Abstract: To address the issue of sensor cross-sensitivity in multi-gas detection for coal storage bunkers, as well as the slow convergence and susceptibility of traditional backpropagation (BP) neural networks to local optima, a multi-gas concentration detection method based on a bird swarm algorithm (BSA)-optimized BP neural network is proposed. A detection system consisting of a sensor array, data acquisition module, data processing unit, and result display module was studied and constructed. The BSA was employed to iteratively optimize the initial weights and biases of the BP neural network. The mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R²), and mean relative error (MRE) were selected as evaluation metrics. BP, particle swarm optimization-based BP (PSO-BP), and genetic algorithm-based BP (GA-BP) models were constructed for comparison. A total of 8,640 gas samples collected from coal storage bunkers were used for model training and performance validation.The results demonstrate that the BSA-BP model outperforms all comparative models across all evaluation metrics. For the prediction of five gases, including CO, CH₄, H₂S, SO₂, and NO₂, the BSA-BP model achieved an MAE of 0.023×10⁻⁶ and an R² value of 0.974. Compared with the traditional BP neural network, the MAE decreased by 73.3%, while the R² value improved by 0.102. The BSA exhibited faster convergence than the PSO and GA algorithms and effectively maintained population diversity to avoid premature convergence. The proposed model showed excellent generalization capability and robustness against interference. It achieved higher prediction accuracy for CH₄ and CO, whereas relatively larger errors occurred in the prediction of low-concentration NO₂ and SO₂. The BSA can efficiently optimize the initial parameters of the BP neural network. The constructed BSA-BP detection method can effectively weaken the cross-sensitivity interference of sensors, and is suitable for accurate multi-gas monitoring under complex working conditions in coal storage silos, with good engineering application value.