在結構主動控制的實務應用上，為了設計控制器以抑制結構之受震反應，常需要對結構物進行系統識別以獲取其數值模型，並根據此數值模型進行控制器之設計，系統識別的結果好壞將直接影響控制器的設計性能。此外，一些控制器於實際應用尚需額外設計狀態估測器，如線性二次調節器(Linear-Quadratic Regulator, LQR)，系統識別及狀態估測器的設計過程，往往仰賴工程師之設計經驗，亦需要額外的時間成本以完成設計。本研究提出主動激振的方式，透過主動控制元件對結構物進行微小激振，以獲得各樓層的加速度反應，並使用外源輸入自迴歸神經網路模型以擬合力量及加速度之間的逆向關係。同時，本研究亦使用生物共生演算法(Symbiotic Organisms Search, SOS)以進行LQR權重矩陣之最佳化，並透過機器學習擬合結構於LQR控制下的頂層加速度至LQR控制力的關係，以減少實務上需額外設計狀態估測器的時間成本及回饋所需的樓層加速度量測訊號。
在數值模擬上使用於頂層配置主動質量阻尼器(Active Mass Damper, AMD)的三種不同樓層數的結構數值模型，分別為9層樓、10層樓及27層樓，並以14組地震加速度歷時進行受控結構之動力歷時分析，以驗證主動激振控制器及擬合LQR控制力加速度回饋控制器之控制性能。模擬結果顯示，經過適當參數調整的主動激振控制器與擬合LQR控制力的加速度回饋控制器其控制性能與傳統LQR控制器相似。最後，使用於頂層配置電動伺服馬達驅動之主動質量阻尼器的三層樓剪力屋架試體進行實驗驗證，實驗結果顯示，使用主動激振及使用頂層加速度擬合LQR控制力之類神經網路控制器，在降低結構加速度反應上之性能指標上有相似的控制性能。

In practical application of structural active control, controller design is mostly based on a structural numerical model which is obtained by conducting system identification. The accuracy of identified model directly affects the performance of controller. Besides, some controllers require additional design of a state estimator in practical applications, such as Linear-Quadratic Regulator (LQR). Generally, respectable system identification results and design of state estimators depend on engineering experiences which may introduce uncertainties into the entire control system. In this study, a novel design procedure for controller design is proposed by employing direct excitation data with machine learning. The structure is excited by an active device to obtain the acceleration response of each floor. Then an artificial neural network named auto regressive with exogenous input (ARX) is used obtain the inverse model between force excitation and acceleration. In addition, Symbiotic Organisms Search (SOS) is used to optimize the LQR weighting matrices. Similarly, an ARX model is utilized to train the relationship between the acceleration at the top floor controlled by the LQR and the LQR control force of the structure. This, additional design of a state estimator can be removed in practice since the control force can be calculated merely based on the top acceleration measurement.
In the numerical simulation, three different structural models with an active mass damper (AMD) at the top floor are adopted including a 9-story, a 10-story, and a 27-story building. A total number of 14 earthquakes are used to excite each building and the associated seismic responses are obtained to verify the control performance of the two aforementioned controllers. The simulation results show that the direct excitation controller with proper adjustment of parameters and the emulated LQR controller have similar control performance which are comparable to conventional LQR. Finally, a 3-story shear building specimen with an active mass damper driven by an electric servo motor at the top floor is used for experimental verification. The experimental results show that the direct excitation controller and the emulated LQR control force have analogous seismic response considering performance indices related to structural acceleration.

[1]. Saaed, T. E., Nikolakopoulos, G., Jonasson, J. E., & Hedlund, H. (2015) ‘‘A state-of-the-art review of structural control systems” Journal of Vibration and Control, 21(5), 919-937.
[2]. Bryan, J.S. (2019) “Metaheuristic Optimization of Weighting Matrices of LQR for Active Structural Control” 國立臺灣科技大學碩士學位論文，陳沛清指導
[3]. Chen, P. C., Sugiarto, B. J., & Chien, K. Y. (2021) ‘‘Performance-based optimization of LQR for active mass damper using symbiotic organisms search” Smart Structures and Systems, 27(4), 705-717.
[4]. 簡楷益(2020)，應用機器學習於結構主動控制律之設計與驗證，國立臺灣科技大學營建工程所碩士論文，陳沛清指導。
[5]. Chen, P. C., & Chien, K. Y. (2020) “Machine-learning based optimal seismic control of structure with active mass damper” Applied Sciences, 10(15), 5342.
