The invention provides a network loss data cleaning method based on multi-dimensional space-time analysis. The method comprises: (1) data property positioning: carrying out property division on a newupper measuring point or a historical measuring point according to a historical access time judgment rule of data; (2) abnormal data identification: abnormal data identification of a new upper measuring point and abnormal data identification of historical measuring points are included, short-term data abnormal identification based on a voltage and current method is applied to the new upper measuring point, and medium-term and long-term abnormal data identification based on a regression analysis method is applied to the historical measuring points; (3) abnormal data correction: carrying out abnormal data correction on the abnormal data identified in the previous step by utilizing a multiple linear interpolation method; and (4) checking the correction data: recalculating the network loss byadopting an electric quantity reduction method, and verifying the accuracy of the correction data according to the network loss. The method has the advantages that the method can be applied to cleaning and screening of multi-type access electric quantity data in a network loss calculation system, an accurate and reliable data source is provided, and scientific and effective network loss optimization is realized.