Crack Cause And Early Warning Analysis Of A Concrete Dam Based On Machine Learning

Dam safety monitoring has always been an important research topic in the dam industry.In order to ensure the safe operation of the dam,the dam construction period is often pre-embedded in a large number of monitoring equipment,these monitoring instruments over time to form a huge database of measured data.Monitoring data analysis is an important but very complex work,the traditional monitoring data analysis methods are generally divided into statistical model method and the classic machine learning methods,statistical model method selection with subjective and high-dimensional non-linear data with poor results,The classical machine learning method has a good effect in the face of the small sample set,but it is prone to convergence speed and easy to fall into the local optimal value when faced with the large quantity sample set.In this paper,the causes and early warning of a concrete dam are studied by using rough set theory,statistical theory,gray system theory,depth learning algorithm,SVM support vector machine algorithm,BP algorithm and typical small probability method.The main results are as follows:(1)The three different reduction algorithms of rough set are studied,and the corresponding program is used to filter their classification ability.The comparison shows that the algcrithm of discretization and attribute reduction of classical rough set has the least number of features,But the accuracy is low.Although the fuzzy information entropy reduction algorithm has high accuracy,but retains more feature quantity.The numerical attribute reduction algorithm based on neighborhood rough set model not only obtains the highest accuracy,Retained fewer features.Therefore,this paper chooses the neighborhood rough set model to establish the early warning index system.(2)The depth learning theory is studied,the depth trust network is used as the framework to establish the crack early warning model and the algorithm is implemented based on matlab.The characteristics of the early warning index system screening are taken as the model input nodes.By training and simulating a concrete dam And the results are compared with the traditional statistical model and the BP network model found that the deep confidence network early warning model has higher accuracy and learning efficiency.(3)The DBN model weighting method is used to obtain the influence degree of different influence factors on the crack opening effect.The rule rule is used to get the rule of the unfavorable load combination of the crack opening degree.The DBN model method is compared with the gray theory The results show that the DBN model method and the typical small probability method can be used to analyze the relationship between the fracture degree of the concrete dam section and the other monitoring physical quantities.And the feasibility of the method of predicting the early warning index of DBN model is verified.