The increased awareness about the ecological status of the waterbodies and, on the other hand, the advances in the treatment technology have acted as driving forces behind the gradual tightening of wastewater purification requirements. Achieving the stringent treatment targets of wastewater treatment plants cost-efficiently is crucially dependent on the high-grade monitoring and control of the process units. Those, in turn, necessitate reliable real-time information about the primary process variables. In spite of the considerable developments of on-line sensors, demanding conditions in biological treatment processes sometimes give rise to an insufficient performance of instruments. The main motivation for this thesis was to design software tools that enable more efficient and safer treatment process operation by complementing conventional instrumentation. Since modern facilities are amply instrumented and there are plenty of accessible historical data, data-derived approaches were used in the studies. The data were processed together with predictive models providing virtual instruments often referred to as soft sensors. In this thesis, the models at the core of the soft sensors are based on multivariate statistics. In particular, principal component analysis with its variants and least-squares-based regression methods were employed in the soft sensor development. The moving-window techniques were applied so as to adapt the models to time-varying wastewater treatment processes. Both linear and nonlinear regression methods were explored.The technical studies of the thesis concern a large-scale municipal wastewater treatment plant. An array of soft sensors for the on-line prediction of nitrate concentrations was developed to support the operation of the biological post-filtration unit. Then, a system that enables the complementary use of the soft sensor estimates and the corresponding hardware instrument measurements was designed. The soft sensors were found to model nitrate concentrations accurately and, especially when integrated with the proposed switching system, to allow for a more secure control of the unit. In addition, a soft sensor for detecting process and instrument anomalies in the activated sludge process was investigated. The presented anomaly detection system motivates a more efficient use of sensors in the process control.It was demonstrated that soft sensors were applicable to the considered tasks and that they have strong potential for providing support to the operations of treatment facilities. The employed multivariate techniques proved to be capable of extracting easily understandable and practicable information from the high-dimensional data.
|Translated title of the contribution||Datapohjaiset virtuaaliset anturit biologisessa jätevedenkäsittelyssä: tilastollisten monimuuttujamenetelmien sovelluksia|
- , Supervisor
- , Advisor
- , Advisor
|Publication status||Published - 2016|
|MoE publication type||G5 Doctoral dissertation (article)|
- data-derived modelling, multivariate statistics, process monitoring, soft sensors, wastewater treatment