A power load prediction method of associated industry chain production resumption based on multi-task LSTM

Abstract

In the power load prediction of industries after production resumption, the single-task prediction is unable to learn the correlation characteristics among industries, making it difficult to further improve the prediction accuracy. A multi-task power load prediction method is proposed in this paper, which is based on multi-task learning mechanism and long and short-term memory network model. The correlation relationship between subtasks is fitted by the parameter sharing layer of multi-task learning, which in turn achieves the effect of improving the prediction accuracy. Firstly, the power consumption correlation between the upstream and downstream industries in the industry chain is analyzed, and multi-task long and short-term memory model is established. Then, according to the structure input of the model, the pre-processed data, such as medical treatment, public opinion, policy, and industry capacity, are input. The sharing layer is used to learn the shared information of upstream and downstream industries of the industry chain, and the validation index is proposed for model accuracy. Finally, compared to MLR, SVR, RNN and single-task LSTM, the multi-task LSTM model respectively improves the weighting accuracy by 2.637%, 2.014%, 1.515% and 1.064%, which verifies the effectiveness of the model.

Keywords

• Multi-task learning
• Long and short-term memory network
• Power load prediction
• Correlation characteristics
• Industry chain