Power consumption prediction method and system based on attention mechanism fusion frequency enhancement

The invention discloses a power consumption long-time prediction method, system and device based on attention mechanism fusion frequency enhancement and a medium, a deep network structure model is constructed based on attention mechanism fusion frequency enhancement, and the deep network structure model comprises an encoder, a frequency-enhanced mixed attention module and a decoder which are connected in sequence. The frequency-enhanced mixed attention module comprises a self-attention module and a frequency-enhanced channel attention module, and predicts the power consumption based on a deep network structure model. According to the electric power consumption prediction method provided by the invention, an attention mechanism is added in the Transform neural network according to the characteristics of the electric power consumption data, so that the prediction result of the model is more stable, and the anti-noise capability of the model in the frequency domain is enhanced through the frequency-enhanced mixe