Seismic fragility analysis of RC frame-shear wall structure under multidimensional performance limit state based on ensemble neural network

•A seismic multidimensional fragility method is proposed based on neural network.•The assumption of distribution type of engineering demand parameter is not required.•An ensemble neural network is proposed through weighted average strategy.•A novel index calculated by limit state function is taken as the network output.•The proposed ensemble neural network shows better prediction ability. An improved seismic fragility analysis method without any assumption on the distribution of the target engineering demand parameter is proposed based on an ensemble neural network. Firstly, a multidimensional performance limit state function is introduced to evaluate both the structural and non-structural performance. Next, an ensemble neural network, which is defined as a weighted average of three artificial neural networks, is trained to predict the index of the structure subjected to any seismic ground motion. A new index derived from the multidimensional performance limit state function is defined as the neural network output. Finally, the fragility curves are obtained by means of Monte Carlo simulation with high computational efficiency with the help of the ensemble neural network. It is applied to a reinforced concrete frame-shear wall structure, and the maximum inter-story drift ratio and the peak floor acceleration are selected as two engineering demand parameters. The results indicate that, the prediction accuracy of the ensemble neural network is higher than other three neural networks. If only a single neural network is used, the obtained fragility curves may have much error. Compared to the traditional multidimensional probabilistic seismic demand model based on the lognormal distribution assumption, the proposed method can give more credible seismic fragility curves.