Neural network-based multi-back-propagation prediction model of a domestic wastewater treatment plant for an under-construction sewer system

Abstract

When the sewer collection of a serviced area has not reached the full design capacity of the domestic wastewater treatment plant, mathematical models have the potential to provide useful information for operating the plant to meet discharge standards and managing the received water system. In this article, a back-propagation neural network (BPNN) was applied for predicting wastewater quantity and quality. Three basic models are included in this network, i.e., A1(PIQQ) for predicting influent quantity and quality, A2(PEQQ) for predicting effluent quantity and quality, and A3(PQWCWS) for predicting the quantity and water content of waste sludge. The multi-model (A1 + A2) system that combines A1 and A2 into a noted multi-BPNN (MBPNN) is used for estimating A2 output parameters directly based on A1 input parameters. The correlation coefficient values (R) are higher than 0.95 for A1, whereas the mean absolute percentage errors are less than 35% for A2 and A3, and 46% for A1 + A2. These results indicate that BPNN and MBPNN are suitable for predicting the wastewater quantity and quality especially for Q, BOD₅, sludge quantity, and water content in an under-construction sewer system.

Keywords:

• multi-back-propagation neural network
• domestic wastewater treatment plant
• mean-square error