Optimisation of wastewater treatment systems with data mining and process modelling

Abstract

Wastewater treatment facilities serve an important purpose in public sanitation. Wastewater treatment plants (WWTP) are built at the end of sewerage systems to purify the wastewater before it enters natural water bodies. Carbon, nitrogen and phosphorus have to be removed from wastewater to avoid oxygen depletion, eutrophication and excretion of toxins. Wastewater treatment has also caused some adverse environmental impacts due to energy consumption, use of chemicals and environmental emissions. In a few modern wastewater treatment plants, resource recovery and water reuse are included as a holistic approach. However, the wastewater treatment process needs to be optimised through advanced control to achieve better performance in terms of economics, effluent quality and environmental impact. The performance of wastewater treatment processes can be affected by disturbances, especially by the variation of influent characteristics. Progress in control laws has been made by researchers and practitioners in wastewater treatment process optimisation, as can be seen from the large amount of publications on control simulation. The lack of reliable and affordable online monitoring equipment and the complexity of the biological treatment process have restricted further implementation of control techniques. This thesis integrates process surveillance, data mining, process modelling and molecular biology to narrow the gap between surveillance and control in practice. Generally, the following works were conducted to optimise the wastewater treatment process: (1) characterisation of influent under the impact of seasonal variation and climate effects; (2) development of a combined approach to achieve advanced control; (3) development of interpretable soft sensors for process surveillance; (4) investigation of interactions between microbial community composition and kinetic modelling. A key step towards optimised wastewater treatment performance is handling the uncertainties of process inputs. In Paper I, a stepwise influent characterisation approach based on data mining methods was developed to characterise influent quality and quantity of a wastewater treatment plant. The seasonal differences of influent quality were compared after eliminating the dilution effect in the cold season. Moreover, the correlation between climate factors and influent characteristics was also investigated. Paper II addresses the core concept of the thesis, where statistical monitoring and process modelling was proposed as a combined approach to achieve model predictive control. The other appended papers focused on either statistical monitoring or process modelling to support this approach. In Paper II, the hard-to-measure variables— chemical oxygen demand (COD) and total phosphorus (TP)— were predicted by statistical models. Furthermore, the predicted values were further used as inputs to the process model. The model outputs of intermediate total suspended solids (TSS) and effluent COD fitted the measured data well, which indicates that the method can be used to control the aeration and chemical dosing of the WWTP. Data mining was tested in a full-scale WWTP for early warning of poor activated sludge settleability. In Paper III, the seasonal variation of activated sludge settleability was investigated by studying the correlation of settleability and process disturbances. Three predictive modelling methods were used to build early warning tools for poor activated sludge settleability. Moreover, the proportion of volatile substances of biomass was found playing a central role in seasonal variation of activated sludge. The storage-biodegradation mechanism explains the reason for poor settleability of activated sludge in the cold season. In Paper IV, the microbial community composition of a lab-scale moving bed biofilm reactor (MBBR) system was investigated by applying high-throughput sequencing. Secondly, the ratios of active heterotrophic biomass and autotrophic biomass in each MBBR chamber were estimated by conducting respiration tests. As a result, the interaction of sequencing results and active biomass ratios led to a novel approach for kinetic model parameter estimation. This approach is useful for the biological process modelling presented in Paper II. Paper V is a continuation of Paper I and Paper II. Soft sensors can be applied to solve surveillance issues in wastewater treatment plants. For influent monitoring, the nonlinearity caused by seasonal variation and climate effect indicates the necessity of nonlinear models for soft sensors. Multivariate Adaptive Regression Splines (MARS) was found as an interpretable nonlinear modelling tool for COD and TP prediction. Wastewater is purified at the cost of energy consumption, chemical usage, environmental emissions and other associated costs. The balance between sufficient treatment and environmentally-friendly performance is always an issue for the control of wastewater treatment processes. In Paper VI, the environmental impacts of wastewater treatment process control strategies were evaluated by conducting Life Cycle Assessments. Significant reductions of climate changing impact and environmental toxicity were achieved by enhancing primary removal of particles and applying model predictive control. The study underlines that environmental impacts should be considered alongside economics and effluent quality when designing control strategies for wastewater treatment processes.

Avløpsrenseanlegg (WWTP) innehar en viktig rolle innen offentlig avfallshåndtering. Avløpsrenseanlegg er plassert på enden av avløpsnettet for å forbedre utløpets vannkvaliteten før det slippes ut til resipientene. Karbon, nitrogen og fosfor må fjernes fra avløpsvannet for å forhindre oksygenutarming, eutrofiering og utslipp av giftstoffer. Behandling av avløpsvann kan også skape negative miljøpåvirkninger på grunn av energiforbruk, kjemikalieforbruk og miljøutslipp. I noen få moderne avløpsrenseanlegg er resursgjenvinning og vanngjenbruk inkludert som en holistisk tilnærming. Avløpsbehandlingsprosessen må optimaliseres for å oppnå bedre ytelse innenfor økonomisk effektivitet, utslippskvalitet og miljøpåvirkning. Ytelsen til avløpsrenseanlegg blir påvirket av forstyrrelser, da spesielt variasjon i karakteristikken til tilløpet. Framskritt i utvikling av metoder for kontrollsystemer i avløpsbehandling har blitt gjort av forskere og praktikere, noe som det kan ses ut ifra de store mengdene publikasjoner innen kontrollsimulering. Mangelen på kostnadseffektivt og pålitelig utstyr for online overvåkning og kompleksiteten på de biologiske behandlingsprosessene har gjort at implementeringen av kontrollteknikker innen avløpsbehandling har vært begrenset. Denne avhandlingen kombinerer prosessovervåkning, datamining, prosessmodellering og molekylærbiologi for å minske gapet mellom overvåkning og prosesskontroll i praksis. Følgende arbeid ble gjennomført for å optimalisere avløpsbehandling: (1) Karakterisering av tilløpskvalitet og -kvantitet med varierende sesong- og klima-forhold; (2) en kombinert tilnærming for å oppnå avansert kontroll; (3) tolkbare virtuelle sensorer for prosessovervåkning; (4) interaksjoner mellom forskjellige mikrobielle komposisjoner og kinetisk modellering.