The intermittence and fluctuation of renewable energy generation pose a huge threat to the frequency stability of modern power system. Thus, the assessment of frequency stability is of great significance to the safety operation of power system. Due to the lack of accurate and efficient probabilistic frequency stability assessment methods and the disadvantages of traditional polynomial chaos expansion methods, this paper proposes a probabilistic frequency stability assessment method based on generalized polynomial chaos expansion. Based on generalized polynomial chaos expansion, the proposed method is suitable for random variables following arbitrary distributions. Moreover, Latin hypercube sampling is applied to improve the sampling efficiency. In addition, the utilization of Cholesky decomposition enables the proposed method to handle correlated random variables. Furthermore, hyperbolic truncation is adopted to alleviate the curse of dimensionality in the traditional polynomial chaos expansion methods. The etficiency and accuracy of the proposed method are verified in the IEEE-39 bus and IEEE-68 bus systems compared with other methods.