Predicting the effluent total coliform count of UV disinfection for wastewater treatment using artificial neural networks

• Main Authors: Liu, Marc Kenzo Wong, Lu, Michael Chin Lumampao, Ngo, Jennings Jervis Te Mew, Que, Jaena Mae Ong
• Format: text
• Language: English
• Published: Animo Repository 2021
• Subjects: Sewage—Purification—Ultraviolet treatment; Enterobacteriaceae; Neural networks (Computer science); Chemical Engineering
• Online Access: https://animorepository.dlsu.edu.ph/etdb_chemeng/15; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1011&context=etdb_chemeng
• Institution: De La Salle University

The use of UV transmittance as the wastewater disinfection method has been increasing among local water utilities. Due to technological advancements in computing, artificial neural networks (ANN) have been proven to be a more appropriate modeling tool than process-based modeling. In this study, two WWTPs were examined as individual case studies. The first case study employed a conventional activated sludge (CAS) system, while the second case study employed a moving bed biofilm reactor system (MBBR) for secondary treatment. Both employ UV transmittance as its disinfection system. For each case study, an ANN model was trained and optimized by varying its network properties, particularly the network architecture and transfer functions in the hidden layers. A total of 80 and 146 data points were used in obtaining the optimal network parameters for CAS and MBBR systems, respectively. The resulting optimal network for both cases employed the hyperbolic tangent transfer function with a network architecture of three hidden layers with ten neurons per hidden layer (5-10-10-10-1). The performance of these networks was evaluated using the correlation coefficient (R) which resulted in significantly high values of 0.9955 and 0.9862 for the CAS and MBBR system, respectively. Mathematical relationships established between the network layers were used to formulate a general equation as a function of the input parameters, weights, and bias weights of each layer. To assess the generalization capability of the optimal networks, these were utilized to predict the effluent total coliform count of a new set of actual wastewater data obtained from the same WWTP for each system. However, this resulted in MAE values of 472 and 261 MPN/100mL for CAS and MBBR, respectively. These errors were attributed to several factors such as the presence of outliers in the simulation data, overfitting, the method of data collection, and the inclusion of repetitive and irrelevant input parameters.