Active structural response control with self-learning mechanism

Using the concept of neural network we have developed an algorithm to control seismic responses of structures. To take into account the uncertainty of dynamic characteristics of structural system the back propagation learning process is applied to identify the structure parameters such as mass damping and stiffness matrices. A new closed-open-loop optimal control scheme that has been derived by minimizing the sum of the quadratic time-dependent performance index and the seismic energy input to the structural system is implemented into the learning process of a layered network. This algorithm is simple and reliable for on-line control operations and effective for a structural system. Numerical examples are worked out to demonstrate the control efficiency of the proposed algorithm.