A comparison of distributed MPC schemes on a hydro‐power plant benchmark

Abstract

SUMMARYIn this paper, we analyze and compare five distributed model predictive control (DMPC) schemes using a hydro-power plant benchmark. Besides being one of the most important sources of renewable power, hydropower plants present very interesting control challenges. The operation of a hydro-power valley involves the coordination of several subsystems over a large geographical area in order to produce the demanded energy while satisfying constraints on water levels and flows. In particular, we test the different DMPC algorithms using a 24-h power tracking scenario in which the hydro-power plant is simulated with an accurate nonlinear model. In this way, it is possible to provide qualitative and quantitative comparisons between different DMPC schemes implemented on a common benchmark, which is a type of assessment rare in the literature.

Distributed model predictive control for continuous‐time nonlinear systems based on suboptimal ADMM

“…Figure shows that the communication effort reduces significantly if a certain degree of suboptimality between 1% and 3% is accepted. A more detailed evaluation of the floats to be transmitted, eg, as considered in other works, requires to keep in mind that the trajectories have to be transmitted in discretized form, ie, the numbers in Figure are to be multiplied by the discretization points (25 in this scenario) and the corresponding variable size depending on single or double precision arithmetics. Note also that the additional transmission of the convergence flags of all agents as pointed out in the ADMM algorithm is negligible regarding the communication load of the local trajectories.…”

“…Figure 4 shows that the communication effort reduces significantly if a certain degree of suboptimality between 1% and 3% is accepted. A more detailed evaluation of the floats to be transmitted, eg, as considered in other works, 19,71,72 requires to keep in mind that the trajectories have to be transmitted in discretized form, ie, the numbers in Figure 4 are to be multiplied by the discretization points (25 in this scenario) and the corresponding variable An interesting question concerns the convergence behavior of the ADMM iterations in view of the linear and R-linear convergence assumptions (32) and (41), respectively. Figure 5 shows the residual norm plotted over the iterations q for different initializations of the ADMM algorithm.…”

Hierarchically coordinated economic MPC plantwide control of mixed continuous‐batch units in process industries with application to a beet sugar plant

“…Recent contributions, for example, deal with hybrid models 24 and event-triggered schemes. 25 In the works of Maestre et al 26 and Alvarado et al, 27 different algorithms are compared as applied to nontrivial benchmarks.…”

“…On the other hand, the term h evap represents the enthalpy of the evaporated water, which is a function of the existing pressure in the chamber (p vap ) and of massecuite temperature (T mc ). The determination of the heat transfer rate Q(W) (26) and the very related balance to the mass of steam in the primary of the heat exchanger (27) are both important elements of the energy balance (25) and of the modeling of the steam consumption behavior of the crystallizers. The term Q results from the difference of temperature at both sides of the shell and tubes exchanger of total area A (m 2 ).…”

Distributed model predictive control

“…The next article, titled A comparison of distributed MPC schemes on a hydro power plant benchmark , is authored by José Maestre, Miguel A. Ridao, Attila Kozma, Carlo Savorgnan, Moritz Diehl, Anna D. Sadowska, Mihn Doan, Tamas Keviczky, Bart De Schutter, Holger Scheu, Wolfgang Marquardt, Felipe Valencia and Jairo Espinosa . This paper gathers the description and application of five DMPC schemes, adapted to be used for power control in the same benchmark model, based on the operation of hydro power plants to be directly compared in a 24‐h power‐tracking scenario.…”