The invention relates to the technical field of
power demand prediction, and discloses a
power demand prediction method and
system fusing adversarial enhancement and a causal
perception mechanism, and the method comprises the steps: collecting and preprocessing the historical data of
power consumption, and dividing the historical data into a historical known variable, a future known variable and a
static variable; performing sample enhancement on the preprocessed data through a Time GAN model, and inputting a historical known variable, a future known variable and a
static variable into a TFT model for model training after the sample enhancement; and performing prediction based on the trained TFT model, and outputting a
power demand quantile prediction result of a future target
time step. According to the invention, by introducing the integrated architecture of the TimeGAN and the TFT, a composite model with structural transparency and long-term prediction capability is constructed while an extreme scene sample is enhanced, so that intelligent and refined prediction of
regional power consumption demands is realized, and support is provided for power dispatching,
load planning and policy analysis.