In light of the inadequacy of the currently used conventional load model for reliability assessment， this paper proposes a new integrative K -means clustering algorithm that takes into account the load characteristics information. Based on the clustering algorithm while considering the bus loads correlation and load forecasting uncertainty， this model adopts the multi-dimensional correlation sampling technique to determine the bus load in the evaluation period. Based on the proposed load model and cumulative probability load model， a new method for frequency index calculation with load shifting taken into account is deduced， solving the problem of frequency index calculation in non-sequential Monte Carlo simulation. Finally， the effects of number of clustering， bus load correlation， load forecast uncertainty and load shifting on reliability indices are discussed and compared by case studies. The feasibility and rationality of the proposed algorithm are proved.