Model predictive control for reference planning in irrigation networks

The aim of this Thesis is that of developing and testing a multiple-level control structure in the context of large-scale irrigation networks. In particular, we will start analysing different possible low-level control schemes for regulating the water-level in every section (called pool) of an irrigation channel to fixed references. To this aim we will show that, using a simple decentralised control scheme, some problems will occur. This will be solved by adding a feedforward compensation and allowing for the decentralised controllers to exchange informations, leading to a distributed control scheme. However, when simulating the behaviour of a whole channel with either decentralised or distributed controllers, a violation of the physical constraints of a channel (i. e., exceeding the maximum/minimum allowed water-level/flow) may occur. As a consequence, we will develop a centralised high-level reference manager, able to generate different references for each pool of a channel using a Model Predictive Control technique, suitable for constrained problems. First, we will schedule the nominal references for a whole season through an off-line optimization, based on the known nominal scheduled demand. Then, since uncertainties on the effective demand could occur, we will need to exploit also reactive robust controllers based on the Receding Horizon principle. After devising the whole control structure for a channel, we will show some simulations results with all the possible controllers. In particular, we will notice that the robust Receding Horizon controller, together with distributed controllers for each pool, yields to positive results with delay-like uncertainties, but not in all the cases. Indeed, we will highlight that there are some scenarios in which we need to relax one of the assumptions of the robust Receding Horizon technique adopted (i. e., the time-invariance of the constraints). By allowing the constraints to be time-varying, this technique will result to be suitable for our scenarios.