非監督式模糊神經網路推理模式於結構主動控制之應用

Abstract

本文將提出一個新的結構主動控制方法，此方法整合了非監督式模糊神經網路推理模式（Unsupervised Fuzzy Neural Network Reasoning Model，UFN）以及非監督式神經網路分類模式（Unsupervised Neural Network Classification Model），將此控制方法稱作UFN控制模式。在減低結構反應累積之脈衝式結構主動控制律的基礎上，建立結構物在某既已發生的地震下之結構響應對控制力的案例庫（Instances Base），當另外之地震擾動發生時，再從案例庫中藉由UFN推理模式找到在此地震擾動下所需要的控制力。UFN控制模式包含了兩個重要的部份：即(1)案例庫的建立與(2)模糊推理。其中，案例庫的建立乃是建立結構響應與控制力之間的關係。首先藉由脈衝式結構主動控制律求得許多的訓練案例（Training Instances），再將這些訓練案例藉由非監督式神經網路分類模式加以分類成若干個叢集（Clusters），而這些叢集的集合就稱之為案例庫；而模糊推理乃是從案例庫求得控制力的過程，採用UFN推理模式加以完成。最後，以受到地震擾動作用之單自由度系統為例，分析UFN控制的可行性。從實例的結果來看，UFN控制確實可降低地震下的結構反應，證明UFN控制於土木結構上的可行性。

In this work, an Unsupervised Fuzzy Neural Network(UFN) reasoning model is applied to the domain of structural control based on active control strategy. The UFN reasoning model is based on a single-layer lateral-connected neural network with an unsupervised competing learning algorithm. The basis of the active UFN reasoning structrural control is based on that response of a structure under excitation is governed by a structural dynamic formula, a function of mass, damper, and stiffness of structure. Given a set of instances, consisting of response of structure and corresponding control force, the UFN reasoning model is used to classify these instances into certain clusters. The instances are similar to each other in the same cluster and dissimilar to the instances in other clusters. Herein, two examples of single-degree-of-freedom (SDOF) under earthquake excitations are used to verify the performance of the UFN reasoning model active control model. The results indicate that the proposed control strategy has the capability of reducing the vibration of structures under earthquake excitations.