Transformer-based power system energy prediction model

With the rapid development of smart grid construction, building energy consumption prediction is gaining more and more attention in energy planning, management, and conservation. Improving the accuracy of energy consumption judgment is a key factor to ensure efficient operation of the energy system. In addition, the model needs to be able to quickly adapt to changes in energy consumption and respond to various emergencies. This paper thus proposes a modified Transformer model based on Multi-head attention and position encoding mechanism. This model makes Transformer focus on the important information which influence the prediction performance by using self-attention mechanism. Compared with LSTM, it achieves the higher prediction accuracy with less training time by the combination of Transformer and position encoding. The final experiment results demonstrate that the proposed method has a good performance in the electronic load prediction task, and it is also suitable for the application scenarios in periodic change prediction such as forecasting sales and price prediction.