The invention relates to a predication method for load peak of a power distribution area. Based on a predication model established by a time sequence of long-term data acquisition, the predication method comprises the following steps of (1) obtaining daily peak data of the load of the area according to the historical load data of the area; (2) carrying out discrimination and restoration on distortion and deletion in the data; (3) establishing a periodicity equation according to an autocorrelation function of the daily peak data of the load; (4) establishing a gradient equation according to the autocorrelation function of the data with periodicity; (5) establishing a predication model of the daily peak of the load by combination of the periodicity equation and the gradient equation; (6) obtaining estimation of model parameters by enabling the quadratic sum of errors to be minimum; and (7) obtaining a load predication value of the daily peak of time points in the future according to the established model. The predication method has the advantages that the blank and the defects in the prior art are filled and overcome, the accuracy is high, the operability is high and the predication of the daily peak of the load of the area is realized.