A Transfer Learning-based Method for the Daily Electricity Consumption Forecasting of Large Industrial Users After Business Expansion

Abstract

With the rapid development of industry, the expansion capacity and frequency of large industrial users continue to increase. However, the traditional static prediction model is difficult to accurately predict the daily electricity consumption of industrial expansion, which is not conducive to the safe and stable operation of the power grid. In response to the above problems, this paper proposes a transfer learning-based method for the daily electricity consumption forecasting of large industrial users after business expansion. Firstly, a dynamic training framework for the prediction model of transfer learning is established, so that the prediction model can dynamically adapt to the capacity change brought about by the expansion of multi-user business. Then, a neural network for predicting daily electricity consumption of industrial users based on multi-resolution time series attention is established, which can deeply mine the characteristics of electricity sequence. Finally, a deep learning model parameter migration and adjustment method considering business expansion is proposed, which can realize efficient migration of prediction models. The effectiveness of the proposed method is demonstrated by comparing it with state-ofthe- art electricity forecasting based on two-year historical data of a specific region. The proposed method is compared with state-of-the-art power forecasting techniques through the validation of local historical data. The obtained results demonstrate the effectiveness of the proposed method.