Research On Anchorage Compactness Identification Based On KPCA And Adaboost

Because of its advantages of simple construction and reliable support performance,bolt anchorage technology is widely used in tunnel construction,stable surrounding rock and mining roadway support engineering construction.Anchor bolt anchorage is susceptible to the impact of the surrounding environment,resulting in quality problems,and its damage has a high degree of concealment,leaving hidden safety hazards for the construction process and future use.The compactness of anchorage will directly affect the overall strength and stiffness of the anchorage system.Therefore,the identification of the compactness of anchorage grouting has become an important task in the quality testing of anchorage.In this paper,the core principal component analysis(KPCA)optimized by the small world particle swarm optimization(SWN-PSO)algorithm is used to reduce the dimension of anchor bolt anchorage system with different grouting compacted degree.Then,the Adaboost strong classifier is constructed by the BP neural network optimized by the genetic algorithm(GA)as a weak classifier to identify the category of grouting compacted degree.The specific contents of this paper are as follows:(1)The significance of the measured frequency response function and the method to obtain it are studied.The measured frequency response function contains all the information contained in the modal parameters of the system,which is the basis for the identification of the anchorage density.(2)The measured frequency response function contains noise and nonlinear components,which will inevitably increase the difficulty of identification.Therefore,KPCA is used to reduce the dimension of the measured frequency response function of the anchor anchorage system.Firstly,the measured frequency response function is projected into the new feature space through the kernel function,and the principal components of each order are calculated in the new feature space.Then,the cumulative contribution rate is calculated.Since the kernel parameters in the sum function will affect the dimensionality reduction effect of KPCA,SWN-PSO algorithm is adopted to optimize it.(3)The theoretical method,structure and factors influencing classification performance of BP neural network are studied.Aiming at the problem that traditional Adaboost algorithm can only deal with binary classification,this paper studies the processing method of multi-classification problem.The BP?Adaboost strong classification model is constructed to improve the classification accuracy of BP network.GA algorithm is adopted to optimize the initial weight and threshold of BP neural network,which can combine the global search ability of GA algorithm with the good classification ability of BP neural network,and overcome the shortcomings of BP network such as local convergence and blindness of optimization.Finally,using GA-BP network as weak classifier,Adaboost algorithm is used to construct GA-BP?Adaboost strong classification model.(4)Nine bolt anchoring models with different bolt compactness are made,and the acquisition system of measured frequency response function of the bolt anchoring system is built to collect the frequency response function of the bolt anchoring model.PCA and KPCA with different kernel parameters are used to classify the measured frequency response function of the anchor bolt model to verify the effectiveness of the dimension reduction method proposed in this paper.The data after dimension reduction are input into BP network,GA-BP network,BP?Adaboost classification model and GA-BP?Adaboost strong classification model for classification,and the results showed that the classification accuracy of GA-BP?Adaboost strong classification model is better than that of the other three classification models.