A Deep Learning–Based Causal Knowledge Extraction Method for Supporting the Development of Nuclear Power Plant Operator Knowledge Base

In human reliability analysis (HRA) for nuclear power plants, cognitive modeling–based approaches require the development of an operator knowledge base to simulate the cognitive processes of operators. However, existing automatic extraction methods fail to provide knowledge that meets the granularity requirements of cognitive modeling for the development of the operator knowledge base.

To address this gap, this paper proposes a deep learning–based extraction method. Specifically, the method utilizes a bidirectional encoder representations from transformers (BERT)–bidirectional long short-term memory (Bi-LSTM)–1conditional random field (CRF) model to perform sequence labeling for extracting fine-grained knowledge, such as entities and their corresponding states, as well as the causal relationships between these pieces of knowledge. Additionally, we define mapping rules to structure the extracted causal knowledge to facilitate the integration of additional knowledge.

To validate the extraction effectiveness of the BERT-Bi-LSTM-CRF model, experiments were conducted on a data set constructed from licensee event reports. The experimental results showed that the model achieved a macro-F1 score of 0.876 on the test set, indicating that the model is capable of effectively extracting the required knowledge and relationships from unstructured text. This method is expected to be applied in the development of operator knowledge bases, potentially reducing the workload involved.