Universal Generating Function (UGF) techniques have been applied to Multi-State System (MSS) reliability analysis, such as long term reserve expansion of power systems with high wind power penetration. However, using simple steady-state distribution models for wind power and large generating units in reliability assessment can yield pessimistic appraisals. To more accurately assess the power system reliability, UGF techniques are extended to dynamic probabilistic simulation analysis on two aspects of modelling improvement. Firstly, a principal component analysis (PCA) combined with a hierarchal clustering algorithm is used to achieve the salient and time-varying patterns of wind power, then a sequential UGF equivalent model of wind power output is established by an apportioning method. Secondly, other than the traditional two-state models, the conventional generator UGF equivalent model is established as a four discrete-state continuous-time Markov model by Lztransform. In the construction process of such a UGF model, the state values are transformed into the integral multiples of one common factor by choosing proper common factors, thus effectively restraining the exponential growth of its state number and alleviating the explosion thereof. The method is suitable for reliability assessment with dynamic probabilistic distributed random variables. In addition, by acquiring the clustering information of wind power, the system reliability indices, such as fuel cost and CO 2 emissions through different seasons and on different workdays, are calculated. Finally, the effectiveness of the method is verified by a modified IEEE-RTS 79 system integrated with several wind farms of historical hourly wind power data of Zhangbei wind farm in North China.