Multi-parameter Identification Of Cable Based On LNN Neural Network And XGBoost Algorithm

Cable-stayed bridge is the main structural form of large-span bridges at present,while cables are the most crucial force-receiving and force-transmitting components in cable-stayed bridges.Therefore,it is extremely important to accurately identify the cable force during the the construction and operation of the bridge.Based on the frequency method theory which is most commonly used in cable force testing,this paper analyzes the complex laws between the cable force,flexual rigidity and the natural frequency.Moreover,this paper uses the LNN neural network and XGBoost algorithm to carry out the research on cable parameter identification based on frequency method.The accuracy and feasibility of the two methods above are also verified through experiments.The main work of the paper is as follows:(1)This paper derives the relationship between the cable force,flexual rigidity and the natural frequency under different boundary conditions.The LNN neural network is constructed to solve the transcendental equations of some special boundary conditions.For cables under complex boundary conditions,the concept of elastic boundary is introduced,and it is assumed that there is rotational constraint stiffness at the two ends of the cable.The cable vibration differential equation considering the elastic boundary is derived,and the LNN neural network is constructed based on this equation.The LNN neural network is used to solve the cable force,the flexual rigidity and the rotation constraint stiffness of the two ends of the cable when the first three frequencies and the length of the cable are known.(2)For the case that the relationship between various parameters of the cable cannot fully satisfy the differential equation of cable vibration,the XGBoost algorithm is used to identify the multi-parameter of the cable.To begin with,the XGBoost algorithm is trained by thousands of simulated cable samples constructed by finite element software based on actual cable data.Then,the XGBoost algorithm is used to predict the test set samples constructed by the finite element software,which proves that the cable parameter identification accuracy of the XGBoost algorithm is above 95%and meets the actual engineering requirements.Meanwhile,this paper uses BP neural network,Extreme Learning Machine and Bayesian method to identify cable parameters through the same training set and test set,which verifies the superiority of the XGBoost algorithm compared to the above three methods.(3)This paper designs the cable scale model experiment to obtain the first three frequencies of the actual cable under different cable lengths and tensile forces.The LNN neural network and the XGBoost algorithm are used to identify the cable force,flexual rigidity and other cable parameters according to the first three frequency and cable length of the experimental cable.Finally,the feasibility and accuracy of the above two methods in the parameter identification of cable with unknown boundary conditions are proved.(4)Combined with the measured engineering data of the cable of a bridge in Hangzhou,a bridge in Fenghua and a bridge in Guizhou Province,this paper utilizes the LNN neural network and XGBoost algorithm to identify the cable force and flexual rigidity according to the first two frequency and the cable length when the boundary condition is known to be fixed at both ends.Finally,the feasibility and accuracy of the above two methods in parameter identification of cable with fixed ends are proved.