The invention relates to a trend recognition method for electric power system predication-like data. The method includes the following steps that the predication-like data are read; discretization is performed on the predication-like data to form time sequence data equal in time interval; the time sequence sampling period, the minimum value, the time point of the minimum value, the maximum value and the time point of the maximum value and the fluctuation range of the time sequence data are determined; a time frame fluctuation threshold is determined; on the basis of the time frame fluctuation threshold, the time sequence data are divided into various trend time frames; whether all the time sequence data are recognized or not is judged; all the trend time frames are judged, the trend time frame smaller than the minimum value of the time frames serves as an invalid time frame, and the invalid frame and the adjacent time frame are merged; trend time frames available within the sampling period of electric power system optimal control are formed. The method analyzes the result of the electric power system predication-like data to recognize a load or power generation increasing trend and a load or power decreasing trend, and therefore trend recognition information needed by optimal control is formed.