The invention discloses a method and a device for electricity sales amount prediction based on random forest regression, wherein a monthly electricity sales amount prediction model based on the random forest regression is established according to historical monthly electricity sales amount data in a preset period and multiple influential factor indexes which are determined in advance. The multiple influential factor indexes in a to-be-predicted month are acquired and calculated and then substituted into the monthly electricity sales amount prediction model so as to obtain monthly electricity sales prediction results in the future to-be-predicted month.