The invention discloses a calculation method suitable for ultra-short-term load prediction. The method comprises the following steps: 1, obtaining system total loads of multiple time points; 2, scanning the multiple system loads according to types of time points to obtain a historical sample; 3, calculating a first load change rate average value delta Liav of each point in the historical sample; 4, performing identification correction on bad data in the historical sample; 5, recalculating a second load change rate average value delta L' iav, comparing the second load change rate average value with the first load change rate average value delta Liav calculated last time, if fabs (delta L' iav-delta Liav) of each point is smaller than 0.01, consuming that identification of the bad data is completed, by use of a current load change rate, entering the sixth step, and otherwise, entering the fourth step; and 6, according to a current load change rate delta Liavf calculated in the fifth step, by use of a load Li of a current time point i of a prediction day, carrying out ultra-short-term load prediction at a time point i+1. The method is simple and practical, and the prediction result is accurate.