The thesis investigates cost optimal control of pumping stations in water supply applications with elevated reservoirs. The report consists of two main parts. The first part covers the mathematical description of Drinking Water Supply Systems(WSSs) with the aim of deriving a model for Model Predictive Control(MPC). The framework on which the modelling is based is motivated by [1] and [2], where a model for WSSs has been derived without considering storages in the network. In the thesis, this framework is extended by assuming a topology with multiple pumping stations and elevated reservoirs. For identification purposes, Neural Networks(NNs) are used to describe the water level in the storage tanks and the inlet pressure at the pumping stations. The NN-based identification is carried out on both simulation data from EPANET and on real data from a distribution grid, provided by Verdo A/S. The second part of the thesis deals with the refinement of the NN model and with finding the appropriate complexity for control. A Linear Time-Varying(LTV) model is established by linearizing the NN in different Operating Points, based on the flow demand in the network. Then, MPC is designed with the assumption that energy prices and the flow demand are forecasted. Compared to the currently used ON/OFF control on Verdo’s WSS, the MPC algorithm has shown better performance by saving money on the actuation and by exploiting the storage capacity of the elevated reservoirs. Additionally to the simulations, the control has been implemented and verified on a small-scale WSS test setup, provided by Aalborg University.

[1] T. N. Jensen, C. S. Kallesøe, and R. Wisniewski, “Adaptive reference control for pressure management in water networks,” European Control Conference(ECC), vol. 3, July 2015.
[2] T. N. Jensen, C. S. Kallesøe, and R. Wisniewski, “Plug-and-play commissionable models for water networks with multiple inlets,” European Control Conference(ECC), submitted 2018.