Data assimilation (DA) has been used as a machine learning approach to estimate a system's state and the unknown parameters in its numerical model by integrating observed data into model predictions. In this paper, we propose using the DA methodology based on the ensemble Kalman filter (EnKF) to improve the accuracy of microstructure prediction using three-dimensional multi-phase-field (3D-MPF) model and estimate the model parameters simultaneously. To demonstrate the applicability of the DA methodology, we performed numerical experiments in which a priori assumed true parameters related to the grain boundary (GB) energy cusp and GB mobility peak of Σ7 coincidence site lattice GB were estimated from synthetic data of time-evolving polycrystalline microstructure. Four model parameters related to the Σ7 GB properties were successfully estimated by assimilating the synthetic microstructure data to the 3D-MPF model predictions using the EnKF-based DA method. Furthermore, we accurately reproduced the preliminarily assumed true shapes of GB energy cusp and GB mobility peak by using the estimated parameters. The results suggest that implementation of the EnKF-based DA method in the MPF model has great potential for identifying unknown material properties and estimating unmeasurable microstructure evolutions in polycrystalline materials based on real time-series 3D microstructure observation data.