Identification Of Near-fault Velocity-pulse Ground Motion Based On Bp Neural Network

According to the existing earthquake disasters,the near-fault ground motions are more destructive to the engineering structure than the ordinary far-field ground motions.The main reason is the influence of velocity pulse.Therefore,in the field of near-fault ground motions,firstly we need to identify the ground motions with velocity pulse.In this paper,from the perspective of representation learning,the BP neural network is used to establish a identification model to study the quantitative identification of near-fault pulse ground motions.Firstly,three kinds of existing velocity pulse ground motions identification methods are analyzed and compared.These existing identification methods have good identification effect in the case of irregular pulse shape and multi-pulse.They still rely on the basic shape of the IMF,and some parameters have empirical problems.Therefore,combining eigenmode vector reconstruction with BP neural network algorithm,and propose a method to identify velocity pulse ground motions by machine learning.The representation learning process is used as a carrier,it can self-drive the corresponding pulse energy,period and some other related parameters based on the data.By using network error as the model goal,this method can identify velocity pulse ground motions effectively.The training process is optimized with reference to the effects of momentum index and self-adaptive learning rate based on the above method.And compare the effect of different activation functions on identification of near-fault pulse ground motions.In this paper,the pre-processed ground motions velocity time history is used as the input of the neural network.And use the learning rules of gradient descent to establish the corresponding neural network identification model.And compare the result of this method with the result of the EMD identification method under the same samples.The result shows that the neural network has good identify ability and can effectively learn features to identify the types of ground motion.The algorithm proposed in this paper helps to quantify and establish a velocity pulse ground motions database.It can promote the improvement of probability-based seismic hazard evaluation and seismic safety evaluation of medium-long period structures.