Source-Load Joint Probability Prediction Based on Transformer Model

Renewable energy power generation prediction, including wind and photovoltaic power prediction, is usually performed separately from load prediction, while there are significant internal relationships between renewable energy power generation and load consumption. Therefore, exploring the relationship between source and load, and implementing sourceload joint prediction may improve the accuracy of source-load prediction simultaneously. This paper proposes a source-load joint probability prediction method, PCMDT (Probability-CNN-Multi-Decoder-Transformer), based on Transformer model. Firstly, we analyzed the possible aspects of the relationship between renewable energy and load. Then the CMDT(CNN-Multi-Decoder-Transformer) model was used to realize the point prediction of renewable energy power generation and load power. Next, the conditional quantile was obtained by optimizing the quantile loss function (Pinball Loss) to realize the joint probability prediction of source and load. Finally, the model was tested on a public state-level dataset. Comparing the results of source-load joint prediction with non-joint prediction, the proposed model can improve the accuracy of renewable energy and load prediction.