To avoid the difficulty of choosing initial cluster centers and clustering number for the k-means algorithm, an improved k- means clustering algorithm is proposed to build multistep load model from the hourly load data. The extended probabilistic load model considering bus load uncertainty and correlation is built. RBTS and IEEE RTS79 reliability test system is used to validate the proposed load model. Case studies show that the result based on proposed load model has high accuracy and it saves the calculation time when adopting the state sampling method; the uncertainty and correlation of bus loads have great effect on the adequacy indices of generating system. The proposed load model is helpful for the generation and composite system’s reliability assessment.