The invention provides an optimized local shape similarity based microgrid ultra-short-term load forecasting method. The method comprises the steps of 1, obtaining a microgrid ultra-short-term load forecasting model with the optimized local shape similarity; 2, determining an optimized local shape similarity curve series, which is the most similar to the variation tendency of a to-be-forecasted real-time curve series, by an optimized local shape similarity coefficient, from the microgrid ultra-short-term load forecasting model; and 3, obtaining a microgrid ultra-short-term load forecasting value according to the optimized local shape similarity curve series and the real-time data of the microgrid ultra-short-term load. The forecasting method is high in fitting degree and accuracy for continuous data, and has significant meaning for the forecasting of the large-area ultra-short-term load in the current microgrids.