Author : Praewa Wongburi
Publisher :
Page : 0 pages
File Size : 28,38 MB
Release : 2021
Category :
ISBN :
In a wastewater treatment plant (WWTP), big data is collected from sensors installed in various unit processes, but limited data is used for operation and regulatory permit requirements. With the advancement in information technology, the data size in wastewater treatment systems has increased significantly. However, WWTPs have not used big data systematically to aid the operation and detect potential operational issues due to the lack of specialized analytical tools.The objectives of the study were to: (1) develop analytics methods suitable for the management of big data generated in WWTPs, (2) interpret analytics results for extracting meaningful information, (3) implement a recurrent neural network (RNN) and Long Short-Term Memory (LSTM) to predict effluent water quality parameters and Sludge Volume Index (SVI), (4) apply an Explainable Artificial Intelligence (AI) algorithm to determine causes of predicted values, and (5) propose a real-time control using a predictive model to monitor and optimize the operation of WWTPs. The predictive AI models in WWTPs were developed by applying big data analytics, statistical analysis, and RNN algorithms with an Explainable AI algorithm. The models successfully and accurately predicted the effluent water quality data and a key operational parameter, SVI. Furthermore, the Explainable AI algorithm provided insight into which influent parameters affected higher predicted effluent concentrations and SVI on a specific day, allowing operators to take corrective actions. From a WWTP's operational data analysis, the RNN model successfully predicted the effluent concentrations of BOD℗Ơ5, total nitrogen (TN) and total phosphorus (TP), and SVI. Furthermore, the Explainable AI analysis found that higher influent NH3N values lead to higher effluent BOD5, and higher influent total suspended solids (TSS) and TP values resulted in lower effluent BOD5, implying the importance of controlling dissolved oxygen (DO) in aeration basins. Since aeration is one of the major energy consumption sources in WWTPs, real-time prediction of the effluent water quality using the self-learning AI system developed in this study can be adopted to lower the energy cost significantly while improving effluent water quality. WWTPs must develop control methods based on the RNN prediction and Explainable AI analysis due to different operational conditions.