To enhance coal mine safety risk identification and supervision capabilities,a model based on Bidirectional Long Short-Term Memory Networks(BiLSTM),Conditional Random Fields(CRF),and Latent Dirichlet Allocation(LDA)is proposed.The BiLSTM-CRF model is trained to split words;the perplexity-var is used to calculate the optimal number of topics for the LDA model;and the BiLSTM-CRF-LDA model is constructed to mine the data of safety hazards in a coal mine in Inner Mongolia.The research findings indicate that the perplexity-variance metric can more accurately determine the number of topics;the word segmentation results of the BiLSTM-CRF model are more precise compared to those of the jieba library;the BiLSTM-CRF-LDA model can accurately identify types of safety hazards,spatial distribution of safety hazards,and the allocation of safety responsibilities in coal mines.These results can provide a reference for coal mine safety risk exam-ination and supervision.
关键词
煤矿安全隐患/BiLSTM/CRF/LDA/困惑度-主题方差
Key words
potential safety hazard data of coal mine/bi-directional long short-term memory(BiLSTM)/conditional random field(CRF)/latent Dirichlet allocation(LDA)/perplexity-topic variance
维普
万方数据