基于人工神经网络的Bouc-Wen-Baber-Noori滞回模型等效线性化研究

Equivalent linearization technique is of great scientific significance for performance-based seismic design of elastoplastic systems. How to determine reasonable equivalent linear parameters and obtain accurate maximum elastoplastic displacement is the key to equivalent linearization. The existing literatures do not provide comprehensive research on the influence mechanism of input seismic wave characteristics on equivalent linear parameters, only focus on simple hysteretic models or stiffness/strength degradation models with specific parameters, and seldom carry out systematic research on equivalent linearization of complex elastoplastic hysteretic models. This project takes Bouc-Wen-Baber-Noori (BWBN) hysteresis model as research object, which can represent hysteresis characteristics of different structures, and adopts Latin hypercube sampling to identify the most sensitive parameters influencing the maximum elastoplastic displacement of BWBN hysteresis model. Artificial neural network is used to build the mapping relationship between the input neurons, such as the characteristics of input seismic wave and the sensitive parameters of BWBN hysteresis model, and the output neurons, such as the maximum elastoplastic displacement and the optimal equivalent linear parameters, so as to clarify the influence mechanisms of input seismic wave characteristics and sensitive hysteresis parameters on equivalent linearization of BWBN hysteresis model. Shaking table test of reinforced concrete shear wall characterized by BWBN model are carried out to verify the accuracy of equivalent linearization model of BWBN hysteresis model. Research results lay the foundation for the application of equivalent linearization method in performance-based seismic design.

等效线性化研究对于弹塑性体系基于性能的抗震设计具有重要的科学意义，如何确定合理的等效线性参数并得到准确的最大弹塑性位移是等效线性化的关键问题。现有文献缺乏输入地震波特性对等效线性参数影响机制的全面研究，大部分仅考虑简单滞回模型或具有特定参数的退化模型，对于复杂弹塑性滞回模型的等效线性化缺少系统研究。本项目以能够表征结构不同滞回特性的Bouc-Wen-Baber-Noori（BWBN）滞回模型为研究对象，采用拉丁超立方体抽样方法识别BWBN模型对于最大弹塑性位移的敏感滞回参数，通过人工神经网络建立输入地震波特性和BWBN滞回模型敏感参数与最大弹塑性位移和最优等效线性参数的映射关系，阐明其对BWBN滞回模型等效线性化的影响机制，并利用钢筋混凝土剪力墙BWBN模型的振动台试验对BWBN等效线性化模型进行验证。研究结果为应用等效线性化方法进行基于性能的抗震设计奠定基础。

等效线性化研究对于弹塑性体系基于性能的抗震设计具有重要的科学意义，如何确定合理的等效线性参数并得到准确的最大弹塑性位移是等效线性化的关键问题。本项目以能够表征结构不同滞回特性的Bouc-Wen-Baber-Noori（BWBN）滞回模型为研究对象，采用拉丁超立方体抽样方法识别BWBN模型对于最大弹塑性位移的敏感滞回参数，通过人工神经网络建立输入地震波特性和BWBN滞回模型敏感参数与最大弹塑性位移和最优等效线性参数的映射关系，阐明其对BWBN滞回模型等效线性化的影响机制，并开发了双线性滞回模型等效线性参数计算软件。根据等效线性化理论，建立了不同地震反应谱的阻尼折减系数模型，为应用等效线性化方法进行基于性能的抗震设计奠定了基础。