Diagnostic Study of the Effect of Nitrogen and Oxygen Content on Membrane Fouling Based on MSCA-Fused-MBConv

Membrane fouling has become the main factor hindering the development of membrane biological wastewater treatment technology. The factors have strong coupling, nonlinearity and time lag, and the content of nitrogen and oxygen in the pool is an important factor directly affecting the development of membrane fouling. Therefore, this paper proposed to use the Multi-scale Spatial and Channel-wise Attention (MSCA) to improve the diagnostic network of the Fused-Mobile inverted Bottleneck Convolution (Fused-MBConv), followed by the stable operation of MBR. Firstly, this paper employed the Principal Component Analysis (PCA) method to identify ammonia nitrogen indices with strong correlations that directly impacted membrane fouling. Secondly, the filtered time-series data were transformed using the Markov Transition Field (MTF) method, which preserved the integration of data features and temporal characteristics. Thirdly, the MSCA attention mechanism was utilized to optimize the structure of Fused-MBConv in EfficientNetV2. Finally, through ablation experiments and comparisons with other related networks, it was concluded that the proposed network achieved an average accuracy of 0.9953 and maintained an average running time of 2.13 seconds. Additionally, 5-fold cross-validation experiments confirmed that the MSCA-Fused-MBConv diagnostic network demonstrated strong robustness and generalization capabilities.