The invention provides a power distribution network loss electric quantity calculation method and device based on a neural network. The method comprises the following steps: acquiring power distribution network data such as a load value, meteorological data and a corresponding date type of each 24 hours in a preset time period in a power distribution network jurisdiction range, and constructing apower distribution network database; screening out related data in the fault area from a power distribution network database; normalizing the screened related data of the fault area; simplifying the normalized non-load value data by using a principal component analysis method, selecting the non-load value data of which the cumulative contribution rate is greater than a preset threshold value as input data of a power distribution loss electric quantity calculation model, and taking a corresponding load value as output until the power distribution loss electric quantity calculation model is trained; calculating a loss load in the fault area by using the trained power distribution loss electric quantity calculation model; and performing reverse normalization processing on the loss load in thefault area to obtain the loss electric quantity in the fault area.