A Feasible Data-Driven Mining System to Optimize Wastewater Treatment Process Design and Operation

Performance assessment of data driven water models using water quality parameters of Wangchu river, Bhutan

Abstract

Multifarious anthropogenic activities triggered by rapid urbanization has led to contamination of water sources at unprecedented rate, with less surveillance, investigation and mitigation. The use of artificial intelligence (AI) in tracking and predicting water quality parameters has surpassed the use of other conventional methods. This study presents the assessment of three main models: adaptive neuro fuzzy inference system (ANFIS), artificial neural network (ANN) and multiple linear regression (MLR) on water quality parameters of Wangchu river located at capital city of Bhutan. The performance and predictive ability of these models are compared and the optimal model for predicting the parameters are recommended based on the coefficient correlation (CC), root mean square error (RMSE) and Nash–Sutcliffe efficiency (NSE) evaluation criteria. Overall NSE and RMSE, the ANN model predicted parameters with maximum efficiency of 97.3 percent and minimum error of 8.57. The efficiency of MLR and ANFIS model are 95.9 percent and 94.1 percent respectively. The overall error generated by MLR and ANFIS are 10.64 and 12.693 respectively. From the analysis made, the ANN is recommended as the most suitable model in predicting the water quality parameters of Wangchu river. From the six-training function of ANN, trainBR (Bayesian Regularization) achieved the CC of 99.8%, NSE of 99.3% and RMSE of 9.822 for next year data prediction. For next location prediction, trainBR achieved CC of 99.2%, NSE of 98.4% and RMSE of 6.485, which is the higher correlation and maximum efficiency with less error compared to rest of the training functions. The study represents first attempt in assessing water quality using AI technology in Bhutan and the results showed a positive conclusion that the traditional means of experiments to check the quality of river water can be substituted with this reliable and realistic data driven water models.

Article highlights

Optimization of water quality monitoring programs by data mining

Water quality monitoring programs are essential planning and management tools, but they face many challenges in the developing world. The scarcity of financial and human resources and the unavailability of infrastructure often make it impossible to meet the legal requirements of water monitoring. Many approaches to optimizing water quality monitoring programs have already been proposed. However, few investigations have developed and tested data mining for this purpose. This article has developed data-based models to reduce the number of water quality parameters of monitoring programs using data mining. The objective was to extract patterns from the database, expressed by association rules, which together with field parameters, measured with automatic probes, can estimate laboratory variables. This approach was applied in 35 monitoring stations along 27 river basins throughout Brazil. The data are from fifty years of monitoring (1971-2021), constituting 6328 observations of 60 water quality parameters investigated in different environmental contexts, water quality, and the structuring of monitoring programs. With the applied approach it was possible to estimate 56% of the laboratory parameters in the monitoring stations investigated. The influence of environmental characteristics on the optimization capacity of monitoring programs was evident. The methodology used was not influenced by different water quality levels and anthropogenic impacts. However, the number of parameters was the most influential element in optimization. Monitoring programs with 20 or more water quality variables have the highest potential (≥44%) of optimization by this methodology. Results demonstrate that this approach is a promising alternative that can reduce the frequency of analyses measured in the laboratory and increase the spatial and temporal coverage of water quality monitoring networks.

Integrated data-driven strategy to optimize the processes configuration for full-scale wastewater treatment plant predesign

Wastewater treatment plants (WWTPs) play an irreplaceable role in eliminating pollutants from domestic and industrial wastewater and contribute to water recycling. Nowadays, the selection of processes configuration of WWTPs mainly depends on the local wastewater treatment standards and the experience of wastewater engineers rather than an intelligent data-driven strategy. In this study, an integrated data-driven strategy consisting of t-distributed stochastic neighbor embedding (t-SNE) and deep neural networks (DNNs) is proposed for optimizing the processes configuration of full-scale WWTP predesign. A large dataset with 14,647 samples collected from 10 full-scale WWTPs with distinct treatment processes is clustered by the t-SNE method based on the influent characteristics, and four meaningful clusters (Clusters I-IV) are identified for the subsequent development of DNN classification models. All four DNN models achieve acceptable classification accuracy (>0.8975) and the maximal testing accuracy is 0.9505. The DNN models are capable of finding the optimized processes configuration of WWTPs under target scenarios. Our results highlight the strength of combining the t-SNE and the DNN models to utilize the relationships between key parameters and processes configuration of WWTPs, and help engineers predesign WWTPs with the optimal processes configuration.

Advanced Supervisory Control System Implemented at Full-Scale WWTP—A Case Study of Optimization and Energy Balance Improvement

In modern and cost-effective Wastewater Treatment Plants (WWTPs), processes such as aeration, chemical feeds and sludge pumping are usually controlled by an operating system integrated with online sensors. The proper verification of these data-driven measurements and the control of different unit operations at the same time has a strong influence on better understanding and accurately optimizing the biochemical processes at WWTP—especially energy-intensive biological parts (e.g., the nitrification zone/aeration system and denitrification zone/internal recirculation). In this study, by integrating a new powerful PreviSys with data driven from the Supervisory Control and Data Acquisition (SCADA) software and advanced algorithms such as Model Predictive Control (MPC) by using the WEST computer platform, it was possible to conduct different operation strategies for optimizing and improving the energy balance at a full-scale “Klimzowiec” WWTP located in Chorzow (Southern Poland). Moreover, the novel concept of double-checking online data-driven measurements (from installed DO, NO3, NH4 sensors, etc.) by mathematical modelling and computer simulation predictions was applied in order to check the data uncertainty and develop a support operator system (SOS)—an additional tool for the widely-used in-operation and control of modern and cost-effective WWTPs. The results showed that by using sophisticated PreviSys technology, a better understanding and accurate optimization of biochemical processes, as well as more sustainable WWTP operation, can be achieved.