Data-driven control design of wastewater treatment systems
The research in this thesis covers three fields of control theory; identification, control design, and real-time control applied to activated sludge wastewater treatment plants. The study is carfied out using simulation and a full-scale implementation in Swinstie wastewater treatment plant from Scottish Water. Subspace algorithms are explored to obtain adequate models for dissolved oxygen and nutrient dynamics. Results presented in this area are the outcome of a number of simulations and full-scale plant experiments, which have lead to the formulation of standard recommendations to the identification of linear models for the activated sludge process. Part of the work has also provided an evaluation of a number of subspace identification algorithms, although this has not been an objective within the thesis. The thesis also contains some insight into the modelling of the activated sludge for an intermittent aeration process. The control design part of the thesis employs a two level hierarchical control approach. The low level control is usually a proportional integral derivative controller (PID) type. This thesis presents the development of three new tuning algorithms for PID type controllers: iterative feedback tuning (IFT), linear quadratic gaussian (LQG) and datadriven. The first two methods are developed for continuous time systems, while the last is a discrete time data-driven method which uses subspace identification. The second control level employs linear model predictive control (MPC). MPC is used for dissolved oxygen and nitrogen removal in a simulation level. Linear models of nutrient removal obtained by identification and by model reduction are used to implement controllers for continuous aeration and intermittent aeration plants. Real-time control is implemented by developing a software platform. The software platform contains algorithms for subspace identification, MPC control design and execution and process monitoring. The software is developed using LabVEEW and MATLAB. The user frontend and the communication with the PLC are implemented in LabVIEW. The PLC communication employs OPC tehcnology. Many of the algorithms required for identification, control design, and process monitoring are programmed in MATLAB and linked to LabVIEW by using several technologies as: Activex and Dynamic Link Libraries (DLL). The thesis finally presents results obtained by real-time execution of the identification and control algorithms.