[6]. Ikeda, Y., Sasaki, K., Sakamoto, M., & Kobori, T. (2001) ‘‘Active mass driver system as the first application of active structural control” Earthquake engineering & structural dynamics, 30(11), 1575-1595.
[7]. Yamamoto, M., Aizawa, S., Higashino, M., & Toyama, K. (2001) “Practical applications of active mass dampers with hydraulic actuator” Earthquake engineering & structural dynamics, 30(11), 1697-1717.
[8]. Samuel, A. L. (2000) ‘‘Some studies in machine learning using the game of checkers” IBM Journal of research and development, 44(1.2), 206-226.
[9]. Amezquita-Sanchez, J. P., Valtierra-Rodriguez, M., Aldwaik, M., & Adeli, H. (2016) “Neurocomputing in civil infrastructure” Scientia Iranica, 23(6), 2417-2428.
[10]. Adeli, H., & Yeh, C. (1989) “Perceptron learning in engineering design” Computer‐Aided Civil and Infrastructure Engineering, 4(4), 247-256.
[11]. Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986) ‘‘Learning representations by back-propagating errors” Nature, 323(6088), 533-536.
[12]. Vanluchene, R. D., & Sun, R. (1990) “Neural networks in structural engineering” Computer‐Aided Civil and Infrastructure Engineering, 5(3), 207-215.
[13]. Huang, C. S., Hung, S. L., Wen, C. M., & Tu, T. T. (2003) “A neural network approach for structural identification and diagnosis of a building from seismic response data” Earthquake engineering & structural dynamics, 32(2), 187-206.
[14]. Brown, A. S., & Yang, H. T. (2001) “Neural networks for multiobjective adaptive structural control” Journal of Structural Engineering, 127(2), 203-210.
[15]. Madan, A. (2005) ‘‘Vibration control of building structures using self-organizing and self-learning neural networks” Journal of sound and vibration, 287(4-5), 759-784.
[16]. Bani‐Hani, K. A. (2007) “Vibration control of wind‐induced response of tall buildings with an active tuned mass damper using neural networks” Structural Control Health Monitoring, 14(1), 83-108.
[17]. Yang, S. M., Chen, C. J., & Huang, W. L. (2006) “Structural vibration suppression by a neural-network controller with a mass-damper actuator” Journal of Vibration and Control, 12(5), 495-508.
[18]. Chen, C. J., & Yang, S. M. (2014) “Application neural network controller and active mass damper in structural vibration suppression” Journal of Intelligent & Fuzzy Systems, 27(6), 2835-2845.
[19]. 賴魁星(2021)，長短期記憶神經網路於提升地震模擬震動台加速度性能之研究，國立臺灣科技大學營建工程所碩士論文，陳沛清指導。
[20]. Kingma, D. P., & Ba, J. (2014) “Adam: A method for stochastic optimization” arXiv preprint arXiv:1412.6980.
[21]. Ferreira, A.A.; Ludermir, T.B. and Aquino, R. (2012) “Comparing recurrent networks for time-series forecasting” The 2012 International Joint Conference on Neural Networks (IJCNN), Brisbane, QLD, 1-8.
[22]. Yohanes, K.S. (2021) “Application of Metaheuristic Optimization for Parameter Design of Building Mass Damper Systems” 國立臺灣科技大學碩士學位論文，陳沛清指導
[23]. Cheng, M. Y., & Prayogo, D. (2014) “Symbiotic organisms search: a new metaheuristic optimization algorithm” Computers & Structures, 139, 98-112.
[24]. Jansen, L.M. and Dyke, S.J. (2000) “Semiactive control strategies for MR dampers: comparative study” Journal of Engineering Mechanics, 126(8): 795-803.
[25]. Chen, P.C.; Ting, G.C. and Li, C.H. (2020) “A versatile small-scale structural laboratory for novel experimental earthquake engineering” Earthquakes and Structures, 18(3): 337-348.
[26]. Amini, F.; Hazaveh, N.K. and Rad, A.A. (2013) “Wavelet PSO-based LQR algorithm for optimal structural control using active tuned mass dampers” Computer-Aided Civil and Infrastructure Engineering, 28(7): 542-557.