ABSTRACT To address the issues of limited data availability, low accuracy, and weak generalization ability in predicting the deformation of curtain wall structures, a dynamic deformation prediction method for irregular aluminum plate curtain wall structures in gymnasiums based on an improved RBF neural network is proposed. The overall architecture is constructed, deformation data is collected using sensors, and it is processed within the intelligent information management platform using the empirical mode decomposition algorithm. A prediction model is developed using the RBF neural network. The intrinsic mode function components of the decomposed deformation data are used as input. The weak learners are enhanced using the AdaBoost algorithm. The number of hidden layer nodes is optimized based on the data distribution mechanism. The monitoring terminal outputs alarms and generates reports based on the model's output. Experimental results show that this method can effectively extract deformation data features through empirical mode decomposition, obtain the intrinsic mode function components, and achieve an average prediction accuracy of 94.7% for key monitoring nodes of the curtain wall structure. For the maximum deformation node G111, the prediction error is only 0.032 mm. This method can accurately predict the deformation values of each node of the curtain wall structure, thereby identifying the main nodes that cause structural instability.
Jin et al. (Wed,) studied this question.