Abstract: To investigate the problem that existing gas emission prediction methods generally do not quantitatively analyze the influence of the data itself, an improved quantum genetic algorithm (IQGA) optimized extreme learning machine (ELM) gas emission prediction model is developed. The conceptual-drifting of absolute gas emission is detected by iForest algorithm, and the outliers generated by the conceptual-drifting of absolute gas emission are corrected by combining convolutional neural network (CNN), bidirectional long-term short-time memory network (BiLSTM), and attention mechanism. The correlation analysis (PCC) dimensionality reduction method was employed to process the input variables and determine the auxiliary variables of the prediction model. Finally, the quantum genetic algorithm (QGA) is improved by dynamically adjusting the quantum rotation angle, quantum crossing, quantum variation and quantum catastrophe operation. Optimization and generalization ability of the algorithm were therefore enhanced. IQGA was used to optimize ELM parameters. Taking the decision coefficient(R²), mean absolute error(MAE), root mean square error(RMSE), and mean absolute percentage error(MAPE) as the indexes to carry out the evaluation, The results show that the IQGA-ELM model had the smallest measurement errors of 0. 985, 0.018, 0.026 and 2.56%, respectively. The improved gas emission prediction model exhibit higher prediction accuracy than other models.