MBR membrane fouling diagnosis based on improved residual neural network

High nonlinearity and dispersion in response to the numerous influencing elements of membrane pollution, lead to challenges in diagnosing and other issues. To increase the accuracy of membrane fouling diagnosis, we suggest a method in this research that uses a residual neural network with an attention mechanism. First, the stacking properties of residual blocks are employed to extract the fault information step by step while avoiding the gradient dispersion problem once the fault data has been extracted by the convolutional neural network. Secondly, at each bottleneck in the residual block, the convolutional and coordinated attention mechanism combination is introduced to extract features from the multi-dimensional refinement and boost the diagnostic precision. Finally, the research object for the experimental examination of fault identification is listed as the membrane fouling data. The results of the experiments demonstrate that the proposed diagnostic method can extract useful features in a wide data range with an average accuracy of 99.42% in model accuracy comparison experiments and 96.67 similar to 97.96% in variable noise experiments, which are higher than other methods, and has the ability to reduce power consumption and maintenance costs, providing a theoretical research basis for practical production.