ชัยวัฒน์ แววศักดิ์. Controlling Anaerobic Hybrid Reactors Using Neural Network and Fuzzy Logic Control (NNFC). Doctoral Degree(Biotechnology). King Mongkut's University of Technology Thonburi. KMUTT Library.. : King Mongkut's University of Technology Thonburi, 2009.
Controlling Anaerobic Hybrid Reactors Using Neural Network and Fuzzy Logic Control (NNFC)
Abstract:
This research purposes to apply the neural network and fuzzy logic in controlling the anaerobic hybrid reactor. This work was divided into three parts. In the first part, the
four-layer feedforward neural network (1 input layer, 2 hidden layer and 1 output layer) based on the backpropagation algorithm was developed in which to predict important
variables of pH, Alkalinity (Alk) and total volatile acids (TVA) in lab-scale anaerobic hybrid reactor. The neural network model was trained and validated by the data from
operating the lab-scale anaerobic hybrid reactor to treat industrial tapioca starch wastewater. The model structure 6-25-20-1 gave the best results from the pH predicted
neural network model in the training model. It had been shown a mean absolute percentage error (MAPE), the root mean square errors (RMSE) and correlation
coefficient (R2) from the prediction of pH were 0.3709%, 0.0417 and 0.9128, respectively. While in the validating model, the MAPE and RMSE were 1.5620% and 0.1369, respectively. The 9-30-20-1 was the model structure of Alk predicted neural network model that presented the satisfied results in the training model with the MAPE,
RMSE and R2 of Alk prediction of 5.1804%, 143.88 mg/l as CaC03 and 0.8193, respectively. In addition, the MAPE and RMSE from validating the Alk model were
11.7774% and 304.7962 mg/l as CaC03, respectively. The model structure 9-35-25-1 of TVA predicted neural network model illustrated that the best results of MAPE, RMSE
and R2 in training model were 4.1257%, 36.8089 mg/l as acetic acid and 0.9198, respectively. The results in validating this model showed the value of MAPE and RMSE were 22.9957% and 149.9983 mg/l as acetic acid, respectively. It indicated that the four-layer neural network model developed is a good model in prediction of pH, Alk and TVA.
Secondly, the fuzzy logic control system had been developed for controlling the influent feed flowrate in the anaerobic hybrid reactor. The process variables were pH, Alk and
TVA and these variables were selected in order to use them as the input of the fuzzy sets and the influent feed flowrate was used as the output of the fuzzy set. This study
selects the type of membership function as namely the generalized bell (gbellmf) which was the membership function of pH, Alk, TVA and influent feed flowrate. The 125
rules of if-then" rules were chosen to use in this fuzzy logic control systenl. During the experimental controlling, the synthesis wastewater was fed continuously with the
concentrations of influent were in range 3-12 g/l. The anaerobic hybrid reactor was fed in range 1.91-3.811/d giving an organic loading rate (OLR) varied in range of 0.53-4.24
gCOD/l.d and the hydraulic retention time (HRT) were in range 2.83-5.64 d. The performance of anaerobic hybrid reactor controlling by the fuzzy logic control system
was evaluated by the change of pH, Alk, TVA, biogas production rate, biogas composition and COD removal efficiency in the reactor during the operating time. From the experiment, pH value was 6.68-7.09 represented the suitable range for methanogens. The Alk were 1386-2764 mg/l as CaC03 showed good buffer capacity of the process,
less TVA accumulation were varied in range 328-1052 mg/l as acetic acid and ratio of TVA to Alk was less than 0.4. Biogas was produced 2.54-16.20 lid during operating the
reactor with CH4 and C02 content were 57-70% and 21-38%, respectively. These results illustrated that the trend of CH4 content generated from anaerobic hybrid reactor
continued constant during the experiment. The majority ofCOD removal efficiency was more than 80% during all of the experiment. The fuzzy controller from this study was
shown to be very good. It was found to be suitable for the controlling these variables of anaerobic hybrid reactor. The fuzzy logic control system can control the influent feed
flowrate and maintain the stability of the anaerobic hybrid reactor.
The third part, the neural-fuzzy control system was designed under the concept of the combination of the neural network model and fuzzy logic control system. The objectives of this study was to develop the neural-fuzzy control system by applying is neural network model for predicting the variables pH, Alk and TVA and fuzzy logic control system to use these predicted variables for calculating the influent feed flowrate of the anaerobic hybrid reactor. The neural network models structure was used in this study were
6-25-20-1 for pH prediction, 9-30-20-1 for Alk prediction and 9-35-25-1 for TVA prediction, respectively. The same fuzzy logic control system has been developed for controlling the influent feed flowrate in the anaerobic hybrid reactor. The pH, Alk and TVA were selected and used them as the input of the fuzzy sets and the influent
feed flowrate was used as the output of the fuzzy set. The 125 rules of if-then" rules were same used and generalized bell (gbellmf) membership function had been chosen
for the purpose of using as the membership function of pH, Alk, TVA and influent feed flowrate. The efficiency of the neural-fuzzy control system was analyzed from the
performance study of anaerobic hybrid reactor that was controlled by this controller. During the experimental controlling, the synthesis wastewater was fed continuously
with the concentrations of influent were in range 3-12 gil. The reactor was fed in range of 1.79-6.75 lid that were giving an OLR of the anaerobic hybrid reactor in range 0.50-5.64 gCOD/l.d and the HRT in range of 1.60-6.02 d. The monitoring factors such as pH, Alk, TVA, biogas production rate, biogas composition and COD removal efficiency
were measured to control the performance and stability of the anaerobic system. The experiment results indicated that the pH values insignificantly changing that were varied
in range of 6.68-6.94 during the operating time. Less TVA accumulation showed in range of 320 to 1277 nlg/1 as acetic acid and Alk varied in range 1532-2502 mg/l as
CaC03, respectively. The TVA to Alk ratio inside the reactor was maintained at the value of less than 0.4 indicated the reactor still maintains the buffer capacity in
sufficient value to maintain the process stability. The biogas production rate increase following the increasing of the OLR, the values were in range of 1.84-19.11 lid during
the experiment time. CH4 and CO2 content quite constant, the values were varied in the range of 62-67% and 22-32%, respectively. The majority of COD removal efficiency
was exceeding more than 84% during the experiment showed that the high treatment perfoffilance of the reactor. The reactor still succeeds in keeping the stability under the
controlling by neural-fuzzy control system.