Abstract: This paper addresses the issue of insufficient selectivity in electrochemical gas sensors within atmospheric miniature monitoring stations, which are susceptible to cross-interference from non-target gases. A data correction approach using a Genetic Algorithm (GA) optimized Back Propagation (BP) neural network (GA-BP model) is proposed. The GA-BP model enhances the BP neural network, adept at tackling nonlinear black-box problems, by integrating GA's global optimization capability. This integration optimizes the neural network's initial parameters, overcoming the BP network's tendency to fall into local minima, and thus enhances the model's overall accuracy and stability. Experimental results demonstrate that the miniature monitoring stations equipped with the GA-BP model can accurately quantify NO₂, CO, O₃, and SO₂ concentrations in mixed gases. The goodness of fit (R²) between the computed values and actual measurements of these four gases exceeded 0.95. Compared to univariate, multivariate linear regressions, and traditional BP neural networks, the GA-BP model shows clear superiority. Moreover, the model exhibits strong generalization capabilities. When applied to new, untrained station data, it achieved R² values above 0.88 for all four gas concentrations, affirming the method's robust applicability. This approach provides a beneficial reference for manufacturers to enhance equipment performance and for users to obtain accurate air pollutant concentration data.