Safety Supervision for Coal Mines Using Machine Learning Methods

Jun Zhang,Yunlong Wang,Dezhi Wang,Zhigang Zhao,Xiong Luo

DOI: https://doi.org/10.1007/978-981-33-4336-8_1

2020-01-01

Abstract:Coals are of great importance to economic development, but the safety problem always affects the efficient production of coal mines. Then, safety supervision is the basis for guaranteeing the productive efficiency of coal mines. The safety supervisions were previously based on handcrafted models of reality. Nowadays, they are able to be achieved using data-driven learning approaches. Hence, to improve the supervision performance in consideration of complex characteristics of coal mine production data, this paper presents a machine learning scheme towards safety supervision. Generally speaking, there are two parts in the developed scheme. On the basis of bidirectional encoder representations from transformers (BERT) model, a regression prediction is firstly used to analyze the coal mine data, with the purpose of finding those possible accident attributions. Meanwhile, to tackle the issues of discovering the relationship in data, the unsupervised Apriori learning algorithm is also developed to mine these association rules in coal mine data. The presented scheme, named machine learning-based safe supervision (MLSS), is experimentally validated by using actual data collected from some coal mines in China. The results demonstrate that our method can achieve satisfactory analysis performance.