Abstract: With the continuous decline in the proportion of synchronous units in receiving-end power grids with high penetration of renewable energy, the inherent frequency regulation capability of power systems is gradually weakened, leading to growing severe risks to frequency security. To investigate the evolution laws of frequency regulation capability and frequency security region for receiving-end power grids, this paper proposes an analysis method based on historical time-series data. Firstly, a multi-machine aggregated frequency response model of the receiving-end power grid is established. The equivalent system inertia and primary frequency regulation coefficient are extracted as key parameters for quantifying the system's frequency regulation capability. The analytical expressions of frequency stability indices under anticipated disturbances are derived. On this basis, critical parameter sets are determined subject to frequency security constraints, and the boundary of the frequency security region of the receiving-end grid under anticipated disturbances is delineated. Secondly, the fractional-order multivariate grey model (F-MGM) is adopted to jointly predict the equivalent inertia and primary frequency regulation coefficient of the power grid in future years, and the prediction accuracy is evaluated. Meanwhile, combining the multilayer perceptron (MLP) with the Koopman operator, the evolution law of the security region boundary is learned from historical boundary data to realize the annual time-series prediction of the frequency security region boundary. Furthermore, the future frequency security margin can be quantified by comparing the predicted frequency regulation capability parameters with the security region boundary. Finally, historical and planned data of the East China Power Grid from 2021 to 2028 are utilized to verify the proposed method and deduce the future frequency regulation capability and frequency security region of the power grid. The results demonstrate that F-MGM outperforms conventional models in the prediction of frequency regulation capability parameter sets. For the prediction of frequency security region boundaries, the MLP-Koopman method can capture the long-term evolution characteristics of boundaries more accurately than the pure MLP approach. In addition, this method reveals the physical connotation underlying the evolution of frequency security regions from the perspective of system dynamics.