Predicting the Seismic Response of Structures with Nonlinear Behavior Using the Combined Fuzzy Inference Model and Multi-verse Optimization Algorithm

One of the effective parameters in performance-based design methods is the determination of lateral displacement demand. Due to the existence of uncertainties in the parameters of analytical models such as mechanical properties of structures and model simplifications, accurate calculation of structural responses is associated with complexities. The use of training-based prediction methods can be a good alternative to accurate analysis in assessing the seismic behavior of a building structure. In this paper, an efficient training approach for modeling and predicting the response of building structures with nonlinear behavior is studied. To perform the training process, an adaptive scheme of fuzzy inference system with the TSK model combined with Multi-Verse optimization algorithm is used to model the seismic behavior of structures. The proposed training model is implemented by optimizing the parameters of the TSK model using the optimization algorithm based on comparing the previous time steps responses. To implement the adaptive design and increase the accuracy of the prediction, three training cases based on the responses of 2, 5, and 10 previous time steps were used. To train this system, the data collected from the results of nonlinear time history analysis under 100 seismic events with different characteristics have been used. Also, 10 events were used to test the inference system. The performance of the proposed design was evaluated on a shear frame structural model with nonlinear hysteresis behavior. The results show that the inference system of the TSK model by combining the optimization method is an efficient computational method for predicting the response of nonlinear structures. The average MSE for the test group ground motions, using three training modes with 2, 5, and 10 previous time steps, 2.817e-03, 1.228e-03, and 2.953e-04, respectively. By increasing the number of time steps from 2 to 10, the prediction error decreases by 89.52%.