Abstract: In order to quickly and accurately predict danger of coal and gas outburst, an IFOA-ELM prediction model based on preprocessing is proposed. Firstly, some measured data of Pingdingshan Eighth Mine was preprocessed, GREA, combining GRA with EWM, was used to remove less influential factors and PCA method was adopted to further simplify factors. Then a coal and gas outburst risk prediction model was constructed, and adaptive step size update strategy and population fitness variance strategy were introduced based on fruit fly optimization algorithm (FOA) to design IFOA which was further utilized to optimize ELM input layer weights and hidden layer thresholds and train and predict preprocessed sample data as well as compare it with that of other models. The results show that prediction results of IFOA-ELM model based on pretreatment completely match with actual results, and its prediction effect is significantly better than that of the unpreprocessed one. And classification accuracy and recall rate of preprocessed IFOA-ELM model are both 100%, obviously higher than other comparison models.

Key words: coal and gas outburst prediction, grey relational entropy analysis (GREA), principal component analysis (PCA), extreme learning machine (ELM), improved fruit fly optimization algorithm (IFOA)