The invention discloses a method for identifying and repairing power load abnormal data based on density clustering and LSTM, and belongs to the technical field of power quality analysis methods. According to the method, a density-based clustering algorithm (Density-based Spatial Clustering of Applications width Noise) and Long Short-Term Memory Neural Network are combined to identify and repair abnormal data. The method comprises the following steps: firstly, carrying out density clustering on data in units of days by utilizing a DSCAN algorithm to obtain abnormal data; then, using a long short-term memory (LSTM) neural network, taking the time series data judged to be abnormal as input of the LSTM neural network, and using the first n pieces of sequence data to predict the next piece ofsequence data; finally, the predicted value of the LSTM serving as an accurate value, setting an up-down floating threshold value is set, if the measured value exceeds the threshold value range, regarding the measured value as an abnormal value, and the predicted value of the LSTM serving as a correction value. According to the method, the time sequence and regularity of the power quality monitoring system data in the actual power grid are fully considered, the specific abnormal value can be accurately detected and repaired, and the method has good actual application value.