Short-Term Net Load Forecasting Model for Microgrid Based on Improved Transformer

In the context of a large number of new energy generation gradually connected to the new power system, the increasing penetration of such uncertain renewable energy sources like wind and solar in the microgrid system increases the uncertainty in the planning of system operation. Various uncertain variables such as combined load, wind, and solar can be combined to form an uncertain variable called net load. To solve the uncertainty of renewable energy generation, the traditional forecasting model can no longer achieve effective accuracy, for this reason, the iTransfomer multivariate load ultrashort-term forecasting model with the improvement of Transformer is established. The model first performs dimensional transposition before embedding the input data, which makes the whole sequence of the same variable mapped into a high-dimensional feature representation, and promotes the dependence between different variables through the interaction of multiple self-attention mechanisms, which is more favorable for multivariate load forecasting; Secondly, the Transformer decoder is discarded, and the decoder is retained to solve the complexity of the Transformer computation to form the iTransformer prediction model. Taking the small microgrid in the suburb of Trondheim, Norway as an example, the comparative analysis between the model established in this paper and the traditional model proves that the model in this paper has a higher prediction accuracy.