Synergistic effect evaluation method of atmospheric emission reduction based on deep learning fusion model

Currently, the absence of a nonlinear response between air quality and industrial emissions, which accurately captures the complex relationship between pollution levels and emission sources, poses a significant challenge in the formulation of effective control policies. For the purpose of effectively managing industrial emissions and mitigating severe pollution peaks, a new method for evaluating the synergistic effects of atmospheric emission reduction was proposed. The objective of this method is to simulate the impact of emissions on air quality. In particular, we have developed a deep learning fusion model, GR-BILSTM, which integrates a generative adversarial network for data enhancement and a ResNet-BILSTM model to effectively address the issue of gradient disappearance in deep networks and capture high-dimensional data features, thereby improving the model's prediction accuracy. This model is used to study the relationship between emissions and air pollution. Subsequently, a perturbation analysis method is employed to provide a quantitative assessment of the impact of varying types and scales of industrial park emissions on PM_2.5 concentrations. By modifying the emission reduction ratio, the PM_2.5 concentration in the subsequent moment is predicted, and the correlation between the predicted value and the measured value is examined. This provides a vital point of reference for the formulation of targeted emission policies in industrial parks. In comparison to the LSTM, BILSTM, and CNN-BILSTM models that are commonly employed in this field, the GR-BILSTM model has been demonstrated to exhibit superior performance in terms of fitting accuracy. Our research indicates that the greatest contributions of SO₂, NO_x, and TSP to PM_2.5 pollution in industrial parks are 16 %, 15 %, and 18 %, respectively.