潘兆东, 刘良坤, 谭平, 周福霖. 大型结构自适应学习率RBF神经网络滑模分散控制研究[J]. 工程力学, 2019, 36(9): 120-127. DOI: 10.6052/j.issn.1000-4750.2018.07.0429
引用本文: 潘兆东, 刘良坤, 谭平, 周福霖. 大型结构自适应学习率RBF神经网络滑模分散控制研究[J]. 工程力学, 2019, 36(9): 120-127. DOI: 10.6052/j.issn.1000-4750.2018.07.0429
PAN Zhao-dong, LIU Liang-kun, TAN Ping, ZHOU Fu-lin. RESEARCH ON SLIDING MODE DECENTRALIZED CONTROL BASED ON ADAPTIVE LEARNING RATE RBF NEURAL NETWORK FOR LARGE-SCALE ENGINEERING STRUCTURES[J]. Engineering Mechanics, 2019, 36(9): 120-127. DOI: 10.6052/j.issn.1000-4750.2018.07.0429
Citation: PAN Zhao-dong, LIU Liang-kun, TAN Ping, ZHOU Fu-lin. RESEARCH ON SLIDING MODE DECENTRALIZED CONTROL BASED ON ADAPTIVE LEARNING RATE RBF NEURAL NETWORK FOR LARGE-SCALE ENGINEERING STRUCTURES[J]. Engineering Mechanics, 2019, 36(9): 120-127. DOI: 10.6052/j.issn.1000-4750.2018.07.0429

大型结构自适应学习率RBF神经网络滑模分散控制研究

RESEARCH ON SLIDING MODE DECENTRALIZED CONTROL BASED ON ADAPTIVE LEARNING RATE RBF NEURAL NETWORK FOR LARGE-SCALE ENGINEERING STRUCTURES

  • 摘要: 为了有效处理土木工程结构分散振动控制中子系统间相互影响力和外界荷载不确定性的影响,提出了自适应学习率RBF神经网络滑模分散控制算法(DALRBFSMC)。首先利用Lyapunov稳定性理论设计了仅依赖于子系统位移和速度响应反馈信息的滑模分散控制律,在此基础上,结合RBF神经网络理论和经典梯度下降法,引入Lyapunov函数,推导了调整RBF网络权值的自适应学习率,进而得到能实时调节滑模分散控制律切换增益项的自适应学习率RBF神经网络滑模分散控制算法(DALRBFSMC)。同时,针对子系统不同划分方式及子控制器之间存在重叠,提出了多种分散控制设计策略。对ASCE 9层Benchmark模型进行多种分散控制和集中控制设计。仿真分析结果表明,该分散控制算法适用于不同的分散控制策略,重叠分散控制策略较传统集中控制策略而言有更好的控制效果;同时能使分散控制系统内各作动器均处于功效最大状态。

     

    Abstract: This paper Proposes a decentralized adaptive learning rate RBF neural network sliding mode control (DALRBFSMC) algorithm for dealing with the influence and the uncertainty of the interaction forces between subsystems and the external loads. Lyapunov stability theory is employed to design the decentralized sliding mode control law which depends only on the displacement and the velocity response of relevant subsystems. Combined with RBF neural network theory and the classical gradient descent method, the adaptive learning rate of RBF network weights-adjustment is derived by using a Lyapunov function. And then the decentralized adaptive learning rate RBF neural network sliding mode control (DALRBFSMC) is designed, which can adjust the switching gain of the sliding mode control law in real time. An ASCE 9-story benchmark building is selected as a numerical example to evaluate the control performances of decentralized control and centralized control. Numerical simulation results indicate that the DALRBFSMC algorithm is suitable for different decentralized control strategy, and that overlapping decentralized control can perform up to a superior control performance when comparing with traditional centralized control, and also guarantee each of the actuators to be operating at maximum efficiency.

     

/

返回文章
返回