基于GA-LSTM的高层建筑结构地震响应的分散控制研究  

Decentralized control for the seismic response of high-rise building structures based on GA-LSTM

作　　者：高经纬 涂建维[1] 刘康生 李召 GAO Jingwei;TU Jianwei;LIU Kangsheng;LI Zhao(Hubei Key Laboratory of Roadway Bridge and Structure Engineering,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学道路桥梁与结构工程湖北省重点实验室,武汉430070

出　　处：《振动与冲击》2021年第10期114-122,共9页Journal of Vibration and Shock

基　　金：国家自然科学基金(51978550);湖北省自然科学基金(2016CFA020)。

摘　　要：高层建筑结构复杂,自由度数庞大,采用主动抗震控制方法时结构的精确模型无法建立,整体控制目标也难以获取。为此,将长短时记忆(LSTM)网络理论和大系统分散控制理论结合,提出一种基于LSTM的智能分散控制方法;利用LSTM深度学习框架构建不同类型的分散控制器,并根据Lyapunov稳定性理论推导分散控制器稳定的充分条件,运用遗传算法(GA)对LSTM框架的初始学习率进行优化,提高分散控制器的收敛速度和预测精度。以20层Benchmark模型作为被控对象,研究GA-LSTM分散控制方法的控制性能,并与集中控制效果进行对比分析。结果表明,基于GA-LSTM的智能分散控制方法可以简化控制器结构,对比集中控制可能产生的整体失效现象,具有更高的可靠性和更好的控制效果。High-rise buildings have complex structures and huge degrees of freedom.When active seismic control method is used,it is difficult to establish a precise structure model,and the overall control target is hard to achieve.Thus,an intelligent decentralized control method based on long short-term memory(LSTM)networks was proposed,based on the LSTM theory combined with the large-scale system decentralized control theory.Different decentralized controller types were constructed using the LSTM deep learning framework,and the sufficient conditions for the stability of decentralized controllers were derived according to the Lyapunov stability theory.A genetic algorithm(GA)was used to optimize the initial learning rate of the LSTM framework to improve the convergence speed and prediction accuracy of the decentralized controller.Taking a 20-layer benchmark model as a controlled object,the control performance of the GA-LSTM decentralized control method was studied and its effect was compared with the centralized control effect.The results show that the intelligent decentralized control method based on GA-LSTM simplifies the controller structure.Compared with the overall failure phenomena that may occur in the centralized control,it has higher reliability and better control effect.

关 键 词：分散控制 长短时记忆(LSTM)网络 LYAPUNOV稳定性理论 遗传算法(GA) 结构振动控制 

分 类 号：TU311.4[建筑科学—结构工程]