Prediction Of Initial Rotational Stiffness Of Semi-rigid Endplate Joints Based On RBF Neural Network

Traditional steel frame connection is not rigid or ideal hinge,but semi-rigid connection.In order to obtain the real performance of semi-rigid connection nodes,scholars at home and abroad have carried out a lot of research on semi-rigid connection.Since the current research method mainly studies semi-rigid nodes from the perspective of a single variable,but ignores the impact of the correlation between variables on semi-rigid nodes,there is still a large deviation when the initial rotational stiffness of nodes is calculated by theoretical analysis.Therefore,considering the good nonlinear mapping ability of neural network and the learning ability of sample data,this paper uses neural network to predict the initial rotational stiffness of semi-rigid endplate joints from the perspective of multivariable,based on the semi-rigid node database.The specific research content of this paper is as follows:(1)Through the study of BP neural network and RBF neural network,the RBF neural network with the best consistent approximation is selected as the tool for subsequent learning and prediction.An improved particle swarm optimization algorithm was proposed to optimize the RBF neural network.Compared with the traditional genetic algorithm and particle swarm optimization algorithm,the results of stiffness prediction have better accuracy and can better represent the change trend of stiffness between nodes.(2)for input variables of neural network,through existing literature and experimental studies,the geometric dimensions of columns,end plates,bolts and beams of semi-rigid endplate joints are preliminarily determined as the main input variables of neural network.A kernel principal component analysis method with parameter optimization was proposed to process variables,which could improve the prediction accuracy of RBF neural network more effectively than the traditional pca method.(3)since there is no unified classification standard for semi-rigid endplate connection nodes,this paper introduces the idea of clustering to classify the node samples.A fuzzy c-means clustering method based on improved particle swarm optimization is proposed.(4)based on the discussion of three groups of semi-rigid end plate connection node,the application of neural network optimization algorithm,clustering analysis method,and kernel principal component analysis method were used to predict the initial rotational stiffness,predict accord well with those of stiffness and stiffness test,thus this method was verified in semi-rigid nodes in the initial rotational stiffness prediction effectiveness and practicability.