Machine learning coupling microbial succession to decipher gas stripping-reinforced sulfate reduction for hydrogen sulfide removal.

Machine learning-assisted gas stripping optimization is showing great potential of removing hydrogen sulfide (H2S) in the acidogenesis phase to enhance sulfate reduction. Herein, response surface methodology (RSM) and artificial neural network (ANN) were employed unprecedentedly to model the sulfate reduction rate after removal of H2S. Under the COD/SO42- ratio of 8.84, reaction temperature of 33.81℃, and gas stripping flow of 0.02 (L gas/min)/L liquid modeled by RSM, a maximum sulfate reduction rate of 92.30 % with 6.38 % error was obtained. Correspondingly, the results of genetic algorithm-based ANN were 8.98, 30.07℃, 0.01 (L gas/min)/L liquid separately, with the maximum sulfate reduction rate of 91.93 % and error of 2.19 %. The low COD/SO42- ratio and gas stripping flow improved the potential competitiveness of Syntrophobacter over Methanosaeta and Methanobacterium. Increasing gas stripping amplificated the gene expression for assimilatory sulfate reduction. This work highlights the importance of optimizing gas stripping to improve sulfate reduction and recycle.