(TSCLOF)method is proposed,which does not need prior assumption,is sensitive to local singular characteristics,and the singularity degree is quantitatively corrected in time and space.An intelligent detection method for deformation singularity of concrete dam based on TSCLOF is established.After accurately detecting the singular values of concrete dam deformation,taking the deep learning supervision model based on convolutional neural networks(CNN)as the core,an intelligent analysis model for the causes of concrete dam deformation singular values based on CNN is constructed,which can intelligently,quickly,and accurately identify the causes of concrete dam deformation singular values.The application of engineering examples shows that the intelligent diagnosis method of concrete dam deformation singular values based on TSCLOF-CNN effectively improves the real-time efficiency of deformation singular value diagnosis.The accuracy of deformation singular value detection and cause analysis are both higher than 90%,and the overall diagnostic performance is better than the traditional singular value diagnosis method.(2)The optimization method of concrete dam deformation influencing factors based on rough set(RS)is established,the high-performance prediction model of concrete dam deformation based on long short-term memory neural network(LSTM)is constructed,and the application evaluation system of the prediction model is developed.Aiming at the subjectivity and redundancy of the selection of concrete dam deformation influencing factors,the rough set theory is introduced to construct the optimization method of concrete dam deformation influencing factors based on RS,and the key factors that have a significant influence on deformation are screened from the empirical influence factors.Based on the optimization result of concrete dam deformation influencing factors,aiming at the shortcomings of traditional concrete dam deformation prediction model in prediction performance,a concrete dam deformation prediction model with good prediction performance is constructed by integrating LSTM neural network model.Aiming at the overlap and limitation of accuracy evaluation indexes,the model prediction performance evaluation index system is constructed from the perspectives of accuracy,robustness,extension,and generalization.The application of engineering examples shows that the root mean square error of the intelligent prediction model of concrete dam deformation based on RS-LSTM is small,and the average absolute percentage error is less than 1.5%,the robustness,extension,and generalization are stable,and the overall prediction performance is better than that of the prediction model of concrete dam deformation based on least squares regression,random forest and support vector machine.(3)The spatiotemporal clustering method deformation measuring points based on SelfOrganizing Maps(SOM)is proposed,the joint distribution function of multi-measuring points is established by Copula theory,and the spatiotemporal joint early warning index of concrete dam deformation based on SOM-Copula is constructed.Aiming at the problems of poor sample representation,vulnerable to interference of unknown factors,poor suitability of overall distribution function,and inconsistent early warning results in the existing formulation of concrete dam deformation early warning indicators,the spatiotemporal clustering method based on SOM is used to cluster the deformation measuring points,which provides a basis for the construction of deformation early warning indexes with spatiotemporal distribution characteristics from the perspective of multiple measuring points firstly.Then the extreme values of deformation under more unfavorable load combinations are obtained by statistical model or hybrid model,and the sample space of measured deformation extreme values under existing unfavorable load combinations is expanded to improve its representativeness.Finally,the Copula theory is used to solve the joint probability distribution function of the strong correlation deformation measuring point group,and the spatiotemporal joint multi-level early warning indexes of concrete dam deformation in multidimensional probability space is established to improve the spatial integrity,effectiveness,and accuracy of the early warning indexes.The application of engineering examples shows that the formulation results of spatiotemporal joint early warning index of concrete dam deformation based on SOM-Copula are close to those of the structural analysis method.The false alarm rate of abnormal early warning is obviously lower than that of mathematical statistics and structural analysis.(4)The intelligent analysis platform for the deformation behavior of concrete dams is developed.The low intelligent level of monitoring information diagnosis,the insufficient internal law mining of monitoring information,and poor reliability of abnormal early warning exist in the current concrete dam deformation analysis platform.With the help of MATLAB App Designer software development platform,the intelligent diagnosis method of concrete dam deformation singular values based on TSCLOF-CNN,the intelligent prediction model of concrete dam deformation based on RS-LSTM,and the formulation method of spatiotemporal joint early warning index of concrete dam deformation based on SOM-Copula are integrated,and the intelligent analysis platform of concrete dam deformation behavior is developed with the functions of automatic management of concrete dam monitoring data,batch statistical drawing,intelligent diagnosis of deformation singular values,intelligent prediction of deformation behavior,spatiotemporal joint early warning of deformation,etc.The functions of the existing concrete dam monitoring system are supplemented to improve the automatic,information,and intelligent management level. |
