Compared with the traditional reliability, the operational reliability of power system can reflect the impacts of short term operation on the reliability. So there is the need to establish a prediction model to forecast short-term load oriented for the operational reliability, which can provide the probability model for adequacy indices evaluation, and provide an accurate model needed for the system security assessment. This paper, an improved fuzzy clustering is proposed to extract similar day, and DB4 wavelet is used to decompose the similar day load sequence in 3 layer wavelet. Then the improved PSO-RVM algorithm has been used to predict wavelet components of forecasted days, and the results of wavelet prediction have been superposed to get the mean value and probability model of load sequences of forecasted days. The simulation results show that the proposed method has better prediction accuracy