| The simulation and prediction of anaerobic bioreactors shocked by toxic load was made with neural network, parallel genetic algorithms (PGA) and fuzzy inference system, and their states were estimated based on the fuzzy set theory. The following achievements were obtained.1. The performance of anaerobic bioreactor subjected to various toxic shocks was extensively studied at mesophilic temperature.(l)When the unacclimated UASB reactor (UASB-R I ), acclimated UASB reactor (UASB-R II) and anerobic filter (AF) were shockcdy by the combined load, the largest increase of the effluent COD under combined load shocking was higher than that of the sum of chloroform (CF) (or 2,4-dinitrophenol, noted DNP) and COD load shocking independently. The largest decrease of the effluent alkalinity (ALK) was less than that of the sum of the two loads shocking independently. The volumetic biogas production rate (VGP) increased firstly, and then decreased under combined load shocking generally. The largest decrease of percent of CH) (%CH4) under COD+CF shocking was higher than that of the sum of CF and COD load shocking independently, but under COD+DNP shocking its most decrease was similar with that of the sum of COD and DNP load shocking independently.(2)The COD-t-CF+DNP+ALK shocking tests were carried out according to the orthogonal design. Under multi-load shocking, the sequences of factors influencing the effluent COD, DNP, VGP, and %CH4 of UASB-R I were CF>COD>DNP>ALK, DNP>CF>COD>ALK, CF>DNP> COD>ALK and CF>DNP>COD>ALK respectively. For the UASB-R II, their sequences were COD>DNP>CF>ALK, DNP>CF>COD>ALK, CF>COD>DNP>ALK and CF>DNP>COD>ALK respectively. The important factor influencing the effluent VFA of UASB-R I was CF, but for the UASB-R II, it is COD. The important factors influencing the effluent VGP of UASB-R I were CF and DNP, but for the UASB-R II, they were CF and COD. The important factors influencing the effluent VGP of UASB-R I were the same as that of UASB-R II.2. The simulation and prediction of three bioreactors shocked by the toxic load was made based on the time serial BP Neural Network (noted TBPNN) and on the different performance parameters BP Neural Network (noted MBPNN). At the same time, the performance of BP neural networks applying different bioreactor was compared.(l)Optimized TBPNN architecture with a hidden layer of 4-9 nodes provided the best performance. When TBPNN-R I was used to simulate the COD, VFA, ALK, DNP, VGP, %CO2 and %CFLi of UASB-R I , the correlation coefficients of observed and predicted values were above 0.850, the relative root-mean-square-errors (RRMSE) were below 30%. Under the 3, 4-dichloronitrobenzene (DCNB) load shocking, the correlation coefficients of the performance parameters were above 0.850, and the RRMSE were below 25%. When TBPNN-R II was used to simulate the different parameters of UASB-R II, the correlation coefficients of observed and predicted values were between 0.778 and 0.916, the RRMSE were below 30%. Under the DCNB load shocking, the correlation coefficients of performance parameters were between 0.749 and 0.911, and the RRMSE were below 25%.(2)The delay effect and accumulated effect must be considered if the MBPNN was applied to simulate and predict the performance of bioreactors. If one parameter was simulated only by MBPNN with the other parameters as input of neural network, the performance of MBPNN was not satisfying. The parameter's history information should be looked as part of input of MBPNN, thus the performance of MBPNN was significantly improved. When MBPNN-R I was used to simulate the VFA, DNP, VGP, and %CH| of UASB-R I , their correlation coefficients were between 0.848 and 0.921, their RRMSE were all below 30%. Under DCNB load shocking, their correlation coefficients were all above 0.850, their RRMSE were all below 20%. When MBPNN-R II was used to simulate the VFA, DNP, VGP, and %CR, of UASB-R II, their correlation coefficients were between 0.834 and 0.936, their RRMSE were mostly below 30%. Under DCNB load shocking, their correlation c...
|
PDF Full Text Request