Knowledge discovery and data mining to understand and optimise the environmental behavior of wastewater treatment processes

Abstract

Direct nitrous oxide (N2O) emissions during the biological nitrogen removal (BNR) processes can significantly increase the carbon footprint of wastewater treatment plant (WWTP) operations. However, quantifying the emissions and understanding the long-term behaviour of N2O fluxes in WWTPs remains challenging and costly. The aim of the current research is to combine wastewater domain knowledge with data-mining techniques to explain the long-term N2O emissions’ behaviour in full-scale biological reactors. A review of the recent full-scale N2O monitoring campaigns is conducted resulting in the development of an emission factor (EF) database with information on configurations, control strategies and operational conditions. The analysis focused on mechanistic model development, molecular biology methods and on the current data management and analysis practices (i.e. visualization techniques, statistical analysis). Sensor and laboratory data acquired from the N2O monitoring campaigns of mainstream and sidestream wastewater processes were used to develop, test and validate a methodological framework for knowledge discovery in wastewater databases. Abnormal events detection, structural changepoint detection, clustering, classification and regression algorithms are used in order to i) translate data into actionable information, ii) link N2O emissions ranges with specific operational conditions, iii) identify and isolate re-occurring system disturbances that affect performance, iv) predict the range of N2O emissions based on operational and environmental conditions and v) provide feedback to monitoring campaigns for the minimisation of sampling requirements. The analysis showed that the relationship of N2O emissions with the operational variables fluctuates in long-term monitoring campaigns; this should be taken into consideration for the development of mitigation measures and during the investigation of triggering operational conditions. Additionally, findings indicate that structural changepoints of operational variables monitored online can be used to detect changes in the behaviour and range of N2O emissions. Finally, data-driven models can reliably estimate N2O behaviour in wastewater processes under given operational conditions. However, fluctuation of dependencies, system disturbances and process-specific characteristics should be taken into consideration.