Research On Bioaugmented Sulfate Reduction Mechanism For Wastewater Treatment And Back Propagation Neural Network Model Simulation

With the development of modern industry, a great amount of acid wastewater drainage contained sulfate produced from some fields, like chemical industrial, pharmacy, metal processing and mining, brings serious hazard to water, soil and atmosphere environments. The treatment of the wastewater contained sulfate has been the subject of extensive attention and recognition.Based on the antecedent works, bioaugment function and mechanism the by zero valent iron (ZVI, Fe~0) in sulfate reducing process were analyzed and discussed in this paper. In addition, the SRB+ Fe~0 system was applied to the treatment of actual wastewater. On this basis, BP neural network models of batch and continue operational environments for SRB and SRB+Fe~0 systems were built, whose performances were evaluated by comparing the model predicted output and experiment data.It was found that, with the present of ZVI, the sulfate reducing system was intensified on the endurance of low temperature and low pH, as well as weakening the dependence on the substance, which make the biotechnology more feasible for acid wastewater treatment. In these conditions, the sulfate reduced rate was intensified by 82.6%, as well as 75% of substance was needed under the same treatment capability.SRB+ Fe~0 system could remove 95% sulfate in 50 hours at 15oC, while almost no sulfate was removal in the same condition. The SRB+ Fe~0 system was also applied to treat the copper mine wastewater. After the treatment, pH value rose from 2.75 to 6.20, and 61% of sulfate from influent 20800 mg·L-1 wastewater was reduced.The bioaugment mechanism of ZVI could be concluded that, the oxidation of ZVI in anaerobic acid environment which causes the transform from H+ to H2 supports energy for the metabolism of HSRB. In this case, HSRB could recapture the domination on reducing the sulphate under growth-limiting condition. BP neural network models of batch and continue operational environments for SRB and SRB+Fe~0 metal treatment systems were built utilizing Matlab software. The train errors of different hidden nodes were evaluated and analyzed in the aspects of convergence, number of training and convergence means to build the optimal neural network models. By comparing the model predicted output and experiment data, we found that, the error scope of simulate models of SRB and SRB+Fe~0 system in the effect of environment factors and the mixed metallic ions are below 10%, which indicates the two models are basically accurate. However, the error scope of simulate models reaches 15%. It was also found that, the predicted error of unknown sample was below 10% by comparing experiment data and model predicted output of unknown samples.