The invention relates to the technical field of power system scheduling and operation, in particular to a micro-grid short-term load prediction method based on long-term and short-term memory and self-adaptive lifting, which comprises the following steps: step 1, calling historical load data; step 2, integrating the data to obtain a training set and a test set; step 3, performing integrated empirical mode decomposition and adjustment on the training set and the test set, and outputting a training sample set and a test sample set; 4, establishing a combined prediction kernel model, and settinghyper-parameter values; 5, inputting the training sample set, and outputting a prediction result; 6, setting the cycle index N, and entering the step 7 when the actual cycle index is greater than N; if the actual cycle index is smaller than N, entering the step 5; 7, calculating a root-mean-square error, judging whether the root-mean-square error is stable or not, entering the step 9 if the root-mean-square error is stable, and entering the step 8 if not ; 8, adjusting hyper-parameters, and entering the step 5; and step 9, inputting the test sample set, and outputting a prediction result. Themethod is high in prediction precision, small in error, high in adaptability and high in practicability.