Bi2-BERT-Based Long Sequence Text Vectorization Method for Power Grid Work Orders

With the continuous development of power systems, power grid work order data provides a large amount of data support for power facilities. However, the long sequence of text in power grid work orders brings challenges to computer processing and analysis. In order to improve the accuracy of power grid work orders, this paper proposes a long sequence text vectorization method for power grid work orders based on leader grey wolf optimization-bidirectional long short-term memory and bidirectional encoder representations from transformers (Bi2-BERT). Firstly, a long sequence text feature extraction method is proposed, which performs the long sequence text feature extraction by BiLSTM, to improve the global optimization performance. Secondly, a text vectorization method for power grid work orders based on increment-stock bilevel-driven BERT is proposed to improve the model learning efficiency and solve the catastrophic forgetting problem. The results show that the proposed algorithm has high accuracy and convergence speed in long sequence text vectorization of power grid work orders.