基于卷积神经网络的高层建筑智能控制算法研究OA北大核心

Traditional shallow neural networks exhibit low prediction accuracy and poor generalization when handling high-dimensional data.To solve these problems,this study proposes an intelligent control algorithm for high-rise buildings based on one-dimensional convolutional neural networks(1D-CNN)and the deep dream visualization algorithm.The proposed method enables high-precision network model training and visualizes data features through 1D-CNN.Using a 20-story benchmark model as a case study,the damping performance of the 1D-CNN-based intelligent control algorithm was analyzed under different conditions and compared with back propagation(BP)and radial basis function(RBF)algorithms.Results show that 1D-CNN can effectively extract deep data features and reduce the dimensionality of massive datasets by virtue of one-dimensional convolution and pooling operations.Under external excitation,the maximum damping rates for acceleration and displacement achieved by the 1D CNN controller were 69.0％and 55.6％respectively,significantly outperforming BP and RBF.Although the control performance of all algorithms decreased under modified excitation conditions,the 1D-CNN consistently exhibited superior performance and the best generalization capability.