<b>AI-based prediction of nonlinear seismic response of inter-story isolation system with supplement inerter</b>

人工智能 （AI） 方法已被用于 非线性时程 结构的 下 强地震激励。然而，在 AI 模型中需要考虑一些影响因素，包括列车数据样本小、结构参数和输入地震对预测性能和准确性、通用性或可解释性的影响因素。预测研究了带附加惰性的层间隔震混合控制系统的地震响应 multi-input 和 multi-output parameter-informed long short-term memory 层间隔振系统 （IIS）惯已使用 响应建模 输入变量包括 5 参数（即下部质量与上部质量的比值、惯性表观质量与上部质量的比值、归一化特征强度、屈服后与屈服前刚度的硬化比和阻尼比）和输入地震动。它准确预测 的 IIS使用数值方法进行基于经典物理学的非线性时程分析。此外，通过 ，成功地证明了预测结果的比较。结果表明，所提出的多输入多输出 PI-LSTM 网络模型是一种计算高效的 结构响应预测方法 IIS 中提供了显着的潜在效率 大规模非线性时程分析 T模型参数对 预测性能位移和加速响应 。研究结果可为预测间隔振系统的预测方法和混合控制Artificial Intelligence (AI) methods have been applied to nonlinear time-history analysis of structures under strong seismic excitation. However, several influencing factors need to be considered in AI models, including the small size of training data samples, and the impacts of structural parameters and input ground motions on prediction performance, accuracy, generalizability or interpretability. This study investigates the seismic response prediction of a layered base-isolated hybrid control system with additional inertia using a multi-input and multi-output parameter-informed long short-term memory (PI-LSTM) network. The Layered Isolation System (IIS) response modeling adopts 5 input parameters, namely the ratio of lower mass to upper mass, the ratio of apparent inertial mass to upper mass, normalized characteristic intensity, hardening ratio of post-yield to pre-yield stiffness, and damping ratio, as well as input ground motions. The proposed model accurately predicts the seismic responses of IIS, while classical physics-based nonlinear time-history analysis was conducted using numerical methods for validation. Additionally, the prediction results were compared with numerical analysis results, successfully verifying the model’s validity. The results demonstrate that the proposed multi-input multi-output PI-LSTM network is a computationally efficient structural response prediction method, which provides significant potential efficiency for large-scale nonlinear time-history analysis of IIS. Model parameters have notable impacts on the prediction performance of displacement and acceleration responses. The findings of this study can provide valuable references for prediction methods and hybrid control of base-isolated systems.