Chinese power dispatching text entity recognition based on a double-layer BiLSTM and multi-feature fusion

A large amount of unstructured data has been accumulated in the daily dispatching work of power systems as the form of text. In order to use these texts effectively, entities in the text need to be recognized, such as names of station and equipment. Because of the complex composition of the power dispatching text, this paper first summarizes the characteristics of the power dispatching text. A character-level entity recognition model based on multiple features is proposed, which is suitable for power text. Our model combines pretrained character embedding, left-neighbour entropy, and part-of-speech to represent the domain characteristics of power dispatching text. And we exploit the fusion method for multiple features inputting. The double-layer BiLSTM proposed in this paper is used to predict character sequence labels, and finally CRF is used to optimize the label predicted. This paper chooses a power outage maintenance application to recognize name entities. The results of experiments show that our model can increase the overall F-1 value by 2.26% compared with the traditional models, and the recognition of lines and stations has increased by 3.88% and 3.99%. The recognition accuracy of each tag has been enhanced. (c) 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 2020 The International Conference on Power Engineering, ICPE, 2020.