A deep learning based dynamic COD prediction model for urban sewage

Abstract

Due to the comprehensive sources of urban sewage, the contents of pollutants in urban sewage are quite complex and fluctuate frequently. The unstable status of both sewage inflow and COD makes it difficult to precisely control the process parameters in wastewater treatment plants (WWTPs). To reach effluent quality standards, WWTPs increase the aeration rate and chemical inputs, resulting in a great waste of resources and energy, rising production costs, and secondary pollution. Reputed for decreasing unnecessary energy and chemical input, a dynamic COD prediction model of urban sewage based on the hybrid CNN-LSTM deep learning algorithm is proposed to support the further development of feed-forward control systems. The prediction results reveal that the hybrid CNN-LSTM prediction model has higher accuracy and better prediction performance than the stand-alone CNN or LSTM model. The error analysis indicates that the prediction performance can satisfy industrial requirements and can be adopted in urban WWTPs.