Self-learning fuzzy control of civil structures

Abstract

The application of ANFIS (Adaptive Network-based Fuzzy Inference System) to the fuzzy control of structures was investigated by the authors in a earlier paper (L. Faravelli and T. Yao, 1994). Through neural network learning, ANFIS can be trained to replace an existing fuzzy controller. The resulting controller makes use of the more efficient Takagi-Sugeno inference scheme instead of COG (center of gravity) and is inherently computationally faster. The next logical step accomplished in this paper is to implement the trajectory adaptive networks (TAN) and stage adaptive networks (SAN) that were proposed to be used with temporal back propagation to achieve a self learning fuzzy controller. This approach should result in a fuzzy controller that is optimized to handle loads of the type used in the self learning training. Because the learning process is goal directed (i.e., a zero vector is the desired displacement behavior), some optimization is introduced.