Biological Treatment of Sanitary Wastewater in a Sequencing Batch Reactor (SBR) by Microalgae Chlorella vulgaris: Effect of Operational Parameters and Artificial Neural Network Modeling

In this work, by constructing a sequencing batch reactor (SBR) and the use of microorganisms and microalgae Chlorella vulgaris, the performance of the system is studied for treating sanitary wastewater of Yazd power plant, Iran. For this purpose, the effect of pH, temperature, influent chemical oxygen demand (COD) concentration and airflow rate is examined on removing of COD and its residual concentration in wastewater. Another aspect of this research is the development of a multilayer feed-forward neural network model to predict the concentration of residual COD during the process of treatment.
The pilot SBR reactor consisted of a wastewater storage tank, an aerobic tank and a settling tank and the used wastewater in this research was sanitary wastewater of Yazd power plant. Each cycle of the reactor was 8 hours including 5 min of filling, 450 min of aeration, 20 min of settling and 5 min of discharge. In order to supply the required microorganisms for the reactor setup, active sludge was prepared from the return line of the sludge of the Yazd power plant wastewater treatment plant. Also due to the quality of the microalgae of Chlorella vulgaris and its accessibility, this microalgae was prepared from the Yazd sewage stabilization pond and both of the microorganisms and microalgae were transferred to the reactor. At the beginning of the operation, the influent COD entered the system with 300 mg/l concentration (minimum organic load). 20 days after the initial setting up of the reactor, the removal efficiency reached its maximum, microorganisms, and microalgae adapted to the existing conditions. Then, at the cycle time of 8 hr the concentrations of residual COD were obtained at different pH, temperature, influent COD concentration and airflow rate. Sampling of the system was done at intervals of one hour for testing.
In this study, a multilayer perceptron artificial neural network (MLP-ANN) was developed via employing Levenberg–Marquardt training algorithm in order to predict the concentration of residual COD. The ANN consisted of three layers, that is, only one hidden layer was used between the input and output layers. Input layer consisted of five neurons, which included pH, time, temperature, airflow rate, and influent COD concentration, and output layer had one neuron, which was residual COD concentration and the optimum number of hidden neurons was obtained by guessing and error. In order to increase the convergence and accuracy of the neural network, the input and output data were normalized and scaled to the range of 0–1. The performance of the ANN model was measured by root mean square error (RMSE) and correlation coefficient (R2) between the predicted values of the network and the experimental values.
One of the important parameters that can have a significant effect on the performance of wastewater treatment is the amount of organic load in the influent wastewater. Increasing of the influent COD concentration, caused to increase the residual COD concentration in the wastewater. It was concluded that according to the destructive effect of increasing the influent COD concentration, control of the influent COD concentration in a constant range is necessary. This will cause the necessary substrate for the growth of microorganisms and microalgae to be provided consistently and sufficiently. The maximum influent COD concentration of the power plant was 1100 mg/l and at least 300 mg/l. Therefore, by establishing a return flow from the effluent wastewater of power plant to the primary wastewater storage tank and adjusting it to the influent COD concentration, the amount of fluctuations in the influent wastewater load was reduced and the influent COD concentration was fixed at 600 mg/l.
One of the effective parameters on the performance of the wastewater treatment system is the pH of influent wastewater. After one hr of testing, the residual concentration of COD was approximately 53 mg/l. After this time, at first, the COD concentration increased at low pH, and even at pH of 4, it reached to 120 mg/l. The reason was that at the beginning of the treatment by microorganisms and microalgae, the digestion of wastewater organic materials was done. The result was the appearance of organic acids that caused the further decrease of pH, thereby reducing the activity of microorganisms and microalgae and increasing of COD concentration. Then, by decomposition of the produced acids and the beginning of the decomposition of proteins and fats, the pH of the system, the activity of microorganisms and microalgae increased and the COD concentration decreased and fixed at about 60 mg/l. At high pH, because of the high activity of microorganisms and microalgae, observed that the residual COD reduction process performed with a suitable gradient during the time and eventually remained constant. As shown in Fig. 1b, the most suitable pH for the activity of microorganisms and microalgae is pH of 8, in which case the residual COD concentration has reached the lowest value of 34 mg/l.
Temperature is one of the important effective parameters on activity of microorganisms. The best temperature for system operation is between 30-35 °C. At high temperatures, because of the reduced solubility of air in the wastewater, enough oxygen and CO2 were not provided for microorganisms and microalgae, which caused to increase the residual COD concentration, compared to the other temperatures. On the other hand, at low temperatures due to the reduced growth of microorganisms and microalgae, increase of residual COD concentration occurred. According to available data from the power plant, the temperature of influent wastewater was almost 30 oC in most seasons. On the other hand, according to Fig. 1c, there is no significant difference between the concentration of residual COD at 30 oC and 35 oC, so the optimum temperature of this process was chosen 30 oC.
In aerobic treatment systems, desirable aeration should be done to create suitable conditions for the growth of microorganisms. Increasing of the aeration generally had a positive effect on the performance of treatment process and with increasing aeration, residual COD concentration has decreased. However, this effect is not noticeable for airflow rate of more than 50 l/min. More aeration will increase the costs, sometimes lead to cell failure, and thus decrease the growth of microorganism and microalgae. Therefore, in this study, the optimum airflow rate of 50 l/min was selected. In this airflow rate, the lowest residual COD concentration was obtained.
In this study, 208 laboratory data was used for modeling. Two thirds of the data were randomly selected for training the network and one third remained for evaluation of modeling accuracy. To determine the optimum number of neurons in hidden layer, 1 to 15 neurons were used. The transfer functions of hidden and output layers were selected tangent sigmoid (tansig) and linear (purelin) respectively. The results indicated that the network with the number of neurons equal to seven had the best performances because the root mean square error (RMSE) had the lowest value and the correlation coefficient (R2) had the closest value to one.
In this study, at first, a pilot SBR reactor was installed and operated by microorganisms and microalgae Chlorella vulgaris. In the next stage, the system was evaluated with variables such as pH, temperature, influent COD concentration and airflow rate. By analyzing the values obtained from the reactor, optimum values of these parameters were determined to achieve the lowest residual COD concentration. The results showed that at a pH of 8, temperature of 30 °C, influent COD concentration of 600 mg/l and airflow rate of 50 l/min, concentration of residual COD was obtained 34 mg/l, which indicates an increase in efficiency of the system. In addition, an artificial neural network model was developed to predict the concentration of residual COD. ANN predicted results were in good agreement with the experimental data with a correlation coefficient and root mean square error of 0.944 and 0.034 respectively.