Joint Application of Intent Recognition and Slot Filling in Query Question-Answering System for Wind Farm Faults

Intelligent question-answering systems have significant advantages in the fault querying for wind farms. However, due to the unpredictability of faults in a wind farm and the scarcity of data, some problem-understanding models in such question-answering systems face challenges when dealing with small datasets for a specific task. This is mainly reflected in the lack of flexibility of the models and the need to optimize too many parameters during the fine-tuning phase. Therefore, this paper proposes a joint model of intent recognition and slot filling with an adapter structure based on a robustly optimized bidirectional encoder representations from transformers pre-training approach. The joint model is optimized by fine-tuning the position of the adapter and the reduction factor. The experimental results show that the proposed model significantly improves the accuracy of intent recognition and slot filling subtasks compared with some other joint models, thus confirming its effectiveness.