During the long-term monitoring process, the data collected by the structural health monitoring system are in-complete due to the presence of many factors such as sensor equipment failure, energy supply interruption, and network transmission problems. To address this problem, combining the advantages of ensemble empirical modal decomposition (EEMD) and bi-directional long and short-term memory network (BiLSTM) in timing processing, this study proposes an EEMD-BiLSTM-based reconstruction method of missing data for structural monitoring. The method utilizes EEMD to adaptively decompose the monitoring time-series data into a set of intrinsic modal components (IMFs) representing different time scales, thereby smoothing the nonlinear, nonsmoothly ordered time-series signals. The IMF components are then input into the BiLSTM network for missing data reconstruction, thus improving the BiLSTM prediction accuracy. The six-story frame structure scale model and Benchmark finite element simulation model are analyzed. The experimental results show that, compared with the mainstream methods of EEMD-LSTM, BiLSTM and LSTM, the EEMD-BiLSTM proposed in this study has the highest pre-diction accuracy. In the case of 5% missing ratio, its R2 metrics is 0.85478 and 0.86245 respectively. Therefore, the preprocessing of unsteady structural acceleration response data by EEMD method can effectively improve the prediction accuracy of BiLSTM, which provides a more adaptable method and theory for the problem of missing data in structural monitoring.