There have been numerous studies on complex structural calculation models represented by damper components, but traditional calculation models often require there are many shortcomings in balancing calculation accuracy, efficiency, and applicability. This article is based on the calculation model of metal dampers, starting from the fine finite element method, and then introducing deep learning technology to explore a new intelligent calculation framework for structures. Firstly, a fine finite element parameterized modeling model of metal dampers based on shell elements was developed. Based on a determined basic strategy, a parameterized modeling model of metal dampers was developed using Notepad++, providing a foundation for the expansion of subsequent datasets in finite element technology. Secondly, a general prediction model for mechanical response of metal dampers based on deep learning was designed. To improve the shortcomings of existing network single level prediction and feature level similarity, a weighted stacked pyramid network structure was used to construct the external framework. Within the unit module, a Pyramid+GA (GRU + X-Attention) model was established to address the three main issues of long sequence problems in hysteresis curve simulation: extremely long sequence, causality, and parameter explosion. The network related parameters were analyzed and discussed to predict the mechanical response of metal dampers.