Analysis of dominant factors initiating microbial community ecological succession based on BPNN model

Adopting acidogenic sulfate-reducing reactor for treating high concentration sulfate waste water, based on the prototype experiment utilizing a algorithm of gradient descent with momentum and adaptive learning rate backpropagation a back-propagation neural network (BPNN) model was established to predict the influence of four key ecological factors of COD/SO₄^2- ratio (C/S), sulfate loading rate (Ns), pH value and alkalinity (ALK) on sulfate removal rate (η). Based on this, three methods of information flow, partitioning connection weights (PCW) and partial derivatives (PaD) were adopted to analyze quantitatively the connection weights between neurons in the different layers of network; thus, that the dominant factors initiating the entire course (stage I + stage II + stage III) of the microbial community ecological succession was C/S. while the roles of different ecological factors in these three stages were different. pH value was the dominant factors of stage I; C/S was that of stage II, and Ns was that of the stage III.