AUTHOR=Tao Peng , Ma Hao , Li Chong , Liu Linqing 

TITLE=Intelligent grid load forecasting based on BERT network model in low-carbon economy

JOURNAL=Frontiers in Energy Research

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/energy-research/articles/10.3389/fenrg.2023.1197024

DOI=10.3389/fenrg.2023.1197024

ISSN=2296-598X

ABSTRACT=In recent years, mitigating the risks and challenges associated with high carbon emissions has become a critical goal for industries worldwide. To achieve this goal, low-carbon transformation has emerged as a key direction for various enterprises and industries. Among them, power systems play a crucial role in the process of low-carbon transformation as they are the main driving force for national development. Therefore, how to help power dispatching departments achieve efficient energy scheduling and utilization has become a core issue in the development of low-carbon economy. With the rapid evolution and development of smart grid and artificial intelligence technology, transformer load forecasting has become the top priority of enterprise power demand management. At the same time, reasonable transformer load forecasting can effectively promote the realization of carbon neutrality goals. Traditional forecasting methods based on regression analysis and support vector machines are difficult to meet the increasingly complex and diverse power load data forecasting. This paper proposes a Bert-based time series power load forecasting method, which combines natural language processing and machine learning methods to improve the accuracy and efficiency of power grid transformer load data forecasting. First, we use Bert to perform data preprocessing, analysis, and feature extraction on the long-term historical load data of power grid transformers. Next, we conduct several rounds of training and fine-tuning on the proposed Bert algorithm among the preprocessed datasets. Finally, we perform load forecasting on the trained model and compare the results with the actual composite data. The results of our future experiments demonstrate that the Bert-based model we have developed is capable of achieving accurate short-term power load forecasting. Furthermore, our Bert-based power load forecasting algorithm can achieve high levels of accuracy, providing strong support for power dispatching departments' resource management and offering theoretical guidance for carbon reduction initiatives.