Abstract: There were some problems of gas emission prediction method for the existing coal mining working face, for example the long prediction time and the low prediction accuracy. According to the above, the gas emission prediction model proposed for the optimized ELM neural network with gravity algorithm. The gravitational search method was improved by employing the optimization strategy and the memory and information exchange function of the particle. IGSA was used to search and optimize network hidden layer node number of ELM neural network. The autocorrelation coefficient method was adopted to screen out eight main factors which affected gas emission quantity. The gas emission prediction model was established based on the IGSA-ELM algorithm, and it was analyzed combined with the historical data from mine monitoring. Experimental results showed that the gas emission prediction model of ELM neural network optimized by IGSA, its prediction accuracy increased by 310%, 60% and 31% respectively, compared with the PSO-ELM neural network, the ACC-ENN and the GSA-ELM neural network prediction model, which provided a new fast method for the gas emission prediction of working face.