Demand-side genus matching based on GAF and generalised self-attention mechanism

The study of matching power consumption attributes of station users plays an important role in smart grids, providing an important basis for tasks such as power dispatch and energy optimisation. Traditional classification tasks usually use hand-designed feature extractors to classify the users in the station area. To solve the problem that its performance is limited by the feature expression, the daily power data of the users are transformed into the form of two-dimensional images by introducing the Gramian Angular Field (GAF) in order to capture their time-varying characteristics, and the image features are learnt by combining with the generalized self-attention mechanism, and the network model based on Transformer is used for the analysis of the generated feature images to achieve a more accurate classification. The generated feature images are analysed to achieve more accurate user category classification. The simulation experiments show that the provided model and method have a good effect of station classification.